Merge remote branch 'main/radeon-rewrite'

120 files changed, 16432 insertions, 18854 deletions

diff --git a/configs/autoconf.in b/configs/autoconf.in
index 77422e3fe5a..83737e3a7df 100644
--- a/configs/autoconf.in
+++ b/configs/autoconf.in
@@ -20,6 +20,8 @@ CXXFLAGS = @CPPFLAGS@ @CXXFLAGS@ \
 	$(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES)
 LDFLAGS = @LDFLAGS@
 EXTRA_LIB_PATH = @EXTRA_LIB_PATH@
+RADEON_CFLAGS = @RADEON_CFLAGS@
+RADEON_LDFLAGS = @RADEON_LDFLAGS@
 
 # Assembler
 MESA_ASM_SOURCES = @MESA_ASM_SOURCES@
diff --git a/configure.ac b/configure.ac
index d4a8222df0d..f34a4740b31 100644
--- a/configure.ac
+++ b/configure.ac
@@ -445,6 +445,8 @@ AC_SUBST([GALLIUM_WINSYS_DRM_DIRS])
 AC_SUBST([GALLIUM_DRIVERS_DIRS])
 AC_SUBST([GALLIUM_AUXILIARY_DIRS])
 AC_SUBST([GALLIUM_STATE_TRACKERS_DIRS])
+AC_SUBST([RADEON_CFLAGS])
+AC_SUBST([RADEON_LDFLAGS])
 
 dnl
 dnl User supplied program configuration
@@ -572,6 +574,13 @@ dri)
     GL_PC_REQ_PRIV="libdrm >= $LIBDRM_REQUIRED dri2proto >= $DRI2PROTO_REQUIRED"
     DRI_PC_REQ_PRIV="libdrm >= $LIBDRM_REQUIRED"
 
+    PKG_CHECK_MODULES([LIBDRM_RADEON], [libdrm_radeon], HAVE_LIBDRM_RADEON=yes, HAVE_LIBDRM_RADEON=no)
+
+    if test "$HAVE_LIBDRM_RADEON" = yes; then
+	RADEON_CFLAGS="-DHAVE_LIBDRM_RADEON=1 $LIBDRM_RADEON_CFLAGS"
+	RADEON_LDFLAGS=$LIBDRM_RADEON_LIBS
+    fi
+
     # find the DRI deps for libGL
     if test "$x11_pkgconfig" = yes; then
         # add xcb modules if necessary
diff --git a/src/mesa/drivers/dri/common/dri_util.c b/src/mesa/drivers/dri/common/dri_util.c
index e1127204718..ae0e61e515b 100644
--- a/src/mesa/drivers/dri/common/dri_util.c
+++ b/src/mesa/drivers/dri/common/dri_util.c
@@ -37,6 +37,9 @@
 typedef GLboolean ( * PFNGLXGETMSCRATEOMLPROC) (__DRIdrawable *drawable, int32_t *numerator, int32_t *denominator);
 #endif
 
+static void dri_get_drawable(__DRIdrawable *pdp);
+static void dri_put_drawable(__DRIdrawable *pdp);
+
 /**
  * This is just a token extension used to signal that the driver
  * supports setting a read drawable.
@@ -127,7 +130,7 @@ static int driUnbindContext(__DRIcontext *pcp)
 	return GL_FALSE;
     }
 
-    pdp->refcount--;
+    dri_put_drawable(pdp);
 
     if (prp != pdp) {
         if (prp->refcount == 0) {
@@ -135,7 +138,7 @@ static int driUnbindContext(__DRIcontext *pcp)
 	    return GL_FALSE;
 	}
 
-	prp->refcount--;
+    	dri_put_drawable(prp);
     }
 
 
@@ -170,10 +173,10 @@ static int driBindContext(__DRIcontext *pcp,
 	pcp->driReadablePriv = prp;
 	if (pdp) {
 	    pdp->driContextPriv = pcp;
-	    pdp->refcount++;
+    	    dri_get_drawable(pdp);
 	}
 	if ( prp && pdp != prp ) {
-	    prp->refcount++;
+    	    dri_get_drawable(prp);
 	}
     }
 
@@ -430,7 +433,7 @@ driCreateNewDrawable(__DRIscreen *psp, const __DRIconfig *config,
 
     pdp->loaderPrivate = data;
     pdp->hHWDrawable = hwDrawable;
-    pdp->refcount = 0;
+    pdp->refcount = 1;
     pdp->pStamp = NULL;
     pdp->lastStamp = 0;
     pdp->index = 0;
@@ -483,12 +486,19 @@ dri2CreateNewDrawable(__DRIscreen *screen,
     return pdraw;
 }
 
-
-static void
-driDestroyDrawable(__DRIdrawable *pdp)
+static void dri_get_drawable(__DRIdrawable *pdp)
+{
+    pdp->refcount++;
+}
+	
+static void dri_put_drawable(__DRIdrawable *pdp)
 {
     __DRIscreenPrivate *psp;
 
+    pdp->refcount--;
+    if (pdp->refcount)
+	return;
+
     if (pdp) {
 	psp = pdp->driScreenPriv;
         (*psp->DriverAPI.DestroyBuffer)(pdp);
@@ -504,6 +514,12 @@ driDestroyDrawable(__DRIdrawable *pdp)
     }
 }
 
+static void
+driDestroyDrawable(__DRIdrawable *pdp)
+{
+    dri_put_drawable(pdp);
+}
+
 /*@}*/
 
 
diff --git a/src/mesa/drivers/dri/r200/Makefile b/src/mesa/drivers/dri/r200/Makefile
index e9144ac75ce..6a246edf7cb 100644
--- a/src/mesa/drivers/dri/r200/Makefile
+++ b/src/mesa/drivers/dri/r200/Makefile
@@ -3,6 +3,8 @@
 TOP = ../../../../..
 include $(TOP)/configs/current
 
+CFLAGS += $(RADEON_CFLAGS)
+
 LIBNAME = r200_dri.so
 
 MINIGLX_SOURCES = server/radeon_dri.c 
@@ -11,25 +13,36 @@ ifeq ($(USING_EGL), 1)
 EGL_SOURCES = server/radeon_egl.c
 endif
 
+RADEON_COMMON_SOURCES = \
+	radeon_texture.c \
+	radeon_common_context.c \
+	radeon_common.c \
+	radeon_dma.c \
+	radeon_lock.c \
+	radeon_bo_legacy.c \
+	radeon_cs_legacy.c \
+	radeon_mipmap_tree.c \
+	radeon_span.c \
+	radeon_fbo.c
+
+
 DRIVER_SOURCES = r200_context.c \
 		 r200_ioctl.c \
-		 r200_lock.c \
 		 r200_state.c \
 		 r200_state_init.c \
 		 r200_cmdbuf.c \
 		 r200_pixel.c \
 		 r200_tex.c \
-		 r200_texmem.c \
 		 r200_texstate.c \
 		 r200_tcl.c \
 		 r200_swtcl.c \
-		 r200_span.c \
 		 r200_maos.c \
 		 r200_sanity.c \
 		 r200_fragshader.c \
 		 r200_vertprog.c \
 		 radeon_screen.c \
-		 $(EGL_SOURCES)
+		 $(EGL_SOURCES) \
+		 $(RADEON_COMMON_SOURCES)
 
 C_SOURCES = $(COMMON_SOURCES) $(DRIVER_SOURCES)
 
@@ -48,7 +61,30 @@ SYMLINKS = \
 COMMON_SYMLINKS = \
 	radeon_chipset.h \
 	radeon_screen.c \
-	radeon_screen.h
+	radeon_screen.h \
+	radeon_bo_legacy.c \
+	radeon_cs_legacy.c \
+	radeon_bo_legacy.h \
+	radeon_cs_legacy.h \
+	radeon_bocs_wrapper.h \
+	radeon_span.h \
+	radeon_span.c \
+	radeon_lock.c \
+	radeon_lock.h \
+	radeon_common.c \
+	radeon_common_context.c \
+	radeon_common_context.h \
+	radeon_common.h \
+	radeon_cmdbuf.h \
+	radeon_mipmap_tree.c \
+	radeon_mipmap_tree.h \
+	radeon_texture.c \
+	radeon_texture.h \
+	radeon_dma.c \
+	radeon_dma.h \
+	radeon_fbo.c
+
+DRI_LIB_DEPS += $(RADEON_LDFLAGS)
 
 ##### TARGETS #####
 
diff --git a/src/mesa/drivers/dri/r200/r200_cmdbuf.c b/src/mesa/drivers/dri/r200/r200_cmdbuf.c
index e1633772a18..df9dd83344c 100644
--- a/src/mesa/drivers/dri/r200/r200_cmdbuf.c
+++ b/src/mesa/drivers/dri/r200/r200_cmdbuf.c
@@ -38,6 +38,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "swrast/swrast.h"
 #include "main/simple_list.h"
 
+#include "radeon_common.h"
 #include "r200_context.h"
 #include "r200_state.h"
 #include "r200_ioctl.h"
@@ -45,18 +46,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_sanity.h"
 #include "radeon_reg.h"
 
-static void print_state_atom( struct r200_state_atom *state )
-{
-   int i;
-
-   fprintf(stderr, "emit %s/%d\n", state->name, state->cmd_size);
-
-   if (0 & R200_DEBUG & DEBUG_VERBOSE) 
-      for (i = 0 ; i < state->cmd_size ; i++) 
-	 fprintf(stderr, "\t%s[%d]: %x\n", state->name, i, state->cmd[i]);
-
-}
-
 /* The state atoms will be emitted in the order they appear in the atom list,
  * so this step is important.
  */
@@ -64,141 +53,85 @@ void r200SetUpAtomList( r200ContextPtr rmesa )
 {
    int i, mtu;
 
-   mtu = rmesa->glCtx->Const.MaxTextureUnits;
-
-   make_empty_list(&rmesa->hw.atomlist);
-   rmesa->hw.atomlist.name = "atom-list";
-
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.ctx );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.set );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.lin );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.msk );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vpt );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vtx );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vap );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vte );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.msc );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.cst );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.zbs );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.tcl );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.msl );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.tcg );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.grd );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.fog );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.tam );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.tf );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.atf );
+   mtu = rmesa->radeon.glCtx->Const.MaxTextureUnits;
+
+   make_empty_list(&rmesa->radeon.hw.atomlist);
+   rmesa->radeon.hw.atomlist.name = "atom-list";
+
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.ctx );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.set );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.lin );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.msk );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vpt );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vtx );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vap );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vte );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.msc );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.cst );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.zbs );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.tcl );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.msl );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.tcg );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.grd );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.fog );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.tam );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.tf );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.atf );
    for (i = 0; i < mtu; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.tex[i] );
+       insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.tex[i] );
    for (i = 0; i < mtu; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.cube[i] );
+       insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.cube[i] );
    for (i = 0; i < 6; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.pix[i] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.afs[0] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.afs[1] );
+       insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.pix[i] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.afs[0] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.afs[1] );
    for (i = 0; i < 8; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.lit[i] );
+       insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.lit[i] );
    for (i = 0; i < 3 + mtu; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.mat[i] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.eye );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.glt );
+       insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.mat[i] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.eye );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.glt );
    for (i = 0; i < 2; ++i)
-      insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.mtl[i] );
+      insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.mtl[i] );
    for (i = 0; i < 6; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.ucp[i] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.spr );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.ptp );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.prf );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.pvs );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vpp[0] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vpp[1] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vpi[0] );
-   insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.vpi[1] );
+       insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.ucp[i] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.spr );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.ptp );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.prf );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.pvs );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vpp[0] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vpp[1] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vpi[0] );
+   insert_at_tail( &rmesa->radeon.hw.atomlist, &rmesa->hw.vpi[1] );
 }
 
-static void r200SaveHwState( r200ContextPtr rmesa )
+void r200EmitScissor(r200ContextPtr rmesa)
 {
-   struct r200_state_atom *atom;
-   char * dest = rmesa->backup_store.cmd_buf;
-
-   if (R200_DEBUG & DEBUG_STATE)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   rmesa->backup_store.cmd_used = 0;
-
-   foreach( atom, &rmesa->hw.atomlist ) {
-      if ( atom->check( rmesa->glCtx, atom->idx ) ) {
-	 int size = atom->cmd_size * 4;
-	 memcpy( dest, atom->cmd, size);
-	 dest += size;
-	 rmesa->backup_store.cmd_used += size;
-	 if (R200_DEBUG & DEBUG_STATE)
-	    print_state_atom( atom );
-      }
-   }
-
-   assert( rmesa->backup_store.cmd_used <= R200_CMD_BUF_SZ );
-   if (R200_DEBUG & DEBUG_STATE)
-      fprintf(stderr, "Returning to r200EmitState\n");
-}
-
-void r200EmitState( r200ContextPtr rmesa )
-{
-   char *dest;
-   int mtu;
-   struct r200_state_atom *atom;
-
-   if (R200_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (rmesa->save_on_next_emit) {
-      r200SaveHwState(rmesa);
-      rmesa->save_on_next_emit = GL_FALSE;
-   }
-
-   if (!rmesa->hw.is_dirty && !rmesa->hw.all_dirty)
-      return;
-
-   mtu = rmesa->glCtx->Const.MaxTextureUnits;
-
-   /* To avoid going across the entire set of states multiple times, just check
-    * for enough space for the case of emitting all state, and inline the
-    * r200AllocCmdBuf code here without all the checks.
-    */
-   r200EnsureCmdBufSpace( rmesa, rmesa->hw.max_state_size );
-
-   /* we need to calculate dest after EnsureCmdBufSpace
-      as we may flush the buffer - airlied */
-   dest = rmesa->store.cmd_buf + rmesa->store.cmd_used;
-   if (R200_DEBUG & DEBUG_STATE) {
-      foreach( atom, &rmesa->hw.atomlist ) {
-	 if ( atom->dirty || rmesa->hw.all_dirty ) {
-	    if ( atom->check( rmesa->glCtx, atom->idx ) )
-	       print_state_atom( atom );
-	    else
-	       fprintf(stderr, "skip state %s\n", atom->name);
-	 }
-      }
-   }
-
-   foreach( atom, &rmesa->hw.atomlist ) {
-      if ( rmesa->hw.all_dirty )
-	 atom->dirty = GL_TRUE;
-      if ( atom->dirty ) {
-	 if ( atom->check( rmesa->glCtx, atom->idx ) ) {
-	    int size = atom->cmd_size * 4;
-	    memcpy( dest, atom->cmd, size);
-	    dest += size;
-	    rmesa->store.cmd_used += size;
-	    atom->dirty = GL_FALSE;
-	 }
-      }
-   }
-
-   assert( rmesa->store.cmd_used <= R200_CMD_BUF_SZ );
-
-   rmesa->hw.is_dirty = GL_FALSE;
-   rmesa->hw.all_dirty = GL_FALSE;
+    BATCH_LOCALS(&rmesa->radeon);
+    if (!rmesa->radeon.radeonScreen->kernel_mm) {
+       return;
+    }
+    if (rmesa->radeon.state.scissor.enabled) {
+        BEGIN_BATCH(8);
+        OUT_BATCH(CP_PACKET0(R200_RE_CNTL, 0));
+        OUT_BATCH(R200_SCISSOR_ENABLE | rmesa->hw.set.cmd[SET_RE_CNTL]);
+        OUT_BATCH(CP_PACKET0(R200_RE_AUX_SCISSOR_CNTL, 0));
+        OUT_BATCH(R200_SCISSOR_ENABLE_0);
+        OUT_BATCH(CP_PACKET0(R200_RE_SCISSOR_TL_0, 0));
+        OUT_BATCH((rmesa->radeon.state.scissor.rect.y1 << 16) |
+                  rmesa->radeon.state.scissor.rect.x1);
+        OUT_BATCH(CP_PACKET0(R200_RE_SCISSOR_BR_0, 0));
+        OUT_BATCH(((rmesa->radeon.state.scissor.rect.y2 - 1) << 16) |
+                  (rmesa->radeon.state.scissor.rect.x2 - 1));
+        END_BATCH();
+    } else {
+        BEGIN_BATCH(4);
+        OUT_BATCH(CP_PACKET0(R200_RE_CNTL, 0));
+        OUT_BATCH(rmesa->hw.set.cmd[SET_RE_CNTL] & ~R200_SCISSOR_ENABLE);
+        OUT_BATCH(CP_PACKET0(R200_RE_AUX_SCISSOR_CNTL, 0));
+        OUT_BATCH(0);
+        END_BATCH();
+    }
 }
 
 /* Fire a section of the retained (indexed_verts) buffer as a regular
@@ -208,51 +141,81 @@ void r200EmitVbufPrim( r200ContextPtr rmesa,
                        GLuint primitive,
                        GLuint vertex_nr )
 {
-   drm_radeon_cmd_header_t *cmd;
+   BATCH_LOCALS(&rmesa->radeon);
 
    assert(!(primitive & R200_VF_PRIM_WALK_IND));
    
-   r200EmitState( rmesa );
+   radeonEmitState(&rmesa->radeon);
    
    if (R200_DEBUG & (DEBUG_IOCTL|DEBUG_PRIMS))
       fprintf(stderr, "%s cmd_used/4: %d prim %x nr %d\n", __FUNCTION__,
 	      rmesa->store.cmd_used/4, primitive, vertex_nr);
-   
-   cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, VBUF_BUFSZ,
-						  __FUNCTION__ );
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
-   cmd[1].i = R200_CP_CMD_3D_DRAW_VBUF_2;
-   cmd[2].i = (primitive | 
-	       R200_VF_PRIM_WALK_LIST |
-	       R200_VF_COLOR_ORDER_RGBA |
-	       (vertex_nr << R200_VF_VERTEX_NUMBER_SHIFT));
+   r200EmitScissor(rmesa);
+ 
+   BEGIN_BATCH(3);
+   OUT_BATCH_PACKET3_CLIP(R200_CP_CMD_3D_DRAW_VBUF_2, 0);
+   OUT_BATCH(primitive | R200_VF_PRIM_WALK_LIST | R200_VF_COLOR_ORDER_RGBA |
+	     (vertex_nr << R200_VF_VERTEX_NUMBER_SHIFT));
+   END_BATCH();
 }
 
+static void r200FireEB(r200ContextPtr rmesa, int vertex_count, int type)
+{
+	BATCH_LOCALS(&rmesa->radeon);
+
+	if (vertex_count > 0) {
+        r200EmitScissor(rmesa);
+		BEGIN_BATCH(8+2);
+		OUT_BATCH_PACKET3_CLIP(R200_CP_CMD_3D_DRAW_INDX_2, 0);
+		OUT_BATCH(R200_VF_PRIM_WALK_IND |
+			  R200_VF_COLOR_ORDER_RGBA | 
+			  ((vertex_count + 0) << 16) |
+			  type);
+		
+		if (!rmesa->radeon.radeonScreen->kernel_mm) {
+			OUT_BATCH_PACKET3(R200_CP_CMD_INDX_BUFFER, 2);
+			OUT_BATCH((0x80 << 24) | (0 << 16) | 0x810);
+			OUT_BATCH_RELOC(rmesa->radeon.tcl.elt_dma_offset,
+					rmesa->radeon.tcl.elt_dma_bo,
+					rmesa->radeon.tcl.elt_dma_offset,
+					RADEON_GEM_DOMAIN_GTT, 0, 0);
+			OUT_BATCH((vertex_count + 1)/2);
+		} else {
+			OUT_BATCH_PACKET3(R200_CP_CMD_INDX_BUFFER, 2);
+			OUT_BATCH((0x80 << 24) | (0 << 16) | 0x810);
+			OUT_BATCH(rmesa->radeon.tcl.elt_dma_offset);
+			OUT_BATCH((vertex_count + 1)/2);
+			radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+					      rmesa->radeon.tcl.elt_dma_bo,
+					      RADEON_GEM_DOMAIN_GTT, 0, 0);
+		}
+		END_BATCH();
+	}
+}
 
-void r200FlushElts( r200ContextPtr rmesa )
+void r200FlushElts(GLcontext *ctx)
 {
-   int *cmd = (int *)(rmesa->store.cmd_buf + rmesa->store.elts_start);
-   int dwords;
-   int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 12)) / 2;
+   r200ContextPtr rmesa = R200_CONTEXT(ctx);
+   int nr, elt_used = rmesa->tcl.elt_used;
 
    if (R200_DEBUG & (DEBUG_IOCTL|DEBUG_PRIMS))
-      fprintf(stderr, "%s\n", __FUNCTION__);
+     fprintf(stderr, "%s %x %d\n", __FUNCTION__, rmesa->tcl.hw_primitive, elt_used);
 
-   assert( rmesa->dma.flush == r200FlushElts );
-   rmesa->dma.flush = NULL;
+   assert( rmesa->radeon.dma.flush == r200FlushElts );
+   rmesa->radeon.dma.flush = NULL;
 
-   /* Cope with odd number of elts:
-    */
-   rmesa->store.cmd_used = (rmesa->store.cmd_used + 2) & ~2;
-   dwords = (rmesa->store.cmd_used - rmesa->store.elts_start) / 4;
+   nr = elt_used / 2;
 
-   cmd[1] |= (dwords - 3) << 16;
-   cmd[2] |= nr << R200_VF_VERTEX_NUMBER_SHIFT;
+   radeon_bo_unmap(rmesa->radeon.tcl.elt_dma_bo);
+
+   r200FireEB(rmesa, nr, rmesa->tcl.hw_primitive);
+
+   radeon_bo_unref(rmesa->radeon.tcl.elt_dma_bo);
+   rmesa->radeon.tcl.elt_dma_bo = NULL;
 
    if (R200_DEBUG & DEBUG_SYNC) {
       fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
-      r200Finish( rmesa->glCtx );
+      radeonFinish( rmesa->radeon.glCtx );
    }
 }
 
@@ -261,7 +224,6 @@ GLushort *r200AllocEltsOpenEnded( r200ContextPtr rmesa,
 				    GLuint primitive,
 				    GLuint min_nr )
 {
-   drm_radeon_cmd_header_t *cmd;
    GLushort *retval;
 
    if (R200_DEBUG & DEBUG_IOCTL)
@@ -269,30 +231,30 @@ GLushort *r200AllocEltsOpenEnded( r200ContextPtr rmesa,
 
    assert((primitive & R200_VF_PRIM_WALK_IND));
    
-   r200EmitState( rmesa );
-   
-   cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, ELTS_BUFSZ(min_nr),
-						__FUNCTION__ );
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
-   cmd[1].i = R200_CP_CMD_3D_DRAW_INDX_2;
-   cmd[2].i = (primitive | 
-	       R200_VF_PRIM_WALK_IND |
-	       R200_VF_COLOR_ORDER_RGBA);
+   radeonEmitState(&rmesa->radeon);
+
+   rmesa->radeon.tcl.elt_dma_bo = radeon_bo_open(rmesa->radeon.radeonScreen->bom,
+					  0, R200_ELT_BUF_SZ, 4,
+					  RADEON_GEM_DOMAIN_GTT, 0);
+   rmesa->radeon.tcl.elt_dma_offset = 0;
+   rmesa->tcl.elt_used = min_nr * 2;
 
+   radeon_validate_bo(&rmesa->radeon, rmesa->radeon.tcl.elt_dma_bo,
+                      RADEON_GEM_DOMAIN_GTT, 0);
+   if (radeon_revalidate_bos(rmesa->radeon.glCtx) == GL_FALSE)
+      fprintf(stderr,"failure to revalidate BOs - badness\n");
+
+   radeon_bo_map(rmesa->radeon.tcl.elt_dma_bo, 1);
+   retval = rmesa->radeon.tcl.elt_dma_bo->ptr + rmesa->radeon.tcl.elt_dma_offset;
    
-   retval = (GLushort *)(cmd+3);
 
    if (R200_DEBUG & DEBUG_PRIMS)
-      fprintf(stderr, "%s: header 0x%x prim %x \n",
-	      __FUNCTION__,
-	      cmd[1].i, primitive);
-
-   assert(!rmesa->dma.flush);
-   rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
-   rmesa->dma.flush = r200FlushElts;
+      fprintf(stderr, "%s: header prim %x \n",
+	      __FUNCTION__, primitive);
 
-   rmesa->store.elts_start = ((char *)cmd) - rmesa->store.cmd_buf;
+   assert(!rmesa->radeon.dma.flush);
+   rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+   rmesa->radeon.dma.flush = r200FlushElts;
 
    return retval;
 }
@@ -300,129 +262,119 @@ GLushort *r200AllocEltsOpenEnded( r200ContextPtr rmesa,
 
 
 void r200EmitVertexAOS( r200ContextPtr rmesa,
-			  GLuint vertex_size,
-			  GLuint offset )
+			GLuint vertex_size,
+ 			struct radeon_bo *bo,
+			GLuint offset )
 {
-   drm_radeon_cmd_header_t *cmd;
+   BATCH_LOCALS(&rmesa->radeon);
 
    if (R200_DEBUG & (DEBUG_PRIMS|DEBUG_IOCTL))
       fprintf(stderr, "%s:  vertex_size 0x%x offset 0x%x \n",
 	      __FUNCTION__, vertex_size, offset);
 
-   cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, VERT_AOS_BUFSZ,
-						  __FUNCTION__ );
 
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
-   cmd[1].i = R200_CP_CMD_3D_LOAD_VBPNTR | (2 << 16);
-   cmd[2].i = 1;
-   cmd[3].i = vertex_size | (vertex_size << 8);
-   cmd[4].i = offset;
+   BEGIN_BATCH(7);
+   OUT_BATCH_PACKET3(R200_CP_CMD_3D_LOAD_VBPNTR, 2);
+   OUT_BATCH(1);
+   OUT_BATCH(vertex_size | (vertex_size << 8));
+   OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+   END_BATCH();
 }
-		       
 
-void r200EmitAOS( r200ContextPtr rmesa,
-		    struct r200_dma_region **component,
-		    GLuint nr,
-		    GLuint offset )
+void r200EmitAOS(r200ContextPtr rmesa, GLuint nr, GLuint offset)
 {
-   drm_radeon_cmd_header_t *cmd;
-   int sz = AOS_BUFSZ(nr);
+   BATCH_LOCALS(&rmesa->radeon);
+   uint32_t voffset;
+   int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2;
    int i;
-   int *tmp;
-
-   if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s nr arrays: %d\n", __FUNCTION__, nr);
-
-   cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, sz, __FUNCTION__ );
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
-   cmd[1].i = R200_CP_CMD_3D_LOAD_VBPNTR | (((sz / sizeof(int)) - 3) << 16);
-   cmd[2].i = nr;
-   tmp = &cmd[0].i;
-   cmd += 3;
-
-   for (i = 0 ; i < nr ; i++) {
-      if (i & 1) {
-	 cmd[0].i |= ((component[i]->aos_stride << 24) | 
-		      (component[i]->aos_size << 16));
-	 cmd[2].i = (component[i]->aos_start + 
-		     offset * component[i]->aos_stride * 4);
-	 cmd += 3;
+   
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__, nr,
+	      offset);
+
+   BEGIN_BATCH(sz+2+ (nr*2));
+   OUT_BATCH_PACKET3(R200_CP_CMD_3D_LOAD_VBPNTR, sz - 1);
+   OUT_BATCH(nr);
+
+    
+   if (!rmesa->radeon.radeonScreen->kernel_mm) {
+      for (i = 0; i + 1 < nr; i += 2) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) |
+		   (rmesa->radeon.tcl.aos[i].stride << 8) |
+		   (rmesa->radeon.tcl.aos[i + 1].components << 16) |
+		   (rmesa->radeon.tcl.aos[i + 1].stride << 24));
+			
+	 voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+	 OUT_BATCH_RELOC(voffset,
+			 rmesa->radeon.tcl.aos[i].bo,
+			 voffset,
+			 RADEON_GEM_DOMAIN_GTT,
+			 0, 0);
+	 voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+	 OUT_BATCH_RELOC(voffset,
+			 rmesa->radeon.tcl.aos[i+1].bo,
+			 voffset,
+			 RADEON_GEM_DOMAIN_GTT,
+			 0, 0);
       }
-      else {
-	 cmd[0].i = ((component[i]->aos_stride << 8) | 
-		     (component[i]->aos_size << 0));
-	 cmd[1].i = (component[i]->aos_start + 
-		     offset * component[i]->aos_stride * 4);
+      
+      if (nr & 1) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) |
+		   (rmesa->radeon.tcl.aos[nr - 1].stride << 8));
+	 voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+	 OUT_BATCH_RELOC(voffset,
+			 rmesa->radeon.tcl.aos[nr - 1].bo,
+			 voffset,
+			 RADEON_GEM_DOMAIN_GTT,
+			 0, 0);
+      }
+   } else {
+      for (i = 0; i + 1 < nr; i += 2) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) |
+		   (rmesa->radeon.tcl.aos[i].stride << 8) |
+		   (rmesa->radeon.tcl.aos[i + 1].components << 16) |
+		   (rmesa->radeon.tcl.aos[i + 1].stride << 24));
+	 
+	 voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+	 OUT_BATCH(voffset);
+	 voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+	 OUT_BATCH(voffset);
+      }
+      
+      if (nr & 1) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) |
+		   (rmesa->radeon.tcl.aos[nr - 1].stride << 8));
+	 voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+	 OUT_BATCH(voffset);
+      }
+      for (i = 0; i + 1 < nr; i += 2) {
+	 voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+	 radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			       rmesa->radeon.tcl.aos[i+0].bo,
+			       RADEON_GEM_DOMAIN_GTT,
+			       0, 0);
+	 voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+	 radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			       rmesa->radeon.tcl.aos[i+1].bo,
+			       RADEON_GEM_DOMAIN_GTT,
+			       0, 0);
+      }
+      if (nr & 1) {
+	 voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+	 radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			       rmesa->radeon.tcl.aos[nr-1].bo,
+			       RADEON_GEM_DOMAIN_GTT,
+			       0, 0);
       }
    }
-
-   if (R200_DEBUG & DEBUG_VERTS) {
-      fprintf(stderr, "%s:\n", __FUNCTION__);
-      for (i = 0 ; i < sz ; i++)
-	 fprintf(stderr, "   %d: %x\n", i, tmp[i]);
-   }
-}
-
-void r200EmitBlit( r200ContextPtr rmesa,
-		   GLuint color_fmt,
-		   GLuint src_pitch,
-		   GLuint src_offset,
-		   GLuint dst_pitch,
-		   GLuint dst_offset,
-		   GLint srcx, GLint srcy,
-		   GLint dstx, GLint dsty,
-		   GLuint w, GLuint h )
-{
-   drm_radeon_cmd_header_t *cmd;
-
-   if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s src %x/%x %d,%d dst: %x/%x %d,%d sz: %dx%d\n",
-	      __FUNCTION__, 
-	      src_pitch, src_offset, srcx, srcy,
-	      dst_pitch, dst_offset, dstx, dsty,
-	      w, h);
-
-   assert( (src_pitch & 63) == 0 );
-   assert( (dst_pitch & 63) == 0 );
-   assert( (src_offset & 1023) == 0 );
-   assert( (dst_offset & 1023) == 0 );
-   assert( w < (1<<16) );
-   assert( h < (1<<16) );
-
-   cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, 8 * sizeof(int),
-						  __FUNCTION__ );
-
-
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
-   cmd[1].i = R200_CP_CMD_BITBLT_MULTI | (5 << 16);
-   cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
-	       RADEON_GMC_DST_PITCH_OFFSET_CNTL |
-	       RADEON_GMC_BRUSH_NONE |
-	       (color_fmt << 8) |
-	       RADEON_GMC_SRC_DATATYPE_COLOR |
-	       RADEON_ROP3_S |
-	       RADEON_DP_SRC_SOURCE_MEMORY |
-	       RADEON_GMC_CLR_CMP_CNTL_DIS |
-	       RADEON_GMC_WR_MSK_DIS );
-
-   cmd[3].i = ((src_pitch/64)<<22) | (src_offset >> 10);
-   cmd[4].i = ((dst_pitch/64)<<22) | (dst_offset >> 10);
-   cmd[5].i = (srcx << 16) | srcy;
-   cmd[6].i = (dstx << 16) | dsty; /* dst */
-   cmd[7].i = (w << 16) | h;
-}
-
-
-void r200EmitWait( r200ContextPtr rmesa, GLuint flags )
-{
-   drm_radeon_cmd_header_t *cmd;
-
-   assert( !(flags & ~(RADEON_WAIT_2D|RADEON_WAIT_3D)) );
-
-   cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, 1 * sizeof(int),
-					   __FUNCTION__ );
-   cmd[0].i = 0;
-   cmd[0].wait.cmd_type = RADEON_CMD_WAIT;
-   cmd[0].wait.flags = flags;
+   END_BATCH();
 }
diff --git a/src/mesa/drivers/dri/r200/r200_context.c b/src/mesa/drivers/dri/r200/r200_context.c
index c06751516eb..8924849d082 100644
--- a/src/mesa/drivers/dri/r200/r200_context.c
+++ b/src/mesa/drivers/dri/r200/r200_context.c
@@ -54,7 +54,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_context.h"
 #include "r200_ioctl.h"
 #include "r200_state.h"
-#include "r200_span.h"
 #include "r200_pixel.h"
 #include "r200_tex.h"
 #include "r200_swtcl.h"
@@ -62,6 +61,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_maos.h"
 #include "r200_vertprog.h"
 
+#include "radeon_span.h"
+
 #define need_GL_ARB_vertex_program
 #define need_GL_ATI_fragment_shader
 #define need_GL_EXT_blend_minmax
@@ -71,6 +72,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define need_GL_EXT_blend_func_separate
 #define need_GL_NV_vertex_program
 #define need_GL_ARB_point_parameters
+#define need_GL_EXT_framebuffer_object
 #include "extension_helper.h"
 
 #define DRIVER_DATE	"20060602"
@@ -78,9 +80,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "vblank.h"
 #include "utils.h"
 #include "xmlpool.h" /* for symbolic values of enum-type options */
-#ifndef R200_DEBUG
-int R200_DEBUG = (0);
-#endif
 
 /* Return various strings for glGetString().
  */
@@ -89,8 +88,8 @@ static const GLubyte *r200GetString( GLcontext *ctx, GLenum name )
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
    static char buffer[128];
    unsigned   offset;
-   GLuint agp_mode = (rmesa->r200Screen->card_type == RADEON_CARD_PCI)? 0 :
-      rmesa->r200Screen->AGPMode;
+   GLuint agp_mode = (rmesa->radeon.radeonScreen->card_type == RADEON_CARD_PCI)? 0 :
+      rmesa->radeon.radeonScreen->AGPMode;
 
    switch ( name ) {
    case GL_VENDOR:
@@ -101,7 +100,7 @@ static const GLubyte *r200GetString( GLcontext *ctx, GLenum name )
 				     agp_mode );
 
       sprintf( & buffer[ offset ], " %sTCL",
-	       !(rmesa->TclFallback & R200_TCL_FALLBACK_TCL_DISABLE)
+	       !(rmesa->radeon.TclFallback & R200_TCL_FALLBACK_TCL_DISABLE)
 	       ? "" : "NO-" );
 
       return (GLubyte *)buffer;
@@ -126,6 +125,7 @@ const struct dri_extension card_extensions[] =
     { "GL_EXT_blend_minmax",               GL_EXT_blend_minmax_functions },
     { "GL_EXT_blend_subtract",             NULL },
     { "GL_EXT_fog_coord",                  GL_EXT_fog_coord_functions },
+    { "GL_EXT_packed_depth_stencil",	   NULL},
     { "GL_EXT_secondary_color",            GL_EXT_secondary_color_functions },
     { "GL_EXT_stencil_wrap",               NULL },
     { "GL_EXT_texture_edge_clamp",         NULL },
@@ -167,6 +167,11 @@ const struct dri_extension point_extensions[] = {
     { NULL,                                NULL }
 };
 
+const struct dri_extension mm_extensions[] = {
+  { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions },
+  { NULL, NULL }
+};
+
 extern const struct tnl_pipeline_stage _r200_render_stage;
 extern const struct tnl_pipeline_stage _r200_tcl_stage;
 
@@ -234,6 +239,39 @@ static const struct dri_debug_control debug_control[] =
     { NULL,    0 }
 };
 
+static void r200_get_lock(radeonContextPtr radeon)
+{
+   r200ContextPtr rmesa = (r200ContextPtr)radeon;
+   drm_radeon_sarea_t *sarea = radeon->sarea;
+
+   R200_STATECHANGE( rmesa, ctx );
+   if (rmesa->radeon.sarea->tiling_enabled) {
+      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= R200_COLOR_TILE_ENABLE;
+   }
+   else rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] &= ~R200_COLOR_TILE_ENABLE;
+
+   if ( sarea->ctx_owner != rmesa->radeon.dri.hwContext ) {
+      sarea->ctx_owner = rmesa->radeon.dri.hwContext;
+      if (!radeon->radeonScreen->kernel_mm)
+         radeon_bo_legacy_texture_age(radeon->radeonScreen->bom);
+   }
+
+}
+
+static void r200_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmesa)
+{
+}
+
+
+static void r200_init_vtbl(radeonContextPtr radeon)
+{
+   radeon->vtbl.get_lock = r200_get_lock;
+   radeon->vtbl.update_viewport_offset = r200UpdateViewportOffset;
+   radeon->vtbl.emit_cs_header = r200_vtbl_emit_cs_header;
+   radeon->vtbl.swtcl_flush = r200_swtcl_flush;
+   radeon->vtbl.fallback = r200Fallback;
+}
+
 
 /* Create the device specific rendering context.
  */
@@ -245,9 +283,9 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
    radeonScreenPtr screen = (radeonScreenPtr)(sPriv->private);
    struct dd_function_table functions;
    r200ContextPtr rmesa;
-   GLcontext *ctx, *shareCtx;
+   GLcontext *ctx;
    int i;
-   int tcl_mode, fthrottle_mode;
+   int tcl_mode;
 
    assert(glVisual);
    assert(driContextPriv);
@@ -257,7 +295,8 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
    rmesa = (r200ContextPtr) CALLOC( sizeof(*rmesa) );
    if ( !rmesa )
       return GL_FALSE;
-      
+
+   r200_init_vtbl(&rmesa->radeon);
    /* init exp fog table data */
    r200InitStaticFogData();
 
@@ -265,12 +304,12 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
     * Do this here so that initialMaxAnisotropy is set before we create
     * the default textures.
     */
-   driParseConfigFiles (&rmesa->optionCache, &screen->optionCache,
+   driParseConfigFiles (&rmesa->radeon.optionCache, &screen->optionCache,
 			screen->driScreen->myNum, "r200");
-   rmesa->initialMaxAnisotropy = driQueryOptionf(&rmesa->optionCache,
-                                                 "def_max_anisotropy");
+   rmesa->radeon.initialMaxAnisotropy = driQueryOptionf(&rmesa->radeon.optionCache,
+							"def_max_anisotropy");
 
-   if ( driQueryOptionb( &rmesa->optionCache, "hyperz" ) ) {
+   if ( driQueryOptionb( &rmesa->radeon.optionCache, "hyperz" ) ) {
       if ( sPriv->drm_version.minor < 13 )
 	 fprintf( stderr, "DRM version 1.%d too old to support HyperZ, "
 			  "disabling.\n", sPriv->drm_version.minor );
@@ -291,59 +330,15 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
    r200InitTextureFuncs(&functions);
    r200InitShaderFuncs(&functions); 
 
-   /* Allocate and initialize the Mesa context */
-   if (sharedContextPrivate)
-      shareCtx = ((r200ContextPtr) sharedContextPrivate)->glCtx;
-   else
-      shareCtx = NULL;
-   rmesa->glCtx = _mesa_create_context(glVisual, shareCtx,
-                                       &functions, (void *) rmesa);
-   if (!rmesa->glCtx) {
-      FREE(rmesa);
-      return GL_FALSE;
-   }
-   driContextPriv->driverPrivate = rmesa;
-
-   /* Init r200 context data */
-   rmesa->dri.context = driContextPriv;
-   rmesa->dri.screen = sPriv;
-   rmesa->dri.drawable = NULL; /* Set by XMesaMakeCurrent */
-   rmesa->dri.hwContext = driContextPriv->hHWContext;
-   rmesa->dri.hwLock = &sPriv->pSAREA->lock;
-   rmesa->dri.fd = sPriv->fd;
-   rmesa->dri.drmMinor = sPriv->drm_version.minor;
-
-   rmesa->r200Screen = screen;
-   rmesa->sarea = (drm_radeon_sarea_t *)((GLubyte *)sPriv->pSAREA +
-				       screen->sarea_priv_offset);
-
-
-   rmesa->dma.buf0_address = rmesa->r200Screen->buffers->list[0].address;
-
-   (void) memset( rmesa->texture_heaps, 0, sizeof( rmesa->texture_heaps ) );
-   make_empty_list( & rmesa->swapped );
-
-   rmesa->nr_heaps = 1 /* screen->numTexHeaps */ ;
-   assert(rmesa->nr_heaps < RADEON_NR_TEX_HEAPS);
-   for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
-      rmesa->texture_heaps[i] = driCreateTextureHeap( i, rmesa,
-	    screen->texSize[i],
-	    12,
-	    RADEON_NR_TEX_REGIONS,
-	    (drmTextureRegionPtr)rmesa->sarea->tex_list[i],
-	    & rmesa->sarea->tex_age[i],
-	    & rmesa->swapped,
-	    sizeof( r200TexObj ),
-	    (destroy_texture_object_t *) r200DestroyTexObj );
+   if (!radeonInitContext(&rmesa->radeon, &functions,
+			  glVisual, driContextPriv,
+			  sharedContextPrivate)) {
+     FREE(rmesa);
+     return GL_FALSE;
    }
-   rmesa->texture_depth = driQueryOptioni (&rmesa->optionCache,
-					   "texture_depth");
-   if (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
-      rmesa->texture_depth = ( screen->cpp == 4 ) ?
-	 DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16;
 
-   rmesa->swtcl.RenderIndex = ~0;
-   rmesa->hw.all_dirty = 1;
+   rmesa->radeon.swtcl.RenderIndex = ~0;
+   rmesa->radeon.hw.all_dirty = 1;
 
    /* Set the maximum texture size small enough that we can guarentee that
     * all texture units can bind a maximal texture and have all of them in
@@ -351,29 +346,17 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
     * setting allow larger textures.
     */
 
-   ctx = rmesa->glCtx;
-   ctx->Const.MaxTextureUnits = driQueryOptioni (&rmesa->optionCache,
+   ctx = rmesa->radeon.glCtx;
+   ctx->Const.MaxTextureUnits = driQueryOptioni (&rmesa->radeon.optionCache,
 						 "texture_units");
    ctx->Const.MaxTextureImageUnits = ctx->Const.MaxTextureUnits;
    ctx->Const.MaxTextureCoordUnits = ctx->Const.MaxTextureUnits;
 
-   i = driQueryOptioni( &rmesa->optionCache, "allow_large_textures");
-
-   driCalculateMaxTextureLevels( rmesa->texture_heaps,
-				 rmesa->nr_heaps,
-				 & ctx->Const,
-				 4,
-				 11, /* max 2D texture size is 2048x2048 */
-#if ENABLE_HW_3D_TEXTURE
-				 8,  /* max 3D texture size is 256^3 */
-#else
-				 0,  /* 3D textures unsupported */
-#endif
-				 11, /* max cube texture size is 2048x2048 */
-				 11, /* max texture rectangle size is 2048x2048 */
-				 12,
-				 GL_FALSE,
-				 i );
+   i = driQueryOptioni( &rmesa->radeon.optionCache, "allow_large_textures");
+
+   /* FIXME: When no memory manager is available we should set this 
+    * to some reasonable value based on texture memory pool size */
+   ctx->Const.MaxTextureLevels = 12;
 
    ctx->Const.MaxTextureMaxAnisotropy = 16.0;
 
@@ -383,7 +366,7 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
    ctx->Const.MinPointSizeAA = 1.0;
    ctx->Const.MaxPointSizeAA = 1.0;
    ctx->Const.PointSizeGranularity = 0.0625;
-   if (rmesa->r200Screen->drmSupportsPointSprites)
+   if (rmesa->radeon.radeonScreen->drmSupportsPointSprites)
       ctx->Const.MaxPointSize = 2047.0;
    else
       ctx->Const.MaxPointSize = 1.0;
@@ -439,32 +422,35 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
    _math_matrix_set_identity( &rmesa->tmpmat );
 
    driInitExtensions( ctx, card_extensions, GL_TRUE );
-   if (!(rmesa->r200Screen->chip_flags & R200_CHIPSET_YCBCR_BROKEN)) {
+
+   if (rmesa->radeon.radeonScreen->kernel_mm)
+     driInitExtensions(ctx, mm_extensions, GL_FALSE);
+   if (!(rmesa->radeon.radeonScreen->chip_flags & R200_CHIPSET_YCBCR_BROKEN)) {
      /* yuv textures don't work with some chips - R200 / rv280 okay so far
 	others get the bit ordering right but don't actually do YUV-RGB conversion */
       _mesa_enable_extension( ctx, "GL_MESA_ycbcr_texture" );
    }
-   if (rmesa->glCtx->Mesa_DXTn) {
+   if (rmesa->radeon.glCtx->Mesa_DXTn) {
       _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
       _mesa_enable_extension( ctx, "GL_S3_s3tc" );
    }
-   else if (driQueryOptionb (&rmesa->optionCache, "force_s3tc_enable")) {
+   else if (driQueryOptionb (&rmesa->radeon.optionCache, "force_s3tc_enable")) {
       _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
    }
 
-   if (rmesa->r200Screen->drmSupportsCubeMapsR200)
+   if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR200)
       _mesa_enable_extension( ctx, "GL_ARB_texture_cube_map" );
-   if (rmesa->r200Screen->drmSupportsBlendColor) {
+   if (rmesa->radeon.radeonScreen->drmSupportsBlendColor) {
        driInitExtensions( ctx, blend_extensions, GL_FALSE );
    }
-   if(rmesa->r200Screen->drmSupportsVertexProgram)
+   if(rmesa->radeon.radeonScreen->drmSupportsVertexProgram)
       driInitSingleExtension( ctx, ARB_vp_extension );
-   if(driQueryOptionb(&rmesa->optionCache, "nv_vertex_program"))
+   if(driQueryOptionb(&rmesa->radeon.optionCache, "nv_vertex_program"))
       driInitSingleExtension( ctx, NV_vp_extension );
 
-   if ((ctx->Const.MaxTextureUnits == 6) && rmesa->r200Screen->drmSupportsFragShader)
+   if ((ctx->Const.MaxTextureUnits == 6) && rmesa->radeon.radeonScreen->drmSupportsFragShader)
       driInitSingleExtension( ctx, ATI_fs_extension );
-   if (rmesa->r200Screen->drmSupportsPointSprites)
+   if (rmesa->radeon.radeonScreen->drmSupportsPointSprites)
       driInitExtensions( ctx, point_extensions, GL_FALSE );
 #if 0
    r200InitDriverFuncs( ctx );
@@ -474,33 +460,16 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
 #endif
    /* plug in a few more device driver functions */
    /* XXX these should really go right after _mesa_init_driver_functions() */
+   radeon_fbo_init(&rmesa->radeon);
+   radeonInitSpanFuncs( ctx );
    r200InitPixelFuncs( ctx );
-   r200InitSpanFuncs( ctx );
    r200InitTnlFuncs( ctx );
    r200InitState( rmesa );
    r200InitSwtcl( ctx );
 
-   fthrottle_mode = driQueryOptioni(&rmesa->optionCache, "fthrottle_mode");
-   rmesa->iw.irq_seq = -1;
-   rmesa->irqsEmitted = 0;
-   rmesa->do_irqs = (fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS &&
-		     rmesa->r200Screen->irq);
-
-   rmesa->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS);
-
-   if (!rmesa->do_irqs)
-      fprintf(stderr,
-	      "IRQ's not enabled, falling back to %s: %d %d\n",
-	      rmesa->do_usleeps ? "usleeps" : "busy waits",
-	      fthrottle_mode,
-	      rmesa->r200Screen->irq);
-
    rmesa->prefer_gart_client_texturing = 
       (getenv("R200_GART_CLIENT_TEXTURES") != 0);
 
-   (*sPriv->systemTime->getUST)( & rmesa->swap_ust );
-
-
 #if DO_DEBUG
    R200_DEBUG  = driParseDebugString( getenv( "R200_DEBUG" ),
 				      debug_control );
@@ -508,202 +477,21 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
 				      debug_control );
 #endif
 
-   tcl_mode = driQueryOptioni(&rmesa->optionCache, "tcl_mode");
-   if (driQueryOptionb(&rmesa->optionCache, "no_rast")) {
+   tcl_mode = driQueryOptioni(&rmesa->radeon.optionCache, "tcl_mode");
+   if (driQueryOptionb(&rmesa->radeon.optionCache, "no_rast")) {
       fprintf(stderr, "disabling 3D acceleration\n");
       FALLBACK(rmesa, R200_FALLBACK_DISABLE, 1);
    }
    else if (tcl_mode == DRI_CONF_TCL_SW || getenv("R200_NO_TCL") ||
-	    !(rmesa->r200Screen->chip_flags & RADEON_CHIPSET_TCL)) {
-      if (rmesa->r200Screen->chip_flags & RADEON_CHIPSET_TCL) {
-	 rmesa->r200Screen->chip_flags &= ~RADEON_CHIPSET_TCL;
+	    !(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
+      if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
+	 rmesa->radeon.radeonScreen->chip_flags &= ~RADEON_CHIPSET_TCL;
 	 fprintf(stderr, "Disabling HW TCL support\n");
       }
-      TCL_FALLBACK(rmesa->glCtx, R200_TCL_FALLBACK_TCL_DISABLE, 1);
+      TCL_FALLBACK(rmesa->radeon.glCtx, R200_TCL_FALLBACK_TCL_DISABLE, 1);
    }
 
    return GL_TRUE;
 }
 
 
-/* Destroy the device specific context.
- */
-/* Destroy the Mesa and driver specific context data.
- */
-void r200DestroyContext( __DRIcontextPrivate *driContextPriv )
-{
-   GET_CURRENT_CONTEXT(ctx);
-   r200ContextPtr rmesa = (r200ContextPtr) driContextPriv->driverPrivate;
-   r200ContextPtr current = ctx ? R200_CONTEXT(ctx) : NULL;
-
-   /* check if we're deleting the currently bound context */
-   if (rmesa == current) {
-      R200_FIREVERTICES( rmesa );
-      _mesa_make_current(NULL, NULL, NULL);
-   }
-
-   /* Free r200 context resources */
-   assert(rmesa); /* should never be null */
-   if ( rmesa ) {
-      GLboolean   release_texture_heaps;
-
-
-      release_texture_heaps = (rmesa->glCtx->Shared->RefCount == 1);
-      _swsetup_DestroyContext( rmesa->glCtx );
-      _tnl_DestroyContext( rmesa->glCtx );
-      _vbo_DestroyContext( rmesa->glCtx );
-      _swrast_DestroyContext( rmesa->glCtx );
-
-      r200DestroySwtcl( rmesa->glCtx );
-      r200ReleaseArrays( rmesa->glCtx, ~0 );
-
-      if (rmesa->dma.current.buf) {
-	 r200ReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ );
-	 r200FlushCmdBuf( rmesa, __FUNCTION__ );
-      }
-
-      if (rmesa->state.scissor.pClipRects) {
-	 FREE(rmesa->state.scissor.pClipRects);
-	 rmesa->state.scissor.pClipRects = NULL;
-      }
-
-      if ( release_texture_heaps ) {
-         /* This share group is about to go away, free our private
-          * texture object data.
-          */
-         int i;
-
-         for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
-	    driDestroyTextureHeap( rmesa->texture_heaps[ i ] );
-	    rmesa->texture_heaps[ i ] = NULL;
-         }
-
-	 assert( is_empty_list( & rmesa->swapped ) );
-      }
-
-      /* free the Mesa context */
-      rmesa->glCtx->DriverCtx = NULL;
-      _mesa_destroy_context( rmesa->glCtx );
-
-      /* free the option cache */
-      driDestroyOptionCache (&rmesa->optionCache);
-
-      FREE( rmesa );
-   }
-}
-
-
-
-
-void
-r200SwapBuffers( __DRIdrawablePrivate *dPriv )
-{
-   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
-      r200ContextPtr rmesa;
-      GLcontext *ctx;
-      rmesa = (r200ContextPtr) dPriv->driContextPriv->driverPrivate;
-      ctx = rmesa->glCtx;
-      if (ctx->Visual.doubleBufferMode) {
-         _mesa_notifySwapBuffers( ctx );  /* flush pending rendering comands */
-         if ( rmesa->doPageFlip ) {
-            r200PageFlip( dPriv );
-         }
-         else {
-	     r200CopyBuffer( dPriv, NULL );
-         }
-      }
-   }
-   else {
-      /* XXX this shouldn't be an error but we can't handle it for now */
-      _mesa_problem(NULL, "%s: drawable has no context!", __FUNCTION__);
-   }
-}
-
-void
-r200CopySubBuffer( __DRIdrawablePrivate *dPriv,
-		   int x, int y, int w, int h )
-{
-   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
-      r200ContextPtr rmesa;
-      GLcontext *ctx;
-      rmesa = (r200ContextPtr) dPriv->driContextPriv->driverPrivate;
-      ctx = rmesa->glCtx;
-      if (ctx->Visual.doubleBufferMode) {
-	 drm_clip_rect_t rect;
-	 rect.x1 = x + dPriv->x;
-	 rect.y1 = (dPriv->h - y - h) + dPriv->y;
-	 rect.x2 = rect.x1 + w;
-	 rect.y2 = rect.y1 + h;
-         _mesa_notifySwapBuffers( ctx );  /* flush pending rendering comands */
-	 r200CopyBuffer( dPriv, &rect );
-      }
-   }
-   else {
-      /* XXX this shouldn't be an error but we can't handle it for now */
-      _mesa_problem(NULL, "%s: drawable has no context!", __FUNCTION__);
-   }
-}
-
-/* Force the context `c' to be the current context and associate with it
- * buffer `b'.
- */
-GLboolean
-r200MakeCurrent( __DRIcontextPrivate *driContextPriv,
-                   __DRIdrawablePrivate *driDrawPriv,
-                   __DRIdrawablePrivate *driReadPriv )
-{
-   if ( driContextPriv ) {
-      r200ContextPtr newCtx = 
-	 (r200ContextPtr) driContextPriv->driverPrivate;
-
-      if (R200_DEBUG & DEBUG_DRI)
-	 fprintf(stderr, "%s ctx %p\n", __FUNCTION__, (void *)newCtx->glCtx);
-
-      newCtx->dri.readable = driReadPriv;
-
-      if ( newCtx->dri.drawable != driDrawPriv ||
-           newCtx->lastStamp != driDrawPriv->lastStamp ) {
-	 if (driDrawPriv->swap_interval == (unsigned)-1) {
-	    driDrawPriv->vblFlags = (newCtx->r200Screen->irq != 0)
-	       ? driGetDefaultVBlankFlags(&newCtx->optionCache)
-	       : VBLANK_FLAG_NO_IRQ;
-
-	    driDrawableInitVBlank( driDrawPriv );
-	 }
-
-	 newCtx->dri.drawable = driDrawPriv;
-
-	 r200SetCliprects(newCtx);
-	 r200UpdateViewportOffset( newCtx->glCtx );
-      }
-
-      _mesa_make_current( newCtx->glCtx,
-			  (GLframebuffer *) driDrawPriv->driverPrivate,
-			  (GLframebuffer *) driReadPriv->driverPrivate );
-
-      _mesa_update_state( newCtx->glCtx );
-      r200ValidateState( newCtx->glCtx );
-
-   } else {
-      if (R200_DEBUG & DEBUG_DRI)
-	 fprintf(stderr, "%s ctx is null\n", __FUNCTION__);
-      _mesa_make_current( NULL, NULL, NULL );
-   }
-
-   if (R200_DEBUG & DEBUG_DRI)
-      fprintf(stderr, "End %s\n", __FUNCTION__);
-   return GL_TRUE;
-}
-
-/* Force the context `c' to be unbound from its buffer.
- */
-GLboolean
-r200UnbindContext( __DRIcontextPrivate *driContextPriv )
-{
-   r200ContextPtr rmesa = (r200ContextPtr) driContextPriv->driverPrivate;
-
-   if (R200_DEBUG & DEBUG_DRI)
-      fprintf(stderr, "%s ctx %p\n", __FUNCTION__, (void *)rmesa->glCtx);
-
-   return GL_TRUE;
-}
diff --git a/src/mesa/drivers/dri/r200/r200_context.h b/src/mesa/drivers/dri/r200/r200_context.h
index 14a1dda46ac..6267293817d 100644
--- a/src/mesa/drivers/dri/r200/r200_context.h
+++ b/src/mesa/drivers/dri/r200/r200_context.h
@@ -53,51 +53,17 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #error This driver requires a newer libdrm to compile
 #endif
 
+#include "radeon_screen.h"
+#include "radeon_common.h"
+
+#include "radeon_lock.h"
+
 struct r200_context;
 typedef struct r200_context r200ContextRec;
 typedef struct r200_context *r200ContextPtr;
 
-/* This union is used to avoid warnings/miscompilation
-   with float to uint32_t casts due to strict-aliasing */
-typedef union { GLfloat f; uint32_t ui32; } float_ui32_type;
-
-#include "r200_lock.h"
-#include "radeon_screen.h"
 #include "main/mm.h"
 
-/* Flags for software fallback cases */
-/* See correponding strings in r200_swtcl.c */
-#define R200_FALLBACK_TEXTURE           0x01
-#define R200_FALLBACK_DRAW_BUFFER       0x02
-#define R200_FALLBACK_STENCIL           0x04
-#define R200_FALLBACK_RENDER_MODE       0x08
-#define R200_FALLBACK_DISABLE           0x10
-#define R200_FALLBACK_BORDER_MODE       0x20
-
-/* The blit width for texture uploads
- */
-#define BLIT_WIDTH_BYTES 1024
-
-/* Use the templated vertex format:
- */
-#define COLOR_IS_RGBA
-#define TAG(x) r200##x
-#include "tnl_dd/t_dd_vertex.h"
-#undef TAG
-
-typedef void (*r200_tri_func)( r200ContextPtr,
-				 r200Vertex *,
-				 r200Vertex *,
-				 r200Vertex * );
-
-typedef void (*r200_line_func)( r200ContextPtr,
-				  r200Vertex *,
-				  r200Vertex * );
-
-typedef void (*r200_point_func)( r200ContextPtr,
-				   r200Vertex * );
-
-
 struct r200_vertex_program {
         struct gl_vertex_program mesa_program; /* Must be first */
         int translated;
@@ -112,93 +78,11 @@ struct r200_vertex_program {
         int fogmode;
 };
 
-struct r200_colorbuffer_state {
-   GLuint clear;
-#if 000
-   GLint drawOffset, drawPitch;
-#endif
-   int roundEnable;
-};
-
-
-struct r200_depthbuffer_state {
-   GLuint clear;
-   GLfloat scale;
-};
-
-#if 000
-struct r200_pixel_state {
-   GLint readOffset, readPitch;
-};
-#endif
-
-struct r200_scissor_state {
-   drm_clip_rect_t rect;
-   GLboolean enabled;
-
-   GLuint numClipRects;			/* Cliprects active */
-   GLuint numAllocedClipRects;		/* Cliprects available */
-   drm_clip_rect_t *pClipRects;
-};
-
-struct r200_stencilbuffer_state {
-   GLboolean hwBuffer;
-   GLuint clear;			/* rb3d_stencilrefmask value */
-};
-
-struct r200_stipple_state {
-   GLuint mask[32];
-};
-
-
-
-#define TEX_0   0x1
-#define TEX_1   0x2
-#define TEX_2	0x4
-#define TEX_3	0x8
-#define TEX_4	0x10
-#define TEX_5	0x20
-#define TEX_ALL 0x3f
-
-typedef struct r200_tex_obj r200TexObj, *r200TexObjPtr;
-
-/* Texture object in locally shared texture space.
- */
-struct r200_tex_obj {
-   driTextureObject   base;
-
-   GLuint bufAddr;			/* Offset to start of locally
-					   shared texture block */
-
-   GLuint dirty_state;		        /* Flags (1 per texunit) for
-					   whether or not this texobj
-					   has dirty hardware state
-					   (pp_*) that needs to be
-					   brought into the
-					   texunit. */
-
-   drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS];
-					/* Six, for the cube faces */
-   GLboolean image_override;		/* Image overridden by GLX_EXT_tfp */
-
-   GLuint pp_txfilter;		        /* hardware register values */
-   GLuint pp_txformat;
-   GLuint pp_txformat_x;
-   GLuint pp_txoffset;		        /* Image location in texmem.
-					   All cube faces follow. */
-   GLuint pp_txsize;		        /* npot only */
-   GLuint pp_txpitch;		        /* npot only */
-   GLuint pp_border_color;
-   GLuint pp_cubic_faces;	        /* cube face 1,2,3,4 log2 sizes */
-
-   GLboolean  border_fallback;
-
-   GLuint tile_bits;			/* hw texture tile bits used on this texture */
-};
+#define R200_TEX_ALL 0x3f
 
 
 struct r200_texture_env_state {
-   r200TexObjPtr texobj;
+   radeonTexObjPtr texobj;
    GLuint outputreg;
    GLuint unitneeded;
 };
@@ -210,19 +94,6 @@ struct r200_texture_state {
 };
 
 
-struct r200_state_atom {
-   struct r200_state_atom *next, *prev;
-   const char *name;		         /* for debug */
-   int cmd_size;		         /* size in bytes */
-   GLuint idx;
-   int *cmd;			         /* one or more cmd's */
-   int *lastcmd;			 /* one or more cmd's */
-   GLboolean dirty;
-   GLboolean (*check)( GLcontext *, int );    /* is this state active? */
-};
-   
-
-
 /* Trying to keep these relatively short as the variables are becoming
  * extravagently long.  Drop the driver name prefix off the front of
  * everything - I think we know which driver we're in by now, and keep the
@@ -597,181 +468,79 @@ struct r200_state_atom {
 
 
 struct r200_hw_state {
-   /* Head of the linked list of state atoms. */
-   struct r200_state_atom atomlist;
-
    /* Hardware state, stored as cmdbuf commands:  
     *   -- Need to doublebuffer for
     *           - reviving state after loss of context
     *           - eliding noop statechange loops? (except line stipple count)
     */
-   struct r200_state_atom ctx;
-   struct r200_state_atom set;
-   struct r200_state_atom vte;
-   struct r200_state_atom lin;
-   struct r200_state_atom msk;
-   struct r200_state_atom vpt;
-   struct r200_state_atom vap;
-   struct r200_state_atom vtx;
-   struct r200_state_atom tcl;
-   struct r200_state_atom msl;
-   struct r200_state_atom tcg;
-   struct r200_state_atom msc;
-   struct r200_state_atom cst;
-   struct r200_state_atom tam;
-   struct r200_state_atom tf;
-   struct r200_state_atom tex[6];
-   struct r200_state_atom cube[6];
-   struct r200_state_atom zbs;
-   struct r200_state_atom mtl[2];
-   struct r200_state_atom mat[9];
-   struct r200_state_atom lit[8]; /* includes vec, scl commands */
-   struct r200_state_atom ucp[6];
-   struct r200_state_atom pix[6]; /* pixshader stages */
-   struct r200_state_atom eye; /* eye pos */
-   struct r200_state_atom grd; /* guard band clipping */
-   struct r200_state_atom fog;
-   struct r200_state_atom glt;
-   struct r200_state_atom prf;
-   struct r200_state_atom afs[2];
-   struct r200_state_atom pvs;
-   struct r200_state_atom vpi[2];
-   struct r200_state_atom vpp[2];
-   struct r200_state_atom atf;
-   struct r200_state_atom spr;
-   struct r200_state_atom ptp;
-
-   int max_state_size;	/* Number of bytes necessary for a full state emit. */
-   GLboolean is_dirty, all_dirty;
+   struct radeon_state_atom ctx;
+   struct radeon_state_atom set;
+   struct radeon_state_atom vte;
+   struct radeon_state_atom lin;
+   struct radeon_state_atom msk;
+   struct radeon_state_atom vpt;
+   struct radeon_state_atom vap;
+   struct radeon_state_atom vtx;
+   struct radeon_state_atom tcl;
+   struct radeon_state_atom msl;
+   struct radeon_state_atom tcg;
+   struct radeon_state_atom msc;
+   struct radeon_state_atom cst;
+   struct radeon_state_atom tam;
+   struct radeon_state_atom tf;
+   struct radeon_state_atom tex[6];
+   struct radeon_state_atom cube[6];
+   struct radeon_state_atom zbs;
+   struct radeon_state_atom mtl[2];
+   struct radeon_state_atom mat[9];
+   struct radeon_state_atom lit[8]; /* includes vec, scl commands */
+   struct radeon_state_atom ucp[6];
+   struct radeon_state_atom pix[6]; /* pixshader stages */
+   struct radeon_state_atom eye; /* eye pos */
+   struct radeon_state_atom grd; /* guard band clipping */
+   struct radeon_state_atom fog;
+   struct radeon_state_atom glt;
+   struct radeon_state_atom prf;
+   struct radeon_state_atom afs[2];
+   struct radeon_state_atom pvs;
+   struct radeon_state_atom vpi[2];
+   struct radeon_state_atom vpp[2];
+   struct radeon_state_atom atf;
+   struct radeon_state_atom spr;
+   struct radeon_state_atom ptp;
 };
 
 struct r200_state {
    /* Derived state for internal purposes:
     */
-   struct r200_colorbuffer_state color;
-   struct r200_depthbuffer_state depth;
-#if 00
-   struct r200_pixel_state pixel;
-#endif
-   struct r200_scissor_state scissor;
-   struct r200_stencilbuffer_state stencil;
-   struct r200_stipple_state stipple;
+   struct radeon_stipple_state stipple;
    struct r200_texture_state texture;
    GLuint envneeded;
 };
 
-/* Need refcounting on dma buffers:
- */
-struct r200_dma_buffer {
-   int refcount;		/* the number of retained regions in buf */
-   drmBufPtr buf;
-};
-
-#define GET_START(rvb) (rmesa->r200Screen->gart_buffer_offset +		\
-			(rvb)->address - rmesa->dma.buf0_address +	\
-			(rvb)->start)
-
-/* A retained region, eg vertices for indexed vertices.
- */
-struct r200_dma_region {
-   struct r200_dma_buffer *buf;
-   char *address;		/* == buf->address */
-   int start, end, ptr;		/* offsets from start of buf */
-   int aos_start;
-   int aos_stride;
-   int aos_size;
-};
-
-
-struct r200_dma {
-   /* Active dma region.  Allocations for vertices and retained
-    * regions come from here.  Also used for emitting random vertices,
-    * these may be flushed by calling flush_current();
-    */
-   struct r200_dma_region current;
-   
-   void (*flush)( r200ContextPtr );
-
-   char *buf0_address;		/* start of buf[0], for index calcs */
-   GLuint nr_released_bufs;	/* flush after so many buffers released */
-};
-
-struct r200_dri_mirror {
-   __DRIcontextPrivate	*context;	/* DRI context */
-   __DRIscreenPrivate	*screen;	/* DRI screen */
-   __DRIdrawablePrivate	*drawable;	/* DRI drawable bound to this ctx */
-   __DRIdrawablePrivate	*readable;	/* DRI readable bound to this ctx */
-
-   drm_context_t hwContext;
-   drm_hw_lock_t *hwLock;
-   int fd;
-   int drmMinor;
-};
-
-
 #define R200_CMD_BUF_SZ  (16*1024) 
 
-struct r200_store {
-   GLuint statenr;
-   GLuint primnr;
-   char cmd_buf[R200_CMD_BUF_SZ];
-   int cmd_used;   
-   int elts_start;
-};
-
-
+#define R200_ELT_BUF_SZ  (16*1024) 
 /* r200_tcl.c
  */
 struct r200_tcl_info {
    GLuint hw_primitive;
 
-/* hw can handle 12 components max */
-   struct r200_dma_region *aos_components[12];
-   GLuint nr_aos_components;
-
    GLuint *Elts;
 
-   struct r200_dma_region indexed_verts;
-   struct r200_dma_region vertex_data[15];
+   int elt_used;
+
 };
 
 
 /* r200_swtcl.c
  */
 struct r200_swtcl_info {
-   GLuint RenderIndex;
-   
-   /**
-    * Size of a hardware vertex.  This is calculated when \c ::vertex_attrs is
-    * installed in the Mesa state vector.
-    */
-   GLuint vertex_size;
 
-   /**
-    * Attributes instructing the Mesa TCL pipeline where / how to put vertex
-    * data in the hardware buffer.
-    */
-   struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
 
-   /**
-    * Number of elements of \c ::vertex_attrs that are actually used.
-    */
-   GLuint vertex_attr_count;
-
-   /**
-    * Cached pointer to the buffer where Mesa will store vertex data.
-    */
-   GLubyte *verts;
-
-   /* Fallback rasterization functions
-    */
-   r200_point_func draw_point;
-   r200_line_func draw_line;
-   r200_tri_func draw_tri;
-
-   GLuint hw_primitive;
-   GLenum render_primitive;
-   GLuint numverts;
+   radeon_point_func draw_point;
+   radeon_line_func draw_line;
+   radeon_tri_func draw_tri;
 
    /**
     * Offset of the 4UB color data within a hardware (swtcl) vertex.
@@ -787,27 +556,10 @@ struct r200_swtcl_info {
     * Should Mesa project vertex data or will the hardware do it?
     */
    GLboolean needproj;
-
-   struct r200_dma_region indexed_verts;
-};
-
-
-struct r200_ioctl {
-   GLuint vertex_offset;
-   GLuint vertex_size;
 };
 
 
 
-#define R200_MAX_PRIMS 64
-
-
-
-struct r200_prim {
-   GLuint start;
-   GLuint end;
-   GLuint prim;
-};
 
    /* A maximum total of 29 elements per vertex:  3 floats for position, 3
     * floats for normal, 4 floats for color, 4 bytes for secondary color,
@@ -822,9 +574,8 @@ struct r200_prim {
 
 #define R200_MAX_VERTEX_SIZE ((3*6)+11)
 
-
 struct r200_context {
-   GLcontext *glCtx;			/* Mesa context */
+   struct radeon_context radeon;
 
    /* Driver and hardware state management
     */
@@ -832,56 +583,15 @@ struct r200_context {
    struct r200_state state;
    struct r200_vertex_program *curr_vp_hw;
 
-   /* Texture object bookkeeping
-    */
-   unsigned              nr_heaps;
-   driTexHeap          * texture_heaps[ RADEON_NR_TEX_HEAPS ];
-   driTextureObject      swapped;
-   int                   texture_depth;
-   float                 initialMaxAnisotropy;
-
-   /* Rasterization and vertex state:
-    */
-   GLuint TclFallback;
-   GLuint Fallback;
-   GLuint NewGLState;
-   DECLARE_RENDERINPUTS(tnl_index_bitset);	/* index of bits for last tnl_install_attrs */
-
    /* Vertex buffers
     */
-   struct r200_ioctl ioctl;
-   struct r200_dma dma;
-   struct r200_store store;
-   /* A full state emit as of the first state emit in the main store, in case
-    * the context is lost.
-    */
-   struct r200_store backup_store;
-
-   /* Page flipping
-    */
-   GLuint doPageFlip;
-
-   /* Busy waiting
-    */
-   GLuint do_usleeps;
-   GLuint do_irqs;
-   GLuint irqsEmitted;
-   drm_radeon_irq_wait_t iw;
+   struct radeon_ioctl ioctl;
+   struct radeon_store store;
 
    /* Clientdata textures;
     */
    GLuint prefer_gart_client_texturing;
 
-   /* Drawable, cliprect and scissor information
-    */
-   GLuint numClipRects;			/* Cliprects for the draw buffer */
-   drm_clip_rect_t *pClipRects;
-   unsigned int lastStamp;
-   GLboolean lost_context;
-   GLboolean save_on_next_emit;
-   radeonScreenPtr r200Screen;	/* Screen private DRI data */
-   drm_radeon_sarea_t *sarea;		/* Private SAREA data */
-
    /* TCL stuff
     */
    GLmatrix TexGenMatrix[R200_MAX_TEXTURE_UNITS];
@@ -893,15 +603,6 @@ struct r200_context {
    GLuint TexGenCompSel;
    GLmatrix tmpmat;
 
-   /* buffer swap
-    */
-   int64_t swap_ust;
-   int64_t swap_missed_ust;
-
-   GLuint swap_count;
-   GLuint swap_missed_count;
-
-
    /* r200_tcl.c
     */
    struct r200_tcl_info tcl;
@@ -910,14 +611,6 @@ struct r200_context {
     */
    struct r200_swtcl_info swtcl;
 
-   /* Mirrors of some DRI state
-    */
-   struct r200_dri_mirror dri;
-
-   /* Configuration cache
-    */
-   driOptionCache optionCache;
-
    GLboolean using_hyperz;
    GLboolean texmicrotile;
 
@@ -927,28 +620,10 @@ struct r200_context {
 #define R200_CONTEXT(ctx)		((r200ContextPtr)(ctx->DriverCtx))
 
 
-static INLINE GLuint r200PackColor( GLuint cpp,
-					GLubyte r, GLubyte g,
-					GLubyte b, GLubyte a )
-{
-   switch ( cpp ) {
-   case 2:
-      return PACK_COLOR_565( r, g, b );
-   case 4:
-      return PACK_COLOR_8888( a, r, g, b );
-   default:
-      return 0;
-   }
-}
-
-
 extern void r200DestroyContext( __DRIcontextPrivate *driContextPriv );
 extern GLboolean r200CreateContext( const __GLcontextModes *glVisual,
 				    __DRIcontextPrivate *driContextPriv,
 				    void *sharedContextPrivate);
-extern void r200SwapBuffers( __DRIdrawablePrivate *dPriv );
-extern void r200CopySubBuffer( __DRIdrawablePrivate * dPriv,
-			       int x, int y, int w, int h );
 extern GLboolean r200MakeCurrent( __DRIcontextPrivate *driContextPriv,
 				  __DRIdrawablePrivate *driDrawPriv,
 				  __DRIdrawablePrivate *driReadPriv );
@@ -957,28 +632,9 @@ extern GLboolean r200UnbindContext( __DRIcontextPrivate *driContextPriv );
 /* ================================================================
  * Debugging:
  */
-#define DO_DEBUG		1
 
-#if DO_DEBUG
-extern int R200_DEBUG;
-#else
-#define R200_DEBUG		0
-#endif
+#define R200_DEBUG RADEON_DEBUG
+
 
-#define DEBUG_TEXTURE	0x001
-#define DEBUG_STATE	0x002
-#define DEBUG_IOCTL	0x004
-#define DEBUG_PRIMS	0x008
-#define DEBUG_VERTS	0x010
-#define DEBUG_FALLBACKS	0x020
-#define DEBUG_VFMT	0x040
-#define DEBUG_CODEGEN	0x080
-#define DEBUG_VERBOSE	0x100
-#define DEBUG_DRI       0x200
-#define DEBUG_DMA       0x400
-#define DEBUG_SANITY    0x800
-#define DEBUG_SYNC      0x1000
-#define DEBUG_PIXEL     0x2000
-#define DEBUG_MEMORY    0x4000
 
 #endif /* __R200_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/r200/r200_fragshader.c b/src/mesa/drivers/dri/r200/r200_fragshader.c
index d514b28219a..85c1b7bdd19 100644
--- a/src/mesa/drivers/dri/r200/r200_fragshader.c
+++ b/src/mesa/drivers/dri/r200/r200_fragshader.c
@@ -522,7 +522,7 @@ static void r200UpdateFSConstants( GLcontext *ctx )
 	 CLAMPED_FLOAT_TO_UBYTE(con_byte[2], ctx->ATIFragmentShader.GlobalConstants[i][2]);
 	 CLAMPED_FLOAT_TO_UBYTE(con_byte[3], ctx->ATIFragmentShader.GlobalConstants[i][3]);
       }
-      rmesa->hw.atf.cmd[ATF_TFACTOR_0 + i] = r200PackColor (
+      rmesa->hw.atf.cmd[ATF_TFACTOR_0 + i] = radeonPackColor (
 	 4, con_byte[0], con_byte[1], con_byte[2], con_byte[3] );
    }
 }
diff --git a/src/mesa/drivers/dri/r200/r200_ioctl.c b/src/mesa/drivers/dri/r200/r200_ioctl.c
index 0741e57af71..0b3398a730e 100644
--- a/src/mesa/drivers/dri/r200/r200_ioctl.c
+++ b/src/mesa/drivers/dri/r200/r200_ioctl.c
@@ -31,7 +31,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * Authors:
  *   Keith Whitwell <[email protected]>
  */
- 
+
 #include <sched.h>
 #include <errno.h>
 
@@ -41,6 +41,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/context.h"
 #include "swrast/swrast.h"
 
+
+
+#include "radeon_common.h"
+#include "radeon_lock.h"
 #include "r200_context.h"
 #include "r200_state.h"
 #include "r200_ioctl.h"
@@ -54,635 +58,19 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R200_TIMEOUT             512
 #define R200_IDLE_RETRY           16
 
-
-static void r200WaitForIdle( r200ContextPtr rmesa );
-
-
-/* At this point we were in FlushCmdBufLocked but we had lost our context, so
- * we need to unwire our current cmdbuf, hook the one with the saved state in
- * it, flush it, and then put the current one back.  This is so commands at the
- * start of a cmdbuf can rely on the state being kept from the previous one.
- */
-static void r200BackUpAndEmitLostStateLocked( r200ContextPtr rmesa )
-{
-   GLuint nr_released_bufs;
-   struct r200_store saved_store;
-
-   if (rmesa->backup_store.cmd_used == 0)
-      return;
-
-   if (R200_DEBUG & DEBUG_STATE)
-      fprintf(stderr, "Emitting backup state on lost context\n");
-
-   rmesa->lost_context = GL_FALSE;
-
-   nr_released_bufs = rmesa->dma.nr_released_bufs;
-   saved_store = rmesa->store;
-   rmesa->dma.nr_released_bufs = 0;
-   rmesa->store = rmesa->backup_store;
-   r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
-   rmesa->dma.nr_released_bufs = nr_released_bufs;
-   rmesa->store = saved_store;
-}
-
-int r200FlushCmdBufLocked( r200ContextPtr rmesa, const char * caller )
-{
-   int ret, i;
-   drm_radeon_cmd_buffer_t cmd;
-
-   if (rmesa->lost_context)
-      r200BackUpAndEmitLostStateLocked( rmesa );
-
-   if (R200_DEBUG & DEBUG_IOCTL) {
-      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); 
-
-      if (0 & R200_DEBUG & DEBUG_VERBOSE) 
-	 for (i = 0 ; i < rmesa->store.cmd_used ; i += 4 )
-	    fprintf(stderr, "%d: %x\n", i/4, 
-		    *(int *)(&rmesa->store.cmd_buf[i]));
-   }
-
-   if (R200_DEBUG & DEBUG_DMA)
-      fprintf(stderr, "%s: Releasing %d buffers\n", __FUNCTION__,
-	      rmesa->dma.nr_released_bufs);
-
-
-   if (R200_DEBUG & DEBUG_SANITY) {
-      if (rmesa->state.scissor.enabled) 
-	 ret = r200SanityCmdBuffer( rmesa, 
-				    rmesa->state.scissor.numClipRects,
-				    rmesa->state.scissor.pClipRects);
-      else
-	 ret = r200SanityCmdBuffer( rmesa, 
-				    rmesa->numClipRects,
-				    rmesa->pClipRects);
-      if (ret) {
-	 fprintf(stderr, "drmSanityCommandWrite: %d\n", ret);	 
-	 goto out;
-      }
-   }
-
-
-   if (R200_DEBUG & DEBUG_MEMORY) {
-      if (! driValidateTextureHeaps( rmesa->texture_heaps, rmesa->nr_heaps,
-				     & rmesa->swapped ) ) {
-	 fprintf( stderr, "%s: texture memory is inconsistent - expect "
-		  "mangled textures\n", __FUNCTION__ );
-      }
-   }
-
-
-   cmd.bufsz = rmesa->store.cmd_used;
-   cmd.buf = rmesa->store.cmd_buf;
-
-   if (rmesa->state.scissor.enabled) {
-      cmd.nbox = rmesa->state.scissor.numClipRects;
-      cmd.boxes = (drm_clip_rect_t *)rmesa->state.scissor.pClipRects;
-   } else {
-      cmd.nbox = rmesa->numClipRects;
-      cmd.boxes = (drm_clip_rect_t *)rmesa->pClipRects;
-   }
-
-   ret = drmCommandWrite( rmesa->dri.fd,
-			  DRM_RADEON_CMDBUF,
-			  &cmd, sizeof(cmd) );
-
-   if (ret)
-      fprintf(stderr, "drmCommandWrite: %d\n", ret);
-
-   if (R200_DEBUG & DEBUG_SYNC) {
-      fprintf(stderr, "\nSyncing in %s\n\n", __FUNCTION__);
-      r200WaitForIdleLocked( rmesa );
-   }
-
-
- out:
-   rmesa->store.primnr = 0;
-   rmesa->store.statenr = 0;
-   rmesa->store.cmd_used = 0;
-   rmesa->dma.nr_released_bufs = 0;
-   rmesa->save_on_next_emit = 1;
-
-   return ret;
-}
-
-
-/* Note: does not emit any commands to avoid recursion on
- * r200AllocCmdBuf.
- */
-void r200FlushCmdBuf( r200ContextPtr rmesa, const char *caller )
-{
-   int ret;
-
-   LOCK_HARDWARE( rmesa );
-
-   ret = r200FlushCmdBufLocked( rmesa, caller );
-
-   UNLOCK_HARDWARE( rmesa );
-
-   if (ret) {
-      fprintf(stderr, "drmRadeonCmdBuffer: %d (exiting)\n", ret);
-      exit(ret);
-   }
-}
-
-
-/* =============================================================
- * Hardware vertex buffer handling
- */
-
-
-void r200RefillCurrentDmaRegion( r200ContextPtr rmesa )
-{
-   struct r200_dma_buffer *dmabuf;
-   int fd = rmesa->dri.fd;
-   int index = 0;
-   int size = 0;
-   drmDMAReq dma;
-   int ret;
-
-   if (R200_DEBUG & (DEBUG_IOCTL|DEBUG_DMA))
-      fprintf(stderr, "%s\n", __FUNCTION__);  
-
-   if (rmesa->dma.flush) {
-      rmesa->dma.flush( rmesa );
-   }
-
-   if (rmesa->dma.current.buf)
-      r200ReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ );
-
-   if (rmesa->dma.nr_released_bufs > 4)
-      r200FlushCmdBuf( rmesa, __FUNCTION__ );
-
-   dma.context = rmesa->dri.hwContext;
-   dma.send_count = 0;
-   dma.send_list = NULL;
-   dma.send_sizes = NULL;
-   dma.flags = 0;
-   dma.request_count = 1;
-   dma.request_size = RADEON_BUFFER_SIZE;
-   dma.request_list = &index;
-   dma.request_sizes = &size;
-   dma.granted_count = 0;
-
-   LOCK_HARDWARE(rmesa);	/* no need to validate */
-
-   while (1) {
-      ret = drmDMA( fd, &dma );
-      if (ret == 0)
-	 break;
-   
-      if (rmesa->dma.nr_released_bufs) {
-	 r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
-      }
-
-      if (rmesa->do_usleeps) {
-	 UNLOCK_HARDWARE( rmesa );
-	 DO_USLEEP( 1 );
-	 LOCK_HARDWARE( rmesa );
-      }
-   }
-
-   UNLOCK_HARDWARE(rmesa);
-
-   if (R200_DEBUG & DEBUG_DMA)
-      fprintf(stderr, "Allocated buffer %d\n", index);
-
-   dmabuf = CALLOC_STRUCT( r200_dma_buffer );
-   dmabuf->buf = &rmesa->r200Screen->buffers->list[index];
-   dmabuf->refcount = 1;
-
-   rmesa->dma.current.buf = dmabuf;
-   rmesa->dma.current.address = dmabuf->buf->address;
-   rmesa->dma.current.end = dmabuf->buf->total;
-   rmesa->dma.current.start = 0;
-   rmesa->dma.current.ptr = 0;
-}
-
-void r200ReleaseDmaRegion( r200ContextPtr rmesa,
-			     struct r200_dma_region *region,
-			     const char *caller )
-{
-   if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); 
-   
-   if (!region->buf)
-      return;
-
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
-
-   if (--region->buf->refcount == 0) {
-      drm_radeon_cmd_header_t *cmd;
-
-      if (R200_DEBUG & (DEBUG_IOCTL|DEBUG_DMA))
-	 fprintf(stderr, "%s -- DISCARD BUF %d\n", __FUNCTION__,
-		 region->buf->buf->idx);  
-      
-      cmd = (drm_radeon_cmd_header_t *)r200AllocCmdBuf( rmesa, sizeof(*cmd), 
-						     __FUNCTION__ );
-      cmd->dma.cmd_type = RADEON_CMD_DMA_DISCARD;
-      cmd->dma.buf_idx = region->buf->buf->idx;
-      FREE(region->buf);
-      rmesa->dma.nr_released_bufs++;
-   }
-
-   region->buf = NULL;
-   region->start = 0;
-}
-
-/* Allocates a region from rmesa->dma.current.  If there isn't enough
- * space in current, grab a new buffer (and discard what was left of current)
- */
-void r200AllocDmaRegion( r200ContextPtr rmesa, 
-			   struct r200_dma_region *region,
-			   int bytes,
-			   int alignment )
-{
-   if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
-
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
-
-   if (region->buf)
-      r200ReleaseDmaRegion( rmesa, region, __FUNCTION__ );
-
-   alignment--;
-   rmesa->dma.current.start = rmesa->dma.current.ptr = 
-      (rmesa->dma.current.ptr + alignment) & ~alignment;
-
-   if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) 
-      r200RefillCurrentDmaRegion( rmesa );
-
-   region->start = rmesa->dma.current.start;
-   region->ptr = rmesa->dma.current.start;
-   region->end = rmesa->dma.current.start + bytes;
-   region->address = rmesa->dma.current.address;
-   region->buf = rmesa->dma.current.buf;
-   region->buf->refcount++;
-
-   rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */
-   rmesa->dma.current.start = 
-      rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;  
-
-   assert( rmesa->dma.current.ptr <= rmesa->dma.current.end );
-}
-
-/* ================================================================
- * SwapBuffers with client-side throttling
- */
-
-static uint32_t r200GetLastFrame(r200ContextPtr rmesa)
-{
-   drm_radeon_getparam_t gp;
-   int ret;
-   uint32_t frame;
-
-   gp.param = RADEON_PARAM_LAST_FRAME;
-   gp.value = (int *)&frame;
-   ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_GETPARAM,
-			      &gp, sizeof(gp) );
-   if ( ret ) {
-      fprintf( stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, ret );
-      exit(1);
-   }
-
-   return frame;
-}
-
-static void r200EmitIrqLocked( r200ContextPtr rmesa )
-{
-   drm_radeon_irq_emit_t ie;
-   int ret;
-
-   ie.irq_seq = &rmesa->iw.irq_seq;
-   ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_IRQ_EMIT, 
-			      &ie, sizeof(ie) );
-   if ( ret ) {
-      fprintf( stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__, ret );
-      exit(1);
-   }
-}
-
-
-static void r200WaitIrq( r200ContextPtr rmesa )
-{
-   int ret;
-
-   do {
-      ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_IRQ_WAIT,
-			     &rmesa->iw, sizeof(rmesa->iw) );
-   } while (ret && (errno == EINTR || errno == EBUSY));
-
-   if ( ret ) {
-      fprintf( stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, ret );
-      exit(1);
-   }
-}
-
-
-static void r200WaitForFrameCompletion( r200ContextPtr rmesa )
-{
-   drm_radeon_sarea_t *sarea = rmesa->sarea;
-
-   if (rmesa->do_irqs) {
-      if (r200GetLastFrame(rmesa) < sarea->last_frame) {
-	 if (!rmesa->irqsEmitted) {
-	    while (r200GetLastFrame (rmesa) < sarea->last_frame)
-	       ;
-	 }
-	 else {
-	    UNLOCK_HARDWARE( rmesa ); 
-	    r200WaitIrq( rmesa );	
-	    LOCK_HARDWARE( rmesa ); 
-	 }
-	 rmesa->irqsEmitted = 10;
-      }
-
-      if (rmesa->irqsEmitted) {
-	 r200EmitIrqLocked( rmesa );
-	 rmesa->irqsEmitted--;
-      }
-   } 
-   else {
-      while (r200GetLastFrame (rmesa) < sarea->last_frame) {
-	 UNLOCK_HARDWARE( rmesa ); 
-	 if (rmesa->do_usleeps) 
-	    DO_USLEEP( 1 );
-	 LOCK_HARDWARE( rmesa ); 
-      }
-   }
-}
-
-
-
-/* Copy the back color buffer to the front color buffer.
- */
-void r200CopyBuffer( __DRIdrawablePrivate *dPriv,
-		      const drm_clip_rect_t	 *rect)
-{
-   r200ContextPtr rmesa;
-   GLint nbox, i, ret;
-   GLboolean   missed_target;
-   int64_t ust;
-   __DRIscreenPrivate *psp = dPriv->driScreenPriv;
-
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   rmesa = (r200ContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   if ( R200_DEBUG & DEBUG_IOCTL ) {
-      fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *)rmesa->glCtx );
-   }
-
-   R200_FIREVERTICES( rmesa );
-
-   LOCK_HARDWARE( rmesa );
-
-
-   /* Throttle the frame rate -- only allow one pending swap buffers
-    * request at a time.
-    */
-   r200WaitForFrameCompletion( rmesa );
-   if (!rect)
-   {
-       UNLOCK_HARDWARE( rmesa );
-       driWaitForVBlank( dPriv, & missed_target );
-       LOCK_HARDWARE( rmesa );
-   }
-
-   nbox = dPriv->numClipRects; /* must be in locked region */
-
-   for ( i = 0 ; i < nbox ; ) {
-      GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox );
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
-      GLint n = 0;
-
-      for ( ; i < nr ; i++ ) {
-
-	  *b = box[i];
-
-	  if (rect)
-	  {
-	     if (rect->x1 > b->x1)
-		 b->x1 = rect->x1;
-	     if (rect->y1 > b->y1)
-		 b->y1 = rect->y1;
-	     if (rect->x2 < b->x2)
-		 b->x2 = rect->x2;
-	     if (rect->y2 < b->y2)
-		 b->y2 = rect->y2;
-
-	     if (b->x1 >= b->x2 || b->y1 >= b->y2)
-		 continue;
-	  }
-
-	  b++;
-	  n++;
-      }
-      rmesa->sarea->nbox = n;
-
-      if (!n)
-	 continue;
-
-      ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP );
-
-      if ( ret ) {
-	 fprintf( stderr, "DRM_R200_SWAP_BUFFERS: return = %d\n", ret );
-	 UNLOCK_HARDWARE( rmesa );
-	 exit( 1 );
-      }
-   }
-
-   UNLOCK_HARDWARE( rmesa );
-   if (!rect)
-   {
-       rmesa->hw.all_dirty = GL_TRUE;
-
-       rmesa->swap_count++;
-       (*psp->systemTime->getUST)( & ust );
-       if ( missed_target ) {
-	   rmesa->swap_missed_count++;
-	   rmesa->swap_missed_ust = ust - rmesa->swap_ust;
-       }
-
-       rmesa->swap_ust = ust;
-
-       sched_yield();
-   }
-}
-
-void r200PageFlip( __DRIdrawablePrivate *dPriv )
+static void r200UserClear(GLcontext *ctx, GLuint mask)
 {
-   r200ContextPtr rmesa;
-   GLint ret;
-   GLboolean   missed_target;
-   __DRIscreenPrivate *psp = dPriv->driScreenPriv;
-
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   rmesa = (r200ContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   if ( R200_DEBUG & DEBUG_IOCTL ) {
-      fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__,
-	      rmesa->sarea->pfCurrentPage);
-   }
-
-   R200_FIREVERTICES( rmesa );
-   LOCK_HARDWARE( rmesa );
-
-   if (!dPriv->numClipRects) {
-      UNLOCK_HARDWARE( rmesa );
-      usleep( 10000 );		/* throttle invisible client 10ms */
-      return;
-   }
-
-   /* Need to do this for the perf box placement:
-    */
-   {
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
-      b[0] = box[0];
-      rmesa->sarea->nbox = 1;
-   }
-
-   /* Throttle the frame rate -- only allow a few pending swap buffers
-    * request at a time.
-    */
-   r200WaitForFrameCompletion( rmesa );
-   UNLOCK_HARDWARE( rmesa );
-   driWaitForVBlank( dPriv, & missed_target );
-   if ( missed_target ) {
-      rmesa->swap_missed_count++;
-      (void) (*psp->systemTime->getUST)( & rmesa->swap_missed_ust );
-   }
-   LOCK_HARDWARE( rmesa );
-
-   ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_FLIP );
-
-   UNLOCK_HARDWARE( rmesa );
-
-   if ( ret ) {
-      fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret );
-      exit( 1 );
-   }
-
-   rmesa->swap_count++;
-   (void) (*psp->systemTime->getUST)( & rmesa->swap_ust );
-
-#if 000
-   if ( rmesa->sarea->pfCurrentPage == 1 ) {
-	 rmesa->state.color.drawOffset = rmesa->r200Screen->frontOffset;
-	 rmesa->state.color.drawPitch  = rmesa->r200Screen->frontPitch;
-   } else {
-	 rmesa->state.color.drawOffset = rmesa->r200Screen->backOffset;
-	 rmesa->state.color.drawPitch  = rmesa->r200Screen->backPitch;
-   }
-
-   R200_STATECHANGE( rmesa, ctx );
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = rmesa->state.color.drawOffset
-					   + rmesa->r200Screen->fbLocation;
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH]  = rmesa->state.color.drawPitch;
-   if (rmesa->sarea->tiling_enabled) {
-      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= R200_COLOR_TILE_ENABLE;
-   }
-#else
-   /* Get ready for drawing next frame.  Update the renderbuffers'
-    * flippedOffset/Pitch fields so we draw into the right place.
-    */
-   driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer,
-                        rmesa->sarea->pfCurrentPage);
-
-
-   r200UpdateDrawBuffer(rmesa->glCtx);
-#endif
+   radeon_clear_tris(ctx, mask);
 }
 
-
-/* ================================================================
- * Buffer clear
- */
-static void r200Clear( GLcontext *ctx, GLbitfield mask )
+static void r200KernelClear(GLcontext *ctx, GLuint flags)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
-   GLuint flags = 0;
-   GLuint color_mask = 0;
-   GLint ret, i;
-   GLint cx, cy, cw, ch;
-
-   if ( R200_DEBUG & DEBUG_IOCTL ) {
-      fprintf( stderr, "r200Clear\n");
-   }
-
-   {
-      LOCK_HARDWARE( rmesa );
-      UNLOCK_HARDWARE( rmesa );
-      if ( dPriv->numClipRects == 0 ) 
-	 return;
-   }
-
-   r200Flush( ctx );
-
-   if ( mask & BUFFER_BIT_FRONT_LEFT ) {
-      flags |= RADEON_FRONT;
-      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
-      mask &= ~BUFFER_BIT_FRONT_LEFT;
-   }
-
-   if ( mask & BUFFER_BIT_BACK_LEFT ) {
-      flags |= RADEON_BACK;
-      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
-      mask &= ~BUFFER_BIT_BACK_LEFT;
-   }
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
+   GLint cx, cy, cw, ch, ret;
+   GLuint i;
 
-   if ( mask & BUFFER_BIT_DEPTH ) {
-      flags |= RADEON_DEPTH;
-      mask &= ~BUFFER_BIT_DEPTH;
-   }
-
-   if ( (mask & BUFFER_BIT_STENCIL) && rmesa->state.stencil.hwBuffer ) {
-      flags |= RADEON_STENCIL;
-      mask &= ~BUFFER_BIT_STENCIL;
-   }
-
-   if ( mask ) {
-      if (R200_DEBUG & DEBUG_FALLBACKS)
-	 fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask);
-      _swrast_Clear( ctx, mask );
-   }
-
-   if ( !flags ) 
-      return;
-
-   if (rmesa->using_hyperz) {
-      flags |= RADEON_USE_COMP_ZBUF;
-/*      if (rmesa->r200Screen->chip_family == CHIP_FAMILY_R200)
-	 flags |= RADEON_USE_HIERZ; */
-      if (!(rmesa->state.stencil.hwBuffer) ||
-	 ((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) &&
-	    ((rmesa->state.stencil.clear & R200_STENCIL_WRITE_MASK) == R200_STENCIL_WRITE_MASK))) {
-	  flags |= RADEON_CLEAR_FASTZ;
-      }
-   }
-
-   LOCK_HARDWARE( rmesa );
-
-   /* compute region after locking: */
-   cx = ctx->DrawBuffer->_Xmin;
-   cy = ctx->DrawBuffer->_Ymin;
-   cw = ctx->DrawBuffer->_Xmax - cx;
-   ch = ctx->DrawBuffer->_Ymax - cy;
-
-   /* Flip top to bottom */
-   cx += dPriv->x;
-   cy  = dPriv->y + dPriv->h - cy - ch;
+   LOCK_HARDWARE( &rmesa->radeon );
 
    /* Throttle the number of clear ioctls we do.
     */
@@ -693,7 +81,7 @@ static void r200Clear( GLcontext *ctx, GLbitfield mask )
 
       gp.param = RADEON_PARAM_LAST_CLEAR;
       gp.value = (int *)&clear;
-      ret = drmCommandWriteRead( rmesa->dri.fd,
+      ret = drmCommandWriteRead( rmesa->radeon.dri.fd,
 		      DRM_RADEON_GETPARAM, &gp, sizeof(gp) );
 
       if ( ret ) {
@@ -703,24 +91,34 @@ static void r200Clear( GLcontext *ctx, GLbitfield mask )
 
       /* Clear throttling needs more thought.
        */
-      if ( rmesa->sarea->last_clear - clear <= 25 ) {
+      if ( rmesa->radeon.sarea->last_clear - clear <= 25 ) {
 	 break;
       }
-      
-      if (rmesa->do_usleeps) {
-	 UNLOCK_HARDWARE( rmesa );
+
+      if (rmesa->radeon.do_usleeps) {
+	 UNLOCK_HARDWARE( &rmesa->radeon );
 	 DO_USLEEP( 1 );
-	 LOCK_HARDWARE( rmesa );
+	 LOCK_HARDWARE( &rmesa->radeon );
       }
    }
 
    /* Send current state to the hardware */
-   r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
+   rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
+
+
+  /* compute region after locking: */
+   cx = ctx->DrawBuffer->_Xmin;
+   cy = ctx->DrawBuffer->_Ymin;
+   cw = ctx->DrawBuffer->_Xmax - cx;
+   ch = ctx->DrawBuffer->_Ymax - cy;
 
+   /* Flip top to bottom */
+   cx += dPriv->x;
+   cy  = dPriv->y + dPriv->h - cy - ch;
    for ( i = 0 ; i < dPriv->numClipRects ; ) {
       GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, dPriv->numClipRects );
       drm_clip_rect_t *box = dPriv->pClipRects;
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
+      drm_clip_rect_t *b = rmesa->radeon.sarea->boxes;
       drm_radeon_clear_t clear;
       drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
       GLint n = 0;
@@ -755,17 +153,17 @@ static void r200Clear( GLcontext *ctx, GLbitfield mask )
 	 }
       }
 
-      rmesa->sarea->nbox = n;
+      rmesa->radeon.sarea->nbox = n;
 
       clear.flags       = flags;
-      clear.clear_color = rmesa->state.color.clear;
-      clear.clear_depth = rmesa->state.depth.clear;	/* needed for hyperz */
+      clear.clear_color = rmesa->radeon.state.color.clear;
+      clear.clear_depth = rmesa->radeon.state.depth.clear;	/* needed for hyperz */
       clear.color_mask  = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
-      clear.depth_mask  = rmesa->state.stencil.clear;
+      clear.depth_mask  = rmesa->radeon.state.stencil.clear;
       clear.depth_boxes = depth_boxes;
 
       n--;
-      b = rmesa->sarea->boxes;
+      b = rmesa->radeon.sarea->boxes;
       for ( ; n >= 0 ; n-- ) {
 	 depth_boxes[n].f[CLEAR_X1] = (float)b[n].x1;
 	 depth_boxes[n].f[CLEAR_Y1] = (float)b[n].y1;
@@ -774,84 +172,91 @@ static void r200Clear( GLcontext *ctx, GLbitfield mask )
 	 depth_boxes[n].f[CLEAR_DEPTH] = ctx->Depth.Clear;
       }
 
-      ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_CLEAR,
+      ret = drmCommandWrite( rmesa->radeon.dri.fd, DRM_RADEON_CLEAR,
 			     &clear, sizeof(clear));
 
 
       if ( ret ) {
-	 UNLOCK_HARDWARE( rmesa );
+	 UNLOCK_HARDWARE( &rmesa->radeon );
 	 fprintf( stderr, "DRM_RADEON_CLEAR: return = %d\n", ret );
 	 exit( 1 );
       }
    }
-
-   UNLOCK_HARDWARE( rmesa );
-   rmesa->hw.all_dirty = GL_TRUE;
+   UNLOCK_HARDWARE( &rmesa->radeon );
 }
+/* ================================================================
+ * Buffer clear
+ */
+static void r200Clear( GLcontext *ctx, GLbitfield mask )
+{
+   r200ContextPtr rmesa = R200_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
+   GLuint flags = 0;
+   GLuint color_mask = 0;
+   GLuint orig_mask = mask;
 
+   if ( R200_DEBUG & DEBUG_IOCTL ) {
+       fprintf( stderr, "r200Clear %x %d\n", mask, rmesa->radeon.sarea->pfCurrentPage);
+   }
 
-void r200WaitForIdleLocked( r200ContextPtr rmesa )
-{
-    int ret;
-    int i = 0;
-    
-    do {
-       ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_CP_IDLE);
-       if (ret) 
-	  DO_USLEEP( 1 );
-    } while (ret && ++i < 100);
-    
-    if ( ret < 0 ) {
-       UNLOCK_HARDWARE( rmesa );
-       fprintf( stderr, "Error: R200 timed out... exiting\n" );
-       exit( -1 );
-    }
-}
+   {
+      LOCK_HARDWARE( &rmesa->radeon );
+      UNLOCK_HARDWARE( &rmesa->radeon );
+      if ( dPriv->numClipRects == 0 )
+	 return;
+   }
 
+   radeonFlush( ctx );
 
-static void r200WaitForIdle( r200ContextPtr rmesa )
-{
-   LOCK_HARDWARE(rmesa);
-   r200WaitForIdleLocked( rmesa );
-   UNLOCK_HARDWARE(rmesa);
-}
+   if ( mask & BUFFER_BIT_FRONT_LEFT ) {
+      flags |= RADEON_FRONT;
+      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      mask &= ~BUFFER_BIT_FRONT_LEFT;
+   }
 
+   if ( mask & BUFFER_BIT_BACK_LEFT ) {
+      flags |= RADEON_BACK;
+      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      mask &= ~BUFFER_BIT_BACK_LEFT;
+   }
 
-void r200Flush( GLcontext *ctx )
-{
-   r200ContextPtr rmesa = R200_CONTEXT( ctx );
+   if ( mask & BUFFER_BIT_DEPTH ) {
+      flags |= RADEON_DEPTH;
+      mask &= ~BUFFER_BIT_DEPTH;
+   }
 
-   if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
+   if ( (mask & BUFFER_BIT_STENCIL) ) {
+      flags |= RADEON_STENCIL;
+      mask &= ~BUFFER_BIT_STENCIL;
+   }
 
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
+   if ( mask ) {
+      if (R200_DEBUG & DEBUG_FALLBACKS)
+	 fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask);
+      _swrast_Clear( ctx, mask );
+   }
 
-   r200EmitState( rmesa );
-   
-   if (rmesa->store.cmd_used)
-      r200FlushCmdBuf( rmesa, __FUNCTION__ );
-}
+   if ( !flags )
+      return;
 
-/* Make sure all commands have been sent to the hardware and have
- * completed processing.
- */
-void r200Finish( GLcontext *ctx )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   r200Flush( ctx );
+   if (rmesa->using_hyperz) {
+      flags |= RADEON_USE_COMP_ZBUF;
+/*      if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200)
+	 flags |= RADEON_USE_HIERZ; */
+      if (!((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) &&
+	    ((rmesa->radeon.state.stencil.clear & R200_STENCIL_WRITE_MASK) == R200_STENCIL_WRITE_MASK))) {
+	  flags |= RADEON_CLEAR_FASTZ;
+      }
+   }
 
-   if (rmesa->do_irqs) {
-      LOCK_HARDWARE( rmesa );
-      r200EmitIrqLocked( rmesa );
-      UNLOCK_HARDWARE( rmesa );
-      r200WaitIrq( rmesa );
+   if (rmesa->radeon.radeonScreen->kernel_mm)
+      r200UserClear(ctx, orig_mask);
+   else {
+      r200KernelClear(ctx, flags);
+      rmesa->radeon.hw.all_dirty = GL_TRUE;
    }
-   else 
-      r200WaitForIdle( rmesa );
 }
 
-
 /* This version of AllocateMemoryMESA allocates only GART memory, and
  * only does so after the point at which the driver has been
  * initialized.
@@ -862,7 +267,7 @@ void r200Finish( GLcontext *ctx )
  * device fd.
  */
 void *r200AllocateMemoryMESA(__DRIscreen *screen, GLsizei size,
-			     GLfloat readfreq, GLfloat writefreq, 
+			     GLfloat readfreq, GLfloat writefreq,
 			     GLfloat priority)
 {
    GET_CURRENT_CONTEXT(ctx);
@@ -872,10 +277,10 @@ void *r200AllocateMemoryMESA(__DRIscreen *screen, GLsizei size,
    int ret;
 
    if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s sz %d %f/%f/%f\n", __FUNCTION__, size, readfreq, 
+      fprintf(stderr, "%s sz %d %f/%f/%f\n", __FUNCTION__, size, readfreq,
 	      writefreq, priority);
 
-   if (!ctx || !(rmesa = R200_CONTEXT(ctx)) || !rmesa->r200Screen->gartTextures.map)
+   if (!ctx || !(rmesa = R200_CONTEXT(ctx)) || !rmesa->radeon.radeonScreen->gartTextures.map)
       return NULL;
 
    if (getenv("R200_NO_ALLOC"))
@@ -886,17 +291,17 @@ void *r200AllocateMemoryMESA(__DRIscreen *screen, GLsizei size,
    alloc.size = size;
    alloc.region_offset = &region_offset;
 
-   ret = drmCommandWriteRead( rmesa->r200Screen->driScreen->fd,
+   ret = drmCommandWriteRead( rmesa->radeon.radeonScreen->driScreen->fd,
 			      DRM_RADEON_ALLOC,
 			      &alloc, sizeof(alloc));
-   
+
    if (ret) {
       fprintf(stderr, "%s: DRM_RADEON_ALLOC ret %d\n", __FUNCTION__, ret);
       return NULL;
    }
-   
+
    {
-      char *region_start = (char *)rmesa->r200Screen->gartTextures.map;
+      char *region_start = (char *)rmesa->radeon.radeonScreen->gartTextures.map;
       return (void *)(region_start + region_offset);
    }
 }
@@ -914,28 +319,28 @@ void r200FreeMemoryMESA(__DRIscreen *screen, GLvoid *pointer)
    if (R200_DEBUG & DEBUG_IOCTL)
       fprintf(stderr, "%s %p\n", __FUNCTION__, pointer);
 
-   if (!ctx || !(rmesa = R200_CONTEXT(ctx)) || !rmesa->r200Screen->gartTextures.map) {
+   if (!ctx || !(rmesa = R200_CONTEXT(ctx)) || !rmesa->radeon.radeonScreen->gartTextures.map) {
       fprintf(stderr, "%s: no context\n", __FUNCTION__);
       return;
    }
 
-   region_offset = (char *)pointer - (char *)rmesa->r200Screen->gartTextures.map;
+   region_offset = (char *)pointer - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
 
-   if (region_offset < 0 || 
-       region_offset > rmesa->r200Screen->gartTextures.size) {
+   if (region_offset < 0 ||
+       region_offset > rmesa->radeon.radeonScreen->gartTextures.size) {
       fprintf(stderr, "offset %d outside range 0..%d\n", region_offset,
-	      rmesa->r200Screen->gartTextures.size);
+	      rmesa->radeon.radeonScreen->gartTextures.size);
       return;
    }
 
    memfree.region = RADEON_MEM_REGION_GART;
    memfree.region_offset = region_offset;
-   
-   ret = drmCommandWrite( rmesa->r200Screen->driScreen->fd,
+
+   ret = drmCommandWrite( rmesa->radeon.radeonScreen->driScreen->fd,
 			  DRM_RADEON_FREE,
 			  &memfree, sizeof(memfree));
-   
-   if (ret) 
+
+   if (ret)
       fprintf(stderr, "%s: DRM_RADEON_FREE ret %d\n", __FUNCTION__, ret);
 }
 
@@ -956,32 +361,32 @@ GLuint r200GetMemoryOffsetMESA(__DRIscreen *screen, const GLvoid *pointer)
 
    card_offset = r200GartOffsetFromVirtual( rmesa, pointer );
 
-   return card_offset - rmesa->r200Screen->gart_base;
+   return card_offset - rmesa->radeon.radeonScreen->gart_base;
 }
 
 GLboolean r200IsGartMemory( r200ContextPtr rmesa, const GLvoid *pointer,
 			   GLint size )
 {
-   ptrdiff_t offset = (char *)pointer - (char *)rmesa->r200Screen->gartTextures.map;
+   ptrdiff_t offset = (char *)pointer - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
    int valid = (size >= 0 &&
 		offset >= 0 &&
-		offset + size < rmesa->r200Screen->gartTextures.size);
+		offset + size < rmesa->radeon.radeonScreen->gartTextures.size);
 
    if (R200_DEBUG & DEBUG_IOCTL)
       fprintf(stderr, "r200IsGartMemory( %p ) : %d\n", pointer, valid );
-   
+
    return valid;
 }
 
 
 GLuint r200GartOffsetFromVirtual( r200ContextPtr rmesa, const GLvoid *pointer )
 {
-   ptrdiff_t offset = (char *)pointer - (char *)rmesa->r200Screen->gartTextures.map;
+   ptrdiff_t offset = (char *)pointer - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
 
-   if (offset < 0 || offset > rmesa->r200Screen->gartTextures.size)
+   if (offset < 0 || offset > rmesa->radeon.radeonScreen->gartTextures.size)
       return ~0;
    else
-      return rmesa->r200Screen->gart_texture_offset + offset;
+      return rmesa->radeon.radeonScreen->gart_texture_offset + offset;
 }
 
 
@@ -989,7 +394,7 @@ GLuint r200GartOffsetFromVirtual( r200ContextPtr rmesa, const GLvoid *pointer )
 void r200InitIoctlFuncs( struct dd_function_table *functions )
 {
     functions->Clear = r200Clear;
-    functions->Finish = r200Finish;
-    functions->Flush = r200Flush;
+    functions->Finish = radeonFinish;
+    functions->Flush = radeonFlush;
 }
 
diff --git a/src/mesa/drivers/dri/r200/r200_ioctl.h b/src/mesa/drivers/dri/r200/r200_ioctl.h
index f7458e4a0ed..2a4b8a11f4c 100644
--- a/src/mesa/drivers/dri/r200/r200_ioctl.h
+++ b/src/mesa/drivers/dri/r200/r200_ioctl.h
@@ -37,65 +37,30 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "main/simple_list.h"
 #include "radeon_dri.h"
-#include "r200_lock.h"
+
+#include "radeon_bocs_wrapper.h"
 
 #include "xf86drm.h"
 #include "drm.h"
 #include "radeon_drm.h"
 
-extern void r200EmitState( r200ContextPtr rmesa );
 extern void r200EmitVertexAOS( r200ContextPtr rmesa,
-				 GLuint vertex_size,
-				 GLuint offset );
+			       GLuint vertex_size,
+			       struct radeon_bo *bo,
+			       GLuint offset );
 
 extern void r200EmitVbufPrim( r200ContextPtr rmesa,
 				GLuint primitive,
 				GLuint vertex_nr );
 
-extern void r200FlushElts( r200ContextPtr rmesa );
+extern void r200FlushElts(GLcontext *ctx);
 
 extern GLushort *r200AllocEltsOpenEnded( r200ContextPtr rmesa,
 					   GLuint primitive,
 					   GLuint min_nr );
 
-extern void r200EmitAOS( r200ContextPtr rmesa,
-			   struct r200_dma_region **regions,
-			   GLuint n,
-			   GLuint offset );
-
-extern void r200EmitBlit( r200ContextPtr rmesa,
-			  GLuint color_fmt,
-			  GLuint src_pitch,
-			  GLuint src_offset,
-			  GLuint dst_pitch,
-			  GLuint dst_offset,
-			  GLint srcx, GLint srcy,
-			  GLint dstx, GLint dsty,
-			  GLuint w, GLuint h );
-
-extern void r200EmitWait( r200ContextPtr rmesa, GLuint flags );
-
-extern void r200FlushCmdBuf( r200ContextPtr rmesa, const char * );
-extern int r200FlushCmdBufLocked( r200ContextPtr rmesa, const char * caller );
-
-extern void r200RefillCurrentDmaRegion( r200ContextPtr rmesa );
-
-extern void r200AllocDmaRegion( r200ContextPtr rmesa,
-				  struct r200_dma_region *region,
-				  int bytes, 
-				  int alignment );
-
-extern void r200ReleaseDmaRegion( r200ContextPtr rmesa,
-				    struct r200_dma_region *region,
-				    const char *caller );
-
-extern void r200CopyBuffer( __DRIdrawablePrivate *drawable,
-			    const drm_clip_rect_t      *rect);
-extern void r200PageFlip( __DRIdrawablePrivate *drawable );
-extern void r200Flush( GLcontext *ctx );
-extern void r200Finish( GLcontext *ctx );
-extern void r200WaitForIdleLocked( r200ContextPtr rmesa );
-extern void r200WaitForVBlank( r200ContextPtr rmesa );
+extern void r200EmitAOS(r200ContextPtr rmesa, GLuint nr, GLuint offset);
+
 extern void r200InitIoctlFuncs( struct dd_function_table *functions );
 
 extern void *r200AllocateMemoryMESA( __DRIscreen *screen, GLsizei size, GLfloat readfreq,
@@ -119,8 +84,8 @@ void r200SetUpAtomList( r200ContextPtr rmesa );
  */
 #define R200_NEWPRIM( rmesa )			\
 do {						\
-   if ( rmesa->dma.flush )			\
-      rmesa->dma.flush( rmesa );	\
+   if ( rmesa->radeon.dma.flush )			\
+      rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	\
 } while (0)
 
 /* Can accomodate several state changes and primitive changes without
@@ -130,7 +95,7 @@ do {						\
 do {								\
    R200_NEWPRIM( rmesa );					\
    rmesa->hw.ATOM.dirty = GL_TRUE;				\
-   rmesa->hw.is_dirty = GL_TRUE;				\
+   rmesa->radeon.hw.is_dirty = GL_TRUE;				\
 } while (0)
 
 #define R200_DB_STATE( ATOM )			        \
@@ -139,13 +104,13 @@ do {								\
 
 static INLINE int R200_DB_STATECHANGE( 
    r200ContextPtr rmesa,
-   struct r200_state_atom *atom )
+   struct radeon_state_atom *atom )
 {
    if (memcmp(atom->cmd, atom->lastcmd, atom->cmd_size*4)) {
-      int *tmp;
+      GLuint *tmp;
       R200_NEWPRIM( rmesa );
       atom->dirty = GL_TRUE;
-      rmesa->hw.is_dirty = GL_TRUE;
+      rmesa->radeon.hw.is_dirty = GL_TRUE;
       tmp = atom->cmd; 
       atom->cmd = atom->lastcmd;
       atom->lastcmd = tmp;
@@ -156,15 +121,6 @@ static INLINE int R200_DB_STATECHANGE(
 }
 
 
-/* Fire the buffered vertices no matter what.
- */
-#define R200_FIREVERTICES( rmesa )			\
-do {							\
-   if ( rmesa->store.cmd_used || rmesa->dma.flush ) {	\
-      r200Flush( rmesa->glCtx );			\
-   }							\
-} while (0)
-
 /* Command lengths.  Note that any time you ensure ELTS_BUFSZ or VBUF_BUFSZ
  * are available, you will also be adding an rmesa->state.max_state_size because
  * r200EmitState is called from within r200EmitVbufPrim and r200FlushElts.
@@ -174,36 +130,36 @@ do {							\
 #define ELTS_BUFSZ(nr)	(12 + nr * 2)
 #define VBUF_BUFSZ	(3 * sizeof(int))
 
-/* Ensure that a minimum amount of space is available in the command buffer.
- * This is used to ensure atomicity of state updates with the rendering requests
- * that rely on them.
- *
- * An alternative would be to implement a "soft lock" such that when the buffer
- * wraps at an inopportune time, we grab the lock, flush the current buffer,
- * and hang on to the lock until the critical section is finished and we flush
- * the buffer again and unlock.
- */
-static INLINE void r200EnsureCmdBufSpace( r200ContextPtr rmesa, int bytes )
+static inline uint32_t cmdpacket3(int cmd_type)
 {
-   if (rmesa->store.cmd_used + bytes > R200_CMD_BUF_SZ)
-      r200FlushCmdBuf( rmesa, __FUNCTION__ );
-   assert( bytes <= R200_CMD_BUF_SZ );
-}
+  drm_radeon_cmd_header_t cmd;
 
-/* Alloc space in the command buffer
- */
-static INLINE char *r200AllocCmdBuf( r200ContextPtr rmesa,
-					 int bytes, const char *where )
-{
-   char * head;
+  cmd.i = 0;
+  cmd.header.cmd_type = cmd_type;
 
-   if (rmesa->store.cmd_used + bytes > R200_CMD_BUF_SZ)
-      r200FlushCmdBuf( rmesa, where );
+  return (uint32_t)cmd.i;
 
-   head = rmesa->store.cmd_buf + rmesa->store.cmd_used;
-   rmesa->store.cmd_used += bytes;
-   assert( rmesa->store.cmd_used <= R200_CMD_BUF_SZ );
-   return head;
 }
 
+#define OUT_BATCH_PACKET3(packet, num_extra) do {	      \
+    if (!b_l_rmesa->radeonScreen->kernel_mm) {		      \
+      OUT_BATCH(cmdpacket3(RADEON_CMD_PACKET3));				      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    } else {						      \
+      OUT_BATCH(CP_PACKET2);				      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    }							      \
+  } while(0)
+
+#define OUT_BATCH_PACKET3_CLIP(packet, num_extra) do {	      \
+    if (!b_l_rmesa->radeonScreen->kernel_mm) {		      \
+      OUT_BATCH(cmdpacket3(RADEON_CMD_PACKET3_CLIP));	      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    } else {						      \
+      OUT_BATCH(CP_PACKET2);				      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    }							      \
+  } while(0)
+
+
 #endif /* __R200_IOCTL_H__ */
diff --git a/src/mesa/drivers/dri/r200/r200_lock.c b/src/mesa/drivers/dri/r200/r200_lock.c
deleted file mode 100644
index 99661a4bfb4..00000000000
--- a/src/mesa/drivers/dri/r200/r200_lock.c
+++ /dev/null
@@ -1,116 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
- 
-#include "r200_context.h"
-#include "r200_lock.h"
-#include "r200_tex.h"
-#include "r200_state.h"
-#include "r200_ioctl.h"
-
-#include "drirenderbuffer.h"
-
-
-#if DEBUG_LOCKING
-char *prevLockFile = NULL;
-int prevLockLine = 0;
-#endif
-
-/* Turn on/off page flipping according to the flags in the sarea:
- */
-static void
-r200UpdatePageFlipping( r200ContextPtr rmesa )
-{
-   rmesa->doPageFlip = rmesa->sarea->pfState;
-   if (rmesa->glCtx->WinSysDrawBuffer) {
-      driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer,
-                           rmesa->sarea->pfCurrentPage);
-   }
-}
-
-
-
-/* Update the hardware state.  This is called if another main/context.has
- * grabbed the hardware lock, which includes the X server.  This
- * function also updates the driver's window state after the X server
- * moves, resizes or restacks a window -- the change will be reflected
- * in the drawable position and clip rects.  Since the X server grabs
- * the hardware lock when it changes the window state, this routine will
- * automatically be called after such a change.
- */
-void r200GetLock( r200ContextPtr rmesa, GLuint flags )
-{
-   __DRIdrawablePrivate *drawable = rmesa->dri.drawable;
-   __DRIdrawablePrivate *readable = rmesa->dri.readable;
-   __DRIscreenPrivate *sPriv = rmesa->dri.screen;
-   drm_radeon_sarea_t *sarea = rmesa->sarea;
-   int i;
-
-   drmGetLock( rmesa->dri.fd, rmesa->dri.hwContext, flags );
-
-   /* The window might have moved, so we might need to get new clip
-    * rects.
-    *
-    * NOTE: This releases and regrabs the hw lock to allow the X server
-    * to respond to the DRI protocol request for new drawable info.
-    * Since the hardware state depends on having the latest drawable
-    * clip rects, all state checking must be done _after_ this call.
-    */
-   DRI_VALIDATE_DRAWABLE_INFO( sPriv, drawable );
-   if (drawable != readable) {
-      DRI_VALIDATE_DRAWABLE_INFO( sPriv, readable );
-   }
-
-   if ( rmesa->lastStamp != drawable->lastStamp ) {
-      r200UpdatePageFlipping( rmesa );
-      r200SetCliprects( rmesa );
-      r200UpdateViewportOffset( rmesa->glCtx );
-      driUpdateFramebufferSize(rmesa->glCtx, drawable);
-   }
-
-   R200_STATECHANGE( rmesa, ctx );
-   if (rmesa->sarea->tiling_enabled) {
-      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= R200_COLOR_TILE_ENABLE;
-   }
-   else rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] &= ~R200_COLOR_TILE_ENABLE;
-
-   if ( sarea->ctx_owner != rmesa->dri.hwContext ) {
-      sarea->ctx_owner = rmesa->dri.hwContext;
-   }
-
-   for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
-      DRI_AGE_TEXTURES( rmesa->texture_heaps[ i ] );
-   }
-
-   rmesa->lost_context = GL_TRUE;
-}
diff --git a/src/mesa/drivers/dri/r200/r200_lock.h b/src/mesa/drivers/dri/r200/r200_lock.h
deleted file mode 100644
index 4ff98907fbf..00000000000
--- a/src/mesa/drivers/dri/r200/r200_lock.h
+++ /dev/null
@@ -1,106 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
-
-#ifndef __R200_LOCK_H__
-#define __R200_LOCK_H__
-
-extern void r200GetLock( r200ContextPtr rmesa, GLuint flags );
-
-/* Turn DEBUG_LOCKING on to find locking conflicts.
- */
-#define DEBUG_LOCKING	0
-
-#if DEBUG_LOCKING
-extern char *prevLockFile;
-extern int prevLockLine;
-
-#define DEBUG_LOCK()							\
-   do {									\
-      prevLockFile = (__FILE__);					\
-      prevLockLine = (__LINE__);					\
-   } while (0)
-
-#define DEBUG_RESET()							\
-   do {									\
-      prevLockFile = 0;							\
-      prevLockLine = 0;							\
-   } while (0)
-
-#define DEBUG_CHECK_LOCK()						\
-   do {									\
-      if ( prevLockFile ) {						\
-	 fprintf( stderr,						\
-		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
-		  prevLockFile, prevLockLine, __FILE__, __LINE__ );	\
-	 exit( 1 );							\
-      }									\
-   } while (0)
-
-#else
-
-#define DEBUG_LOCK()
-#define DEBUG_RESET()
-#define DEBUG_CHECK_LOCK()
-
-#endif
-
-/*
- * !!! We may want to separate locks from locks with validation.  This
- * could be used to improve performance for those things commands that
- * do not do any drawing !!!
- */
-
-
-/* Lock the hardware and validate our state.
- */
-#define LOCK_HARDWARE( rmesa )					\
-   do {								\
-      char __ret = 0;						\
-      DEBUG_CHECK_LOCK();					\
-      DRM_CAS( rmesa->dri.hwLock, rmesa->dri.hwContext,		\
-	       (DRM_LOCK_HELD | rmesa->dri.hwContext), __ret );	\
-      if ( __ret )						\
-	 r200GetLock( rmesa, 0 );				\
-      DEBUG_LOCK();						\
-   } while (0)
-
-#define UNLOCK_HARDWARE( rmesa )					\
-   do {									\
-      DRM_UNLOCK( rmesa->dri.fd,					\
-		  rmesa->dri.hwLock,					\
-		  rmesa->dri.hwContext );				\
-      DEBUG_RESET();							\
-   } while (0)
-
-#endif /* __R200_LOCK_H__ */
diff --git a/src/mesa/drivers/dri/r200/r200_maos.h b/src/mesa/drivers/dri/r200/r200_maos.h
index d3ed06d4021..16a70475e18 100644
--- a/src/mesa/drivers/dri/r200/r200_maos.h
+++ b/src/mesa/drivers/dri/r200/r200_maos.h
@@ -38,6 +38,5 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_context.h"
 
 extern void r200EmitArrays( GLcontext *ctx, GLubyte *vimap_rev );
-extern void r200ReleaseArrays( GLcontext *ctx, GLuint newinputs );
 
 #endif
diff --git a/src/mesa/drivers/dri/r200/r200_maos_arrays.c b/src/mesa/drivers/dri/r200/r200_maos_arrays.c
index 8512b9af478..383a0c4b0d3 100644
--- a/src/mesa/drivers/dri/r200/r200_maos_arrays.c
+++ b/src/mesa/drivers/dri/r200/r200_maos_arrays.c
@@ -50,110 +50,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_maos.h"
 #include "r200_tcl.h"
 
-
-#if 0
-/* Usage:
- *   - from r200_tcl_render
- *   - call r200EmitArrays to ensure uptodate arrays in dma
- *   - emit primitives (new type?) which reference the data
- *       -- need to use elts for lineloop, quads, quadstrip/flat
- *       -- other primitives are all well-formed (need tristrip-1,fake-poly)
- *
- */
-static void emit_ubyte_rgba3( GLcontext *ctx,
-		       struct r200_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   r200_color_t *out = (r200_color_t *)(rvb->start + rvb->address);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d out %p\n",
-	      __FUNCTION__, count, stride, (void *)out);
-
-   for (i = 0; i < count; i++) {
-      out->red   = *data;
-      out->green = *(data+1);
-      out->blue  = *(data+2);
-      out->alpha = 0xFF;
-      out++;
-      data += stride;
-   }
-}
-
-static void emit_ubyte_rgba4( GLcontext *ctx,
-			      struct r200_dma_region *rvb,
-			      char *data,
-			      int stride,
-			      int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 4) {
-      for (i = 0; i < count; i++)
-	 ((int *)out)[i] = LE32_TO_CPU(((int *)data)[i]);
-   } else {
-      for (i = 0; i < count; i++) {
-	 *(int *)out++ = LE32_TO_CPU(*(int *)data);
-	 data += stride;
-      }
-   }
-}
-
-
-static void emit_ubyte_rgba( GLcontext *ctx,
-			     struct r200_dma_region *rvb,
-			     char *data,
-			     int size,
-			     int stride,
-			     int count )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size);
-
-   assert (!rvb->buf);
-
-   if (stride == 0) {
-      r200AllocDmaRegion( rmesa, rvb, 4, 4 );
-      count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = 1;
-   }
-   else {
-      r200AllocDmaRegion( rmesa, rvb, 4 * count, 4 );	/* alignment? */
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 1;
-      rvb->aos_size = 1;
-   }
-
-   /* Emit the data
-    */
-   switch (size) {
-   case 3:
-      emit_ubyte_rgba3( ctx, rvb, data, stride, count );
-      break;
-   case 4:
-      emit_ubyte_rgba4( ctx, rvb, data, stride, count );
-      break;
-   default:
-      assert(0);
-      exit(1);
-      break;
-   }
-}
-#endif
-
-
 #if defined(USE_X86_ASM)
 #define COPY_DWORDS( dst, src, nr )					\
 do {									\
@@ -174,204 +70,34 @@ do {						\
 } while (0)
 #endif
 
-
-static void emit_vecfog( GLcontext *ctx,
-			 struct r200_dma_region *rvb,
-			 char *data,
-			 int stride,
-			 int count )
+static void r200_emit_vecfog(GLcontext *ctx, struct radeon_aos *aos,
+			     GLvoid *data, int stride, int count)
 {
-   int i;
-   GLfloat *out;
-
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   assert (!rvb->buf);
-
-   if (stride == 0) {
-      r200AllocDmaRegion( rmesa, rvb, 4, 4 );
-      count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = 1;
-   }
-   else {
-      r200AllocDmaRegion( rmesa, rvb, count * 4, 4 );	/* alignment? */
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 1;
-      rvb->aos_size = 1;
-   }
-
-   /* Emit the data
-    */
-   out = (GLfloat *)(rvb->address + rvb->start);
-   for (i = 0; i < count; i++) {
-      out[0] = r200ComputeFogBlendFactor( ctx, *(GLfloat *)data );
-      out++;
-      data += stride;
-   }
-
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	uint32_t *out;
+	int i;
+	int size = 1;
+
+	if (stride == 0) {
+		radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32);
+		count = 1;
+		aos->stride = 0;
+	} else {
+		radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32);
+		aos->stride = size;
+	}
+
+	aos->components = size;
+	aos->count = count;
+
+	out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
+	for (i = 0; i < count; i++) {
+	  out[0] = r200ComputeFogBlendFactor( ctx, *(GLfloat *)data );
+	  out++;
+	  data += stride;
+	}
 }
 
-
-static void emit_vec4( GLcontext *ctx,
-		       struct r200_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 4)
-      COPY_DWORDS( out, data, count );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out++;
-	 data += stride;
-      }
-}
-
-
-static void emit_vec8( GLcontext *ctx,
-		       struct r200_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 8)
-      COPY_DWORDS( out, data, count*2 );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out[1] = *(int *)(data+4);
-	 out += 2;
-	 data += stride;
-      }
-}
-
-static void emit_vec12( GLcontext *ctx,
-		       struct r200_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d out %p data %p\n",
-	      __FUNCTION__, count, stride, (void *)out, (void *)data);
-
-   if (stride == 12)
-      COPY_DWORDS( out, data, count*3 );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out[1] = *(int *)(data+4);
-	 out[2] = *(int *)(data+8);
-	 out += 3;
-	 data += stride;
-      }
-}
-
-static void emit_vec16( GLcontext *ctx,
-			struct r200_dma_region *rvb,
-			char *data,
-			int stride,
-			int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 16)
-      COPY_DWORDS( out, data, count*4 );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out[1] = *(int *)(data+4);
-	 out[2] = *(int *)(data+8);
-	 out[3] = *(int *)(data+12);
-	 out += 4;
-	 data += stride;
-      }
-}
-
-
-static void emit_vector( GLcontext *ctx,
-			 struct r200_dma_region *rvb,
-			 char *data,
-			 int size,
-			 int stride,
-			 int count )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if (R200_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d size %d stride %d\n",
-	      __FUNCTION__, count, size, stride);
-
-   assert (!rvb->buf);
-
-   if (stride == 0) {
-      r200AllocDmaRegion( rmesa, rvb, size * 4, 4 );
-      count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = size;
-   }
-   else {
-      r200AllocDmaRegion( rmesa, rvb, size * count * 4, 4 );	/* alignment? */
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = size;
-      rvb->aos_size = size;
-   }
-
-   /* Emit the data
-    */
-   switch (size) {
-   case 1:
-      emit_vec4( ctx, rvb, data, stride, count );
-      break;
-   case 2:
-      emit_vec8( ctx, rvb, data, stride, count );
-      break;
-   case 3:
-      emit_vec12( ctx, rvb, data, stride, count );
-      break;
-   case 4:
-      emit_vec16( ctx, rvb, data, stride, count );
-      break;
-   default:
-      assert(0);
-      exit(1);
-      break;
-   }
-
-}
-
-
-
 /* Emit any changed arrays to new GART memory, re-emit a packet to
  * update the arrays.  
  */
@@ -379,12 +105,12 @@ void r200EmitArrays( GLcontext *ctx, GLubyte *vimap_rev )
 {
    r200ContextPtr rmesa = R200_CONTEXT( ctx );
    struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb;
-   struct r200_dma_region **component = rmesa->tcl.aos_components;
    GLuint nr = 0;
    GLuint vfmt0 = 0, vfmt1 = 0;
    GLuint count = VB->Count;
    GLuint i, emitsize;
 
+   //   fprintf(stderr,"emit arrays\n");
    for ( i = 0; i < 15; i++ ) {
       GLubyte attrib = vimap_rev[i];
       if (attrib != 255) {
@@ -416,20 +142,20 @@ void r200EmitArrays( GLcontext *ctx, GLubyte *vimap_rev )
 	 case 3:
 	    /* special handling to fix up fog. Will get us into trouble with vbos...*/
 	    assert(attrib == VERT_ATTRIB_FOG);
-	    if (!rmesa->tcl.vertex_data[i].buf) {
+	    if (!rmesa->radeon.tcl.aos[i].bo) {
 	       if (ctx->VertexProgram._Enabled)
-		  emit_vector( ctx,
-			 &(rmesa->tcl.vertex_data[i]),
-			 (char *)VB->AttribPtr[attrib]->data,
-			 1,
-			 VB->AttribPtr[attrib]->stride,
-			 count);
+		  rcommon_emit_vector( ctx,
+				       &(rmesa->radeon.tcl.aos[nr]),
+				       (char *)VB->AttribPtr[attrib]->data,
+				       1,
+				       VB->AttribPtr[attrib]->stride,
+				       count);
 	       else
-		  emit_vecfog( ctx,
-			 &(rmesa->tcl.vertex_data[i]),
-			 (char *)VB->AttribPtr[attrib]->data,
-			 VB->AttribPtr[attrib]->stride,
-			 count);
+		 r200_emit_vecfog( ctx,
+				   &(rmesa->radeon.tcl.aos[nr]),
+				   (char *)VB->AttribPtr[attrib]->data,
+				   VB->AttribPtr[attrib]->stride,
+				   count);
 	    }
 	    vfmt0 |= R200_VTX_DISCRETE_FOG;
 	    goto after_emit;
@@ -473,17 +199,17 @@ void r200EmitArrays( GLcontext *ctx, GLubyte *vimap_rev )
 	 default:
 	    assert(0);
 	 }
-	 if (!rmesa->tcl.vertex_data[i].buf) {
-	    emit_vector( ctx,
-			 &(rmesa->tcl.vertex_data[i]),
-			 (char *)VB->AttribPtr[attrib]->data,
-			 emitsize,
-			 VB->AttribPtr[attrib]->stride,
-			 count );
+	 if (!rmesa->radeon.tcl.aos[nr].bo) {
+	   rcommon_emit_vector( ctx,
+				&(rmesa->radeon.tcl.aos[nr]),
+				(char *)VB->AttribPtr[attrib]->data,
+				emitsize,
+				VB->AttribPtr[attrib]->stride,
+				count );
 	 }
 after_emit:
 	 assert(nr < 12);
-	 component[nr++] = &rmesa->tcl.vertex_data[i];
+	 nr++;
       }
    }
 
@@ -494,19 +220,6 @@ after_emit:
       rmesa->hw.vtx.cmd[VTX_VTXFMT_1] = vfmt1;
    }
 
-   rmesa->tcl.nr_aos_components = nr;
+   rmesa->radeon.tcl.aos_count = nr;
 }
 
-
-void r200ReleaseArrays( GLcontext *ctx, GLuint newinputs )
-{
-   r200ContextPtr rmesa = R200_CONTEXT( ctx );
-
-   /* only do it for changed inputs ? */
-   int i;
-   for (i = 0; i < 15; i++) {
-      if (newinputs & (1 << i))
-	 r200ReleaseDmaRegion( rmesa,
-	    &rmesa->tcl.vertex_data[i], __FUNCTION__ );
-   }
-}
diff --git a/src/mesa/drivers/dri/r200/r200_pixel.c b/src/mesa/drivers/dri/r200/r200_pixel.c
index 2797cbb3dc0..654f2c6ae98 100644
--- a/src/mesa/drivers/dri/r200/r200_pixel.c
+++ b/src/mesa/drivers/dri/r200/r200_pixel.c
@@ -51,7 +51,7 @@ check_color( const GLcontext *ctx, GLenum type, GLenum format,
 	     const void *pixels, GLint sz, GLint pitch )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLuint cpp = rmesa->r200Screen->cpp;
+   GLuint cpp = rmesa->radeon.radeonScreen->cpp;
 
    if (R200_DEBUG & DEBUG_PIXEL)
       fprintf(stderr, "%s\n", __FUNCTION__);
@@ -65,8 +65,8 @@ check_color( const GLcontext *ctx, GLenum type, GLenum format,
       return GL_FALSE;
    }
 
-   if ( type == GL_UNSIGNED_INT_8_8_8_8_REV && 
-	cpp == 4 && 
+   if ( type == GL_UNSIGNED_INT_8_8_8_8_REV &&
+	cpp == 4 &&
 	format == GL_BGRA ) {
       if (R200_DEBUG & DEBUG_PIXEL)
 	 fprintf(stderr, "%s: passed 2\n", __FUNCTION__);
@@ -83,7 +83,7 @@ static GLboolean
 check_color_per_fragment_ops( const GLcontext *ctx )
 {
    int result;
-   result = (!(     ctx->Color.AlphaEnabled || 
+   result = (!(     ctx->Color.AlphaEnabled ||
 		    ctx->Depth.Test ||
 		    ctx->Fog.Enabled ||
 		    ctx->Scissor.Enabled ||
@@ -96,7 +96,7 @@ check_color_per_fragment_ops( const GLcontext *ctx )
 		    ctx->Texture._EnabledUnits
            ) &&
 	   ctx->Current.RasterPosValid);
-   
+
    return result;
 }
 
@@ -137,8 +137,8 @@ clip_pixelrect( const GLcontext *ctx,
    if (*height <= 0)
       return GL_FALSE;
 
-   *size = ((*y + *height - 1) * rmesa->r200Screen->frontPitch +
-	    (*x + *width - 1) * rmesa->r200Screen->cpp);
+   *size = ((*y + *height - 1) * rmesa->radeon.radeonScreen->frontPitch +
+	    (*x + *width - 1) * rmesa->radeon.radeonScreen->cpp);
 
    return GL_TRUE;
 }
@@ -153,19 +153,20 @@ r200TryReadPixels( GLcontext *ctx,
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
    GLint pitch = pack->RowLength ? pack->RowLength : width;
    GLint blit_format;
-   GLuint cpp = rmesa->r200Screen->cpp;
+   GLuint cpp = rmesa->radeon.radeonScreen->cpp;
    GLint size = width * height * cpp;
 
+   return GL_FALSE;
+#if 0
    if (R200_DEBUG & DEBUG_PIXEL)
       fprintf(stderr, "%s\n", __FUNCTION__);
 
    /* Only accelerate reading to GART buffers.
     */
-   if ( !r200IsGartMemory(rmesa, pixels, 
-			 pitch * height * rmesa->r200Screen->cpp ) ) {
+   if ( !r200IsGartMemory(rmesa, pixels,
+			 pitch * height * rmesa->radeon.radeonScreen->cpp ) ) {
       if (R200_DEBUG & DEBUG_PIXEL)
 	 fprintf(stderr, "%s: dest not GART\n", __FUNCTION__);
-      return GL_FALSE;
    }
 
    /* Need GL_PACK_INVERT_MESA to cope with upsidedown results from
@@ -180,7 +181,7 @@ r200TryReadPixels( GLcontext *ctx,
    if (!check_color(ctx, type, format, pack, pixels, size, pitch))
       return GL_FALSE;
 
-   switch ( rmesa->r200Screen->cpp ) {
+   switch ( rmesa->radeon.radeonScreen->cpp ) {
    case 4:
       blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
       break;
@@ -197,14 +198,14 @@ r200TryReadPixels( GLcontext *ctx,
     * a full command buffer expects to be called unlocked.  As a
     * workaround, immediately flush the buffer on aquiring the lock.
     */
-   LOCK_HARDWARE( rmesa );
+   LOCK_HARDWARE( &rmesa->radeon );
 
    if (rmesa->store.cmd_used)
-      r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
+      rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
 
    if (!clip_pixelrect(ctx, ctx->ReadBuffer, &x, &y, &width, &height,
 		       &size)) {
-      UNLOCK_HARDWARE( rmesa );
+      UNLOCK_HARDWARE( &rmesa->radeon );
       if (R200_DEBUG & DEBUG_PIXEL)
 	 fprintf(stderr, "%s totally clipped -- nothing to do\n",
 		 __FUNCTION__);
@@ -212,18 +213,18 @@ r200TryReadPixels( GLcontext *ctx,
    }
 
    {
-      __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+      __DRIdrawablePrivate *dPriv = rmesa->radeon.dri.drawable;
       driRenderbuffer *drb = (driRenderbuffer *) ctx->ReadBuffer->_ColorReadBuffer;
       int nbox = dPriv->numClipRects;
       int src_offset = drb->offset
-		     + rmesa->r200Screen->fbLocation;
+		     + rmesa->radeon.radeonScreen->fbLocation;
       int src_pitch = drb->pitch * drb->cpp;
       int dst_offset = r200GartOffsetFromVirtual( rmesa, pixels );
-      int dst_pitch = pitch * rmesa->r200Screen->cpp;
+      int dst_pitch = pitch * rmesa->radeon.radeonScreen->cpp;
       drm_clip_rect_t *box = dPriv->pClipRects;
       int i;
 
-      r200EmitWait( rmesa, RADEON_WAIT_3D ); 
+      r200EmitWait( rmesa, RADEON_WAIT_3D );
 
       y = dPriv->h - y - height;
       x += dPriv->x;
@@ -240,7 +241,7 @@ r200TryReadPixels( GLcontext *ctx,
 	 GLint by = box[i].y1;
 	 GLint bw = box[i].x2 - bx;
 	 GLint bh = box[i].y2 - by;
-	 
+
 	 if (bx < x) bw -= x - bx, bx = x;
 	 if (by < y) bh -= y - by, by = y;
 	 if (bx + bw > x + width) bw = x + width - bx;
@@ -257,12 +258,12 @@ r200TryReadPixels( GLcontext *ctx,
 		       bw, bh );
       }
 
-      r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
+      rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
    }
-   UNLOCK_HARDWARE( rmesa );
-
-   r200Finish( ctx ); /* required by GL */
+   UNLOCK_HARDWARE( &rmesa->radeon );
 
+   radeonFinish( ctx ); /* required by GL */
+#endif
    return GL_TRUE;
 }
 
@@ -276,9 +277,9 @@ r200ReadPixels( GLcontext *ctx,
    if (R200_DEBUG & DEBUG_PIXEL)
       fprintf(stderr, "%s\n", __FUNCTION__);
 
-   if (!r200TryReadPixels( ctx, x, y, width, height, format, type, pack, 
+   if (!r200TryReadPixels( ctx, x, y, width, height, format, type, pack,
 			   pixels))
-      _swrast_ReadPixels( ctx, x, y, width, height, format, type, pack, 
+      _swrast_ReadPixels( ctx, x, y, width, height, format, type, pack,
 			  pixels);
 }
 
@@ -292,7 +293,7 @@ static void do_draw_pix( GLcontext *ctx,
 			 GLuint planemask)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
    drm_clip_rect_t *box = dPriv->pClipRects;
    struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorDrawBuffers[0];
    driRenderbuffer *drb = (driRenderbuffer *) rb;
@@ -301,12 +302,12 @@ static void do_draw_pix( GLcontext *ctx,
    int blit_format;
    int size;
    int src_offset = r200GartOffsetFromVirtual( rmesa, pixels );
-   int src_pitch = pitch * rmesa->r200Screen->cpp;
+   int src_pitch = pitch * rmesa->radeon.radeonScreen->cpp;
 
    if (R200_DEBUG & DEBUG_PIXEL)
       fprintf(stderr, "%s\n", __FUNCTION__);
-
-   switch ( rmesa->r200Screen->cpp ) {
+#if 0
+   switch ( rmesa->radeon.radeonScreen->cpp ) {
    case 2:
       blit_format = R200_CP_COLOR_FORMAT_RGB565;
       break;
@@ -318,17 +319,17 @@ static void do_draw_pix( GLcontext *ctx,
    }
 
 
-   LOCK_HARDWARE( rmesa );
+   LOCK_HARDWARE( &rmesa->radeon );
 
    if (rmesa->store.cmd_used)
-      r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
+      rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
 
    y -= height;			/* cope with pixel zoom */
-   
+
    if (!clip_pixelrect(ctx, ctx->DrawBuffer,
 		       &x, &y, &width, &height,
 		       &size)) {
-      UNLOCK_HARDWARE( rmesa );
+      UNLOCK_HARDWARE( &rmesa->radeon );
       return;
    }
 
@@ -357,15 +358,16 @@ static void do_draw_pix( GLcontext *ctx,
 		    blit_format,
 		    src_pitch, src_offset,
 		    drb->pitch * drb->cpp,
-		    drb->offset + rmesa->r200Screen->fbLocation,
+		    drb->offset + rmesa->radeon.radeonScreen->fbLocation,
 		    bx - x, by - y,
 		    bx, by,
 		    bw, bh );
    }
 
-   r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
-   r200WaitForIdleLocked( rmesa ); /* required by GL */
-   UNLOCK_HARDWARE( rmesa );
+   rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
+   radeonWaitForIdleLocked( &rmesa->radeon ); /* required by GL */
+   UNLOCK_HARDWARE( &rmesa->radeon );
+#endif
 }
 
 
@@ -381,7 +383,7 @@ r200TryDrawPixels( GLcontext *ctx,
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
    GLint pitch = unpack->RowLength ? unpack->RowLength : width;
    GLuint planemask;
-   GLuint cpp = rmesa->r200Screen->cpp;
+   GLuint cpp = rmesa->radeon.radeonScreen->cpp;
    GLint size = height * pitch * cpp;
 
    if (R200_DEBUG & DEBUG_PIXEL)
@@ -395,7 +397,7 @@ r200TryDrawPixels( GLcontext *ctx,
    case GL_RGB:
    case GL_RGBA:
    case GL_BGRA:
-      planemask = r200PackColor(cpp,
+      planemask = radeonPackColor(cpp,
 				ctx->Color.ColorMask[RCOMP],
 				ctx->Color.ColorMask[GCOMP],
 				ctx->Color.ColorMask[BCOMP],
@@ -407,7 +409,7 @@ r200TryDrawPixels( GLcontext *ctx,
       if (planemask != ~0)
 	 return GL_FALSE;	/* fix me -- should be possible */
 
-      /* Can't do conversions on GART reads/draws. 
+      /* Can't do conversions on GART reads/draws.
        */
       if ( !r200IsGartMemory( rmesa, pixels, size ) ) {
 	 if (R200_DEBUG & DEBUG_PIXEL)
@@ -431,7 +433,7 @@ r200TryDrawPixels( GLcontext *ctx,
       return GL_FALSE;
    }
 
-   if ( r200IsGartMemory(rmesa, pixels, size) )
+   if (0)// r200IsGartMemory(rmesa, pixels, size) )
    {
       do_draw_pix( ctx, x, y, width, height, pitch, pixels, planemask );
       return GL_TRUE;
@@ -471,7 +473,7 @@ r200Bitmap( GLcontext *ctx, GLint px, GLint py,
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 
-   if (rmesa->Fallback)
+   if (rmesa->radeon.Fallback)
       _swrast_Bitmap( ctx, px, py, width, height, unpack, bitmap );
    else
       r200PointsBitmap( ctx, px, py, width, height, unpack, bitmap );
@@ -482,9 +484,9 @@ r200Bitmap( GLcontext *ctx, GLint px, GLint py,
 void r200InitPixelFuncs( GLcontext *ctx )
 {
    if (!getenv("R200_NO_BLITS")) {
-      ctx->Driver.ReadPixels = r200ReadPixels;  
-      ctx->Driver.DrawPixels = r200DrawPixels; 
-      if (getenv("R200_HW_BITMAP")) 
+      ctx->Driver.ReadPixels = r200ReadPixels;
+      ctx->Driver.DrawPixels = r200DrawPixels;
+      if (getenv("R200_HW_BITMAP"))
 	 ctx->Driver.Bitmap = r200Bitmap;
    }
 }
diff --git a/src/mesa/drivers/dri/r200/r200_reg.h b/src/mesa/drivers/dri/r200/r200_reg.h
index 5ce287f7a5f..526a624b697 100644
--- a/src/mesa/drivers/dri/r200/r200_reg.h
+++ b/src/mesa/drivers/dri/r200/r200_reg.h
@@ -463,8 +463,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define     R200_VSC_UPDATE_USER_COLOR_1_ENABLE    0x00020000
 /* gap */
 #define R200_SE_TCL_VECTOR_INDX_REG                0x2200
+#       define RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT  16
+#       define RADEON_VEC_INDX_DWORD_COUNT_SHIFT     28
 #define R200_SE_TCL_VECTOR_DATA_REG                0x2204
 #define R200_SE_TCL_SCALAR_INDX_REG                0x2208
+#       define RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT  16
 #define R200_SE_TCL_SCALAR_DATA_REG                0x220c
 /* gap */
 #define R200_SE_TCL_MATRIX_SEL_0                   0x2230
@@ -949,6 +952,9 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define     R200_LOD_BIAS_MASK                        (0xfff80000)
 #define     R200_LOD_BIAS_SHIFT                       19
 #define R200_PP_TXSIZE_0                  0x2c0c /* NPOT only */
+#define R200_PP_TX_WIDTHMASK_SHIFT 0
+#define R200_PP_TX_HEIGHTMASK_SHIFT 16
+
 #define R200_PP_TXPITCH_0                 0x2c10 /* NPOT only */
 #define R200_PP_BORDER_COLOR_0            0x2c14
 #define R200_PP_CUBIC_FACES_0             0x2c18
diff --git a/src/mesa/drivers/dri/r200/r200_span.c b/src/mesa/drivers/dri/r200/r200_span.c
deleted file mode 100644
index 9783678028a..00000000000
--- a/src/mesa/drivers/dri/r200/r200_span.c
+++ /dev/null
@@ -1,307 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/colormac.h"
-#include "swrast/swrast.h"
-
-#include "r200_context.h"
-#include "r200_ioctl.h"
-#include "r200_state.h"
-#include "r200_span.h"
-#include "r200_tex.h"
-
-#define DBG 0
-
-/*
- * Note that all information needed to access pixels in a renderbuffer
- * should be obtained through the gl_renderbuffer parameter, not per-context
- * information.
- */
-#define LOCAL_VARS						\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   const __DRIdrawablePrivate *dPriv = drb->dPriv;		\
-   const GLuint bottom = dPriv->h - 1;				\
-   GLubyte *buf = (GLubyte *) drb->flippedData			\
-      + (dPriv->y * drb->flippedPitch + dPriv->x) * drb->cpp;	\
-   GLuint p;							\
-   (void) p;
-
-#define LOCAL_DEPTH_VARS				\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;	\
-   const __DRIdrawablePrivate *dPriv = drb->dPriv;	\
-   const GLuint bottom = dPriv->h - 1;			\
-   GLuint xo = dPriv->x;				\
-   GLuint yo = dPriv->y;				\
-   GLubyte *buf = (GLubyte *) drb->Base.Data;
-
-#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
-
-#define Y_FLIP(Y) (bottom - (Y))
-
-#define HW_LOCK() 
-
-#define HW_UNLOCK()							
-
-
-
-/* ================================================================
- * Color buffer
- */
-
-/* 16 bit, RGB565 color spanline and pixel functions
- */
-#define SPANTMP_PIXEL_FMT GL_RGB
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-
-#define TAG(x)    r200##x##_RGB565
-#define TAG2(x,y) r200##x##_RGB565##y
-#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 2)
-#include "spantmp2.h"
-
-/* 32 bit, ARGB8888 color spanline and pixel functions
- */
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    r200##x##_ARGB8888
-#define TAG2(x,y) r200##x##_ARGB8888##y
-#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 4)
-#include "spantmp2.h"
-
-
-/* ================================================================
- * Depth buffer
- */
-
-/* The Radeon family has depth tiling on all the time, so we have to convert
- * the x,y coordinates into the memory bus address (mba) in the same
- * manner as the engine.  In each case, the linear block address (ba)
- * is calculated, and then wired with x and y to produce the final
- * memory address.
- * The chip will do address translation on its own if the surface registers
- * are set up correctly. It is not quite enough to get it working with hyperz too...
- */
-
-/* extract bit 'b' of x, result is zero or one */
-#define BIT(x,b) ((x & (1<<b))>>b)
-
-static GLuint
-r200_mba_z32( driRenderbuffer *drb, GLint x, GLint y )
-{
-   GLuint pitch = drb->pitch;
-   if (drb->depthHasSurface) {
-      return 4 * (x + y * pitch);
-   }
-   else {
-      GLuint b = ((y & 0x7FF) >> 4) * ((pitch & 0xFFF) >> 5) + ((x & 0x7FF) >> 5);
-      GLuint a = 
-         (BIT(x,0) << 2) |
-         (BIT(y,0) << 3) |
-         (BIT(x,1) << 4) |
-         (BIT(y,1) << 5) |
-         (BIT(x,3) << 6) |
-         (BIT(x,4) << 7) |
-         (BIT(x,2) << 8) |
-         (BIT(y,2) << 9) |
-         (BIT(y,3) << 10) |
-         (((pitch & 0x20) ? (b & 0x01) : ((b & 0x01) ^ (BIT(y,4)))) << 11) |
-         ((b >> 1) << 12);
-      return a;
-   }
-}
-
-static GLuint
-r200_mba_z16( driRenderbuffer *drb, GLint x, GLint y )
-{
-   GLuint pitch = drb->pitch;
-   if (drb->depthHasSurface) {
-      return 2 * (x + y * pitch);
-   }
-   else {
-      GLuint b = ((y & 0x7FF) >> 4) * ((pitch & 0xFFF) >> 6) + ((x & 0x7FF) >> 6);
-      GLuint a = 
-         (BIT(x,0) << 1) |
-         (BIT(y,0) << 2) |
-         (BIT(x,1) << 3) |
-         (BIT(y,1) << 4) |
-         (BIT(x,2) << 5) |
-         (BIT(x,4) << 6) |
-         (BIT(x,5) << 7) |
-         (BIT(x,3) << 8) |
-         (BIT(y,2) << 9) |
-         (BIT(y,3) << 10) |
-         (((pitch & 0x40) ? (b & 0x01) : ((b & 0x01) ^ (BIT(y,4)))) << 11) |
-         ((b >> 1) << 12);
-      return a;
-   }
-}
-
-
-/* 16-bit depth buffer functions
- */
-#define VALUE_TYPE GLushort
-
-#define WRITE_DEPTH( _x, _y, d )					\
-   *(GLushort *)(buf + r200_mba_z16( drb, _x + xo, _y + yo )) = d;
-
-#define READ_DEPTH( d, _x, _y )						\
-   d = *(GLushort *)(buf + r200_mba_z16( drb, _x + xo, _y + yo ));
-
-#define TAG(x) r200##x##_z16
-#include "depthtmp.h"
-
-
-/* 24 bit depth, 8 bit stencil depthbuffer functions
- */
-#define VALUE_TYPE GLuint
-
-#define WRITE_DEPTH( _x, _y, d )					\
-do {									\
-   GLuint offset = r200_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0xff000000;							\
-   tmp |= ((d) & 0x00ffffff);						\
-   *(GLuint *)(buf + offset) = tmp;					\
-} while (0)
-
-#define READ_DEPTH( d, _x, _y )						\
-   d = *(GLuint *)(buf + r200_mba_z32( drb, _x + xo,			\
-					 _y + yo )) & 0x00ffffff;
-
-#define TAG(x) r200##x##_z24_s8
-#include "depthtmp.h"
-
-
-/* ================================================================
- * Stencil buffer
- */
-
-/* 24 bit depth, 8 bit stencil depthbuffer functions
- */
-#define WRITE_STENCIL( _x, _y, d )					\
-do {									\
-   GLuint offset = r200_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0x00ffffff;							\
-   tmp |= (((d) & 0xff) << 24);						\
-   *(GLuint *)(buf + offset) = tmp;					\
-} while (0)
-
-#define READ_STENCIL( d, _x, _y )					\
-do {									\
-   GLuint offset = r200_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0xff000000;							\
-   d = tmp >> 24;							\
-} while (0)
-
-#define TAG(x) r200##x##_z24_s8
-#include "stenciltmp.h"
-
-
-/* Move locking out to get reasonable span performance (10x better
- * than doing this in HW_LOCK above).  WaitForIdle() is the main
- * culprit.
- */
-
-static void r200SpanRenderStart( GLcontext *ctx )
-{
-   r200ContextPtr rmesa = R200_CONTEXT( ctx );
-
-   R200_FIREVERTICES( rmesa );
-   LOCK_HARDWARE( rmesa );
-   r200WaitForIdleLocked( rmesa );
-
-   /* Read & rewrite the first pixel in the frame buffer.  This should
-    * be a noop, right?  In fact without this conform fails as reading
-    * from the framebuffer sometimes produces old results -- the
-    * on-card read cache gets mixed up and doesn't notice that the
-    * framebuffer has been updated.
-    *
-    * In the worst case this is buggy too as p might get the wrong
-    * value first time, so really need a hidden pixel somewhere for this.
-    */
-   {
-      int p;
-      driRenderbuffer *drb =
-	 (driRenderbuffer *) ctx->WinSysDrawBuffer->_ColorDrawBuffers[0];
-      volatile int *buf =
-	 (volatile int *)(rmesa->dri.screen->pFB + drb->offset);
-      p = *buf;
-      *buf = p;
-   }
-}
-
-static void r200SpanRenderFinish( GLcontext *ctx )
-{
-   r200ContextPtr rmesa = R200_CONTEXT( ctx );
-   _swrast_flush( ctx );
-   UNLOCK_HARDWARE( rmesa );
-}
-
-void r200InitSpanFuncs( GLcontext *ctx )
-{
-   struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
-   swdd->SpanRenderStart          = r200SpanRenderStart;
-   swdd->SpanRenderFinish         = r200SpanRenderFinish; 
-}
-
-
-
-/**
- * Plug in the Get/Put routines for the given driRenderbuffer.
- */
-void
-radeonSetSpanFunctions(driRenderbuffer *drb, const GLvisual *vis)
-{
-   if (drb->Base.InternalFormat == GL_RGBA) {
-      if (vis->redBits == 5 && vis->greenBits == 6 && vis->blueBits == 5) {
-         r200InitPointers_RGB565(&drb->Base);
-      }
-      else {
-         r200InitPointers_ARGB8888(&drb->Base);
-      }
-   }
-   else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) {
-      r200InitDepthPointers_z16(&drb->Base);
-   }
-   else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) {
-      r200InitDepthPointers_z24_s8(&drb->Base);
-   }
-   else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) {
-      r200InitStencilPointers_z24_s8(&drb->Base);
-   }
-}
diff --git a/src/mesa/drivers/dri/r200/r200_span.h b/src/mesa/drivers/dri/r200/r200_span.h
deleted file mode 100644
index bae56443092..00000000000
--- a/src/mesa/drivers/dri/r200/r200_span.h
+++ /dev/null
@@ -1,45 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
-
-#ifndef __R200_SPAN_H__
-#define __R200_SPAN_H__
-
-#include "drirenderbuffer.h"
-
-extern void r200InitSpanFuncs( GLcontext *ctx );
-
-extern void
-radeonSetSpanFunctions(driRenderbuffer *rb, const GLvisual *vis);
-
-#endif
diff --git a/src/mesa/drivers/dri/r200/r200_state.c b/src/mesa/drivers/dri/r200/r200_state.c
index 81ee1ed022c..f8ebe0df57e 100644
--- a/src/mesa/drivers/dri/r200/r200_state.c
+++ b/src/mesa/drivers/dri/r200/r200_state.c
@@ -47,6 +47,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/t_pipeline.h"
 #include "swrast_setup/swrast_setup.h"
 
+#include "radeon_common.h"
+#include "radeon_mipmap_tree.h"
 #include "r200_context.h"
 #include "r200_ioctl.h"
 #include "r200_state.h"
@@ -77,7 +79,7 @@ static void r200AlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref )
 
    switch ( func ) {
    case GL_NEVER:
-      pp_misc |= R200_ALPHA_TEST_FAIL; 
+      pp_misc |= R200_ALPHA_TEST_FAIL;
       break;
    case GL_LESS:
       pp_misc |= R200_ALPHA_TEST_LESS;
@@ -114,8 +116,8 @@ static void r200BlendColor( GLcontext *ctx, const GLfloat cf[4] )
    CLAMPED_FLOAT_TO_UBYTE(color[1], cf[1]);
    CLAMPED_FLOAT_TO_UBYTE(color[2], cf[2]);
    CLAMPED_FLOAT_TO_UBYTE(color[3], cf[3]);
-   if (rmesa->r200Screen->drmSupportsBlendColor)
-      rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCOLOR] = r200PackColor( 4, color[0], color[1], color[2], color[3] );
+   if (rmesa->radeon.radeonScreen->drmSupportsBlendColor)
+      rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCOLOR] = radeonPackColor( 4, color[0], color[1], color[2], color[3] );
 }
 
 /**
@@ -213,7 +215,7 @@ static void r200_set_blend_state( GLcontext * ctx )
 
    R200_STATECHANGE( rmesa, ctx );
 
-   if (rmesa->r200Screen->drmSupportsBlendColor) {
+   if (rmesa->radeon.radeonScreen->drmSupportsBlendColor) {
       if (ctx->Color.ColorLogicOpEnabled) {
          rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] =  cntl | R200_ROP_ENABLE;
          rmesa->hw.ctx.cmd[CTX_RB3D_ABLENDCNTL] = eqn | func;
@@ -278,7 +280,7 @@ static void r200_set_blend_state( GLcontext * ctx )
       return;
    }
 
-   if (!rmesa->r200Screen->drmSupportsBlendColor) {
+   if (!rmesa->radeon.radeonScreen->drmSupportsBlendColor) {
       rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = eqn | func;
       return;
    }
@@ -383,10 +385,10 @@ static void r200ClearDepth( GLcontext *ctx, GLclampd d )
 
    switch ( format ) {
    case R200_DEPTH_FORMAT_16BIT_INT_Z:
-      rmesa->state.depth.clear = d * 0x0000ffff;
+      rmesa->radeon.state.depth.clear = d * 0x0000ffff;
       break;
    case R200_DEPTH_FORMAT_24BIT_INT_Z:
-      rmesa->state.depth.clear = d * 0x00ffffff;
+      rmesa->radeon.state.depth.clear = d * 0x00ffffff;
       break;
    }
 }
@@ -477,10 +479,10 @@ static void r200Fogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
 	 }
       }
       break;
-   case GL_FOG_COLOR: 
+   case GL_FOG_COLOR:
       R200_STATECHANGE( rmesa, ctx );
       UNCLAMPED_FLOAT_TO_RGB_CHAN( col, ctx->Fog.Color );
-      i = r200PackColor( 4, col[0], col[1], col[2], 0 );
+      i = radeonPackColor( 4, col[0], col[1], col[2], 0 );
       rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] &= ~R200_FOG_COLOR_MASK;
       rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] |= i;
       break;
@@ -505,7 +507,7 @@ static void r200Fogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
 
       if (out_0 != rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0]) {
 	 R200_STATECHANGE( rmesa, vtx );
-	 rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] = out_0;	 
+	 rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] = out_0;
       }
 
       break;
@@ -521,102 +523,6 @@ static void r200Fogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
    }
 }
 
-
-/* =============================================================
- * Scissoring
- */
-
-
-static GLboolean intersect_rect( drm_clip_rect_t *out,
-				 drm_clip_rect_t *a,
-				 drm_clip_rect_t *b )
-{
-   *out = *a;
-   if ( b->x1 > out->x1 ) out->x1 = b->x1;
-   if ( b->y1 > out->y1 ) out->y1 = b->y1;
-   if ( b->x2 < out->x2 ) out->x2 = b->x2;
-   if ( b->y2 < out->y2 ) out->y2 = b->y2;
-   if ( out->x1 >= out->x2 ) return GL_FALSE;
-   if ( out->y1 >= out->y2 ) return GL_FALSE;
-   return GL_TRUE;
-}
-
-
-void r200RecalcScissorRects( r200ContextPtr rmesa )
-{
-   drm_clip_rect_t *out;
-   int i;
-
-   /* Grow cliprect store?
-    */
-   if (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) {
-      while (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) {
-	 rmesa->state.scissor.numAllocedClipRects += 1;	/* zero case */
-	 rmesa->state.scissor.numAllocedClipRects *= 2;
-      }
-
-      if (rmesa->state.scissor.pClipRects)
-	 FREE(rmesa->state.scissor.pClipRects);
-
-      rmesa->state.scissor.pClipRects = 
-	 MALLOC( rmesa->state.scissor.numAllocedClipRects * 
-		 sizeof(drm_clip_rect_t) );
-
-      if ( rmesa->state.scissor.pClipRects == NULL ) {
-	 rmesa->state.scissor.numAllocedClipRects = 0;
-	 return;
-      }
-   }
-   
-   out = rmesa->state.scissor.pClipRects;
-   rmesa->state.scissor.numClipRects = 0;
-
-   for ( i = 0 ; i < rmesa->numClipRects ;  i++ ) {
-      if ( intersect_rect( out, 
-			   &rmesa->pClipRects[i], 
-			   &rmesa->state.scissor.rect ) ) {
-	 rmesa->state.scissor.numClipRects++;
-	 out++;
-      }
-   }
-}
-
-
-static void r200UpdateScissor( GLcontext *ctx )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if ( rmesa->dri.drawable ) {
-      __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
-
-      int x = ctx->Scissor.X;
-      int y = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height;
-      int w = ctx->Scissor.X + ctx->Scissor.Width - 1;
-      int h = dPriv->h - ctx->Scissor.Y - 1;
-
-      rmesa->state.scissor.rect.x1 = x + dPriv->x;
-      rmesa->state.scissor.rect.y1 = y + dPriv->y;
-      rmesa->state.scissor.rect.x2 = w + dPriv->x + 1;
-      rmesa->state.scissor.rect.y2 = h + dPriv->y + 1;
-
-      r200RecalcScissorRects( rmesa );
-   }
-}
-
-
-static void r200Scissor( GLcontext *ctx,
-			   GLint x, GLint y, GLsizei w, GLsizei h )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if ( ctx->Scissor.Enabled ) {
-      R200_FIREVERTICES( rmesa );	/* don't pipeline cliprect changes */
-      r200UpdateScissor( ctx );
-   }
-
-}
-
-
 /* =============================================================
  * Culling
  */
@@ -790,7 +696,7 @@ static void r200LineStipple( GLcontext *ctx, GLint factor, GLushort pattern )
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 
    R200_STATECHANGE( rmesa, lin );
-   rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] = 
+   rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] =
       ((((GLuint)factor & 0xff) << 16) | ((GLuint)pattern));
 }
 
@@ -803,7 +709,7 @@ static void r200ColorMask( GLcontext *ctx,
 			   GLboolean b, GLboolean a )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLuint mask = r200PackColor( rmesa->r200Screen->cpp,
+   GLuint mask = radeonPackColor( rmesa->radeon.radeonScreen->cpp,
 				ctx->Color.ColorMask[RCOMP],
 				ctx->Color.ColorMask[GCOMP],
 				ctx->Color.ColorMask[BCOMP],
@@ -814,10 +720,10 @@ static void r200ColorMask( GLcontext *ctx,
    if (!(r && g && b && a))
       flag |= R200_PLANE_MASK_ENABLE;
 
-   if ( rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] != flag ) { 
-      R200_STATECHANGE( rmesa, ctx ); 
-      rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = flag; 
-   } 
+   if ( rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] != flag ) {
+      R200_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = flag;
+   }
 
    if ( rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK] != mask ) {
       R200_STATECHANGE( rmesa, msk );
@@ -834,7 +740,8 @@ static void r200PolygonOffset( GLcontext *ctx,
 			       GLfloat factor, GLfloat units )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   float_ui32_type constant =  { units * rmesa->state.depth.scale };
+   const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
+   float_ui32_type constant =  { units * depthScale };
    float_ui32_type factoru = { factor };
 
 /*    factor *= 2; */
@@ -861,15 +768,15 @@ static void r200PolygonStipple( GLcontext *ctx, const GLubyte *mask )
 
    /* TODO: push this into cmd mechanism
     */
-   R200_FIREVERTICES( rmesa );
-   LOCK_HARDWARE( rmesa );
+   radeon_firevertices(&rmesa->radeon);
+   LOCK_HARDWARE( &rmesa->radeon );
 
    /* FIXME: Use window x,y offsets into stipple RAM.
     */
    stipple.mask = rmesa->state.stipple.mask;
-   drmCommandWrite( rmesa->dri.fd, DRM_RADEON_STIPPLE, 
+   drmCommandWrite( rmesa->radeon.dri.fd, DRM_RADEON_STIPPLE,
                     &stipple, sizeof(stipple) );
-   UNLOCK_HARDWARE( rmesa );
+   UNLOCK_HARDWARE( &rmesa->radeon );
 }
 
 static void r200PolygonMode( GLcontext *ctx, GLenum face, GLenum mode )
@@ -878,10 +785,10 @@ static void r200PolygonMode( GLcontext *ctx, GLenum face, GLenum mode )
    GLboolean flag = (ctx->_TriangleCaps & DD_TRI_UNFILLED) != 0;
 
    /* Can't generally do unfilled via tcl, but some good special
-    * cases work. 
+    * cases work.
     */
    TCL_FALLBACK( ctx, R200_TCL_FALLBACK_UNFILLED, flag);
-   if (rmesa->TclFallback) {
+   if (rmesa->radeon.TclFallback) {
       r200ChooseRenderState( ctx );
       r200ChooseVertexState( ctx );
    }
@@ -920,34 +827,34 @@ static void r200UpdateSpecular( GLcontext *ctx )
 
    if (ctx->Light.Enabled &&
        ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR) {
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |= 
+      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |=
 	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_0_SHIFT) |
-	  (R200_VTX_FP_RGBA << R200_VTX_COLOR_1_SHIFT));	
+	  (R200_VTX_FP_RGBA << R200_VTX_COLOR_1_SHIFT));
       rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_COMPSEL] |= R200_OUTPUT_COLOR_0;
       rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_COMPSEL] |= R200_OUTPUT_COLOR_1;
       rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] |= R200_LIGHTING_ENABLE;
       p |=  R200_SPECULAR_ENABLE;
-      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] &= 
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] &=
 	 ~R200_DIFFUSE_SPECULAR_COMBINE;
    }
    else if (ctx->Light.Enabled) {
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |= 
-	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_0_SHIFT));	
+      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |=
+	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_0_SHIFT));
       rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_COMPSEL] |= R200_OUTPUT_COLOR_0;
       rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] |= R200_LIGHTING_ENABLE;
    } else if (ctx->Fog.ColorSumEnabled ) {
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |= 
+      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |=
 	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_0_SHIFT) |
-	  (R200_VTX_FP_RGBA << R200_VTX_COLOR_1_SHIFT));	
+	  (R200_VTX_FP_RGBA << R200_VTX_COLOR_1_SHIFT));
       p |=  R200_SPECULAR_ENABLE;
    } else {
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |= 
-	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_0_SHIFT));	
+      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |=
+	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_0_SHIFT));
    }
 
    if (ctx->Fog.Enabled) {
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |= 
-	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_1_SHIFT));	
+      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_0] |=
+	 ((R200_VTX_FP_RGBA << R200_VTX_COLOR_1_SHIFT));
       rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_COMPSEL] |= R200_OUTPUT_COLOR_1;
    }
 
@@ -958,7 +865,7 @@ static void r200UpdateSpecular( GLcontext *ctx )
 
    /* Update vertex/render formats
     */
-   if (rmesa->TclFallback) { 
+   if (rmesa->radeon.TclFallback) {
       r200ChooseRenderState( ctx );
       r200ChooseVertexState( ctx );
    }
@@ -970,7 +877,7 @@ static void r200UpdateSpecular( GLcontext *ctx )
  */
 
 
-/* Update on colormaterial, material emmissive/ambient, 
+/* Update on colormaterial, material emmissive/ambient,
  * lightmodel.globalambient
  */
 static void update_global_ambient( GLcontext *ctx )
@@ -984,23 +891,23 @@ static void update_global_ambient( GLcontext *ctx )
     */
    if ((rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_1] &
        ((3 << R200_FRONT_EMISSIVE_SOURCE_SHIFT) |
-	(3 << R200_FRONT_AMBIENT_SOURCE_SHIFT))) == 0) 
+	(3 << R200_FRONT_AMBIENT_SOURCE_SHIFT))) == 0)
    {
-      COPY_3V( &fcmd[GLT_RED], 
+      COPY_3V( &fcmd[GLT_RED],
 	       ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_EMISSION]);
       ACC_SCALE_3V( &fcmd[GLT_RED],
 		   ctx->Light.Model.Ambient,
 		   ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT]);
-   } 
+   }
    else
    {
       COPY_3V( &fcmd[GLT_RED], ctx->Light.Model.Ambient );
    }
-   
+
    R200_DB_STATECHANGE(rmesa, &rmesa->hw.glt);
 }
 
-/* Update on change to 
+/* Update on change to
  *    - light[p].colors
  *    - light[p].enabled
  */
@@ -1014,10 +921,10 @@ static void update_light_colors( GLcontext *ctx, GLuint p )
       r200ContextPtr rmesa = R200_CONTEXT(ctx);
       float *fcmd = (float *)R200_DB_STATE( lit[p] );
 
-      COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient );	 
+      COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient );
       COPY_4V( &fcmd[LIT_DIFFUSE_RED], l->Diffuse );
       COPY_4V( &fcmd[LIT_SPECULAR_RED], l->Specular );
-      
+
       R200_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] );
    }
 }
@@ -1037,7 +944,7 @@ static void r200ColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
 
    if (ctx->Light.ColorMaterialEnabled) {
       GLuint mask = ctx->Light.ColorMaterialBitmask;
-   
+
       if (mask & MAT_BIT_FRONT_EMISSION) {
 	 light_model_ctl1 |= (R200_LM1_SOURCE_VERTEX_COLOR_0 <<
 			     R200_FRONT_EMISSIVE_SOURCE_SHIFT);
@@ -1053,7 +960,7 @@ static void r200ColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
       else
          light_model_ctl1 |= (R200_LM1_SOURCE_MATERIAL_0 <<
 			     R200_FRONT_AMBIENT_SOURCE_SHIFT);
-	 
+
       if (mask & MAT_BIT_FRONT_DIFFUSE) {
 	 light_model_ctl1 |= (R200_LM1_SOURCE_VERTEX_COLOR_0 <<
 			     R200_FRONT_DIFFUSE_SOURCE_SHIFT);
@@ -1061,7 +968,7 @@ static void r200ColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
       else
          light_model_ctl1 |= (R200_LM1_SOURCE_MATERIAL_0 <<
 			     R200_FRONT_DIFFUSE_SOURCE_SHIFT);
-   
+
       if (mask & MAT_BIT_FRONT_SPECULAR) {
 	 light_model_ctl1 |= (R200_LM1_SOURCE_VERTEX_COLOR_0 <<
 			     R200_FRONT_SPECULAR_SOURCE_SHIFT);
@@ -1070,7 +977,7 @@ static void r200ColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
          light_model_ctl1 |= (R200_LM1_SOURCE_MATERIAL_0 <<
 			     R200_FRONT_SPECULAR_SOURCE_SHIFT);
       }
-   
+
       if (mask & MAT_BIT_BACK_EMISSION) {
 	 light_model_ctl1 |= (R200_LM1_SOURCE_VERTEX_COLOR_0 <<
 			     R200_BACK_EMISSIVE_SOURCE_SHIFT);
@@ -1120,8 +1027,8 @@ static void r200ColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
       R200_STATECHANGE( rmesa, tcl );
       rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_1] = light_model_ctl1;
    }
-   
-   
+
+
 }
 
 void r200UpdateMaterial( GLcontext *ctx )
@@ -1131,7 +1038,7 @@ void r200UpdateMaterial( GLcontext *ctx )
    GLfloat *fcmd = (GLfloat *)R200_DB_STATE( mtl[0] );
    GLfloat *fcmd2 = (GLfloat *)R200_DB_STATE( mtl[1] );
    GLuint mask = ~0;
-   
+
    /* Might be possible and faster to update everything unconditionally? */
    if (ctx->Light.ColorMaterialEnabled)
       mask &= ~ctx->Light.ColorMaterialBitmask;
@@ -1217,7 +1124,7 @@ void r200UpdateMaterial( GLcontext *ctx )
  *
  * which are calculated in light.c and are correct for the current
  * lighting space (model or eye), hence dependencies on _NEW_MODELVIEW
- * and _MESA_NEW_NEED_EYE_COORDS.  
+ * and _MESA_NEW_NEED_EYE_COORDS.
  */
 static void update_light( GLcontext *ctx )
 {
@@ -1234,8 +1141,8 @@ static void update_light( GLcontext *ctx )
 	 tmp &= ~R200_LIGHT_IN_MODELSPACE;
       else
 	 tmp |= R200_LIGHT_IN_MODELSPACE;
-      
-      if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0]) 
+
+      if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0])
       {
 	 R200_STATECHANGE( rmesa, tcl );
 	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] = tmp;
@@ -1259,10 +1166,10 @@ static void update_light( GLcontext *ctx )
 	 if (ctx->Light.Light[p].Enabled) {
 	    struct gl_light *l = &ctx->Light.Light[p];
 	    GLfloat *fcmd = (GLfloat *)R200_DB_STATE( lit[p] );
-	    
+
 	    if (l->EyePosition[3] == 0.0) {
-	       COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm ); 
-	       COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm ); 
+	       COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm );
+	       COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm );
 	       fcmd[LIT_POSITION_W] = 0;
 	       fcmd[LIT_DIRECTION_W] = 0;
 	    } else {
@@ -1286,21 +1193,21 @@ static void r200Lightfv( GLcontext *ctx, GLenum light,
    GLint p = light - GL_LIGHT0;
    struct gl_light *l = &ctx->Light.Light[p];
    GLfloat *fcmd = (GLfloat *)rmesa->hw.lit[p].cmd;
-   
+
 
    switch (pname) {
-   case GL_AMBIENT:		
+   case GL_AMBIENT:
    case GL_DIFFUSE:
    case GL_SPECULAR:
       update_light_colors( ctx, p );
       break;
 
-   case GL_SPOT_DIRECTION: 
-      /* picked up in update_light */	
+   case GL_SPOT_DIRECTION:
+      /* picked up in update_light */
       break;
 
    case GL_POSITION: {
-      /* positions picked up in update_light, but can do flag here */	
+      /* positions picked up in update_light, but can do flag here */
       GLuint flag = (p&1)? R200_LIGHT_1_IS_LOCAL : R200_LIGHT_0_IS_LOCAL;
       GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
 
@@ -1416,7 +1323,7 @@ static void r200LightModelfv( GLcontext *ctx, GLenum pname,
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 
    switch (pname) {
-      case GL_LIGHT_MODEL_AMBIENT: 
+      case GL_LIGHT_MODEL_AMBIENT:
 	 update_global_ambient( ctx );
 	 break;
 
@@ -1430,7 +1337,7 @@ static void r200LightModelfv( GLcontext *ctx, GLenum pname,
 	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] |= R200_LIGHT_TWOSIDE;
 	 else
 	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] &= ~(R200_LIGHT_TWOSIDE);
-	 if (rmesa->TclFallback) {
+	 if (rmesa->radeon.TclFallback) {
 	    r200ChooseRenderState( ctx );
 	    r200ChooseVertexState( ctx );
 	 }
@@ -1675,7 +1582,7 @@ static void r200ClearStencil( GLcontext *ctx, GLint s )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 
-   rmesa->state.stencil.clear = 
+   rmesa->radeon.state.stencil.clear =
       ((GLuint) (ctx->Stencil.Clear & 0xff) |
        (0xff << R200_STENCIL_MASK_SHIFT) |
        ((ctx->Stencil.WriteMask[0] & 0xff) << R200_STENCIL_WRITEMASK_SHIFT));
@@ -1700,19 +1607,29 @@ static void r200ClearStencil( GLcontext *ctx, GLint s )
 void r200UpdateWindow( GLcontext *ctx )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
-   GLfloat xoffset = (GLfloat)dPriv->x;
-   GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h;
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
+   GLfloat xoffset = dPriv ? (GLfloat) dPriv->x : 0;
+   GLfloat yoffset = dPriv ? (GLfloat) dPriv->y + dPriv->h : 0;
    const GLfloat *v = ctx->Viewport._WindowMap.m;
+   const GLboolean render_to_fbo = (ctx->DrawBuffer ? (ctx->DrawBuffer->Name != 0) : 0);
+   const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
+   GLfloat y_scale, y_bias;
+
+   if (render_to_fbo) {
+      y_scale = 1.0;
+      y_bias = 0;
+   } else {
+      y_scale = -1.0;
+      y_bias = yoffset;
+   }
 
    float_ui32_type sx = { v[MAT_SX] };
    float_ui32_type tx = { v[MAT_TX] + xoffset + SUBPIXEL_X };
-   float_ui32_type sy = { - v[MAT_SY] };
-   float_ui32_type ty = { (- v[MAT_TY]) + yoffset + SUBPIXEL_Y };
-   float_ui32_type sz = { v[MAT_SZ] * rmesa->state.depth.scale };
-   float_ui32_type tz = { v[MAT_TZ] * rmesa->state.depth.scale };
+   float_ui32_type sy = { v[MAT_SY] * y_scale };
+   float_ui32_type ty = { (v[MAT_TY] * y_scale) + y_bias + SUBPIXEL_Y };
+   float_ui32_type sz = { v[MAT_SZ] * depthScale };
+   float_ui32_type tz = { v[MAT_TZ] * depthScale };
 
-   R200_FIREVERTICES( rmesa );
    R200_STATECHANGE( rmesa, vpt );
 
    rmesa->hw.vpt.cmd[VPT_SE_VPORT_XSCALE]  = sx.ui32;
@@ -1733,6 +1650,8 @@ static void r200Viewport( GLcontext *ctx, GLint x, GLint y,
     * values, or keep the originals hanging around.
     */
    r200UpdateWindow( ctx );
+
+   radeon_viewport(ctx, x, y, width, height);
 }
 
 static void r200DepthRange( GLcontext *ctx, GLclampd nearval,
@@ -1744,7 +1663,7 @@ static void r200DepthRange( GLcontext *ctx, GLclampd nearval,
 void r200UpdateViewportOffset( GLcontext *ctx )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
    GLfloat xoffset = (GLfloat)dPriv->x;
    GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h;
    const GLfloat *v = ctx->Viewport._WindowMap.m;
@@ -1774,8 +1693,8 @@ void r200UpdateViewportOffset( GLcontext *ctx )
                 R200_STIPPLE_Y_OFFSET_MASK);
 
          /* add magic offsets, then invert */
-         stx = 31 - ((rmesa->dri.drawable->x - 1) & R200_STIPPLE_COORD_MASK);
-         sty = 31 - ((rmesa->dri.drawable->y + rmesa->dri.drawable->h - 1)
+         stx = 31 - ((dPriv->x - 1) & R200_STIPPLE_COORD_MASK);
+         sty = 31 - ((dPriv->y + dPriv->h - 1)
                      & R200_STIPPLE_COORD_MASK);
 
          m |= ((stx << R200_STIPPLE_X_OFFSET_SHIFT) |
@@ -1788,7 +1707,7 @@ void r200UpdateViewportOffset( GLcontext *ctx )
       }
    }
 
-   r200UpdateScissor( ctx );
+   radeonUpdateScissor( ctx );
 }
 
 
@@ -1805,7 +1724,7 @@ static void r200ClearColor( GLcontext *ctx, const GLfloat c[4] )
    CLAMPED_FLOAT_TO_UBYTE(color[1], c[1]);
    CLAMPED_FLOAT_TO_UBYTE(color[2], c[2]);
    CLAMPED_FLOAT_TO_UBYTE(color[3], c[3]);
-   rmesa->state.color.clear = r200PackColor( rmesa->r200Screen->cpp,
+   rmesa->radeon.state.color.clear = radeonPackColor( rmesa->radeon.radeonScreen->cpp,
                                              color[0], color[1],
                                              color[2], color[3] );
 }
@@ -1848,96 +1767,6 @@ static void r200LogicOpCode( GLcontext *ctx, GLenum opcode )
    rmesa->hw.msk.cmd[MSK_RB3D_ROPCNTL] = r200_rop_tab[rop];
 }
 
-
-/*
- * Set up the cliprects for either front or back-buffer drawing.
- */
-void r200SetCliprects( r200ContextPtr rmesa )
-{
-   __DRIdrawablePrivate *const drawable = rmesa->dri.drawable;
-   __DRIdrawablePrivate *const readable = rmesa->dri.readable;
-   GLframebuffer *const draw_fb = (GLframebuffer*) drawable->driverPrivate;
-   GLframebuffer *const read_fb = (GLframebuffer*) readable->driverPrivate;
-
-   if (draw_fb->_ColorDrawBufferIndexes[0] == BUFFER_BIT_BACK_LEFT) {
-      /* Can't ignore 2d windows if we are page flipping.
-       */
-      if ( drawable->numBackClipRects == 0 || rmesa->doPageFlip ) {
-         rmesa->numClipRects = drawable->numClipRects;
-         rmesa->pClipRects = drawable->pClipRects;
-      }
-      else {
-         rmesa->numClipRects = drawable->numBackClipRects;
-         rmesa->pClipRects = drawable->pBackClipRects;
-      }
-   }
-   else {
-     /* front buffer (or none, or multiple buffers) */
-     rmesa->numClipRects = drawable->numClipRects;
-     rmesa->pClipRects = drawable->pClipRects;
-  }
-
-   if ((draw_fb->Width != drawable->w) || (draw_fb->Height != drawable->h)) {
-      _mesa_resize_framebuffer(rmesa->glCtx, draw_fb,
-			       drawable->w, drawable->h);
-      draw_fb->Initialized = GL_TRUE;
-   }
-
-   if (drawable != readable) {
-      if ((read_fb->Width != readable->w) ||
-	  (read_fb->Height != readable->h)) {
-	 _mesa_resize_framebuffer(rmesa->glCtx, read_fb,
-				  readable->w, readable->h);
-	 read_fb->Initialized = GL_TRUE;
-      }
-   }
-
-   if (rmesa->state.scissor.enabled)
-      r200RecalcScissorRects( rmesa );
-
-   rmesa->lastStamp = drawable->lastStamp;
-}
-
-
-static void r200DrawBuffer( GLcontext *ctx, GLenum mode )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if (R200_DEBUG & DEBUG_DRI)
-      fprintf(stderr, "%s %s\n", __FUNCTION__,
-	      _mesa_lookup_enum_by_nr( mode ));
-
-   R200_FIREVERTICES(rmesa);	/* don't pipeline cliprect changes */
-
-   if (ctx->DrawBuffer->_NumColorDrawBuffers != 1) {
-      /* 0 (GL_NONE) buffers or multiple color drawing buffers */
-      FALLBACK( rmesa, R200_FALLBACK_DRAW_BUFFER, GL_TRUE );
-      return;
-   }
-
-   switch ( ctx->DrawBuffer->_ColorDrawBufferIndexes[0] ) {
-   case BUFFER_FRONT_LEFT:
-   case BUFFER_BACK_LEFT:
-      FALLBACK( rmesa, R200_FALLBACK_DRAW_BUFFER, GL_FALSE );
-      break;
-   default:
-      FALLBACK( rmesa, R200_FALLBACK_DRAW_BUFFER, GL_TRUE );
-      return;
-   }
-
-   r200SetCliprects( rmesa );
-
-   /* We'll set the drawing engine's offset/pitch parameters later
-    * when we update other state.
-    */
-}
-
-
-static void r200ReadBuffer( GLcontext *ctx, GLenum mode )
-{
-   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
-}
-
 /* =============================================================
  * State enable/disable
  */
@@ -1979,7 +1808,7 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_CLIP_PLANE2:
    case GL_CLIP_PLANE3:
    case GL_CLIP_PLANE4:
-   case GL_CLIP_PLANE5: 
+   case GL_CLIP_PLANE5:
       p = cap-GL_CLIP_PLANE0;
       R200_STATECHANGE( rmesa, tcl );
       if (state) {
@@ -2013,10 +1842,10 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
       R200_STATECHANGE(rmesa, ctx );
       if ( state ) {
 	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  R200_DITHER_ENABLE;
-	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~rmesa->state.color.roundEnable;
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~rmesa->radeon.state.color.roundEnable;
       } else {
 	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~R200_DITHER_ENABLE;
-	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  rmesa->state.color.roundEnable;
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  rmesa->radeon.state.color.roundEnable;
       }
       break;
 
@@ -2031,7 +1860,7 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
 	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~R200_TCL_FOG_MASK;
       }
       r200UpdateSpecular( ctx ); /* for PK_SPEC */
-      if (rmesa->TclFallback) 
+      if (rmesa->radeon.TclFallback)
 	 r200ChooseVertexState( ctx );
       _mesa_allow_light_in_model( ctx, !state );
       break;
@@ -2046,13 +1875,13 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_LIGHT7:
       R200_STATECHANGE(rmesa, tcl);
       p = cap - GL_LIGHT0;
-      if (p&1) 
+      if (p&1)
 	 flag = (R200_LIGHT_1_ENABLE |
-		 R200_LIGHT_1_ENABLE_AMBIENT | 
+		 R200_LIGHT_1_ENABLE_AMBIENT |
 		 R200_LIGHT_1_ENABLE_SPECULAR);
       else
 	 flag = (R200_LIGHT_0_ENABLE |
-		 R200_LIGHT_0_ENABLE_AMBIENT | 
+		 R200_LIGHT_0_ENABLE_AMBIENT |
 		 R200_LIGHT_0_ENABLE_SPECULAR);
 
       if (state)
@@ -2060,7 +1889,7 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
       else
 	 rmesa->hw.tcl.cmd[p/2 + TCL_PER_LIGHT_CTL_0] &= ~flag;
 
-      /* 
+      /*
        */
       update_light_colors( ctx, p );
       break;
@@ -2068,7 +1897,7 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_LIGHTING:
       r200UpdateSpecular(ctx);
       /* for reflection map fixup - might set recheck_texgen for all units too */
-      rmesa->NewGLState |= _NEW_TEXTURE;
+      rmesa->radeon.NewGLState |= _NEW_TEXTURE;
       break;
 
    case GL_LINE_SMOOTH:
@@ -2181,21 +2010,30 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
    }
 
    case GL_SCISSOR_TEST:
-      R200_FIREVERTICES( rmesa );
-      rmesa->state.scissor.enabled = state;
-      r200UpdateScissor( ctx );
+      radeon_firevertices(&rmesa->radeon);
+      rmesa->radeon.state.scissor.enabled = state;
+      radeonUpdateScissor( ctx );
       break;
 
    case GL_STENCIL_TEST:
-      if ( rmesa->state.stencil.hwBuffer ) {
-	 R200_STATECHANGE( rmesa, ctx );
-	 if ( state ) {
-	    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  R200_STENCIL_ENABLE;
+      {
+	 GLboolean hw_stencil = GL_FALSE;
+	 if (ctx->DrawBuffer) {
+	    struct radeon_renderbuffer *rrbStencil
+	       = radeon_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL);
+	    hw_stencil = (rrbStencil && rrbStencil->bo);
+	 }
+
+	 if (hw_stencil) {
+	    R200_STATECHANGE( rmesa, ctx );
+	    if ( state ) {
+	       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  R200_STENCIL_ENABLE;
+	    } else {
+	       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~R200_STENCIL_ENABLE;
+	    }
 	 } else {
-	    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~R200_STENCIL_ENABLE;
+	    FALLBACK( rmesa, R200_FALLBACK_STENCIL, state );
 	 }
-      } else {
-	 FALLBACK( rmesa, R200_FALLBACK_STENCIL, state );
       }
       break;
 
@@ -2205,7 +2043,7 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_TEXTURE_GEN_T:
       /* Picked up in r200UpdateTextureState.
        */
-      rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE; 
+      rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE;
       break;
 
    case GL_COLOR_SUM_EXT:
@@ -2322,7 +2160,7 @@ void r200LightingSpaceChange( GLcontext *ctx )
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
    GLboolean tmp;
 
-   if (R200_DEBUG & DEBUG_STATE) 
+   if (R200_DEBUG & DEBUG_STATE)
       fprintf(stderr, "%s %d BEFORE %x\n", __FUNCTION__, ctx->_NeedEyeCoords,
 	      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0]);
 
@@ -2338,7 +2176,7 @@ void r200LightingSpaceChange( GLcontext *ctx )
       rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0] &= ~R200_RESCALE_NORMALS;
    }
 
-   if (R200_DEBUG & DEBUG_STATE) 
+   if (R200_DEBUG & DEBUG_STATE)
       fprintf(stderr, "%s %d AFTER %x\n", __FUNCTION__, ctx->_NeedEyeCoords,
 	      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL_0]);
 }
@@ -2381,7 +2219,7 @@ static void update_texturematrix( GLcontext *ctx )
    GLuint compsel = rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_COMPSEL];
    int unit;
 
-   if (R200_DEBUG & DEBUG_STATE) 
+   if (R200_DEBUG & DEBUG_STATE)
       fprintf(stderr, "%s before COMPSEL: %x\n", __FUNCTION__,
 	      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_COMPSEL]);
 
@@ -2389,7 +2227,7 @@ static void update_texturematrix( GLcontext *ctx )
    rmesa->TexMatCompSel = 0;
 
    for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++) {
-      if (!ctx->Texture.Unit[unit]._ReallyEnabled) 
+      if (!ctx->Texture.Unit[unit]._ReallyEnabled)
 	 continue;
 
       if (ctx->TextureMatrixStack[unit].Top->type != MATRIX_IDENTITY) {
@@ -2399,21 +2237,21 @@ static void update_texturematrix( GLcontext *ctx )
 	 rmesa->TexMatCompSel |= R200_OUTPUT_TEX_0 << unit;
 
 	 if (rmesa->TexGenEnabled & (R200_TEXMAT_0_ENABLE << unit)) {
-	    /* Need to preconcatenate any active texgen 
+	    /* Need to preconcatenate any active texgen
 	     * obj/eyeplane matrices:
 	     */
 	    _math_matrix_mul_matrix( &rmesa->tmpmat,
-				     ctx->TextureMatrixStack[unit].Top, 
+				     ctx->TextureMatrixStack[unit].Top,
 				     &rmesa->TexGenMatrix[unit] );
 	    upload_matrix( rmesa, rmesa->tmpmat.m, R200_MTX_TEX0+unit );
-	 } 
+	 }
 	 else {
-	    upload_matrix( rmesa, ctx->TextureMatrixStack[unit].Top->m, 
+	    upload_matrix( rmesa, ctx->TextureMatrixStack[unit].Top->m,
 			   R200_MTX_TEX0+unit );
 	 }
       }
       else if (rmesa->TexGenEnabled & (R200_TEXMAT_0_ENABLE << unit)) {
-	 upload_matrix( rmesa, rmesa->TexGenMatrix[unit].m, 
+	 upload_matrix( rmesa, rmesa->TexGenMatrix[unit].m,
 			R200_MTX_TEX0+unit );
       }
    }
@@ -2432,69 +2270,78 @@ static void update_texturematrix( GLcontext *ctx )
    }
 }
 
-
-
-/**
- * Tell the card where to render (offset, pitch).
- * Effected by glDrawBuffer, etc
- */
-void
-r200UpdateDrawBuffer(GLcontext *ctx)
+static GLboolean r200ValidateBuffers(GLcontext *ctx)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_framebuffer *fb = ctx->DrawBuffer;
-   driRenderbuffer *drb;
+   struct radeon_renderbuffer *rrb;
+   int i;
 
-   if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) {
-      /* draw to front */
-      drb = (driRenderbuffer *) fb->Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
-   }
-   else if (fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) {
-      /* draw to back */
-      drb = (driRenderbuffer *) fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+   radeon_validate_reset_bos(&rmesa->radeon);
+
+   rrb = radeon_get_colorbuffer(&rmesa->radeon);
+   /* color buffer */
+   if (rrb && rrb->bo) {
+     radeon_validate_bo(&rmesa->radeon, rrb->bo,
+			0, RADEON_GEM_DOMAIN_VRAM);
    }
-   else {
-      /* drawing to multiple buffers, or none */
-      return;
+
+   /* depth buffer */
+   rrb = radeon_get_depthbuffer(&rmesa->radeon);
+   /* color buffer */
+   if (rrb && rrb->bo) {
+     radeon_validate_bo(&rmesa->radeon, rrb->bo,
+			0, RADEON_GEM_DOMAIN_VRAM);
    }
 
-   assert(drb);
-   assert(drb->flippedPitch);
+   for (i = 0; i < ctx->Const.MaxTextureImageUnits; ++i) {
+      radeonTexObj *t;
 
-   R200_STATECHANGE( rmesa, ctx );
+      if (!ctx->Texture.Unit[i]._ReallyEnabled)
+	 continue;
 
-   /* Note: we used the (possibly) page-flipped values */
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET]
-     = ((drb->flippedOffset + rmesa->r200Screen->fbLocation)
-	& R200_COLOROFFSET_MASK);
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = drb->flippedPitch;
-   if (rmesa->sarea->tiling_enabled) {
-      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= R200_COLOR_TILE_ENABLE;
+      t = radeon_tex_obj(ctx->Texture.Unit[i]._Current);
+      if (t->image_override && t->bo)
+	radeon_validate_bo(&rmesa->radeon, t->bo,
+			   RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
+      else if (t->mt->bo)
+	radeon_validate_bo(&rmesa->radeon, t->mt->bo,
+			   RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
    }
-}
 
+   if (rmesa->radeon.dma.current)
+      radeon_validate_bo(&rmesa->radeon, rmesa->radeon.dma.current, RADEON_GEM_DOMAIN_GTT, 0);
 
+   return radeon_revalidate_bos(ctx);
+}
 
-void r200ValidateState( GLcontext *ctx )
+GLboolean r200ValidateState( GLcontext *ctx )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLuint new_state = rmesa->NewGLState;
+   GLuint new_state = rmesa->radeon.NewGLState;
 
    if (new_state & _NEW_BUFFERS) {
-     r200UpdateDrawBuffer(ctx);
+      _mesa_update_framebuffer(ctx);
+      /* this updates the DrawBuffer's Width/Height if it's a FBO */
+      _mesa_update_draw_buffer_bounds(ctx);
+
+      R200_STATECHANGE(rmesa, ctx);
    }
 
    if (new_state & (_NEW_TEXTURE | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS)) {
       r200UpdateTextureState( ctx );
-      new_state |= rmesa->NewGLState; /* may add TEXTURE_MATRIX */
+      new_state |= rmesa->radeon.NewGLState; /* may add TEXTURE_MATRIX */
       r200UpdateLocalViewer( ctx );
    }
 
+   /* we need to do a space check here */
+   if (!r200ValidateBuffers(ctx))
+     return GL_FALSE;
+
 /* FIXME: don't really need most of these when vertex progs are enabled */
 
    /* Need an event driven matrix update?
     */
-   if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION)) 
+   if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION))
       upload_matrix( rmesa, ctx->_ModelProjectMatrix.m, R200_MTX_MVP );
 
    /* Need these for lighting (shouldn't upload otherwise)
@@ -2518,7 +2365,7 @@ void r200ValidateState( GLcontext *ctx )
    /* emit all active clip planes if projection matrix changes.
     */
    if (new_state & (_NEW_PROJECTION)) {
-      if (ctx->Transform.ClipPlanesEnabled) 
+      if (ctx->Transform.ClipPlanesEnabled)
 	 r200UpdateClipPlanes( ctx );
    }
 
@@ -2534,7 +2381,8 @@ void r200ValidateState( GLcontext *ctx )
       else TCL_FALLBACK(ctx, R200_TCL_FALLBACK_VERTEX_PROGRAM, 0);
    }
 
-   rmesa->NewGLState = 0;
+   rmesa->radeon.NewGLState = 0;
+   return GL_TRUE;
 }
 
 
@@ -2545,7 +2393,7 @@ static void r200InvalidateState( GLcontext *ctx, GLuint new_state )
    _vbo_InvalidateState( ctx, new_state );
    _tnl_InvalidateState( ctx, new_state );
    _ae_invalidate_state( ctx, new_state );
-   R200_CONTEXT(ctx)->NewGLState |= new_state;
+   R200_CONTEXT(ctx)->radeon.NewGLState |= new_state;
 }
 
 /* A hack.  The r200 can actually cope just fine with materials
@@ -2574,12 +2422,13 @@ static void r200WrapRunPipeline( GLcontext *ctx )
    GLboolean has_material;
 
    if (0)
-      fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->NewGLState);
+      fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->radeon.NewGLState);
 
    /* Validate state:
     */
-   if (rmesa->NewGLState)
-      r200ValidateState( ctx );
+   if (rmesa->radeon.NewGLState)
+      if (!r200ValidateState( ctx ))
+	 FALLBACK(rmesa, RADEON_FALLBACK_TEXTURE, GL_TRUE);
 
    has_material = !ctx->VertexProgram._Enabled && ctx->Light.Enabled && check_material( ctx );
 
@@ -2588,7 +2437,7 @@ static void r200WrapRunPipeline( GLcontext *ctx )
    }
 
    /* Run the pipeline.
-    */ 
+    */
    _tnl_run_pipeline( ctx );
 
    if (has_material) {
@@ -2604,8 +2453,8 @@ void r200InitStateFuncs( struct dd_function_table *functions )
    functions->UpdateState		= r200InvalidateState;
    functions->LightingSpaceChange	= r200LightingSpaceChange;
 
-   functions->DrawBuffer		= r200DrawBuffer;
-   functions->ReadBuffer		= r200ReadBuffer;
+   functions->DrawBuffer		= radeonDrawBuffer;
+   functions->ReadBuffer		= radeonReadBuffer;
 
    functions->AlphaFunc			= r200AlphaFunc;
    functions->BlendColor		= r200BlendColor;
@@ -2637,7 +2486,7 @@ void r200InitStateFuncs( struct dd_function_table *functions )
    functions->PointParameterfv		= r200PointParameter;
    functions->PointSize			= r200PointSize;
    functions->RenderMode		= r200RenderMode;
-   functions->Scissor			= r200Scissor;
+   functions->Scissor			= radeonScissor;
    functions->ShadeModel		= r200ShadeModel;
    functions->StencilFuncSeparate	= r200StencilFuncSeparate;
    functions->StencilMaskSeparate	= r200StencilMaskSeparate;
diff --git a/src/mesa/drivers/dri/r200/r200_state.h b/src/mesa/drivers/dri/r200/r200_state.h
index a917163a00a..23cf8aea667 100644
--- a/src/mesa/drivers/dri/r200/r200_state.h
+++ b/src/mesa/drivers/dri/r200/r200_state.h
@@ -43,23 +43,17 @@ extern void r200InitTnlFuncs( GLcontext *ctx );
 
 extern void r200UpdateMaterial( GLcontext *ctx );
 
-extern void r200SetCliprects( r200ContextPtr rmesa );
-extern void r200RecalcScissorRects( r200ContextPtr rmesa );
 extern void r200UpdateViewportOffset( GLcontext *ctx );
 extern void r200UpdateWindow( GLcontext *ctx );
 extern void r200UpdateDrawBuffer(GLcontext *ctx);
 
-extern void r200ValidateState( GLcontext *ctx );
-
-extern void r200PrintDirty( r200ContextPtr rmesa,
-			      const char *msg );
-
+extern GLboolean r200ValidateState( GLcontext *ctx );
 
 extern void r200Fallback( GLcontext *ctx, GLuint bit, GLboolean mode );
 #define FALLBACK( rmesa, bit, mode ) do {				\
    if ( 0 ) fprintf( stderr, "FALLBACK in %s: #%d=%d\n",		\
 		     __FUNCTION__, bit, mode );				\
-   r200Fallback( rmesa->glCtx, bit, mode );				\
+   r200Fallback( rmesa->radeon.glCtx, bit, mode );				\
 } while (0)
 
 extern void r200LightingSpaceChange( GLcontext *ctx );
diff --git a/src/mesa/drivers/dri/r200/r200_state_init.c b/src/mesa/drivers/dri/r200/r200_state_init.c
index 9e4677eda40..83920093378 100644
--- a/src/mesa/drivers/dri/r200/r200_state_init.c
+++ b/src/mesa/drivers/dri/r200/r200_state_init.c
@@ -43,6 +43,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/t_pipeline.h"
 #include "swrast_setup/swrast_setup.h"
 
+#include "radeon_common.h"
+#include "radeon_mipmap_tree.h"
 #include "r200_context.h"
 #include "r200_ioctl.h"
 #include "r200_state.h"
@@ -52,31 +54,129 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "xmlpool.h"
 
+/* New (1.3) state mechanism.  3 commands (packet, scalar, vector) in
+ * 1.3 cmdbuffers allow all previous state to be updated as well as
+ * the tcl scalar and vector areas.
+ */
+static struct {
+	int start;
+	int len;
+	const char *name;
+} packet[RADEON_MAX_STATE_PACKETS] = {
+	{RADEON_PP_MISC, 7, "RADEON_PP_MISC"},
+	{RADEON_PP_CNTL, 3, "RADEON_PP_CNTL"},
+	{RADEON_RB3D_COLORPITCH, 1, "RADEON_RB3D_COLORPITCH"},
+	{RADEON_RE_LINE_PATTERN, 2, "RADEON_RE_LINE_PATTERN"},
+	{RADEON_SE_LINE_WIDTH, 1, "RADEON_SE_LINE_WIDTH"},
+	{RADEON_PP_LUM_MATRIX, 1, "RADEON_PP_LUM_MATRIX"},
+	{RADEON_PP_ROT_MATRIX_0, 2, "RADEON_PP_ROT_MATRIX_0"},
+	{RADEON_RB3D_STENCILREFMASK, 3, "RADEON_RB3D_STENCILREFMASK"},
+	{RADEON_SE_VPORT_XSCALE, 6, "RADEON_SE_VPORT_XSCALE"},
+	{RADEON_SE_CNTL, 2, "RADEON_SE_CNTL"},
+	{RADEON_SE_CNTL_STATUS, 1, "RADEON_SE_CNTL_STATUS"},
+	{RADEON_RE_MISC, 1, "RADEON_RE_MISC"},
+	{RADEON_PP_TXFILTER_0, 6, "RADEON_PP_TXFILTER_0"},
+	{RADEON_PP_BORDER_COLOR_0, 1, "RADEON_PP_BORDER_COLOR_0"},
+	{RADEON_PP_TXFILTER_1, 6, "RADEON_PP_TXFILTER_1"},
+	{RADEON_PP_BORDER_COLOR_1, 1, "RADEON_PP_BORDER_COLOR_1"},
+	{RADEON_PP_TXFILTER_2, 6, "RADEON_PP_TXFILTER_2"},
+	{RADEON_PP_BORDER_COLOR_2, 1, "RADEON_PP_BORDER_COLOR_2"},
+	{RADEON_SE_ZBIAS_FACTOR, 2, "RADEON_SE_ZBIAS_FACTOR"},
+	{RADEON_SE_TCL_OUTPUT_VTX_FMT, 11, "RADEON_SE_TCL_OUTPUT_VTX_FMT"},
+	{RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, 17,
+		    "RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED"},
+	{R200_PP_TXCBLEND_0, 4, "R200_PP_TXCBLEND_0"},
+	{R200_PP_TXCBLEND_1, 4, "R200_PP_TXCBLEND_1"},
+	{R200_PP_TXCBLEND_2, 4, "R200_PP_TXCBLEND_2"},
+	{R200_PP_TXCBLEND_3, 4, "R200_PP_TXCBLEND_3"},
+	{R200_PP_TXCBLEND_4, 4, "R200_PP_TXCBLEND_4"},
+	{R200_PP_TXCBLEND_5, 4, "R200_PP_TXCBLEND_5"},
+	{R200_PP_TXCBLEND_6, 4, "R200_PP_TXCBLEND_6"},
+	{R200_PP_TXCBLEND_7, 4, "R200_PP_TXCBLEND_7"},
+	{R200_SE_TCL_LIGHT_MODEL_CTL_0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0"},
+	{R200_PP_TFACTOR_0, 6, "R200_PP_TFACTOR_0"},
+	{R200_SE_VTX_FMT_0, 4, "R200_SE_VTX_FMT_0"},
+	{R200_SE_VAP_CNTL, 1, "R200_SE_VAP_CNTL"},
+	{R200_SE_TCL_MATRIX_SEL_0, 5, "R200_SE_TCL_MATRIX_SEL_0"},
+	{R200_SE_TCL_TEX_PROC_CTL_2, 5, "R200_SE_TCL_TEX_PROC_CTL_2"},
+	{R200_SE_TCL_UCP_VERT_BLEND_CTL, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL"},
+	{R200_PP_TXFILTER_0, 6, "R200_PP_TXFILTER_0"},
+	{R200_PP_TXFILTER_1, 6, "R200_PP_TXFILTER_1"},
+	{R200_PP_TXFILTER_2, 6, "R200_PP_TXFILTER_2"},
+	{R200_PP_TXFILTER_3, 6, "R200_PP_TXFILTER_3"},
+	{R200_PP_TXFILTER_4, 6, "R200_PP_TXFILTER_4"},
+	{R200_PP_TXFILTER_5, 6, "R200_PP_TXFILTER_5"},
+	{R200_PP_TXOFFSET_0, 1, "R200_PP_TXOFFSET_0"},
+	{R200_PP_TXOFFSET_1, 1, "R200_PP_TXOFFSET_1"},
+	{R200_PP_TXOFFSET_2, 1, "R200_PP_TXOFFSET_2"},
+	{R200_PP_TXOFFSET_3, 1, "R200_PP_TXOFFSET_3"},
+	{R200_PP_TXOFFSET_4, 1, "R200_PP_TXOFFSET_4"},
+	{R200_PP_TXOFFSET_5, 1, "R200_PP_TXOFFSET_5"},
+	{R200_SE_VTE_CNTL, 1, "R200_SE_VTE_CNTL"},
+	{R200_SE_TCL_OUTPUT_VTX_COMP_SEL, 1,
+	 "R200_SE_TCL_OUTPUT_VTX_COMP_SEL"},
+	{R200_PP_TAM_DEBUG3, 1, "R200_PP_TAM_DEBUG3"},
+	{R200_PP_CNTL_X, 1, "R200_PP_CNTL_X"},
+	{R200_RB3D_DEPTHXY_OFFSET, 1, "R200_RB3D_DEPTHXY_OFFSET"},
+	{R200_RE_AUX_SCISSOR_CNTL, 1, "R200_RE_AUX_SCISSOR_CNTL"},
+	{R200_RE_SCISSOR_TL_0, 2, "R200_RE_SCISSOR_TL_0"},
+	{R200_RE_SCISSOR_TL_1, 2, "R200_RE_SCISSOR_TL_1"},
+	{R200_RE_SCISSOR_TL_2, 2, "R200_RE_SCISSOR_TL_2"},
+	{R200_SE_VAP_CNTL_STATUS, 1, "R200_SE_VAP_CNTL_STATUS"},
+	{R200_SE_VTX_STATE_CNTL, 1, "R200_SE_VTX_STATE_CNTL"},
+	{R200_RE_POINTSIZE, 1, "R200_RE_POINTSIZE"},
+	{R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0, 4,
+		    "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0"},
+	{R200_PP_CUBIC_FACES_0, 1, "R200_PP_CUBIC_FACES_0"},	/* 61 */
+	{R200_PP_CUBIC_OFFSET_F1_0, 5, "R200_PP_CUBIC_OFFSET_F1_0"}, /* 62 */
+	{R200_PP_CUBIC_FACES_1, 1, "R200_PP_CUBIC_FACES_1"},
+	{R200_PP_CUBIC_OFFSET_F1_1, 5, "R200_PP_CUBIC_OFFSET_F1_1"},
+	{R200_PP_CUBIC_FACES_2, 1, "R200_PP_CUBIC_FACES_2"},
+	{R200_PP_CUBIC_OFFSET_F1_2, 5, "R200_PP_CUBIC_OFFSET_F1_2"},
+	{R200_PP_CUBIC_FACES_3, 1, "R200_PP_CUBIC_FACES_3"},
+	{R200_PP_CUBIC_OFFSET_F1_3, 5, "R200_PP_CUBIC_OFFSET_F1_3"},
+	{R200_PP_CUBIC_FACES_4, 1, "R200_PP_CUBIC_FACES_4"},
+	{R200_PP_CUBIC_OFFSET_F1_4, 5, "R200_PP_CUBIC_OFFSET_F1_4"},
+	{R200_PP_CUBIC_FACES_5, 1, "R200_PP_CUBIC_FACES_5"},
+	{R200_PP_CUBIC_OFFSET_F1_5, 5, "R200_PP_CUBIC_OFFSET_F1_5"},
+	{RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0"},
+	{RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1"},
+	{RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_2"},
+	{R200_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"},
+	{R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"},
+	{RADEON_PP_CUBIC_FACES_0, 1, "RADEON_PP_CUBIC_FACES_0"},
+	{RADEON_PP_CUBIC_OFFSET_T0_0, 5, "RADEON_PP_CUBIC_OFFSET_T0_0"},
+	{RADEON_PP_CUBIC_FACES_1, 1, "RADEON_PP_CUBIC_FACES_1"},
+	{RADEON_PP_CUBIC_OFFSET_T1_0, 5, "RADEON_PP_CUBIC_OFFSET_T1_0"},
+	{RADEON_PP_CUBIC_FACES_2, 1, "RADEON_PP_CUBIC_FACES_2"},
+	{RADEON_PP_CUBIC_OFFSET_T2_0, 5, "RADEON_PP_CUBIC_OFFSET_T2_0"},
+	{R200_PP_TRI_PERF, 2, "R200_PP_TRI_PERF"},
+	{R200_PP_TXCBLEND_8, 32, "R200_PP_AFS_0"},     /* 85 */
+	{R200_PP_TXCBLEND_0, 32, "R200_PP_AFS_1"},
+	{R200_PP_TFACTOR_0, 8, "R200_ATF_TFACTOR"},
+	{R200_PP_TXFILTER_0, 8, "R200_PP_TXCTLALL_0"},
+	{R200_PP_TXFILTER_1, 8, "R200_PP_TXCTLALL_1"},
+	{R200_PP_TXFILTER_2, 8, "R200_PP_TXCTLALL_2"},
+	{R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"},
+	{R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"},
+	{R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"},
+	{R200_VAP_PVS_CNTL_1, 2, "R200_VAP_PVS_CNTL"},
+};
+
 /* =============================================================
  * State initialization
  */
-
-void r200PrintDirty( r200ContextPtr rmesa, const char *msg )
+static int cmdpkt( r200ContextPtr rmesa, int id ) 
 {
-   struct r200_state_atom *l;
-
-   fprintf(stderr, msg);
-   fprintf(stderr, ": ");
+   drm_radeon_cmd_header_t h;
 
-   foreach(l, &rmesa->hw.atomlist) {
-      if (l->dirty || rmesa->hw.all_dirty)
-	 fprintf(stderr, "%s, ", l->name);
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+     return CP_PACKET0(packet[id].start, packet[id].len - 1);
+   } else {
+     h.i = 0;
+     h.packet.cmd_type = RADEON_CMD_PACKET;
+     h.packet.packet_id = id;
    }
-
-   fprintf(stderr, "\n");
-}
-
-static int cmdpkt( int id ) 
-{
-   drm_radeon_cmd_header_t h;
-   h.i = 0;
-   h.packet.cmd_type = RADEON_CMD_PACKET;
-   h.packet.packet_id = id;
    return h.i;
 }
 
@@ -127,150 +227,467 @@ static int cmdscl2( int offset, int stride, int count )
 }
 
 #define CHECK( NM, FLAG )				\
-static GLboolean check_##NM( GLcontext *ctx, int idx )	\
+static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom) \
 {							\
    r200ContextPtr rmesa = R200_CONTEXT(ctx);		\
-   (void) idx;						\
    (void) rmesa;					\
-   return FLAG;						\
+   return (FLAG) ? atom->cmd_size : 0;			\
 }
 
 #define TCL_CHECK( NM, FLAG )				\
-static GLboolean check_##NM( GLcontext *ctx, int idx )	\
-{							\
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);		\
-   (void) idx;						\
-   return !rmesa->TclFallback && !ctx->VertexProgram._Enabled && (FLAG);	\
+static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom) \
+{									\
+   r200ContextPtr rmesa = R200_CONTEXT(ctx);				\
+   return (!rmesa->radeon.TclFallback && !ctx->VertexProgram._Enabled && (FLAG)) ? atom->cmd_size : 0; \
 }
 
 #define TCL_OR_VP_CHECK( NM, FLAG )			\
-static GLboolean check_##NM( GLcontext *ctx, int idx )	\
+static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \
 {							\
    r200ContextPtr rmesa = R200_CONTEXT(ctx);		\
-   (void) idx;						\
-   return !rmesa->TclFallback && (FLAG);		\
+   return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size : 0;	\
 }
 
 #define VP_CHECK( NM, FLAG )				\
-static GLboolean check_##NM( GLcontext *ctx, int idx )	\
-{							\
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);		\
-   (void) idx;						\
-   return !rmesa->TclFallback && ctx->VertexProgram._Enabled && (FLAG);		\
+static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom ) \
+{									\
+   r200ContextPtr rmesa = R200_CONTEXT(ctx);				\
+   (void) atom;								\
+   return (!rmesa->radeon.TclFallback && ctx->VertexProgram._Enabled && (FLAG)) ? atom->cmd_size : 0; \
 }
 
-
 CHECK( always, GL_TRUE )
 CHECK( never, GL_FALSE )
 CHECK( tex_any, ctx->Texture._EnabledUnits )
 CHECK( tf, (ctx->Texture._EnabledUnits && !ctx->ATIFragmentShader._Enabled) );
-CHECK( tex_pair, (rmesa->state.texture.unit[idx].unitneeded | rmesa->state.texture.unit[idx & ~1].unitneeded) )
-CHECK( tex, rmesa->state.texture.unit[idx].unitneeded )
+CHECK( tex_pair, (rmesa->state.texture.unit[atom->idx].unitneeded | rmesa->state.texture.unit[atom->idx & ~1].unitneeded) )
+CHECK( tex, rmesa->state.texture.unit[atom->idx].unitneeded )
 CHECK( pix_zero, !ctx->ATIFragmentShader._Enabled )
-CHECK( texenv, (rmesa->state.envneeded & (1 << idx) && !ctx->ATIFragmentShader._Enabled) )
+   CHECK( texenv, (rmesa->state.envneeded & (1 << (atom->idx)) && !ctx->ATIFragmentShader._Enabled) )
 CHECK( afs_pass1, (ctx->ATIFragmentShader._Enabled && (ctx->ATIFragmentShader.Current->NumPasses > 1)) )
 CHECK( afs, ctx->ATIFragmentShader._Enabled )
-CHECK( tex_cube, rmesa->state.texture.unit[idx].unitneeded & TEXTURE_CUBE_BIT )
+CHECK( tex_cube, rmesa->state.texture.unit[atom->idx].unitneeded & TEXTURE_CUBE_BIT )
 TCL_CHECK( tcl_fog, ctx->Fog.Enabled )
 TCL_CHECK( tcl, GL_TRUE )
-TCL_CHECK( tcl_tex, rmesa->state.texture.unit[idx].unitneeded )
+TCL_CHECK( tcl_tex, rmesa->state.texture.unit[atom->idx].unitneeded )
 TCL_CHECK( tcl_lighting, ctx->Light.Enabled )
-TCL_CHECK( tcl_light, ctx->Light.Enabled && ctx->Light.Light[idx].Enabled )
-TCL_OR_VP_CHECK( tcl_ucp, (ctx->Transform.ClipPlanesEnabled & (1 << idx)) )
+TCL_CHECK( tcl_light, ctx->Light.Enabled && ctx->Light.Light[atom->idx].Enabled )
+TCL_OR_VP_CHECK( tcl_ucp, (ctx->Transform.ClipPlanesEnabled & (1 << (atom->idx))) )
 TCL_OR_VP_CHECK( tcl_or_vp, GL_TRUE )
 VP_CHECK( tcl_vp, GL_TRUE )
 VP_CHECK( tcl_vp_size, ctx->VertexProgram.Current->Base.NumNativeInstructions > 64 )
 VP_CHECK( tcl_vpp_size, ctx->VertexProgram.Current->Base.NumNativeParameters > 96 )
 
+#define OUT_VEC(hdr, data) do {			\
+    drm_radeon_cmd_header_t h;					\
+    h.i = hdr;								\
+    OUT_BATCH(CP_PACKET0(RADEON_SE_TCL_STATE_FLUSH, 0));		\
+    OUT_BATCH(0);							\
+    OUT_BATCH(CP_PACKET0(R200_SE_TCL_VECTOR_INDX_REG, 0));		\
+    OUT_BATCH(h.vectors.offset | (h.vectors.stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT)); \
+    OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_VECTOR_DATA_REG, h.vectors.count - 1));	\
+    OUT_BATCH_TABLE((data), h.vectors.count);				\
+  } while(0)
+
+#define OUT_VECLINEAR(hdr, data) do {					\
+    drm_radeon_cmd_header_t h;						\
+    uint32_t _start, _sz;						\
+    h.i = hdr;								\
+    _start = h.veclinear.addr_lo | (h.veclinear.addr_hi << 8);		\
+    _sz = h.veclinear.count * 4;					\
+    if (r200->radeon.radeonScreen->kernel_mm && _sz) { \
+    BEGIN_BATCH_NO_AUTOSTATE(dwords); \
+    OUT_BATCH(CP_PACKET0(RADEON_SE_TCL_STATE_FLUSH, 0));		\
+    OUT_BATCH(0);							\
+    OUT_BATCH(CP_PACKET0(R200_SE_TCL_VECTOR_INDX_REG, 0));		\
+    OUT_BATCH(_start | (1 << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));	\
+    OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_VECTOR_DATA_REG, _sz - 1));	\
+    OUT_BATCH_TABLE((data), _sz);					\
+    END_BATCH(); \
+    } \
+  } while(0)
+
+#define OUT_SCL(hdr, data) do {					\
+    drm_radeon_cmd_header_t h;						\
+    h.i = hdr;								\
+    OUT_BATCH(CP_PACKET0(R200_SE_TCL_SCALAR_INDX_REG, 0));		\
+    OUT_BATCH((h.scalars.offset) | (h.scalars.stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT)); \
+    OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_SCALAR_DATA_REG, h.scalars.count - 1));	\
+    OUT_BATCH_TABLE((data), h.scalars.count);				\
+  } while(0)
+
+#define OUT_SCL2(hdr, data) do {					\
+    drm_radeon_cmd_header_t h;						\
+    h.i = hdr;								\
+    OUT_BATCH(CP_PACKET0(R200_SE_TCL_SCALAR_INDX_REG, 0));		\
+    OUT_BATCH((h.scalars.offset + 0x100) | (h.scalars.stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT)); \
+    OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_SCALAR_DATA_REG, h.scalars.count - 1));	\
+    OUT_BATCH_TABLE((data), h.scalars.count);				\
+  } while(0)
+
+static void mtl_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 6;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_VEC(atom->cmd[MTL_CMD_0], (atom->cmd+1));
+   OUT_SCL2(atom->cmd[MTL_CMD_1], (atom->cmd + 18));
+   END_BATCH();
+}
 
-/* Initialize the context's hardware state.
- */
-void r200InitState( r200ContextPtr rmesa )
+static void lit_emit(GLcontext *ctx, struct radeon_state_atom *atom)
 {
-   GLcontext *ctx = rmesa->glCtx;
-   GLuint color_fmt, depth_fmt, i;
-   GLint drawPitch, drawOffset;
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 8;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_VEC(atom->cmd[LIT_CMD_0], atom->cmd+1);
+   OUT_VEC(atom->cmd[LIT_CMD_1], atom->cmd+LIT_CMD_1+1);
+   END_BATCH();
+}
 
-   switch ( rmesa->r200Screen->cpp ) {
-   case 2:
-      color_fmt = R200_COLOR_FORMAT_RGB565;
-      break;
-   case 4:
-      color_fmt = R200_COLOR_FORMAT_ARGB8888;
-      break;
-   default:
-      fprintf( stderr, "Error: Unsupported pixel depth... exiting\n" );
-      exit( -1 );
+static void ptp_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 8;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_VEC(atom->cmd[PTP_CMD_0], atom->cmd+1);
+   OUT_VEC(atom->cmd[PTP_CMD_1], atom->cmd+PTP_CMD_1+1);
+   END_BATCH();
+}
+
+static void veclinear_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 4;
+   OUT_VECLINEAR(atom->cmd[0], atom->cmd+1);
+}
+
+static void scl_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 2;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_SCL(atom->cmd[0], atom->cmd+1);
+   END_BATCH();
+}
+
+
+static void vec_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 4;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_VEC(atom->cmd[0], atom->cmd+1);
+   END_BATCH();
+}
+
+static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   struct radeon_renderbuffer *rrb;
+   uint32_t cbpitch;
+   uint32_t zbpitch, depth_fmt;
+   uint32_t dwords = atom->cmd_size;
+
+   /* output the first 7 bytes of context */
+   BEGIN_BATCH_NO_AUTOSTATE(dwords+2+2);
+   OUT_BATCH_TABLE(atom->cmd, 5);
+
+   rrb = radeon_get_depthbuffer(&r200->radeon);
+   if (!rrb) {
+     OUT_BATCH(0);
+     OUT_BATCH(0);
+   } else {
+     zbpitch = (rrb->pitch / rrb->cpp);
+     if (r200->using_hyperz)
+       zbpitch |= RADEON_DEPTH_HYPERZ;
+     OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+     OUT_BATCH(zbpitch);
+     if (rrb->cpp == 4) 
+       depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z; 
+     else 
+       depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z; 
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK; 
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt; 
+   }
+     
+   OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]);
+   OUT_BATCH(atom->cmd[CTX_CMD_1]);
+   OUT_BATCH(atom->cmd[CTX_PP_CNTL]);
+
+   rrb = radeon_get_colorbuffer(&r200->radeon);
+   if (!rrb || !rrb->bo) {
+     OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
+     OUT_BATCH(atom->cmd[CTX_RB3D_COLOROFFSET]);
+   } else {
+     atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10); 
+     if (rrb->cpp == 4) 
+       atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888; 
+     else 
+       atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565; 
+ 
+     OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]); 
+     OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
    }
 
-   rmesa->state.color.clear = 0x00000000;
+   OUT_BATCH(atom->cmd[CTX_CMD_2]);
 
-   switch ( ctx->Visual.depthBits ) {
-   case 16:
-      rmesa->state.depth.clear = 0x0000ffff;
-      rmesa->state.depth.scale = 1.0 / (GLfloat)0xffff;
-      depth_fmt = R200_DEPTH_FORMAT_16BIT_INT_Z;
-      rmesa->state.stencil.clear = 0x00000000;
-      break;
-   case 24:
-      rmesa->state.depth.clear = 0x00ffffff;
-      rmesa->state.depth.scale = 1.0 / (GLfloat)0xffffff;
-      depth_fmt = R200_DEPTH_FORMAT_24BIT_INT_Z;
-      rmesa->state.stencil.clear = 0xffff0000;
-      break;
-   default:
-      fprintf( stderr, "Error: Unsupported depth %d... exiting\n",
-	       ctx->Visual.depthBits );
-      exit( -1 );
+   if (!rrb || !rrb->bo) {
+     OUT_BATCH(atom->cmd[CTX_RB3D_COLORPITCH]);
+   } else {
+     cbpitch = (rrb->pitch / rrb->cpp);
+     if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
+       cbpitch |= R200_COLOR_TILE_ENABLE;
+     OUT_BATCH(cbpitch);
    }
 
-   /* Only have hw stencil when depth buffer is 24 bits deep */
-   rmesa->state.stencil.hwBuffer = ( ctx->Visual.stencilBits > 0 &&
-				     ctx->Visual.depthBits == 24 );
+   if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM)
+     OUT_BATCH_TABLE((atom->cmd + 14), 4);
 
-   rmesa->Fallback = 0;
+   END_BATCH();
+}
 
-   if ( ctx->Visual.doubleBufferMode && rmesa->sarea->pfCurrentPage == 0 ) {
-      drawOffset = rmesa->r200Screen->backOffset;
-      drawPitch  = rmesa->r200Screen->backPitch;
-   } else {
-      drawOffset = rmesa->r200Screen->frontOffset;
-      drawPitch  = rmesa->r200Screen->frontPitch;
+static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   struct radeon_renderbuffer *rrb, *drb;
+   uint32_t cbpitch = 0;
+   uint32_t zbpitch = 0;
+   uint32_t dwords = atom->cmd_size;
+   uint32_t depth_fmt;
+
+   rrb = radeon_get_colorbuffer(&r200->radeon);
+   if (!rrb || !rrb->bo) {
+      return;
+   }
+
+   atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10);
+   if (rrb->cpp == 4)
+	atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888;
+   else
+	atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565;
+
+   cbpitch = (rrb->pitch / rrb->cpp);
+   if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
+       cbpitch |= R200_COLOR_TILE_ENABLE;
+
+   drb = radeon_get_depthbuffer(&r200->radeon);
+   if (drb) {
+     zbpitch = (drb->pitch / drb->cpp);
+     if (drb->cpp == 4)
+        depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
+     else
+        depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK;
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt;
+   }
+
+   dwords = 10;
+   if (drb)
+     dwords += 6;
+   if (rrb)
+     dwords += 6;
+   if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM)
+     dwords += 4;
+
+   /* output the first 7 bytes of context */
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+
+   /* In the CS case we need to split this up */
+   OUT_BATCH(CP_PACKET0(packet[0].start, 3));
+   OUT_BATCH_TABLE((atom->cmd + 1), 4);
+
+   if (drb) {
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHOFFSET, 0));
+     OUT_BATCH_RELOC(0, drb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHPITCH, 0));
+     OUT_BATCH(zbpitch);
+   }
+
+   OUT_BATCH(CP_PACKET0(RADEON_RB3D_ZSTENCILCNTL, 0));
+   OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]);
+   OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 1));
+   OUT_BATCH(atom->cmd[CTX_PP_CNTL]);
+   OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
+
+
+   if (rrb) {
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLOROFFSET, 0));
+     OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0));
+     OUT_BATCH(cbpitch);
+   }
+
+   if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) {
+     OUT_BATCH_TABLE((atom->cmd + 14), 4);
    }
-#if 000
-   if ( ctx->Visual.doubleBufferMode && rmesa->sarea->pfCurrentPage == 0 ) {
-      rmesa->state.color.drawOffset = rmesa->r200Screen->backOffset;
-      rmesa->state.color.drawPitch  = rmesa->r200Screen->backPitch;
+
+   END_BATCH();
+}
+
+static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+   int i = atom->idx;
+   radeonTexObj *t = r200->state.texture.unit[i].texobj;
+   radeon_mipmap_level *lvl;
+
+   if (t && t->mt && !t->image_override)
+     dwords += 2;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   /* is this ok even with drm older than 1.18? */
+   OUT_BATCH_TABLE(atom->cmd, 10);
+
+   if (t && t->mt && !t->image_override) {
+     OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+		  RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+   } else if (!t) {
+     /* workaround for old CS mechanism */
+     OUT_BATCH(r200->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]);
    } else {
-      rmesa->state.color.drawOffset = rmesa->r200Screen->frontOffset;
-      rmesa->state.color.drawPitch  = rmesa->r200Screen->frontPitch;
+     OUT_BATCH(t->override_offset);
    }
 
-   rmesa->state.pixel.readOffset = rmesa->state.color.drawOffset;
-   rmesa->state.pixel.readPitch  = rmesa->state.color.drawPitch;
-#endif
+   END_BATCH();
+}
+
+static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = atom->cmd_size;
+   int i = atom->idx;
+   radeonTexObj *t = r200->state.texture.unit[i].texobj;
+   radeon_mipmap_level *lvl;
+   int hastexture = 1;
+
+   if (!t)
+	hastexture = 0;
+   else {
+	if (!t->mt && !t->bo)
+		hastexture = 0;
+   }
+
+   if (hastexture)
+     dwords += 2;
+   else
+     dwords -= 2;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+
+   OUT_BATCH(CP_PACKET0(R200_PP_TXFILTER_0 + (32 * i), 7));
+   OUT_BATCH_TABLE((atom->cmd + 1), 8);
+
+   if (hastexture) {
+     OUT_BATCH(CP_PACKET0(R200_PP_TXOFFSET_0 + (24 * i), 0));
+     if (t->mt && !t->image_override) {
+        OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+		  RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+      } else {
+	if (t->bo)
+            OUT_BATCH_RELOC(t->tile_bits, t->bo, 0,
+                            RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+      }
+   }
+   END_BATCH();
+}
 
-   rmesa->hw.max_state_size = 0;
+
+static void cube_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r200ContextPtr r200 = R200_CONTEXT(ctx);
+   BATCH_LOCALS(&r200->radeon);
+   uint32_t dwords = 2;
+   int i = atom->idx, j;
+   radeonTexObj *t = r200->state.texture.unit[i].texobj;
+   radeon_mipmap_level *lvl;
+
+   BEGIN_BATCH_NO_AUTOSTATE(dwords + (4 * 5));
+   OUT_BATCH_TABLE(atom->cmd, 2);
+
+   if (t && !t->image_override) {
+     lvl = &t->mt->levels[0];
+     for (j = 1; j <= 5; j++) {
+       OUT_BATCH(CP_PACKET0(R200_PP_CUBIC_OFFSET_F1_0 + (24*i) + (4 * (j-1)), 0));
+       OUT_BATCH_RELOC(lvl->faces[j].offset, t->mt->bo, lvl->faces[j].offset,
+			RADEON_GEM_DOMAIN_VRAM, 0, 0);
+     }
+   }
+   END_BATCH();
+}
+
+/* Initialize the context's hardware state.
+ */
+void r200InitState( r200ContextPtr rmesa )
+{
+   GLcontext *ctx = rmesa->radeon.glCtx;
+   GLuint i;
+
+   rmesa->radeon.state.color.clear = 0x00000000;
+
+   switch ( ctx->Visual.depthBits ) {
+   case 16:
+      rmesa->radeon.state.depth.clear = 0x0000ffff;
+      rmesa->radeon.state.stencil.clear = 0x00000000;
+      break;
+   case 24:
+   default:
+      rmesa->radeon.state.depth.clear = 0x00ffffff;
+      rmesa->radeon.state.stencil.clear = 0xffff0000;
+      break;
+   }
+
+   rmesa->radeon.Fallback = 0;
+
+   rmesa->radeon.hw.max_state_size = 0;
 
 #define ALLOC_STATE( ATOM, CHK, SZ, NM, IDX )				\
    do {								\
       rmesa->hw.ATOM.cmd_size = SZ;				\
-      rmesa->hw.ATOM.cmd = (int *)CALLOC(SZ * sizeof(int));	\
-      rmesa->hw.ATOM.lastcmd = (int *)CALLOC(SZ * sizeof(int));	\
+      rmesa->hw.ATOM.cmd = (GLuint *)CALLOC(SZ * sizeof(int));	\
+      rmesa->hw.ATOM.lastcmd = (GLuint *)CALLOC(SZ * sizeof(int));	\
       rmesa->hw.ATOM.name = NM;					\
       rmesa->hw.ATOM.idx = IDX;					\
       rmesa->hw.ATOM.check = check_##CHK;			\
       rmesa->hw.ATOM.dirty = GL_FALSE;				\
-      rmesa->hw.max_state_size += SZ * sizeof(int);		\
+      rmesa->radeon.hw.max_state_size += SZ * sizeof(int);		\
    } while (0)
 
 
    /* Allocate state buffers:
     */
-   if (rmesa->r200Screen->drmSupportsBlendColor)
+   if (rmesa->radeon.radeonScreen->drmSupportsBlendColor)
       ALLOC_STATE( ctx, always, CTX_STATE_SIZE_NEWDRM, "CTX/context", 0 );
    else
       ALLOC_STATE( ctx, always, CTX_STATE_SIZE_OLDDRM, "CTX/context", 0 );
+
+   if (rmesa->radeon.radeonScreen->kernel_mm)
+     rmesa->hw.ctx.emit = ctx_emit_cs;
+   else
+     rmesa->hw.ctx.emit = ctx_emit;
    ALLOC_STATE( set, always, SET_STATE_SIZE, "SET/setup", 0 );
    ALLOC_STATE( lin, always, LIN_STATE_SIZE, "LIN/line", 0 );
    ALLOC_STATE( msk, always, MSK_STATE_SIZE, "MSK/mask", 0 );
@@ -282,8 +699,8 @@ void r200InitState( r200ContextPtr rmesa )
    ALLOC_STATE( cst, always, CST_STATE_SIZE, "CST/constant", 0 );
    ALLOC_STATE( zbs, always, ZBS_STATE_SIZE, "ZBS/zbias", 0 );
    ALLOC_STATE( tf, tf, TF_STATE_SIZE, "TF/tfactor", 0 );
-   if (rmesa->r200Screen->drmSupportsFragShader) {
-      if (rmesa->r200Screen->chip_family == CHIP_FAMILY_R200) {
+   if (rmesa->radeon.radeonScreen->drmSupportsFragShader) {
+      if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) {
       /* make sure texture units 0/1 are emitted pair-wise for r200 t0 hang workaround */
 	 ALLOC_STATE( tex[0], tex_pair, TEX_STATE_SIZE_NEWDRM, "TEX/tex-0", 0 );
 	 ALLOC_STATE( tex[1], tex_pair, TEX_STATE_SIZE_NEWDRM, "TEX/tex-1", 1 );
@@ -303,7 +720,7 @@ void r200InitState( r200ContextPtr rmesa )
       ALLOC_STATE( afs[1], afs, AFS_STATE_SIZE, "AFS/afsinst-1", 1 );
    }
    else {
-      if (rmesa->r200Screen->chip_family == CHIP_FAMILY_R200) {
+      if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) {
 	 ALLOC_STATE( tex[0], tex_pair, TEX_STATE_SIZE_OLDDRM, "TEX/tex-0", 0 );
 	 ALLOC_STATE( tex[1], tex_pair, TEX_STATE_SIZE_OLDDRM, "TEX/tex-1", 1 );
 	 ALLOC_STATE( tam, tex_any, TAM_STATE_SIZE, "TAM/tam", 0 );
@@ -321,13 +738,21 @@ void r200InitState( r200ContextPtr rmesa )
       ALLOC_STATE( afs[0], never, AFS_STATE_SIZE, "AFS/afsinst-0", 0 );
       ALLOC_STATE( afs[1], never, AFS_STATE_SIZE, "AFS/afsinst-1", 1 );
    }
-   if (rmesa->r200Screen->drmSupportsCubeMapsR200) {
+
+   for (i = 0; i < 5; i++)
+      if (rmesa->radeon.radeonScreen->kernel_mm)
+          rmesa->hw.tex[i].emit = tex_emit_cs;
+      else
+          rmesa->hw.tex[i].emit = tex_emit;
+   if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR200) {
       ALLOC_STATE( cube[0], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-0", 0 );
       ALLOC_STATE( cube[1], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-1", 1 );
       ALLOC_STATE( cube[2], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-2", 2 );
       ALLOC_STATE( cube[3], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-3", 3 );
       ALLOC_STATE( cube[4], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-4", 4 );
       ALLOC_STATE( cube[5], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-5", 5 );
+      for (i = 0; i < 5; i++)
+	rmesa->hw.cube[i].emit = cube_emit;
    }
    else {
       ALLOC_STATE( cube[0], never, CUBE_STATE_SIZE, "CUBE/tex-0", 0 );
@@ -337,7 +762,8 @@ void r200InitState( r200ContextPtr rmesa )
       ALLOC_STATE( cube[4], never, CUBE_STATE_SIZE, "CUBE/tex-4", 4 );
       ALLOC_STATE( cube[5], never, CUBE_STATE_SIZE, "CUBE/tex-5", 5 );
    }
-   if (rmesa->r200Screen->drmSupportsVertexProgram) {
+
+   if (rmesa->radeon.radeonScreen->drmSupportsVertexProgram) {
       ALLOC_STATE( pvs, tcl_vp, PVS_STATE_SIZE, "PVS/pvscntl", 0 );
       ALLOC_STATE( vpi[0], tcl_vp, VPI_STATE_SIZE, "VP/vertexprog-0", 0 );
       ALLOC_STATE( vpi[1], tcl_vp_size, VPI_STATE_SIZE, "VP/vertexprog-1", 1 );
@@ -390,13 +816,13 @@ void r200InitState( r200ContextPtr rmesa )
    ALLOC_STATE( pix[3], texenv, PIX_STATE_SIZE, "PIX/pixstage-3", 3 );
    ALLOC_STATE( pix[4], texenv, PIX_STATE_SIZE, "PIX/pixstage-4", 4 );
    ALLOC_STATE( pix[5], texenv, PIX_STATE_SIZE, "PIX/pixstage-5", 5 );
-   if (rmesa->r200Screen->drmSupportsTriPerf) {
+   if (rmesa->radeon.radeonScreen->drmSupportsTriPerf) {
       ALLOC_STATE( prf, always, PRF_STATE_SIZE, "PRF/performance-tri", 0 );
    }
    else {
       ALLOC_STATE( prf, never, PRF_STATE_SIZE, "PRF/performance-tri", 0 );
    }
-   if (rmesa->r200Screen->drmSupportsPointSprites) {
+   if (rmesa->radeon.radeonScreen->drmSupportsPointSprites) {
       ALLOC_STATE( spr, always, SPR_STATE_SIZE, "SPR/pointsprite", 0 );
       ALLOC_STATE( ptp, tcl, PTP_STATE_SIZE, "PTP/pointparams", 0 );
    }
@@ -409,87 +835,115 @@ void r200InitState( r200ContextPtr rmesa )
 
    /* Fill in the packet headers:
     */
-   rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(RADEON_EMIT_PP_MISC);
-   rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(RADEON_EMIT_PP_CNTL);
-   rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(RADEON_EMIT_RB3D_COLORPITCH);
-   if (rmesa->r200Screen->drmSupportsBlendColor)
-      rmesa->hw.ctx.cmd[CTX_CMD_3] = cmdpkt(R200_EMIT_RB3D_BLENDCOLOR);
-   rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(RADEON_EMIT_RE_LINE_PATTERN);
-   rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(RADEON_EMIT_SE_LINE_WIDTH);
-   rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(RADEON_EMIT_RB3D_STENCILREFMASK);
-   rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(RADEON_EMIT_SE_VPORT_XSCALE);
-   rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(RADEON_EMIT_SE_CNTL);
-   rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(RADEON_EMIT_RE_MISC);
-   rmesa->hw.cst.cmd[CST_CMD_0] = cmdpkt(R200_EMIT_PP_CNTL_X);
-   rmesa->hw.cst.cmd[CST_CMD_1] = cmdpkt(R200_EMIT_RB3D_DEPTHXY_OFFSET);
-   rmesa->hw.cst.cmd[CST_CMD_2] = cmdpkt(R200_EMIT_RE_AUX_SCISSOR_CNTL);
-   rmesa->hw.cst.cmd[CST_CMD_3] = cmdpkt(R200_EMIT_RE_SCISSOR_TL_0);
-   rmesa->hw.cst.cmd[CST_CMD_4] = cmdpkt(R200_EMIT_SE_VAP_CNTL_STATUS);
-   rmesa->hw.cst.cmd[CST_CMD_5] = cmdpkt(R200_EMIT_RE_POINTSIZE);
-   rmesa->hw.cst.cmd[CST_CMD_6] = cmdpkt(R200_EMIT_TCL_INPUT_VTX_VECTOR_ADDR_0);
-   rmesa->hw.tam.cmd[TAM_CMD_0] = cmdpkt(R200_EMIT_PP_TAM_DEBUG3);
-   rmesa->hw.tf.cmd[TF_CMD_0] = cmdpkt(R200_EMIT_TFACTOR_0);
-   if (rmesa->r200Screen->drmSupportsFragShader) {
-      rmesa->hw.atf.cmd[ATF_CMD_0] = cmdpkt(R200_EMIT_ATF_TFACTOR);
-      rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCTLALL_0);
-      rmesa->hw.tex[0].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_0);
-      rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCTLALL_1);
-      rmesa->hw.tex[1].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_1);
-      rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCTLALL_2);
-      rmesa->hw.tex[2].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_2);
-      rmesa->hw.tex[3].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCTLALL_3);
-      rmesa->hw.tex[3].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_3);
-      rmesa->hw.tex[4].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCTLALL_4);
-      rmesa->hw.tex[4].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_4);
-      rmesa->hw.tex[5].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCTLALL_5);
-      rmesa->hw.tex[5].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_5);
+   rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_MISC);
+   rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CNTL);
+   rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(rmesa, RADEON_EMIT_RB3D_COLORPITCH);
+   if (rmesa->radeon.radeonScreen->drmSupportsBlendColor)
+      rmesa->hw.ctx.cmd[CTX_CMD_3] = cmdpkt(rmesa, R200_EMIT_RB3D_BLENDCOLOR);
+   rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RE_LINE_PATTERN);
+   rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_SE_LINE_WIDTH);
+   rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RB3D_STENCILREFMASK);
+   rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_VPORT_XSCALE);
+   rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_CNTL);
+   rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RE_MISC);
+   rmesa->hw.cst.cmd[CST_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CNTL_X);
+   rmesa->hw.cst.cmd[CST_CMD_1] = cmdpkt(rmesa, R200_EMIT_RB3D_DEPTHXY_OFFSET);
+   rmesa->hw.cst.cmd[CST_CMD_2] = cmdpkt(rmesa, R200_EMIT_RE_AUX_SCISSOR_CNTL);
+   rmesa->hw.cst.cmd[CST_CMD_3] = cmdpkt(rmesa, R200_EMIT_RE_SCISSOR_TL_0);
+   rmesa->hw.cst.cmd[CST_CMD_4] = cmdpkt(rmesa, R200_EMIT_SE_VAP_CNTL_STATUS);
+   rmesa->hw.cst.cmd[CST_CMD_5] = cmdpkt(rmesa, R200_EMIT_RE_POINTSIZE);
+   rmesa->hw.cst.cmd[CST_CMD_6] = cmdpkt(rmesa, R200_EMIT_TCL_INPUT_VTX_VECTOR_ADDR_0);
+   rmesa->hw.tam.cmd[TAM_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TAM_DEBUG3);
+   rmesa->hw.tf.cmd[TF_CMD_0] = cmdpkt(rmesa, R200_EMIT_TFACTOR_0);
+   if (rmesa->radeon.radeonScreen->drmSupportsFragShader) {
+      rmesa->hw.atf.cmd[ATF_CMD_0] = cmdpkt(rmesa, R200_EMIT_ATF_TFACTOR);
+      rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCTLALL_0);
+      rmesa->hw.tex[0].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_0);
+      rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCTLALL_1);
+      rmesa->hw.tex[1].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_1);
+      rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCTLALL_2);
+      rmesa->hw.tex[2].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_2);
+      rmesa->hw.tex[3].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCTLALL_3);
+      rmesa->hw.tex[3].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_3);
+      rmesa->hw.tex[4].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCTLALL_4);
+      rmesa->hw.tex[4].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_4);
+      rmesa->hw.tex[5].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCTLALL_5);
+      rmesa->hw.tex[5].cmd[TEX_CMD_1_NEWDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_5);
    } else {
-      rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXFILTER_0);
-      rmesa->hw.tex[0].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_0);
-      rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXFILTER_1);
-      rmesa->hw.tex[1].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_1);
-      rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXFILTER_2);
-      rmesa->hw.tex[2].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_2);
-      rmesa->hw.tex[3].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXFILTER_3);
-      rmesa->hw.tex[3].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_3);
-      rmesa->hw.tex[4].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXFILTER_4);
-      rmesa->hw.tex[4].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_4);
-      rmesa->hw.tex[5].cmd[TEX_CMD_0] = cmdpkt(R200_EMIT_PP_TXFILTER_5);
-      rmesa->hw.tex[5].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(R200_EMIT_PP_TXOFFSET_5);
-   }
-   rmesa->hw.afs[0].cmd[AFS_CMD_0] = cmdpkt(R200_EMIT_PP_AFS_0);
-   rmesa->hw.afs[1].cmd[AFS_CMD_0] = cmdpkt(R200_EMIT_PP_AFS_1);
-   rmesa->hw.pvs.cmd[PVS_CMD_0] = cmdpkt(R200_EMIT_VAP_PVS_CNTL);
-   rmesa->hw.cube[0].cmd[CUBE_CMD_0] = cmdpkt(R200_EMIT_PP_CUBIC_FACES_0);
-   rmesa->hw.cube[0].cmd[CUBE_CMD_1] = cmdpkt(R200_EMIT_PP_CUBIC_OFFSETS_0);
-   rmesa->hw.cube[1].cmd[CUBE_CMD_0] = cmdpkt(R200_EMIT_PP_CUBIC_FACES_1);
-   rmesa->hw.cube[1].cmd[CUBE_CMD_1] = cmdpkt(R200_EMIT_PP_CUBIC_OFFSETS_1);
-   rmesa->hw.cube[2].cmd[CUBE_CMD_0] = cmdpkt(R200_EMIT_PP_CUBIC_FACES_2);
-   rmesa->hw.cube[2].cmd[CUBE_CMD_1] = cmdpkt(R200_EMIT_PP_CUBIC_OFFSETS_2);
-   rmesa->hw.cube[3].cmd[CUBE_CMD_0] = cmdpkt(R200_EMIT_PP_CUBIC_FACES_3);
-   rmesa->hw.cube[3].cmd[CUBE_CMD_1] = cmdpkt(R200_EMIT_PP_CUBIC_OFFSETS_3);
-   rmesa->hw.cube[4].cmd[CUBE_CMD_0] = cmdpkt(R200_EMIT_PP_CUBIC_FACES_4);
-   rmesa->hw.cube[4].cmd[CUBE_CMD_1] = cmdpkt(R200_EMIT_PP_CUBIC_OFFSETS_4);
-   rmesa->hw.cube[5].cmd[CUBE_CMD_0] = cmdpkt(R200_EMIT_PP_CUBIC_FACES_5);
-   rmesa->hw.cube[5].cmd[CUBE_CMD_1] = cmdpkt(R200_EMIT_PP_CUBIC_OFFSETS_5);
-   rmesa->hw.pix[0].cmd[PIX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCBLEND_0);
-   rmesa->hw.pix[1].cmd[PIX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCBLEND_1);
-   rmesa->hw.pix[2].cmd[PIX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCBLEND_2);
-   rmesa->hw.pix[3].cmd[PIX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCBLEND_3);
-   rmesa->hw.pix[4].cmd[PIX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCBLEND_4);
-   rmesa->hw.pix[5].cmd[PIX_CMD_0] = cmdpkt(R200_EMIT_PP_TXCBLEND_5);
-   rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(RADEON_EMIT_SE_ZBIAS_FACTOR);
-   rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(R200_EMIT_TCL_LIGHT_MODEL_CTL_0);
-   rmesa->hw.tcl.cmd[TCL_CMD_1] = cmdpkt(R200_EMIT_TCL_UCP_VERT_BLEND_CTL);
-   rmesa->hw.tcg.cmd[TCG_CMD_0] = cmdpkt(R200_EMIT_TEX_PROC_CTL_2);
-   rmesa->hw.msl.cmd[MSL_CMD_0] = cmdpkt(R200_EMIT_MATRIX_SELECT_0);
-   rmesa->hw.vap.cmd[VAP_CMD_0] = cmdpkt(R200_EMIT_VAP_CTL);
-   rmesa->hw.vtx.cmd[VTX_CMD_0] = cmdpkt(R200_EMIT_VTX_FMT_0);
-   rmesa->hw.vtx.cmd[VTX_CMD_1] = cmdpkt(R200_EMIT_OUTPUT_VTX_COMP_SEL);
-   rmesa->hw.vtx.cmd[VTX_CMD_2] = cmdpkt(R200_EMIT_SE_VTX_STATE_CNTL);
-   rmesa->hw.vte.cmd[VTE_CMD_0] = cmdpkt(R200_EMIT_VTE_CNTL);
-   rmesa->hw.prf.cmd[PRF_CMD_0] = cmdpkt(R200_EMIT_PP_TRI_PERF_CNTL);
-   rmesa->hw.spr.cmd[SPR_CMD_0] = cmdpkt(R200_EMIT_TCL_POINT_SPRITE_CNTL);
+      rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXFILTER_0);
+      rmesa->hw.tex[0].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_0);
+      rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXFILTER_1);
+      rmesa->hw.tex[1].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_1);
+      rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXFILTER_2);
+      rmesa->hw.tex[2].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_2);
+      rmesa->hw.tex[3].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXFILTER_3);
+      rmesa->hw.tex[3].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_3);
+      rmesa->hw.tex[4].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXFILTER_4);
+      rmesa->hw.tex[4].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_4);
+      rmesa->hw.tex[5].cmd[TEX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXFILTER_5);
+      rmesa->hw.tex[5].cmd[TEX_CMD_1_OLDDRM] = cmdpkt(rmesa, R200_EMIT_PP_TXOFFSET_5);
+   }
+   rmesa->hw.afs[0].cmd[AFS_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_AFS_0);
+   rmesa->hw.afs[1].cmd[AFS_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_AFS_1);
+   rmesa->hw.pvs.cmd[PVS_CMD_0] = cmdpkt(rmesa, R200_EMIT_VAP_PVS_CNTL);
+   rmesa->hw.cube[0].cmd[CUBE_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_FACES_0);
+   rmesa->hw.cube[0].cmd[CUBE_CMD_1] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_OFFSETS_0);
+   rmesa->hw.cube[1].cmd[CUBE_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_FACES_1);
+   rmesa->hw.cube[1].cmd[CUBE_CMD_1] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_OFFSETS_1);
+   rmesa->hw.cube[2].cmd[CUBE_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_FACES_2);
+   rmesa->hw.cube[2].cmd[CUBE_CMD_1] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_OFFSETS_2);
+   rmesa->hw.cube[3].cmd[CUBE_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_FACES_3);
+   rmesa->hw.cube[3].cmd[CUBE_CMD_1] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_OFFSETS_3);
+   rmesa->hw.cube[4].cmd[CUBE_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_FACES_4);
+   rmesa->hw.cube[4].cmd[CUBE_CMD_1] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_OFFSETS_4);
+   rmesa->hw.cube[5].cmd[CUBE_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_FACES_5);
+   rmesa->hw.cube[5].cmd[CUBE_CMD_1] = cmdpkt(rmesa, R200_EMIT_PP_CUBIC_OFFSETS_5);
+   rmesa->hw.pix[0].cmd[PIX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCBLEND_0);
+   rmesa->hw.pix[1].cmd[PIX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCBLEND_1);
+   rmesa->hw.pix[2].cmd[PIX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCBLEND_2);
+   rmesa->hw.pix[3].cmd[PIX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCBLEND_3);
+   rmesa->hw.pix[4].cmd[PIX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCBLEND_4);
+   rmesa->hw.pix[5].cmd[PIX_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TXCBLEND_5);
+   rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_ZBIAS_FACTOR);
+   rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(rmesa, R200_EMIT_TCL_LIGHT_MODEL_CTL_0);
+   rmesa->hw.tcl.cmd[TCL_CMD_1] = cmdpkt(rmesa, R200_EMIT_TCL_UCP_VERT_BLEND_CTL);
+   rmesa->hw.tcg.cmd[TCG_CMD_0] = cmdpkt(rmesa, R200_EMIT_TEX_PROC_CTL_2);
+   rmesa->hw.msl.cmd[MSL_CMD_0] = cmdpkt(rmesa, R200_EMIT_MATRIX_SELECT_0);
+   rmesa->hw.vap.cmd[VAP_CMD_0] = cmdpkt(rmesa, R200_EMIT_VAP_CTL);
+   rmesa->hw.vtx.cmd[VTX_CMD_0] = cmdpkt(rmesa, R200_EMIT_VTX_FMT_0);
+   rmesa->hw.vtx.cmd[VTX_CMD_1] = cmdpkt(rmesa, R200_EMIT_OUTPUT_VTX_COMP_SEL);
+   rmesa->hw.vtx.cmd[VTX_CMD_2] = cmdpkt(rmesa, R200_EMIT_SE_VTX_STATE_CNTL);
+   rmesa->hw.vte.cmd[VTE_CMD_0] = cmdpkt(rmesa, R200_EMIT_VTE_CNTL);
+   rmesa->hw.prf.cmd[PRF_CMD_0] = cmdpkt(rmesa, R200_EMIT_PP_TRI_PERF_CNTL);
+   rmesa->hw.spr.cmd[SPR_CMD_0] = cmdpkt(rmesa, R200_EMIT_TCL_POINT_SPRITE_CNTL);
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+	rmesa->hw.mtl[0].emit = mtl_emit;
+	rmesa->hw.mtl[1].emit = mtl_emit;
+
+	rmesa->hw.vpi[0].emit = veclinear_emit;
+	rmesa->hw.vpi[1].emit = veclinear_emit;
+	rmesa->hw.vpp[0].emit = veclinear_emit;
+	rmesa->hw.vpp[1].emit = veclinear_emit;
+
+	rmesa->hw.grd.emit = scl_emit;
+	rmesa->hw.fog.emit = vec_emit;
+	rmesa->hw.glt.emit = vec_emit;
+	rmesa->hw.eye.emit = vec_emit;
+
+	for (i = R200_MTX_MV; i <= R200_MTX_TEX5; i++)
+	  rmesa->hw.mat[i].emit = vec_emit;
+
+	for (i = 0; i < 8; i++)
+	  rmesa->hw.lit[i].emit = lit_emit;
+
+	for (i = 0; i < 6; i++)
+	  rmesa->hw.ucp[i].emit = vec_emit;
+
+	rmesa->hw.ptp.emit = ptp_emit;
+   }
+
+
+   
    rmesa->hw.mtl[0].cmd[MTL_CMD_0] = 
       cmdvec( R200_VS_MAT_0_EMISS, 1, 16 );
    rmesa->hw.mtl[0].cmd[MTL_CMD_1] = 
@@ -567,7 +1021,7 @@ void r200InitState( r200ContextPtr rmesa )
 				(R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
 				(R200_BLEND_GL_ZERO << R200_DST_BLEND_SHIFT));
 
-   if (rmesa->r200Screen->drmSupportsBlendColor) {
+   if (rmesa->radeon.radeonScreen->drmSupportsBlendColor) {
       rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCOLOR] = 0x00000000;
       rmesa->hw.ctx.cmd[CTX_RB3D_ABLENDCNTL] = (R200_COMB_FCN_ADD_CLAMP |
 				(R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
@@ -578,18 +1032,17 @@ void r200InitState( r200ContextPtr rmesa )
    }
 
    rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHOFFSET] =
-      rmesa->r200Screen->depthOffset + rmesa->r200Screen->fbLocation;
+      rmesa->radeon.radeonScreen->depthOffset + rmesa->radeon.radeonScreen->fbLocation;
 
    rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] = 
-      ((rmesa->r200Screen->depthPitch &
+      ((rmesa->radeon.radeonScreen->depthPitch &
 	R200_DEPTHPITCH_MASK) |
        R200_DEPTH_ENDIAN_NO_SWAP);
    
    if (rmesa->using_hyperz)
       rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] |= R200_DEPTH_HYPERZ;
 
-   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (depth_fmt |
-					       R200_Z_TEST_LESS |
+   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (R200_Z_TEST_LESS |
 					       R200_STENCIL_TEST_ALWAYS |
 					       R200_STENCIL_FAIL_KEEP |
 					       R200_STENCIL_ZPASS_KEEP |
@@ -599,15 +1052,14 @@ void r200InitState( r200ContextPtr rmesa )
    if (rmesa->using_hyperz) {
       rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= R200_Z_COMPRESSION_ENABLE |
 						  R200_Z_DECOMPRESSION_ENABLE;
-/*      if (rmesa->r200Screen->chip_family == CHIP_FAMILY_R200)
+/*      if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200)
 	 rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_HIERARCHY_ENABLE;*/
    }
 
    rmesa->hw.ctx.cmd[CTX_PP_CNTL] = (R200_ANTI_ALIAS_NONE 
  				     | R200_TEX_BLEND_0_ENABLE);
 
-   rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = color_fmt;
-   switch ( driQueryOptioni( &rmesa->optionCache, "dither_mode" ) ) {
+   switch ( driQueryOptioni( &rmesa->radeon.optionCache, "dither_mode" ) ) {
    case DRI_CONF_DITHER_XERRORDIFFRESET:
       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= R200_DITHER_INIT;
       break;
@@ -615,41 +1067,19 @@ void r200InitState( r200ContextPtr rmesa )
       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= R200_SCALE_DITHER_ENABLE;
       break;
    }
-   if ( driQueryOptioni( &rmesa->optionCache, "round_mode" ) ==
+   if ( driQueryOptioni( &rmesa->radeon.optionCache, "round_mode" ) ==
 	DRI_CONF_ROUND_ROUND )
-      rmesa->state.color.roundEnable = R200_ROUND_ENABLE;
+      rmesa->radeon.state.color.roundEnable = R200_ROUND_ENABLE;
    else
-      rmesa->state.color.roundEnable = 0;
-   if ( driQueryOptioni (&rmesa->optionCache, "color_reduction" ) ==
+      rmesa->radeon.state.color.roundEnable = 0;
+   if ( driQueryOptioni (&rmesa->radeon.optionCache, "color_reduction" ) ==
 	DRI_CONF_COLOR_REDUCTION_DITHER )
       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= R200_DITHER_ENABLE;
    else
-      rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->state.color.roundEnable;
-
-#if 000
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = ((rmesa->state.color.drawOffset +
-					       rmesa->r200Screen->fbLocation)
-					      & R200_COLOROFFSET_MASK);
-
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = ((rmesa->state.color.drawPitch &
-					      R200_COLORPITCH_MASK) |
-					     R200_COLOR_ENDIAN_NO_SWAP);
-#else
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = ((drawOffset +
-					       rmesa->r200Screen->fbLocation)
-					      & R200_COLOROFFSET_MASK);
-
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = ((drawPitch &
-					      R200_COLORPITCH_MASK) |
-					     R200_COLOR_ENDIAN_NO_SWAP);
-#endif
-   /* (fixed size) sarea is initialized to zero afaics so can omit version check. Phew! */
-   if (rmesa->sarea->tiling_enabled) {
-      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= R200_COLOR_TILE_ENABLE;
-   }
+      rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->radeon.state.color.roundEnable;
 
    rmesa->hw.prf.cmd[PRF_PP_TRI_PERF] = R200_TRI_CUTOFF_MASK - R200_TRI_CUTOFF_MASK * 
-			driQueryOptionf (&rmesa->optionCache,"texture_blend_quality");
+			driQueryOptionf (&rmesa->radeon.optionCache,"texture_blend_quality");
    rmesa->hw.prf.cmd[PRF_PP_PERF_CNTL] = 0;
 
    rmesa->hw.set.cmd[SET_SE_CNTL] = (R200_FFACE_CULL_CCW |
@@ -704,7 +1134,7 @@ void r200InitState( r200ContextPtr rmesa )
 						R200_VC_NO_SWAP;
 #endif
 
-   if (!(rmesa->r200Screen->chip_flags & RADEON_CHIPSET_TCL)) {
+   if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
       /* Bypass TCL */
       rmesa->hw.cst.cmd[CST_SE_VAP_CNTL_STATUS] |= (1<<8);
    }
@@ -743,28 +1173,28 @@ void r200InitState( r200ContextPtr rmesa )
       rmesa->hw.tex[i].cmd[TEX_PP_TXFORMAT_X] =
          (/* R200_TEXCOORD_PROJ | */
           0x100000);	/* Small default bias */
-      if (rmesa->r200Screen->drmSupportsFragShader) {
+      if (rmesa->radeon.radeonScreen->drmSupportsFragShader) {
 	 rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET_NEWDRM] =
-	     rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	     rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
 	 rmesa->hw.tex[i].cmd[TEX_PP_CUBIC_FACES] = 0;
 	 rmesa->hw.tex[i].cmd[TEX_PP_TXMULTI_CTL] = 0;
       }
       else {
 	  rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET_OLDDRM] =
-	     rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	     rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
      }
 
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_FACES] = 0;
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_F1] =
-         rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+         rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_F2] =
-         rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+         rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_F3] =
-         rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+         rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_F4] =
-         rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+         rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_F5] =
-         rmesa->r200Screen->texOffset[RADEON_LOCAL_TEX_HEAP];
+         rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
 
       rmesa->hw.pix[i].cmd[PIX_PP_TXCBLEND] =
          (R200_TXC_ARG_A_ZERO |
@@ -967,5 +1397,7 @@ void r200InitState( r200ContextPtr rmesa )
 
    r200LightingSpaceChange( ctx );
 
-   rmesa->hw.all_dirty = GL_TRUE;
+   rmesa->radeon.hw.all_dirty = GL_TRUE;
+
+   rcommonInitCmdBuf(&rmesa->radeon);
 }
diff --git a/src/mesa/drivers/dri/r200/r200_swtcl.c b/src/mesa/drivers/dri/r200/r200_swtcl.c
index b25f0282445..83e70b586d7 100644
--- a/src/mesa/drivers/dri/r200/r200_swtcl.c
+++ b/src/mesa/drivers/dri/r200/r200_swtcl.c
@@ -55,27 +55,24 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_tcl.h"
 
 
-static void flush_last_swtcl_prim( r200ContextPtr rmesa  );
-
-
 /***********************************************************************
- *                         Initialization 
+ *                         Initialization
  ***********************************************************************/
 
 #define EMIT_ATTR( ATTR, STYLE, F0 )					\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = (ATTR);	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = (STYLE);	\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR);	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE);	\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
    fmt_0 |= F0;								\
 } while (0)
 
 #define EMIT_PAD( N )							\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = 0;		\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = EMIT_PAD;	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].offset = (N);		\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0;		\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD;	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N);		\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
 } while (0)
 
 static void r200SetVertexFormat( GLcontext *ctx )
@@ -100,7 +97,7 @@ static void r200SetVertexFormat( GLcontext *ctx )
    }
 
    assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
-   rmesa->swtcl.vertex_attr_count = 0;
+   rmesa->radeon.swtcl.vertex_attr_count = 0;
 
    /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
     * build up a hardware vertex.
@@ -121,7 +118,7 @@ static void r200SetVertexFormat( GLcontext *ctx )
    }
 
    rmesa->swtcl.coloroffset = offset;
-#if MESA_LITTLE_ENDIAN 
+#if MESA_LITTLE_ENDIAN
    EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA, (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT) );
 #else
    EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR, (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT) );
@@ -132,7 +129,7 @@ static void r200SetVertexFormat( GLcontext *ctx )
    if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 ) ||
        RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_FOG )) {
 
-#if MESA_LITTLE_ENDIAN 
+#if MESA_LITTLE_ENDIAN
       if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 )) {
 	 rmesa->swtcl.specoffset = offset;
 	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_RGB, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) );
@@ -185,7 +182,7 @@ static void r200SetVertexFormat( GLcontext *ctx )
       rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] |= R200_FOG_USE_SPEC_ALPHA;
    }
 
-   if (!RENDERINPUTS_EQUAL( rmesa->tnl_index_bitset, index_bitset ) ||
+   if (!RENDERINPUTS_EQUAL( rmesa->radeon.tnl_index_bitset, index_bitset ) ||
 	(rmesa->hw.vtx.cmd[VTX_VTXFMT_0] != fmt_0) ||
 	(rmesa->hw.vtx.cmd[VTX_VTXFMT_1] != fmt_1) ) {
       R200_NEWPRIM(rmesa);
@@ -193,26 +190,20 @@ static void r200SetVertexFormat( GLcontext *ctx )
       rmesa->hw.vtx.cmd[VTX_VTXFMT_0] = fmt_0;
       rmesa->hw.vtx.cmd[VTX_VTXFMT_1] = fmt_1;
 
-      rmesa->swtcl.vertex_size =
+      rmesa->radeon.swtcl.vertex_size =
 	  _tnl_install_attrs( ctx,
-			      rmesa->swtcl.vertex_attrs, 
-			      rmesa->swtcl.vertex_attr_count,
+			      rmesa->radeon.swtcl.vertex_attrs,
+			      rmesa->radeon.swtcl.vertex_attr_count,
 			      NULL, 0 );
-      rmesa->swtcl.vertex_size /= 4;
-      RENDERINPUTS_COPY( rmesa->tnl_index_bitset, index_bitset );
+      rmesa->radeon.swtcl.vertex_size /= 4;
+      RENDERINPUTS_COPY( rmesa->radeon.tnl_index_bitset, index_bitset );
    }
 }
 
 
 static void r200RenderStart( GLcontext *ctx )
 {
-   r200ContextPtr rmesa = R200_CONTEXT( ctx );
-
    r200SetVertexFormat( ctx );
-
-   if (rmesa->dma.flush != 0 && 
-       rmesa->dma.flush != flush_last_swtcl_prim)
-      rmesa->dma.flush( rmesa );
 }
 
 
@@ -232,7 +223,7 @@ void r200ChooseVertexState( GLcontext *ctx )
     * rasterization fallback.  As this function will be called again when we
     * leave a rasterization fallback, we can just skip it for now.
     */
-   if (rmesa->Fallback != 0)
+   if (rmesa->radeon.Fallback != 0)
       return;
 
    vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL];
@@ -273,78 +264,27 @@ void r200ChooseVertexState( GLcontext *ctx )
    }
 }
 
-
-/* Flush vertices in the current dma region.
- */
-static void flush_last_swtcl_prim( r200ContextPtr rmesa  )
+void r200_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
 {
-   if (R200_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   rmesa->dma.flush = NULL;
-
-   if (rmesa->dma.current.buf) {
-      struct r200_dma_region *current = &rmesa->dma.current;
-      GLuint current_offset = (rmesa->r200Screen->gart_buffer_offset +
-			       current->buf->buf->idx * RADEON_BUFFER_SIZE + 
-			       current->start);
-
-      assert (!(rmesa->swtcl.hw_primitive & R200_VF_PRIM_WALK_IND));
-
-      assert (current->start + 
-	      rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-	      current->ptr);
-
-      if (rmesa->dma.current.start != rmesa->dma.current.ptr) {
-	 r200EnsureCmdBufSpace( rmesa, VERT_AOS_BUFSZ +
-			        rmesa->hw.max_state_size + VBUF_BUFSZ );
-	 r200EmitVertexAOS( rmesa,
-			      rmesa->swtcl.vertex_size,
-			      current_offset);
-
-	 r200EmitVbufPrim( rmesa,
-			   rmesa->swtcl.hw_primitive,
-			   rmesa->swtcl.numverts);
-      }
-
-      rmesa->swtcl.numverts = 0;
-      current->start = current->ptr;
-   }
-}
-
-
-/* Alloc space in the current dma region.
- */
-static INLINE void *
-r200AllocDmaLowVerts( r200ContextPtr rmesa, int nverts, int vsize )
-{
-   GLuint bytes = vsize * nverts;
-
-   if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) 
-      r200RefillCurrentDmaRegion( rmesa );
+   r200ContextPtr rmesa = R200_CONTEXT(ctx);
+   rcommonEnsureCmdBufSpace(&rmesa->radeon,
+			    rmesa->radeon.hw.max_state_size + (12*sizeof(int)),
+			    __FUNCTION__);
 
-   if (!rmesa->dma.flush) {
-      rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
-      rmesa->dma.flush = flush_last_swtcl_prim;
-   }
 
-   ASSERT( vsize == rmesa->swtcl.vertex_size * 4 );
-   ASSERT( rmesa->dma.flush == flush_last_swtcl_prim );
-   ASSERT( rmesa->dma.current.start + 
-	   rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-	   rmesa->dma.current.ptr );
+   radeonEmitState(&rmesa->radeon);
+   r200EmitVertexAOS( rmesa,
+		      rmesa->radeon.swtcl.vertex_size,
+		      rmesa->radeon.dma.current,
+		      current_offset);
 
 
-   {
-      GLubyte *head = (GLubyte *) (rmesa->dma.current.address + rmesa->dma.current.ptr);
-      rmesa->dma.current.ptr += bytes;
-      rmesa->swtcl.numverts += nverts;
-      return head;
-   }
+   r200EmitVbufPrim( rmesa,
+		     rmesa->radeon.swtcl.hw_primitive,
+		     rmesa->radeon.swtcl.numverts);
 
 }
 
-
 /**************************************************************************/
 
 
@@ -392,13 +332,13 @@ static void r200ResetLineStipple( GLcontext *ctx );
 #undef LOCAL_VARS
 #undef ALLOC_VERTS
 #define CTX_ARG r200ContextPtr rmesa
-#define GET_VERTEX_DWORDS() rmesa->swtcl.vertex_size
-#define ALLOC_VERTS( n, size ) r200AllocDmaLowVerts( rmesa, n, size * 4 )
+#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
+#define ALLOC_VERTS( n, size ) rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 )
 #define LOCAL_VARS						\
    r200ContextPtr rmesa = R200_CONTEXT(ctx);		\
-   const char *r200verts = (char *)rmesa->swtcl.verts;
-#define VERT(x) (r200Vertex *)(r200verts + ((x) * vertsize * sizeof(int)))
-#define VERTEX r200Vertex 
+   const char *r200verts = (char *)rmesa->radeon.swtcl.verts;
+#define VERT(x) (radeonVertex *)(r200verts + ((x) * vertsize * sizeof(int)))
+#define VERTEX radeonVertex
 #define DO_DEBUG_VERTS (1 && (R200_DEBUG & DEBUG_VERTS))
 
 #undef TAG
@@ -456,11 +396,11 @@ static struct {
 #define VERT_Y(_v) _v->v.y
 #define VERT_Z(_v) _v->v.z
 #define AREA_IS_CCW( a ) (a < 0)
-#define GET_VERTEX(e) (rmesa->swtcl.verts + (e*rmesa->swtcl.vertex_size*sizeof(int)))
+#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + (e*rmesa->radeon.swtcl.vertex_size*sizeof(int)))
 
 #define VERT_SET_RGBA( v, c )  					\
 do {								\
-   r200_color_t *color = (r200_color_t *)&((v)->ui[coloroffset]);	\
+   radeon_color_t *color = (radeon_color_t *)&((v)->ui[coloroffset]);	\
    UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]);		\
    UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]);		\
    UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]);		\
@@ -472,7 +412,7 @@ do {								\
 #define VERT_SET_SPEC( v, c )					\
 do {								\
    if (specoffset) {						\
-      r200_color_t *spec = (r200_color_t *)&((v)->ui[specoffset]);	\
+      radeon_color_t *spec = (radeon_color_t *)&((v)->ui[specoffset]);	\
       UNCLAMPED_FLOAT_TO_UBYTE(spec->red, (c)[0]);	\
       UNCLAMPED_FLOAT_TO_UBYTE(spec->green, (c)[1]);	\
       UNCLAMPED_FLOAT_TO_UBYTE(spec->blue, (c)[2]);	\
@@ -481,8 +421,8 @@ do {								\
 #define VERT_COPY_SPEC( v0, v1 )			\
 do {							\
    if (specoffset) {					\
-      r200_color_t *spec0 = (r200_color_t *)&((v0)->ui[specoffset]);	\
-      r200_color_t *spec1 = (r200_color_t *)&((v1)->ui[specoffset]);	\
+      radeon_color_t *spec0 = (radeon_color_t *)&((v0)->ui[specoffset]);	\
+      radeon_color_t *spec1 = (radeon_color_t *)&((v1)->ui[specoffset]);	\
       spec0->red   = spec1->red;	\
       spec0->green = spec1->green;	\
       spec0->blue  = spec1->blue; 	\
@@ -513,7 +453,7 @@ do {							\
  ***********************************************************************/
 
 #define RASTERIZE(x) r200RasterPrimitive( ctx, reduced_hw_prim(ctx, x) )
-#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive
+#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive
 #undef TAG
 #define TAG(x) x
 #include "tnl_dd/t_dd_unfilled.h"
@@ -569,8 +509,8 @@ static void init_rast_tab( void )
 #undef LOCAL_VARS
 #define LOCAL_VARS						\
    r200ContextPtr rmesa = R200_CONTEXT(ctx);		\
-   const GLuint vertsize = rmesa->swtcl.vertex_size;		\
-   const char *r200verts = (char *)rmesa->swtcl.verts;		\
+   const GLuint vertsize = rmesa->radeon.swtcl.vertex_size;		\
+   const char *r200verts = (char *)rmesa->radeon.swtcl.verts;		\
    const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
    const GLboolean stipple = ctx->Line.StippleFlag;		\
    (void) elt; (void) stipple;
@@ -599,13 +539,13 @@ void r200ChooseRenderState( GLcontext *ctx )
    GLuint index = 0;
    GLuint flags = ctx->_TriangleCaps;
 
-   if (!rmesa->TclFallback || rmesa->Fallback) 
+   if (!rmesa->radeon.TclFallback || rmesa->radeon.Fallback)
       return;
 
    if (flags & DD_TRI_LIGHT_TWOSIDE) index |= R200_TWOSIDE_BIT;
    if (flags & DD_TRI_UNFILLED)      index |= R200_UNFILLED_BIT;
 
-   if (index != rmesa->swtcl.RenderIndex) {
+   if (index != rmesa->radeon.swtcl.RenderIndex) {
       tnl->Driver.Render.Points = rast_tab[index].points;
       tnl->Driver.Render.Line = rast_tab[index].line;
       tnl->Driver.Render.ClippedLine = rast_tab[index].line;
@@ -622,7 +562,7 @@ void r200ChooseRenderState( GLcontext *ctx )
 	 tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
       }
 
-      rmesa->swtcl.RenderIndex = index;
+      rmesa->radeon.swtcl.RenderIndex = index;
    }
 }
 
@@ -636,7 +576,7 @@ static void r200RasterPrimitive( GLcontext *ctx, GLuint hwprim )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 
-   if (rmesa->swtcl.hw_primitive != hwprim) {
+   if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
       /* need to disable perspective-correct texturing for point sprites */
       if ((hwprim & 0xf) == R200_VF_PRIM_POINT_SPRITES && ctx->Point.PointSprite) {
 	 if (rmesa->hw.set.cmd[SET_RE_CNTL] & R200_PERSPECTIVE_ENABLE) {
@@ -649,15 +589,15 @@ static void r200RasterPrimitive( GLcontext *ctx, GLuint hwprim )
 	 rmesa->hw.set.cmd[SET_RE_CNTL] |= R200_PERSPECTIVE_ENABLE;
       }
       R200_NEWPRIM( rmesa );
-      rmesa->swtcl.hw_primitive = hwprim;
+      rmesa->radeon.swtcl.hw_primitive = hwprim;
    }
 }
 
 static void r200RenderPrimitive( GLcontext *ctx, GLenum prim )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   rmesa->swtcl.render_primitive = prim;
-   if (prim < GL_TRIANGLES || !(ctx->_TriangleCaps & DD_TRI_UNFILLED)) 
+   rmesa->radeon.swtcl.render_primitive = prim;
+   if (prim < GL_TRIANGLES || !(ctx->_TriangleCaps & DD_TRI_UNFILLED))
       r200RasterPrimitive( ctx, reduced_hw_prim(ctx, prim) );
 }
 
@@ -701,15 +641,15 @@ void r200Fallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
-   GLuint oldfallback = rmesa->Fallback;
+   GLuint oldfallback = rmesa->radeon.Fallback;
 
    if (mode) {
-      rmesa->Fallback |= bit;
+      rmesa->radeon.Fallback |= bit;
       if (oldfallback == 0) {
-	 R200_FIREVERTICES( rmesa );
+	 radeon_firevertices(&rmesa->radeon);
 	 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_RASTER, GL_TRUE );
 	 _swsetup_Wakeup( ctx );
-	 rmesa->swtcl.RenderIndex = ~0;
+	 rmesa->radeon.swtcl.RenderIndex = ~0;
          if (R200_DEBUG & DEBUG_FALLBACKS) {
             fprintf(stderr, "R200 begin rasterization fallback: 0x%x %s\n",
                     bit, getFallbackString(bit));
@@ -717,7 +657,7 @@ void r200Fallback( GLcontext *ctx, GLuint bit, GLboolean mode )
       }
    }
    else {
-      rmesa->Fallback &= ~bit;
+      rmesa->radeon.Fallback &= ~bit;
       if (oldfallback == bit) {
 
 	 _swrast_flush( ctx );
@@ -731,14 +671,14 @@ void r200Fallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 
 	 tnl->Driver.Render.ResetLineStipple = r200ResetLineStipple;
 	 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_RASTER, GL_FALSE );
-	 if (rmesa->TclFallback) {
-	    /* These are already done if rmesa->TclFallback goes to
+	 if (rmesa->radeon.TclFallback) {
+	    /* These are already done if rmesa->radeon.TclFallback goes to
 	     * zero above. But not if it doesn't (R200_NO_TCL for
 	     * example?)
 	     */
 	    _tnl_invalidate_vertex_state( ctx, ~0 );
 	    _tnl_invalidate_vertices( ctx, ~0 );
-	    RENDERINPUTS_ZERO( rmesa->tnl_index_bitset );
+	    RENDERINPUTS_ZERO( rmesa->radeon.tnl_index_bitset );
 	    r200ChooseVertexState( ctx );
 	    r200ChooseRenderState( ctx );
 	 }
@@ -755,7 +695,7 @@ void r200Fallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 
 /**
  * Cope with depth operations by drawing individual pixels as points.
- * 
+ *
  * \todo
  * The way the vertex state is set in this routine is hokey.  It seems to
  * work, but it's very hackish.  This whole routine is pretty hackish.  If
@@ -770,14 +710,14 @@ r200PointsBitmap( GLcontext *ctx, GLint px, GLint py,
 		  const GLubyte *bitmap )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   const GLfloat *rc = ctx->Current.RasterColor; 
+   const GLfloat *rc = ctx->Current.RasterColor;
    GLint row, col;
-   r200Vertex vert;
+   radeonVertex vert;
    GLuint orig_vte;
    GLuint h;
 
 
-   /* Turn off tcl.  
+   /* Turn off tcl.
     */
    TCL_FALLBACK( ctx, R200_TCL_FALLBACK_BITMAP, 1 );
 
@@ -794,7 +734,7 @@ r200PointsBitmap( GLcontext *ctx, GLint px, GLint py,
       vte |= R200_VTX_W0_FMT;
       vap &= ~R200_VAP_FORCE_W_TO_ONE;
 
-      rmesa->swtcl.vertex_size = 5;
+      rmesa->radeon.swtcl.vertex_size = 5;
 
       if ( (rmesa->hw.vtx.cmd[VTX_VTXFMT_0] != fmt_0)
 	   || (rmesa->hw.vtx.cmd[VTX_VTXFMT_1] != fmt_1) ) {
@@ -828,7 +768,7 @@ r200PointsBitmap( GLcontext *ctx, GLint px, GLint py,
 					   R200_VPORT_Z_SCALE_ENA |
 					   R200_VPORT_X_OFFSET_ENA |
 					   R200_VPORT_Y_OFFSET_ENA |
-					   R200_VPORT_Z_OFFSET_ENA); 
+					   R200_VPORT_Z_OFFSET_ENA);
 
    /* Turn off other stuff:  Stipple?, texture?, blending?, etc.
     */
@@ -871,16 +811,16 @@ r200PointsBitmap( GLcontext *ctx, GLint px, GLint py,
 
    /* Update window height
     */
-   LOCK_HARDWARE( rmesa );
-   UNLOCK_HARDWARE( rmesa );
-   h = rmesa->dri.drawable->h + rmesa->dri.drawable->y;
-   px += rmesa->dri.drawable->x;
+   LOCK_HARDWARE( &rmesa->radeon );
+   UNLOCK_HARDWARE( &rmesa->radeon );
+   h = radeon_get_drawable(&rmesa->radeon)->h + radeon_get_drawable(&rmesa->radeon)->y;
+   px += radeon_get_drawable(&rmesa->radeon)->x;
 
    /* Clipping handled by existing mechansims in r200_ioctl.c?
     */
    for (row=0; row<height; row++) {
-      const GLubyte *src = (const GLubyte *) 
-	 _mesa_image_address2d(unpack, bitmap, width, height, 
+      const GLubyte *src = (const GLubyte *)
+	 _mesa_image_address2d(unpack, bitmap, width, height,
                                GL_COLOR_INDEX, GL_BITMAP, row, 0 );
 
       if (unpack->LsbFirst) {
@@ -929,7 +869,7 @@ r200PointsBitmap( GLcontext *ctx, GLint px, GLint py,
 
    /* Need to restore vertexformat?
     */
-   if (rmesa->TclFallback)
+   if (rmesa->radeon.TclFallback)
       r200ChooseVertexState( ctx );
 }
 
@@ -959,20 +899,12 @@ void r200InitSwtcl( GLcontext *ctx )
    tnl->Driver.Render.Interp = _tnl_interp;
 
    /* FIXME: what are these numbers? */
-   _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 
+   _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12,
 		       36 * sizeof(GLfloat) );
-   
-   rmesa->swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
-   rmesa->swtcl.RenderIndex = ~0;
-   rmesa->swtcl.render_primitive = GL_TRIANGLES;
-   rmesa->swtcl.hw_primitive = 0;
-}
-
 
-void r200DestroySwtcl( GLcontext *ctx )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if (rmesa->swtcl.indexed_verts.buf) 
-      r200ReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, __FUNCTION__ );
+   rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
+   rmesa->radeon.swtcl.RenderIndex = ~0;
+   rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES;
+   rmesa->radeon.swtcl.hw_primitive = 0;
 }
+
diff --git a/src/mesa/drivers/dri/r200/r200_swtcl.h b/src/mesa/drivers/dri/r200/r200_swtcl.h
index 8c29fd0c999..b0905879d7a 100644
--- a/src/mesa/drivers/dri/r200/r200_swtcl.h
+++ b/src/mesa/drivers/dri/r200/r200_swtcl.h
@@ -39,7 +39,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_context.h"
 
 extern void r200InitSwtcl( GLcontext *ctx );
-extern void r200DestroySwtcl( GLcontext *ctx );
 
 extern void r200ChooseRenderState( GLcontext *ctx );
 extern void r200ChooseVertexState( GLcontext *ctx );
@@ -52,15 +51,11 @@ extern void r200BuildVertices( GLcontext *ctx, GLuint start, GLuint count,
 extern void r200PrintSetupFlags(char *msg, GLuint flags );
 
 
-extern void r200_emit_indexed_verts( GLcontext *ctx,
-				       GLuint start,
-				       GLuint count );
-
 extern void r200_translate_vertex( GLcontext *ctx, 
-				     const r200Vertex *src, 
+				     const radeonVertex *src, 
 				     SWvertex *dst );
 
-extern void r200_print_vertex( GLcontext *ctx, const r200Vertex *v );
+extern void r200_print_vertex( GLcontext *ctx, const radeonVertex *v );
 
 extern void r200_import_float_colors( GLcontext *ctx );
 extern void r200_import_float_spec_colors( GLcontext *ctx );
@@ -70,5 +65,5 @@ extern void r200PointsBitmap( GLcontext *ctx, GLint px, GLint py,
 			      const struct gl_pixelstore_attrib *unpack,
 			      const GLubyte *bitmap );
 
-
+void r200_swtcl_flush(GLcontext *ctx, uint32_t current_offset);
 #endif
diff --git a/src/mesa/drivers/dri/r200/r200_tcl.c b/src/mesa/drivers/dri/r200/r200_tcl.c
index 99aecfe1e90..580370933ee 100644
--- a/src/mesa/drivers/dri/r200/r200_tcl.c
+++ b/src/mesa/drivers/dri/r200/r200_tcl.c
@@ -123,7 +123,7 @@ static GLboolean discrete_prim[0x10] = {
 
 #define RESET_STIPPLE() do {			\
    R200_STATECHANGE( rmesa, lin );		\
-   r200EmitState( rmesa );			\
+   radeonEmitState(&rmesa->radeon);			\
 } while (0)
 
 #define AUTO_STIPPLE( mode )  do {		\
@@ -134,7 +134,7 @@ static GLboolean discrete_prim[0x10] = {
    else						\
       rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &=	\
 	 ~R200_LINE_PATTERN_AUTO_RESET;	\
-   r200EmitState( rmesa );			\
+   radeonEmitState(&rmesa->radeon);			\
 } while (0)
 
 
@@ -142,26 +142,24 @@ static GLboolean discrete_prim[0x10] = {
 
 static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr ) 
 {
-   if (rmesa->dma.flush == r200FlushElts &&
-       rmesa->store.cmd_used + nr*2 < R200_CMD_BUF_SZ) {
+   if (rmesa->radeon.dma.flush == r200FlushElts &&
+       rmesa->tcl.elt_used + nr*2 < R200_ELT_BUF_SZ) {
 
-      GLushort *dest = (GLushort *)(rmesa->store.cmd_buf +
-				    rmesa->store.cmd_used);
+      GLushort *dest = (GLushort *)(rmesa->radeon.tcl.elt_dma_bo->ptr +
+				    rmesa->tcl.elt_used);
 
-      rmesa->store.cmd_used += nr*2;
+      rmesa->tcl.elt_used += nr*2;
 
       return dest;
    }
    else {
-      if (rmesa->dma.flush)
-	 rmesa->dma.flush( rmesa );
+      if (rmesa->radeon.dma.flush)
+	 rmesa->radeon.dma.flush( rmesa->radeon.glCtx );
 
-      r200EnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
-			     rmesa->hw.max_state_size + ELTS_BUFSZ(nr) );
+      rcommonEnsureCmdBufSpace(&rmesa->radeon, AOS_BUFSZ(rmesa->radeon.tcl.aos_count), __FUNCTION__);
 
       r200EmitAOS( rmesa,
-		   rmesa->tcl.aos_components,
-		   rmesa->tcl.nr_aos_components, 0 );
+		   rmesa->radeon.tcl.aos_count, 0 );
 
       return r200AllocEltsOpenEnded( rmesa, rmesa->tcl.hw_primitive, nr );
    }
@@ -188,13 +186,14 @@ static void r200EmitPrim( GLcontext *ctx,
    r200ContextPtr rmesa = R200_CONTEXT( ctx );
    r200TclPrimitive( ctx, prim, hwprim );
    
-   r200EnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
-			  rmesa->hw.max_state_size + VBUF_BUFSZ );
+   //   fprintf(stderr,"Emit prim %d\n", rmesa->radeon.tcl.aos_count);
+   rcommonEnsureCmdBufSpace( &rmesa->radeon,
+			     AOS_BUFSZ(rmesa->radeon.tcl.aos_count) +
+			     rmesa->radeon.hw.max_state_size + VBUF_BUFSZ, __FUNCTION__ );
 
    r200EmitAOS( rmesa,
-		  rmesa->tcl.aos_components,
-		  rmesa->tcl.nr_aos_components,
-		  start );
+		rmesa->radeon.tcl.aos_count,
+		start );
    
    /* Why couldn't this packet have taken an offset param?
     */
@@ -394,7 +393,7 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
 
    /* TODO: separate this from the swtnl pipeline 
     */
-   if (rmesa->TclFallback)
+   if (rmesa->radeon.TclFallback)
       return GL_TRUE;	/* fallback to software t&l */
 
    if (R200_DEBUG & DEBUG_PRIMS)
@@ -405,8 +404,9 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
 
    /* Validate state:
     */
-   if (rmesa->NewGLState)
-      r200ValidateState( ctx );
+   if (rmesa->radeon.NewGLState)
+      if (!r200ValidateState( ctx ))
+         return GL_TRUE; /* fallback to sw t&l */
 
    if (!ctx->VertexProgram._Enabled) {
    /* NOTE: inputs != tnl->render_inputs - these are the untransformed
@@ -481,7 +481,7 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
 
    /* Do the actual work:
     */
-   r200ReleaseArrays( ctx, ~0 /* stage->changed_inputs */ );
+   radeonReleaseArrays( ctx, ~0 /* stage->changed_inputs */ );
    r200EmitArrays( ctx, vimap_rev );
 
    rmesa->tcl.Elts = VB->Elts;
@@ -545,7 +545,7 @@ static void transition_to_swtnl( GLcontext *ctx )
    tnl->Driver.NotifyMaterialChange = 
       _mesa_validate_all_lighting_tables;
 
-   r200ReleaseArrays( ctx, ~0 );
+   radeonReleaseArrays( ctx, ~0 );
 
    /* Still using the D3D based hardware-rasterizer from the radeon;
     * need to put the card into D3D mode to make it work:
@@ -565,15 +565,11 @@ static void transition_to_hwtnl( GLcontext *ctx )
 
    tnl->Driver.NotifyMaterialChange = r200UpdateMaterial;
 
-   if ( rmesa->dma.flush )			
-      rmesa->dma.flush( rmesa );	
+   if ( rmesa->radeon.dma.flush )			
+      rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	
 
-   rmesa->dma.flush = NULL;
+   rmesa->radeon.dma.flush = NULL;
    
-   if (rmesa->swtcl.indexed_verts.buf) 
-      r200ReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, 
-			      __FUNCTION__ );
-
    R200_STATECHANGE( rmesa, vap );
    rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] |= R200_VAP_TCL_ENABLE;
    rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] &= ~R200_VAP_FORCE_W_TO_ONE;
@@ -631,10 +627,10 @@ static char *getFallbackString(GLuint bit)
 void r200TclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLuint oldfallback = rmesa->TclFallback;
+   GLuint oldfallback = rmesa->radeon.TclFallback;
 
    if (mode) {
-      rmesa->TclFallback |= bit;
+      rmesa->radeon.TclFallback |= bit;
       if (oldfallback == 0) {
 	 if (R200_DEBUG & DEBUG_FALLBACKS) 
 	    fprintf(stderr, "R200 begin tcl fallback %s\n",
@@ -643,7 +639,7 @@ void r200TclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
       }
    }
    else {
-      rmesa->TclFallback &= ~bit;
+      rmesa->radeon.TclFallback &= ~bit;
       if (oldfallback == bit) {
 	 if (R200_DEBUG & DEBUG_FALLBACKS) 
 	    fprintf(stderr, "R200 end tcl fallback %s\n",
diff --git a/src/mesa/drivers/dri/r200/r200_tex.c b/src/mesa/drivers/dri/r200/r200_tex.c
index 259f35a34c3..9f791579158 100644
--- a/src/mesa/drivers/dri/r200/r200_tex.c
+++ b/src/mesa/drivers/dri/r200/r200_tex.c
@@ -43,8 +43,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/teximage.h"
 #include "main/texobj.h"
 
-#include "texmem.h"
-
+#include "radeon_mipmap_tree.h"
 #include "r200_context.h"
 #include "r200_state.h"
 #include "r200_ioctl.h"
@@ -63,10 +62,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * \param twrap Wrap mode for the \a t texture coordinate
  */
 
-static void r200SetTexWrap( r200TexObjPtr t, GLenum swrap, GLenum twrap, GLenum rwrap )
+static void r200SetTexWrap( radeonTexObjPtr t, GLenum swrap, GLenum twrap, GLenum rwrap )
 {
    GLboolean  is_clamp = GL_FALSE;
    GLboolean  is_clamp_to_border = GL_FALSE;
+   struct gl_texture_object *tObj = &t->base;
 
    t->pp_txfilter &= ~(R200_CLAMP_S_MASK | R200_CLAMP_T_MASK | R200_BORDER_MODE_D3D);
 
@@ -103,7 +103,7 @@ static void r200SetTexWrap( r200TexObjPtr t, GLenum swrap, GLenum twrap, GLenum
       _mesa_problem(NULL, "bad S wrap mode in %s", __FUNCTION__);
    }
 
-   if (t->base.tObj->Target != GL_TEXTURE_1D) {
+   if (tObj->Target != GL_TEXTURE_1D) {
       switch ( twrap ) {
       case GL_REPEAT:
          t->pp_txfilter |= R200_CLAMP_T_WRAP;
@@ -180,7 +180,7 @@ static void r200SetTexWrap( r200TexObjPtr t, GLenum swrap, GLenum twrap, GLenum
    t->border_fallback = (is_clamp && is_clamp_to_border);
 }
 
-static void r200SetTexMaxAnisotropy( r200TexObjPtr t, GLfloat max )
+static void r200SetTexMaxAnisotropy( radeonTexObjPtr t, GLfloat max )
 {
    t->pp_txfilter &= ~R200_MAX_ANISO_MASK;
 
@@ -205,10 +205,13 @@ static void r200SetTexMaxAnisotropy( r200TexObjPtr t, GLfloat max )
  * \param magf Texture magnification mode
  */
 
-static void r200SetTexFilter( r200TexObjPtr t, GLenum minf, GLenum magf )
+static void r200SetTexFilter( radeonTexObjPtr t, GLenum minf, GLenum magf )
 {
    GLuint anisotropy = (t->pp_txfilter & R200_MAX_ANISO_MASK);
 
+   /* Force revalidation to account for switches from/to mipmapping. */
+   t->validated = GL_FALSE;
+
    t->pp_txfilter &= ~(R200_MIN_FILTER_MASK | R200_MAG_FILTER_MASK);
    t->pp_txformat_x &= ~R200_VOLUME_FILTER_MASK;
 
@@ -267,701 +270,16 @@ static void r200SetTexFilter( r200TexObjPtr t, GLenum minf, GLenum magf )
    }
 }
 
-static void r200SetTexBorderColor( r200TexObjPtr t, const GLfloat color[4] )
+static void r200SetTexBorderColor( radeonTexObjPtr t, const GLfloat color[4] )
 {
    GLubyte c[4];
    CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]);
    CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]);
    CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]);
    CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]);
-   t->pp_border_color = r200PackColor( 4, c[0], c[1], c[2], c[3] );
-}
-
-
-/**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-static r200TexObjPtr r200AllocTexObj( struct gl_texture_object *texObj )
-{
-   r200TexObjPtr t;
-
-   t = CALLOC_STRUCT( r200_tex_obj );
-   texObj->DriverData = t;
-   if ( t != NULL ) {
-      if ( R200_DEBUG & DEBUG_TEXTURE ) {
-	 fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, (void *)texObj, 
-		  (void *)t );
-      }
-
-      /* Initialize non-image-dependent parts of the state:
-       */
-      t->base.tObj = texObj;
-      t->border_fallback = GL_FALSE;
-
-      make_empty_list( & t->base );
-
-      r200SetTexWrap( t, texObj->WrapS, texObj->WrapT, texObj->WrapR );
-      r200SetTexMaxAnisotropy( t, texObj->MaxAnisotropy );
-      r200SetTexFilter( t, texObj->MinFilter, texObj->MagFilter );
-      r200SetTexBorderColor( t, texObj->BorderColor );
-   }
-
-   return t;
-}
-
-/* try to find a format which will only need a memcopy */
-static const struct gl_texture_format *
-r200Choose8888TexFormat( GLenum srcFormat, GLenum srcType )
-{
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
-
-   if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
-       (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
-       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
-       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && littleEndian)) {
-      return &_mesa_texformat_rgba8888;
-   }
-   else if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
-       (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && littleEndian) ||
-       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
-       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && !littleEndian)) {
-      return &_mesa_texformat_rgba8888_rev;
-   }
-   else return _dri_texformat_argb8888;
-}
-
-static const struct gl_texture_format *
-r200ChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
-                           GLenum format, GLenum type )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   const GLboolean do32bpt =
-       ( rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32 );
-   const GLboolean force16bpt =
-       ( rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16 );
-   (void) format;
-
-   switch ( internalFormat ) {
-   case 4:
-   case GL_RGBA:
-   case GL_COMPRESSED_RGBA:
-      switch ( type ) {
-      case GL_UNSIGNED_INT_10_10_10_2:
-      case GL_UNSIGNED_INT_2_10_10_10_REV:
-	 return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_argb1555;
-      case GL_UNSIGNED_SHORT_4_4_4_4:
-      case GL_UNSIGNED_SHORT_4_4_4_4_REV:
-	 return _dri_texformat_argb4444;
-      case GL_UNSIGNED_SHORT_5_5_5_1:
-      case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-	 return _dri_texformat_argb1555;
-      default:
-         return do32bpt ?
-	    r200Choose8888TexFormat(format, type) : _dri_texformat_argb4444;
-      }
-
-   case 3:
-   case GL_RGB:
-   case GL_COMPRESSED_RGB:
-      switch ( type ) {
-      case GL_UNSIGNED_SHORT_4_4_4_4:
-      case GL_UNSIGNED_SHORT_4_4_4_4_REV:
-	 return _dri_texformat_argb4444;
-      case GL_UNSIGNED_SHORT_5_5_5_1:
-      case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-	 return _dri_texformat_argb1555;
-      case GL_UNSIGNED_SHORT_5_6_5:
-      case GL_UNSIGNED_SHORT_5_6_5_REV:
-	 return _dri_texformat_rgb565;
-      default:
-         return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_rgb565;
-      }
-
-   case GL_RGBA8:
-   case GL_RGB10_A2:
-   case GL_RGBA12:
-   case GL_RGBA16:
-      return !force16bpt ?
-	  r200Choose8888TexFormat(format, type) : _dri_texformat_argb4444;
-
-   case GL_RGBA4:
-   case GL_RGBA2:
-      return _dri_texformat_argb4444;
-
-   case GL_RGB5_A1:
-      return _dri_texformat_argb1555;
-
-   case GL_RGB8:
-   case GL_RGB10:
-   case GL_RGB12:
-   case GL_RGB16:
-      return !force16bpt ? _dri_texformat_argb8888 : _dri_texformat_rgb565;
-
-   case GL_RGB5:
-   case GL_RGB4:
-   case GL_R3_G3_B2:
-      return _dri_texformat_rgb565;
-
-   case GL_ALPHA:
-   case GL_ALPHA4:
-   case GL_ALPHA8:
-   case GL_ALPHA12:
-   case GL_ALPHA16:
-   case GL_COMPRESSED_ALPHA:
-   /* can't use a8 format since interpreting hw I8 as a8 would result
-      in wrong rgb values (same as alpha value instead of 0). */
-      return _dri_texformat_al88;
-
-   case 1:
-   case GL_LUMINANCE:
-   case GL_LUMINANCE4:
-   case GL_LUMINANCE8:
-   case GL_LUMINANCE12:
-   case GL_LUMINANCE16:
-   case GL_COMPRESSED_LUMINANCE:
-      return _dri_texformat_l8;
-
-   case 2:
-   case GL_LUMINANCE_ALPHA:
-   case GL_LUMINANCE4_ALPHA4:
-   case GL_LUMINANCE6_ALPHA2:
-   case GL_LUMINANCE8_ALPHA8:
-   case GL_LUMINANCE12_ALPHA4:
-   case GL_LUMINANCE12_ALPHA12:
-   case GL_LUMINANCE16_ALPHA16:
-   case GL_COMPRESSED_LUMINANCE_ALPHA:
-      return _dri_texformat_al88;
-
-   case GL_INTENSITY:
-   case GL_INTENSITY4:
-   case GL_INTENSITY8:
-   case GL_INTENSITY12:
-   case GL_INTENSITY16:
-   case GL_COMPRESSED_INTENSITY:
-       return _dri_texformat_i8;
-
-   case GL_YCBCR_MESA:
-      if (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
-          type == GL_UNSIGNED_BYTE)
-         return &_mesa_texformat_ycbcr;
-      else
-         return &_mesa_texformat_ycbcr_rev;
-
-   case GL_RGB_S3TC:
-   case GL_RGB4_S3TC:
-   case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
-      return &_mesa_texformat_rgb_dxt1;
-
-   case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
-      return &_mesa_texformat_rgba_dxt1;
-
-   case GL_RGBA_S3TC:
-   case GL_RGBA4_S3TC:
-   case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
-      return &_mesa_texformat_rgba_dxt3;
-
-   case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
-      return &_mesa_texformat_rgba_dxt5;
-
-   default:
-      _mesa_problem(ctx,
-         "unexpected internalFormat 0x%x in r200ChooseTextureFormat",
-         (int) internalFormat);
-      return NULL;
-   }
-
-   return NULL; /* never get here */
-}
-
-
-static GLboolean
-r200ValidateClientStorage( GLcontext *ctx, GLenum target,
-			   GLint internalFormat,
-			   GLint srcWidth, GLint srcHeight, 
-                           GLenum format, GLenum type,  const void *pixels,
-			   const struct gl_pixelstore_attrib *packing,
-			   struct gl_texture_object *texObj,
-			   struct gl_texture_image *texImage)
-
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if ( R200_DEBUG & DEBUG_TEXTURE )
-      fprintf(stderr, "intformat %s format %s type %s\n",
-	      _mesa_lookup_enum_by_nr( internalFormat ),
-	      _mesa_lookup_enum_by_nr( format ),
-	      _mesa_lookup_enum_by_nr( type ));
-
-   if (!ctx->Unpack.ClientStorage)
-      return 0;
-
-   if (ctx->_ImageTransferState ||
-       texImage->IsCompressed ||
-       texObj->GenerateMipmap)
-      return 0;
-
-
-   /* This list is incomplete, may be different on ppc???
-    */
-   switch ( internalFormat ) {
-   case GL_RGBA:
-      if ( format == GL_BGRA && type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
-	 texImage->TexFormat = _dri_texformat_argb8888;
-      }
-      else
-	 return 0;
-      break;
-
-   case GL_RGB:
-      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
-	 texImage->TexFormat = _dri_texformat_rgb565;
-      }
-      else
-	 return 0;
-      break;
-
-   case GL_YCBCR_MESA:
-      if ( format == GL_YCBCR_MESA && 
-	   type == GL_UNSIGNED_SHORT_8_8_REV_APPLE ) {
-	 texImage->TexFormat = &_mesa_texformat_ycbcr_rev;
-      }
-      else if ( format == GL_YCBCR_MESA && 
-		(type == GL_UNSIGNED_SHORT_8_8_APPLE || 
-		 type == GL_UNSIGNED_BYTE)) {
-	 texImage->TexFormat = &_mesa_texformat_ycbcr;
-      }
-      else
-	 return 0;
-      break;
-
-   default:
-      return 0;
-   }
-
-   /* Could deal with these packing issues, but currently don't:
-    */
-   if (packing->SkipPixels || 
-       packing->SkipRows || 
-       packing->SwapBytes ||
-       packing->LsbFirst) {
-      return 0;
-   }
-
-   {      
-      GLint srcRowStride = _mesa_image_row_stride(packing, srcWidth,
-						  format, type);
-
-      
-      if ( R200_DEBUG & DEBUG_TEXTURE )
-	 fprintf(stderr, "%s: srcRowStride %d/%x\n", 
-		 __FUNCTION__, srcRowStride, srcRowStride);
-
-      /* Could check this later in upload, pitch restrictions could be
-       * relaxed, but would need to store the image pitch somewhere,
-       * as packing details might change before image is uploaded:
-       */
-      if (!r200IsGartMemory( rmesa, pixels, srcHeight * srcRowStride ) ||
-	  (srcRowStride & 63))
-	 return 0;
-
-
-      /* Have validated that _mesa_transfer_teximage would be a straight
-       * memcpy at this point.  NOTE: future calls to TexSubImage will
-       * overwrite the client data.  This is explicitly mentioned in the
-       * extension spec.
-       */
-      texImage->Data = (void *)pixels;
-      texImage->IsClientData = GL_TRUE;
-      texImage->RowStride = srcRowStride / texImage->TexFormat->TexelBytes;
-
-      return 1;
-   }
-}
-
-
-static void r200TexImage1D( GLcontext *ctx, GLenum target, GLint level,
-                              GLint internalFormat,
-                              GLint width, GLint border,
-                              GLenum format, GLenum type, const GLvoid *pixels,
-                              const struct gl_pixelstore_attrib *packing,
-                              struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
-         return;
-      }
-   }
-
-   /* Note, this will call ChooseTextureFormat */
-   _mesa_store_teximage1d(ctx, target, level, internalFormat,
-                          width, border, format, type, pixels,
-                          &ctx->Unpack, texObj, texImage);
-
-   t->dirty_images[0] |= (1 << level);
-}
-
-
-static void r200TexSubImage1D( GLcontext *ctx, GLenum target, GLint level,
-                                 GLint xoffset,
-                                 GLsizei width,
-                                 GLenum format, GLenum type,
-                                 const GLvoid *pixels,
-                                 const struct gl_pixelstore_attrib *packing,
-                                 struct gl_texture_object *texObj,
-                                 struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D");
-         return;
-      }
-   }
-
-   _mesa_store_texsubimage1d(ctx, target, level, xoffset, width,
-			     format, type, pixels, packing, texObj,
-			     texImage);
-
-   t->dirty_images[0] |= (1 << level);
-}
-
-
-static void r200TexImage2D( GLcontext *ctx, GLenum target, GLint level,
-                              GLint internalFormat,
-                              GLint width, GLint height, GLint border,
-                              GLenum format, GLenum type, const GLvoid *pixels,
-                              const struct gl_pixelstore_attrib *packing,
-                              struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   if ( t != NULL ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
-         return;
-      }
-   }
-
-   texImage->IsClientData = GL_FALSE;
-
-   if (r200ValidateClientStorage( ctx, target, 
-				  internalFormat, 
-				  width, height, 
-				  format, type, pixels, 
-				  packing, texObj, texImage)) {
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using client storage\n", __FUNCTION__); 
-   }
-   else {
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using normal storage\n", __FUNCTION__); 
-
-      /* Normal path: copy (to cached memory) and eventually upload
-       * via another copy to GART memory and then a blit...  Could
-       * eliminate one copy by going straight to (permanent) GART.
-       *
-       * Note, this will call r200ChooseTextureFormat.
-       */
-      _mesa_store_teximage2d(ctx, target, level, internalFormat,
-			     width, height, border, format, type, pixels,
-			     &ctx->Unpack, texObj, texImage);
-      
-      t->dirty_images[face] |= (1 << level);
-   }
-}
-
-
-static void r200TexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
-                                 GLint xoffset, GLint yoffset,
-                                 GLsizei width, GLsizei height,
-                                 GLenum format, GLenum type,
-                                 const GLvoid *pixels,
-                                 const struct gl_pixelstore_attrib *packing,
-                                 struct gl_texture_object *texObj,
-                                 struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D");
-         return;
-      }
-   }
-
-   _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
-			     height, format, type, pixels, packing, texObj,
-			     texImage);
-
-   t->dirty_images[face] |= (1 << level);
-}
-
-
-static void r200CompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level,
-                              GLint internalFormat,
-                              GLint width, GLint height, GLint border,
-                              GLsizei imageSize, const GLvoid *data,
-                              struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   if ( t != NULL ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage2D");
-         return;
-      }
-   }
-
-   texImage->IsClientData = GL_FALSE;
-/* can't call this, different parameters. Would never evaluate to true anyway currently
-   if (r200ValidateClientStorage( ctx, target, 
-				  internalFormat,
-				  width, height,
-				  format, type, pixels,
-				  packing, texObj, texImage)) {
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using client storage\n", __FUNCTION__);
-   }
-   else */{
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using normal storage\n", __FUNCTION__);
-
-      /* Normal path: copy (to cached memory) and eventually upload
-       * via another copy to GART memory and then a blit...  Could
-       * eliminate one copy by going straight to (permanent) GART.
-       *
-       * Note, this will call r200ChooseTextureFormat.
-       */
-      _mesa_store_compressed_teximage2d(ctx, target, level, internalFormat, width,
-                                 height, border, imageSize, data, texObj, texImage);
-
-      t->dirty_images[face] |= (1 << level);
-   }
+   t->pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
 }
 
-
-static void r200CompressedTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
-                                 GLint xoffset, GLint yoffset,
-                                 GLsizei width, GLsizei height,
-                                 GLenum format,
-                                 GLsizei imageSize, const GLvoid *data,
-                                 struct gl_texture_object *texObj,
-                                 struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexSubImage2D");
-         return;
-      }
-   }
-
-   _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
-                            height, format, imageSize, data, texObj, texImage);
-
-   t->dirty_images[face] |= (1 << level);
-}
-
-
-#if ENABLE_HW_3D_TEXTURE
-static void r200TexImage3D( GLcontext *ctx, GLenum target, GLint level,
-                            GLint internalFormat,
-                            GLint width, GLint height, GLint depth,
-                            GLint border,
-                            GLenum format, GLenum type, const GLvoid *pixels,
-                            const struct gl_pixelstore_attrib *packing,
-                            struct gl_texture_object *texObj,
-                            struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage3D");
-         return;
-      }
-   }
-
-   texImage->IsClientData = GL_FALSE;
-
-#if 0
-   if (r200ValidateClientStorage( ctx, target, 
-				  internalFormat, 
-				  width, height, 
-				  format, type, pixels, 
-				  packing, texObj, texImage)) {
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using client storage\n", __FUNCTION__); 
-   }
-   else
-#endif
-   {
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using normal storage\n", __FUNCTION__); 
-
-      /* Normal path: copy (to cached memory) and eventually upload
-       * via another copy to GART memory and then a blit...  Could
-       * eliminate one copy by going straight to (permanent) GART.
-       *
-       * Note, this will call r200ChooseTextureFormat.
-       */
-      _mesa_store_teximage3d(ctx, target, level, internalFormat,
-			     width, height, depth, border,
-                             format, type, pixels,
-			     &ctx->Unpack, texObj, texImage);
-      
-      t->dirty_images[0] |= (1 << level);
-   }
-}
-#endif
-
-
-#if ENABLE_HW_3D_TEXTURE
-static void
-r200TexSubImage3D( GLcontext *ctx, GLenum target, GLint level,
-                   GLint xoffset, GLint yoffset, GLint zoffset,
-                   GLsizei width, GLsizei height, GLsizei depth,
-                   GLenum format, GLenum type,
-                   const GLvoid *pixels,
-                   const struct gl_pixelstore_attrib *packing,
-                   struct gl_texture_object *texObj,
-                   struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-/*     fprintf(stderr, "%s\n", __FUNCTION__); */
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) r200AllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage3D");
-         return;
-      }
-      texObj->DriverData = t;
-   }
-
-   _mesa_store_texsubimage3d(ctx, target, level, xoffset, yoffset, zoffset,
-                             width, height, depth,
-                             format, type, pixels, packing, texObj, texImage);
-
-   t->dirty_images[0] |= (1 << level);
-}
-#endif
-
-
-
 static void r200TexEnv( GLcontext *ctx, GLenum target,
 			  GLenum pname, const GLfloat *param )
 {
@@ -983,7 +301,7 @@ static void r200TexEnv( GLcontext *ctx, GLenum target,
       GLubyte c[4];
       GLuint envColor;
       UNCLAMPED_FLOAT_TO_RGBA_CHAN( c, texUnit->EnvColor );
-      envColor = r200PackColor( 4, c[0], c[1], c[2], c[3] );
+      envColor = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
       if ( rmesa->hw.tf.cmd[TF_TFACTOR_0 + unit] != envColor ) {
 	 R200_STATECHANGE( rmesa, tf );
 	 rmesa->hw.tf.cmd[TF_TFACTOR_0 + unit] = envColor;
@@ -1002,7 +320,7 @@ static void r200TexEnv( GLcontext *ctx, GLenum target,
        * NOTE: Add a small bias to the bias for conform mipsel.c test.
        */
       bias = *param + .01;
-      min = driQueryOptionb (&rmesa->optionCache, "no_neg_lod_bias") ?
+      min = driQueryOptionb (&rmesa->radeon.optionCache, "no_neg_lod_bias") ?
 	  0.0 : -16.0;
       bias = CLAMP( bias, min, 16.0 );
       b = (int)(bias * fixed_one) & R200_LOD_BIAS_MASK;
@@ -1039,7 +357,7 @@ static void r200TexParameter( GLcontext *ctx, GLenum target,
 				struct gl_texture_object *texObj,
 				GLenum pname, const GLfloat *params )
 {
-   r200TexObjPtr t = (r200TexObjPtr) texObj->DriverData;
+   radeonTexObj* t = radeon_tex_obj(texObj);
 
    if ( R200_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
       fprintf( stderr, "%s( %s )\n", __FUNCTION__,
@@ -1073,59 +391,46 @@ static void r200TexParameter( GLcontext *ctx, GLenum target,
        * we just have to rely on loading the right subset of mipmap levels
        * to simulate a clamped LOD.
        */
-      driSwapOutTextureObject( (driTextureObject *) t );
+      if (t->mt) {
+         radeon_miptree_unreference(t->mt);
+	 t->mt = 0;
+	 t->validated = GL_FALSE;
+      }
       break;
 
    default:
       return;
    }
-
-   /* Mark this texobj as dirty (one bit per tex unit)
-    */
-   t->dirty_state = TEX_ALL;
 }
 
 
-
-static void r200BindTexture( GLcontext *ctx, GLenum target,
-			       struct gl_texture_object *texObj )
-{
-   if ( R200_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
-      fprintf( stderr, "%s( %p ) unit=%d\n", __FUNCTION__, (void *)texObj,
-	       ctx->Texture.CurrentUnit );
-   }
-
-   if ( (target == GL_TEXTURE_1D)
-	|| (target == GL_TEXTURE_2D) 
-#if ENABLE_HW_3D_TEXTURE
-	|| (target == GL_TEXTURE_3D)
-#endif
-	|| (target == GL_TEXTURE_CUBE_MAP)
-	|| (target == GL_TEXTURE_RECTANGLE_NV) ) {
-      assert( texObj->DriverData != NULL );
-   }
-}
-
-
-static void r200DeleteTexture( GLcontext *ctx,
-				 struct gl_texture_object *texObj )
+static void r200DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   if ( R200_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
-      fprintf( stderr, "%s( %p (target = %s) )\n", __FUNCTION__, (void *)texObj,
-	       _mesa_lookup_enum_by_nr( texObj->Target ) );
+   radeonTexObj* t = radeon_tex_obj(texObj);
+
+   if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) {
+      fprintf(stderr, "%s( %p (target = %s) )\n", __FUNCTION__,
+	      (void *)texObj,
+	      _mesa_lookup_enum_by_nr(texObj->Target));
+   }
+   
+   if (rmesa) {
+      int i;
+      radeon_firevertices(&rmesa->radeon);
+      for ( i = 0 ; i < rmesa->radeon.glCtx->Const.MaxTextureUnits ; i++ ) {
+	 if ( t == rmesa->state.texture.unit[i].texobj ) {
+	    rmesa->state.texture.unit[i].texobj = NULL;
+	    rmesa->hw.tex[i].dirty = GL_FALSE;
+	    rmesa->hw.cube[i].dirty = GL_FALSE;
+	 }
+      }      
    }
-
-   if ( t != NULL ) {
-      if ( rmesa ) {
-         R200_FIREVERTICES( rmesa );
-      }
-
-      driDestroyTextureObject( t );
+   
+   if (t->mt) {
+      radeon_miptree_unreference(t->mt);
+      t->mt = 0;
    }
-   /* Free mipmap images and the texture object itself */
    _mesa_delete_texture_object(ctx, texObj);
 }
 
@@ -1155,46 +460,59 @@ static void r200TexGen( GLcontext *ctx,
  * Called via ctx->Driver.NewTextureObject.
  * Note: this function will be called during context creation to
  * allocate the default texture objects.
- * Note: we could use containment here to 'derive' the driver-specific
- * texture object from the core mesa gl_texture_object.  Not done at this time.
  * Fixup MaxAnisotropy according to user preference.
  */
-static struct gl_texture_object *
-r200NewTextureObject( GLcontext *ctx, GLuint name, GLenum target )
+static struct gl_texture_object *r200NewTextureObject(GLcontext * ctx,
+						      GLuint name,
+						      GLenum target)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_texture_object *obj;
-   obj = _mesa_new_texture_object(ctx, name, target);
-   if (!obj)
-      return NULL;
-   obj->MaxAnisotropy = rmesa->initialMaxAnisotropy;
-   r200AllocTexObj( obj );
-   return obj;
+   radeonTexObj* t = CALLOC_STRUCT(radeon_tex_obj);
+
+
+   if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) {
+     fprintf(stderr, "%s( %p (target = %s) )\n", __FUNCTION__,
+	     t, _mesa_lookup_enum_by_nr(target));
+   }
+
+   _mesa_initialize_texture_object(&t->base, name, target);
+   t->base.MaxAnisotropy = rmesa->radeon.initialMaxAnisotropy;
+
+   /* Initialize hardware state */
+   r200SetTexWrap( t, t->base.WrapS, t->base.WrapT, t->base.WrapR );
+   r200SetTexMaxAnisotropy( t, t->base.MaxAnisotropy );
+   r200SetTexFilter(t, t->base.MinFilter, t->base.MagFilter);
+   r200SetTexBorderColor(t, t->base.BorderColor);
+
+   return &t->base;
 }
 
 
+
 void r200InitTextureFuncs( struct dd_function_table *functions )
 {
    /* Note: we only plug in the functions we implement in the driver
     * since _mesa_init_driver_functions() was already called.
     */
-   functions->ChooseTextureFormat	= r200ChooseTextureFormat;
-   functions->TexImage1D		= r200TexImage1D;
-   functions->TexImage2D		= r200TexImage2D;
+   functions->ChooseTextureFormat	= radeonChooseTextureFormat_mesa;
+   functions->TexImage1D		= radeonTexImage1D;
+   functions->TexImage2D		= radeonTexImage2D;
 #if ENABLE_HW_3D_TEXTURE
-   functions->TexImage3D		= r200TexImage3D;
+   functions->TexImage3D		= radeonTexImage3D;
 #else
    functions->TexImage3D		= _mesa_store_teximage3d;
 #endif
-   functions->TexSubImage1D		= r200TexSubImage1D;
-   functions->TexSubImage2D		= r200TexSubImage2D;
+   functions->TexSubImage1D		= radeonTexSubImage1D;
+   functions->TexSubImage2D		= radeonTexSubImage2D;
 #if ENABLE_HW_3D_TEXTURE
-   functions->TexSubImage3D		= r200TexSubImage3D;
+   functions->TexSubImage3D		= radeonTexSubImage3D;
 #else
    functions->TexSubImage3D		= _mesa_store_texsubimage3d;
 #endif
+   functions->GetTexImage               = radeonGetTexImage;
+   functions->GetCompressedTexImage     = radeonGetCompressedTexImage;
    functions->NewTextureObject		= r200NewTextureObject;
-   functions->BindTexture		= r200BindTexture;
+   //   functions->BindTexture		= r200BindTexture;
    functions->DeleteTexture		= r200DeleteTexture;
    functions->IsTextureResident		= driIsTextureResident;
 
@@ -1202,22 +520,16 @@ void r200InitTextureFuncs( struct dd_function_table *functions )
    functions->TexParameter		= r200TexParameter;
    functions->TexGen			= r200TexGen;
 
-   functions->CompressedTexImage2D	= r200CompressedTexImage2D;
-   functions->CompressedTexSubImage2D	= r200CompressedTexSubImage2D;
+   functions->CompressedTexImage2D	= radeonCompressedTexImage2D;
+   functions->CompressedTexSubImage2D	= radeonCompressedTexSubImage2D;
 
-   driInitTextureFormats();
+   functions->GenerateMipmap = radeonGenerateMipmap;
 
-#if 000
-   /* moved or obsolete code */
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   driInitTextureObjects( ctx, & rmesa->swapped,
-			  DRI_TEXMGR_DO_TEXTURE_1D
-			  | DRI_TEXMGR_DO_TEXTURE_2D );
+   functions->NewTextureImage = radeonNewTextureImage;
+   functions->FreeTexImageData = radeonFreeTexImageData;
+   functions->MapTexture = radeonMapTexture;
+   functions->UnmapTexture = radeonUnmapTexture;
+
+   driInitTextureFormats();
 
-   /* Hack: r200NewTextureObject is not yet installed when the
-    * default textures are created. Therefore set MaxAnisotropy of the
-    * default 2D texture now. */
-   ctx->Shared->Default2D->MaxAnisotropy = driQueryOptionf (&rmesa->optionCache,
-							    "def_max_anisotropy");
-#endif
 }
diff --git a/src/mesa/drivers/dri/r200/r200_tex.h b/src/mesa/drivers/dri/r200/r200_tex.h
index 10ff8e8a660..e122de6e5ed 100644
--- a/src/mesa/drivers/dri/r200/r200_tex.h
+++ b/src/mesa/drivers/dri/r200/r200_tex.h
@@ -35,15 +35,18 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifndef __R200_TEX_H__
 #define __R200_TEX_H__
 
+extern void r200SetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv);
+extern void r200SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format,
+			      __DRIdrawable *dPriv);
 extern void r200SetTexOffset(__DRIcontext *pDRICtx, GLint texname,
 			     unsigned long long offset, GLint depth,
 			     GLuint pitch);
 
 extern void r200UpdateTextureState( GLcontext *ctx );
 
-extern int r200UploadTexImages( r200ContextPtr rmesa, r200TexObjPtr t, GLuint face );
+extern int r200UploadTexImages( r200ContextPtr rmesa, radeonTexObjPtr t, GLuint face );
 
-extern void r200DestroyTexObj( r200ContextPtr rmesa, r200TexObjPtr t );
+extern void r200DestroyTexObj( r200ContextPtr rmesa, radeonTexObjPtr t );
 
 extern void r200InitTextureFuncs( struct dd_function_table *functions );
 
diff --git a/src/mesa/drivers/dri/r200/r200_texmem.c b/src/mesa/drivers/dri/r200/r200_texmem.c
deleted file mode 100644
index 3b81ac0c802..00000000000
--- a/src/mesa/drivers/dri/r200/r200_texmem.c
+++ /dev/null
@@ -1,530 +0,0 @@
-/**************************************************************************
-
-Copyright (C) Tungsten Graphics 2002.  All Rights Reserved.  
-The Weather Channel, Inc. funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86
-license. This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation on the rights to use, copy, modify, merge, publish,
-distribute, sub license, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-NON-INFRINGEMENT. IN NO EVENT SHALL ATI, VA LINUX SYSTEMS AND/OR THEIR
-SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR
-IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Kevin E. Martin <[email protected]>
- *   Gareth Hughes <[email protected]>
- *
- */
- 
-#include <errno.h>
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/context.h"
-#include "main/colormac.h"
-#include "main/macros.h"
-#include "r200_context.h"
-#include "r200_ioctl.h"
-#include "r200_tex.h"
-#include "radeon_reg.h"
-
-#include <unistd.h>  /* for usleep() */
-
-
-/**
- * Destroy any device-dependent state associated with the texture.  This may
- * include NULLing out hardware state that points to the texture.
- */
-void
-r200DestroyTexObj( r200ContextPtr rmesa, r200TexObjPtr t )
-{
-   if ( R200_DEBUG & DEBUG_TEXTURE ) {
-      fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, 
-	       (void *)t, (void *)t->base.tObj );
-   }
-
-   if ( rmesa != NULL ) {
-      unsigned   i;
-
-
-      for ( i = 0 ; i < rmesa->glCtx->Const.MaxTextureUnits ; i++ ) {
-	 if ( t == rmesa->state.texture.unit[i].texobj ) {
-	    rmesa->state.texture.unit[i].texobj = NULL;
-	    rmesa->hw.tex[i].dirty = GL_FALSE;
-	    rmesa->hw.cube[i].dirty = GL_FALSE;
-	 }
-      }
-   }
-}
-
-
-/* ------------------------------------------------------------
- * Texture image conversions
- */
-
-
-static void r200UploadGARTClientSubImage( r200ContextPtr rmesa,
-					  r200TexObjPtr t, 
-					  struct gl_texture_image *texImage,
-					  GLint hwlevel,
-					  GLint x, GLint y, 
-					  GLint width, GLint height )
-{
-   const struct gl_texture_format *texFormat = texImage->TexFormat;
-   GLuint srcPitch, dstPitch;
-   int blit_format;
-   int srcOffset;
-
-   /*
-    * XXX it appears that we always upload the full image, not a subimage.
-    * I.e. x==0, y==0, width=texWidth, height=texWidth.  If this is ever
-    * changed, the src pitch will have to change.
-    */
-   switch ( texFormat->TexelBytes ) {
-   case 1:
-      blit_format = R200_CP_COLOR_FORMAT_CI8;
-      srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-      dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-      break;
-   case 2:
-      blit_format = R200_CP_COLOR_FORMAT_RGB565;
-      srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-      dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-      break;
-   case 4:
-      blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
-      srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-      dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-      break;
-   default:
-      return;
-   }
-
-   t->image[0][hwlevel].data = texImage->Data;
-   srcOffset = r200GartOffsetFromVirtual( rmesa, texImage->Data );
-
-   assert( srcOffset != ~0 );
-
-   /* Don't currently need to cope with small pitches?
-    */
-   width = texImage->Width;
-   height = texImage->Height;
-
-   r200EmitWait( rmesa, RADEON_WAIT_3D );
-
-   r200EmitBlit( rmesa, blit_format, 
-		 srcPitch,  
-		 srcOffset,   
-		 dstPitch,
-		 t->bufAddr,
-		 x, 
-		 y, 
-		 t->image[0][hwlevel].x + x,
-		 t->image[0][hwlevel].y + y, 
-		 width,
-		 height );
-
-   r200EmitWait( rmesa, RADEON_WAIT_2D );
-}
-
-static void r200UploadRectSubImage( r200ContextPtr rmesa,
-				    r200TexObjPtr t, 
-				    struct gl_texture_image *texImage,
-				    GLint x, GLint y, 
-				    GLint width, GLint height )
-{
-   const struct gl_texture_format *texFormat = texImage->TexFormat;
-   int blit_format, dstPitch, done;
-
-   switch ( texFormat->TexelBytes ) {
-   case 1:
-      blit_format = R200_CP_COLOR_FORMAT_CI8;
-      break;
-   case 2:
-      blit_format = R200_CP_COLOR_FORMAT_RGB565;
-      break;
-   case 4:
-      blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
-      break;
-   default:
-      return;
-   }
-
-   t->image[0][0].data = texImage->Data;
-
-   /* Currently don't need to cope with small pitches.
-    */
-   width = texImage->Width;
-   height = texImage->Height;
-   dstPitch = t->pp_txpitch + 32;
-
-   if (rmesa->prefer_gart_client_texturing && texImage->IsClientData) {
-      /* In this case, could also use GART texturing.  This is
-       * currently disabled, but has been tested & works.
-       */
-      if ( !t->image_override )
-         t->pp_txoffset = r200GartOffsetFromVirtual( rmesa, texImage->Data );
-      t->pp_txpitch = texImage->RowStride * texFormat->TexelBytes - 32;
-
-      if (R200_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, 
-		 "Using GART texturing for rectangular client texture\n");
-
-      /* Release FB memory allocated for this image:
-       */
-      /* FIXME This may not be correct as driSwapOutTextureObject sets
-       * FIXME dirty_images.  It may be fine, though.
-       */
-      if ( t->base.memBlock ) {
-	 driSwapOutTextureObject( (driTextureObject *) t );
-      }
-   }
-   else if (texImage->IsClientData) {
-      /* Data already in GART memory, with usable pitch.
-       */
-      GLuint srcPitch;
-      srcPitch = texImage->RowStride * texFormat->TexelBytes;
-      r200EmitBlit( rmesa, 
-		    blit_format, 
-		    srcPitch,
-		    r200GartOffsetFromVirtual( rmesa, texImage->Data ),   
-		    dstPitch, t->bufAddr,
-		    0, 0, 
-		    0, 0, 
-		    width, height );
-   }
-   else {
-      /* Data not in GART memory, or bad pitch.
-       */
-      for (done = 0; done < height ; ) {
-	 struct r200_dma_region region;
-	 int lines = MIN2( height - done, RADEON_BUFFER_SIZE / dstPitch );
-	 int src_pitch;
-	 char *tex;
-
-         src_pitch = texImage->RowStride * texFormat->TexelBytes;
-
-	 tex = (char *)texImage->Data + done * src_pitch;
-
-	 memset(&region, 0, sizeof(region));
-	 r200AllocDmaRegion( rmesa, &region, lines * dstPitch, 1024 );
-
-	 /* Copy texdata to dma:
-	  */
-	 if (0)
-	    fprintf(stderr, "%s: src_pitch %d dst_pitch %d\n",
-		    __FUNCTION__, src_pitch, dstPitch);
-
-	 if (src_pitch == dstPitch) {
-	    memcpy( region.address + region.start, tex, lines * src_pitch );
-	 } 
-	 else {
-	    char *buf = region.address + region.start;
-	    int i;
-	    for (i = 0 ; i < lines ; i++) {
-	       memcpy( buf, tex, src_pitch );
-	       buf += dstPitch;
-	       tex += src_pitch;
-	    }
-	 }
-
-	 r200EmitWait( rmesa, RADEON_WAIT_3D );
-
-	 /* Blit to framebuffer
-	  */
-	 r200EmitBlit( rmesa,
-		       blit_format,
-		       dstPitch, GET_START( &region ),
-		       dstPitch | (t->tile_bits >> 16),
-		       t->bufAddr,
-		       0, 0,
-		       0, done,
-		       width, lines );
-	 
-	 r200EmitWait( rmesa, RADEON_WAIT_2D );
-
-	 r200ReleaseDmaRegion( rmesa, &region, __FUNCTION__ );
-	 done += lines;
-      }
-   }
-}
-
-
-/**
- * Upload the texture image associated with texture \a t at the specified
- * level at the address relative to \a start.
- */
-static void uploadSubImage( r200ContextPtr rmesa, r200TexObjPtr t, 
-			    GLint hwlevel,
-			    GLint x, GLint y, GLint width, GLint height,
-			    GLuint face )
-{
-   struct gl_texture_image *texImage = NULL;
-   GLuint offset;
-   GLint imageWidth, imageHeight;
-   GLint ret;
-   drm_radeon_texture_t tex;
-   drm_radeon_tex_image_t tmp;
-   const int level = hwlevel + t->base.firstLevel;
-
-   if ( R200_DEBUG & DEBUG_TEXTURE ) {
-      fprintf( stderr, "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n", 
-	       __FUNCTION__, (void *)t, (void *)t->base.tObj,
-	       level, width, height, face );
-   }
-
-   ASSERT(face < 6);
-
-   /* Ensure we have a valid texture to upload */
-   if ( ( hwlevel < 0 ) || ( hwlevel >= RADEON_MAX_TEXTURE_LEVELS ) ) {
-      _mesa_problem(NULL, "bad texture level in %s", __FUNCTION__);
-      return;
-   }
-
-   texImage = t->base.tObj->Image[face][level];
-
-   if ( !texImage ) {
-      if ( R200_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: texImage %d is NULL!\n", __FUNCTION__, level );
-      return;
-   }
-   if ( !texImage->Data ) {
-      if ( R200_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is NULL!\n", __FUNCTION__ );
-      return;
-   }
-
-
-   if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-      assert(level == 0);
-      assert(hwlevel == 0);
-      if ( R200_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is rectangular\n", __FUNCTION__);
-      r200UploadRectSubImage( rmesa, t, texImage, x, y, width, height );
-      return;
-   }
-   else if (texImage->IsClientData) {
-      if ( R200_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is in GART client storage\n",
-		  __FUNCTION__);
-      r200UploadGARTClientSubImage( rmesa, t, texImage, hwlevel,
-				   x, y, width, height );
-      return;
-   }
-   else if ( R200_DEBUG & DEBUG_TEXTURE )
-      fprintf( stderr, "%s: image data is in normal memory\n",
-	       __FUNCTION__);
-      
-
-   imageWidth = texImage->Width;
-   imageHeight = texImage->Height;
-
-   offset = t->bufAddr + t->base.totalSize / 6 * face;
-
-   if ( R200_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
-      GLint imageX = 0;
-      GLint imageY = 0;
-      GLint blitX = t->image[face][hwlevel].x;
-      GLint blitY = t->image[face][hwlevel].y;
-      GLint blitWidth = t->image[face][hwlevel].width;
-      GLint blitHeight = t->image[face][hwlevel].height;
-      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
-	       imageWidth, imageHeight, imageX, imageY );
-      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
-	       blitWidth, blitHeight, blitX, blitY );
-      fprintf( stderr, "       blit ofs: 0x%07x level: %d/%d\n",
-	       (GLuint)offset, hwlevel, level );
-   }
-
-   t->image[face][hwlevel].data = texImage->Data;
-
-   /* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct.
-    * NOTE: we're always use a 1KB-wide blit and I8 texture format.
-    * We used to use 1, 2 and 4-byte texels and used to use the texture
-    * width to dictate the blit width - but that won't work for compressed
-    * textures. (Brian)
-    * NOTE: can't do that with texture tiling. (sroland)
-    */
-   tex.offset = offset;
-   tex.image = &tmp;
-   /* copy (x,y,width,height,data) */
-   memcpy( &tmp, &t->image[face][hwlevel], sizeof(tmp) );
-   
-   if (texImage->TexFormat->TexelBytes) {
-      /* use multi-byte upload scheme */
-      tex.height = imageHeight;
-      tex.width = imageWidth;
-      tex.format = t->pp_txformat & R200_TXFORMAT_FORMAT_MASK;
-      if (tex.format == R200_TXFORMAT_ABGR8888) {
-	 /* drm will refuse abgr8888 textures. */
-	 tex.format = R200_TXFORMAT_ARGB8888;
-      }
-      tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1);
-      tex.offset += tmp.x & ~1023;
-      tmp.x = tmp.x % 1024;
-      if (t->tile_bits & R200_TXO_MICRO_TILE) {
-	 /* need something like "tiled coordinates" ? */
-	 tmp.y = tmp.x / (tex.pitch * 128) * 2;
-	 tmp.x = tmp.x % (tex.pitch * 128) / 2 / texImage->TexFormat->TexelBytes;
-	 tex.pitch |= RADEON_DST_TILE_MICRO >> 22;
-      }
-      else {
-	 tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1);
-      }
-      if ((t->tile_bits & R200_TXO_MACRO_TILE) &&
-	 (texImage->Width * texImage->TexFormat->TexelBytes >= 256) &&
-	 ((!(t->tile_bits & R200_TXO_MICRO_TILE) && (texImage->Height >= 8)) ||
-	    (texImage->Height >= 16))) {
-	 /* weird: R200 disables macro tiling if mip width is smaller than 256 bytes,
-	    OR if height is smaller than 8 automatically, but if micro tiling is active
-	    the limit is height 16 instead ? */
-	 tex.pitch |= RADEON_DST_TILE_MACRO >> 22;
-      }
-   }
-   else {
-      /* In case of for instance 8x8 texture (2x2 dxt blocks), padding after the first two blocks is
-         needed (only with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */
-      /* set tex.height to 1/4 since 1 "macropixel" (dxt-block) has 4 real pixels. Needed
-         so the kernel module reads the right amount of data. */
-      tex.format = R200_TXFORMAT_I8; /* any 1-byte texel format */
-      tex.pitch = (BLIT_WIDTH_BYTES / 64);
-      tex.height = (imageHeight + 3) / 4;
-      tex.width = (imageWidth + 3) / 4;
-      switch (t->pp_txformat & R200_TXFORMAT_FORMAT_MASK) {
-      case R200_TXFORMAT_DXT1:
-           tex.width *= 8;
-           break;
-      case R200_TXFORMAT_DXT23:
-      case R200_TXFORMAT_DXT45:
-           tex.width *= 16;
-           break;
-      default:
-          fprintf(stderr, "unknown compressed tex format in uploadSubImage\n");
-      }
-   }
-
-   LOCK_HARDWARE( rmesa );
-   do {
-      ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_TEXTURE,
-                                 &tex, sizeof(drm_radeon_texture_t) );
-      if (ret) {
-	 if (R200_DEBUG & DEBUG_IOCTL)
-	    fprintf(stderr, "DRM_RADEON_TEXTURE:  again!\n");
-	 usleep(1);
-      }
-   } while ( ret == -EAGAIN );
-
-   UNLOCK_HARDWARE( rmesa );
-
-   if ( ret ) {
-      fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret );
-      fprintf( stderr, "   offset=0x%08x\n",
-	       offset );
-      fprintf( stderr, "   image width=%d height=%d\n",
-	       imageWidth, imageHeight );
-      fprintf( stderr, "    blit width=%d height=%d data=%p\n",
-	       t->image[face][hwlevel].width, t->image[face][hwlevel].height,
-	       t->image[face][hwlevel].data );
-      exit( 1 );
-   }
-}
-
-
-/**
- * Upload the texture images associated with texture \a t.  This might
- * require the allocation of texture memory.
- * 
- * \param rmesa Context pointer
- * \param t Texture to be uploaded
- * \param face Cube map face to be uploaded.  Zero for non-cube maps.
- */
-
-int r200UploadTexImages( r200ContextPtr rmesa, r200TexObjPtr t, GLuint face )
-{
-   const int numLevels = t->base.lastLevel - t->base.firstLevel + 1;
-
-   if ( R200_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
-      fprintf( stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__,
-	       (void *)rmesa->glCtx, (void *)t->base.tObj, t->base.totalSize,
-	       t->base.firstLevel, t->base.lastLevel );
-   }
-
-   if ( !t || t->base.totalSize == 0 || t->image_override )
-      return 0;
-
-   if (R200_DEBUG & DEBUG_SYNC) {
-      fprintf(stderr, "%s: Syncing\n", __FUNCTION__ );
-      r200Finish( rmesa->glCtx );
-   }
-
-   LOCK_HARDWARE( rmesa );
-
-   if ( t->base.memBlock == NULL ) {
-      int heap;
-
-      heap = driAllocateTexture( rmesa->texture_heaps, rmesa->nr_heaps,
-				 (driTextureObject *) t );
-      if ( heap == -1 ) {
-	 UNLOCK_HARDWARE( rmesa );
-	 return -1;
-      }
-
-      /* Set the base offset of the texture image */
-      t->bufAddr = rmesa->r200Screen->texOffset[heap] 
-	   + t->base.memBlock->ofs;
-      t->pp_txoffset = t->bufAddr;
-       
-      if (!(t->base.tObj->Image[0][0]->IsClientData)) {
-	 /* hope it's safe to add that here... */
-	 t->pp_txoffset |= t->tile_bits;
-      }
-
-      /* Mark this texobj as dirty on all units:
-       */
-      t->dirty_state = TEX_ALL;
-   }
-
-   /* Let the world know we've used this memory recently.
-    */
-   driUpdateTextureLRU( (driTextureObject *) t );
-   UNLOCK_HARDWARE( rmesa );
-
-   /* Upload any images that are new */
-   if (t->base.dirty_images[face]) {
-      int i;
-      for ( i = 0 ; i < numLevels ; i++ ) {
-         if ( (t->base.dirty_images[face] & (1 << (i+t->base.firstLevel))) != 0 ) {
-            uploadSubImage( rmesa, t, i, 0, 0, t->image[face][i].width,
-			    t->image[face][i].height, face );
-         }
-      }
-      t->base.dirty_images[face] = 0;
-   }
-
-
-   if (R200_DEBUG & DEBUG_SYNC) {
-      fprintf(stderr, "%s: Syncing\n", __FUNCTION__ );
-      r200Finish( rmesa->glCtx );
-   }
-
-   return 0;
-}
diff --git a/src/mesa/drivers/dri/r200/r200_texstate.c b/src/mesa/drivers/dri/r200/r200_texstate.c
index 0ad5651cd47..ed1995e147c 100644
--- a/src/mesa/drivers/dri/r200/r200_texstate.c
+++ b/src/mesa/drivers/dri/r200/r200_texstate.c
@@ -37,9 +37,12 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/context.h"
 #include "main/macros.h"
 #include "main/texformat.h"
+#include "main/teximage.h"
 #include "main/texobj.h"
 #include "main/enums.h"
 
+#include "radeon_common.h"
+#include "radeon_mipmap_tree.h"
 #include "r200_context.h"
 #include "r200_state.h"
 #include "r200_ioctl.h"
@@ -139,257 +142,6 @@ static const struct tx_table tx_table_le[] =
 #undef _ALPHA
 #undef _INVALID
 
-/**
- * This function computes the number of bytes of storage needed for
- * the given texture object (all mipmap levels, all cube faces).
- * The \c image[face][level].x/y/width/height parameters for upload/blitting
- * are computed here.  \c pp_txfilter, \c pp_txformat, etc. will be set here
- * too.
- * 
- * \param rmesa Context pointer
- * \param tObj GL texture object whose images are to be posted to
- *                 hardware state.
- */
-static void r200SetTexImages( r200ContextPtr rmesa,
-			      struct gl_texture_object *tObj )
-{
-   r200TexObjPtr t = (r200TexObjPtr)tObj->DriverData;
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   GLint curOffset, blitWidth;
-   GLint i, texelBytes;
-   GLint numLevels;
-   GLint log2Width, log2Height, log2Depth;
-
-   /* Set the hardware texture format
-    */
-   if ( !t->image_override ) {
-      if ( VALID_FORMAT( baseImage->TexFormat->MesaFormat ) ) {
-	 const struct tx_table *table = _mesa_little_endian() ? tx_table_le :
-								tx_table_be;
-
-         t->pp_txformat &= ~(R200_TXFORMAT_FORMAT_MASK |
-                             R200_TXFORMAT_ALPHA_IN_MAP);
-         t->pp_txfilter &= ~R200_YUV_TO_RGB;
-
-	 t->pp_txformat |= table[ baseImage->TexFormat->MesaFormat ].format;
-	 t->pp_txfilter |= table[ baseImage->TexFormat->MesaFormat ].filter;
-      }
-      else {
-         _mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__);
-         return;
-      }
-   }
-
-   texelBytes = baseImage->TexFormat->TexelBytes;
-
-   /* Compute which mipmap levels we really want to send to the hardware.
-    */
-
-   driCalculateTextureFirstLastLevel( (driTextureObject *) t );
-   log2Width  = tObj->Image[0][t->base.firstLevel]->WidthLog2;
-   log2Height = tObj->Image[0][t->base.firstLevel]->HeightLog2;
-   log2Depth  = tObj->Image[0][t->base.firstLevel]->DepthLog2;
-
-   numLevels = t->base.lastLevel - t->base.firstLevel + 1;
-
-   assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS);
-
-   /* Calculate mipmap offsets and dimensions for blitting (uploading)
-    * The idea is that we lay out the mipmap levels within a block of
-    * memory organized as a rectangle of width BLIT_WIDTH_BYTES.
-    */
-   curOffset = 0;
-   blitWidth = BLIT_WIDTH_BYTES;
-   t->tile_bits = 0;
-
-   /* figure out if this texture is suitable for tiling. */
-   if (texelBytes) {
-      if (rmesa->texmicrotile  && (tObj->Target != GL_TEXTURE_RECTANGLE_NV) &&
-      /* texrect might be able to use micro tiling too in theory? */
-	 (baseImage->Height > 1)) {
-	 /* allow 32 (bytes) x 1 mip (which will use two times the space
-	 the non-tiled version would use) max if base texture is large enough */
-	 if ((numLevels == 1) ||
-	   (((baseImage->Width * texelBytes / baseImage->Height) <= 32) &&
-	       (baseImage->Width * texelBytes > 64)) ||
-	    ((baseImage->Width * texelBytes / baseImage->Height) <= 16)) {
-	    t->tile_bits |= R200_TXO_MICRO_TILE;
-	 }
-      }
-      if (tObj->Target != GL_TEXTURE_RECTANGLE_NV) {
-	 /* we can set macro tiling even for small textures, they will be untiled anyway */
-	 t->tile_bits |= R200_TXO_MACRO_TILE;
-      }
-   }
-
-   for (i = 0; i < numLevels; i++) {
-      const struct gl_texture_image *texImage;
-      GLuint size;
-
-      texImage = tObj->Image[0][i + t->base.firstLevel];
-      if ( !texImage )
-	 break;
-
-      /* find image size in bytes */
-      if (texImage->IsCompressed) {
-      /* need to calculate the size AFTER padding even though the texture is
-         submitted without padding.
-         Only handle pot textures currently - don't know if npot is even possible,
-         size calculation would certainly need (trivial) adjustments.
-         Align (and later pad) to 32byte, not sure what that 64byte blit width is
-         good for? */
-         if ((t->pp_txformat & R200_TXFORMAT_FORMAT_MASK) == R200_TXFORMAT_DXT1) {
-            /* RGB_DXT1/RGBA_DXT1, 8 bytes per block */
-            if ((texImage->Width + 3) < 8) /* width one block */
-               size = texImage->CompressedSize * 4;
-            else if ((texImage->Width + 3) < 16)
-               size = texImage->CompressedSize * 2;
-            else size = texImage->CompressedSize;
-         }
-         else /* DXT3/5, 16 bytes per block */
-            if ((texImage->Width + 3) < 8)
-               size = texImage->CompressedSize * 2;
-            else size = texImage->CompressedSize;
-      }
-      else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-	 size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height;
-      }
-      else if (t->tile_bits & R200_TXO_MICRO_TILE) {
-	 /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
-	    though the actual offset may be different (if texture is less than
-	    32 bytes width) to the untiled case */
-	 int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
-	 size = (w * ((texImage->Height + 1) / 2)) * texImage->Depth;
-	 blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-      }
-      else {
-	 int w = (texImage->Width * texelBytes + 31) & ~31;
-	 size = w * texImage->Height * texImage->Depth;
-	 blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-      }
-      assert(size > 0);
-
-      /* Align to 32-byte offset.  It is faster to do this unconditionally
-       * (no branch penalty).
-       */
-
-      curOffset = (curOffset + 0x1f) & ~0x1f;
-
-      if (texelBytes) {
-	 t->image[0][i].x = curOffset; /* fix x and y coords up later together with offset */
-	 t->image[0][i].y = 0;
-	 t->image[0][i].width = MIN2(size / texelBytes, blitWidth);
-	 t->image[0][i].height = (size / texelBytes) / t->image[0][i].width;
-      }
-      else {
-         t->image[0][i].x = curOffset % BLIT_WIDTH_BYTES;
-         t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES;
-         t->image[0][i].width  = MIN2(size, BLIT_WIDTH_BYTES);
-         t->image[0][i].height = size / t->image[0][i].width;     
-      }
-
-#if 0
-      /* for debugging only and only  applicable to non-rectangle targets */
-      assert(size % t->image[0][i].width == 0);
-      assert(t->image[0][i].x == 0
-             || (size < BLIT_WIDTH_BYTES && t->image[0][i].height == 1));
-#endif
-
-      if (0)
-         fprintf(stderr,
-                 "level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
-                 i, texImage->Width, texImage->Height,
-                 t->image[0][i].x, t->image[0][i].y,
-                 t->image[0][i].width, t->image[0][i].height, size, curOffset);
-
-      curOffset += size;
-
-   }
-
-   /* Align the total size of texture memory block.
-    */
-   t->base.totalSize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
-
-   /* Setup remaining cube face blits, if needed */
-   if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-      const GLuint faceSize = t->base.totalSize;
-      GLuint face;
-      /* reuse face 0 x/y/width/height - just update the offset when uploading */
-      for (face = 1; face < 6; face++) {
-         for (i = 0; i < numLevels; i++) {
-            t->image[face][i].x =  t->image[0][i].x;
-            t->image[face][i].y =  t->image[0][i].y;
-            t->image[face][i].width  = t->image[0][i].width;
-            t->image[face][i].height = t->image[0][i].height;
-         }
-      }
-      t->base.totalSize = 6 * faceSize; /* total texmem needed */
-   }
-
-
-   /* Hardware state:
-    */
-   t->pp_txfilter &= ~R200_MAX_MIP_LEVEL_MASK;
-   t->pp_txfilter |= (numLevels - 1) << R200_MAX_MIP_LEVEL_SHIFT;
-
-   t->pp_txformat &= ~(R200_TXFORMAT_WIDTH_MASK |
-		       R200_TXFORMAT_HEIGHT_MASK |
-                       R200_TXFORMAT_CUBIC_MAP_ENABLE |
-                       R200_TXFORMAT_F5_WIDTH_MASK |
-                       R200_TXFORMAT_F5_HEIGHT_MASK);
-   t->pp_txformat |= ((log2Width << R200_TXFORMAT_WIDTH_SHIFT) |
-		      (log2Height << R200_TXFORMAT_HEIGHT_SHIFT));
-
-   t->pp_txformat_x &= ~(R200_DEPTH_LOG2_MASK | R200_TEXCOORD_MASK);
-   if (tObj->Target == GL_TEXTURE_3D) {
-      t->pp_txformat_x |= (log2Depth << R200_DEPTH_LOG2_SHIFT);
-      t->pp_txformat_x |= R200_TEXCOORD_VOLUME;
-   }
-   else if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-      ASSERT(log2Width == log2Height);
-      t->pp_txformat |= ((log2Width << R200_TXFORMAT_F5_WIDTH_SHIFT) |
-                         (log2Height << R200_TXFORMAT_F5_HEIGHT_SHIFT) |
-/* don't think we need this bit, if it exists at all - fglrx does not set it */
-                         (R200_TXFORMAT_CUBIC_MAP_ENABLE));
-      t->pp_txformat_x |= R200_TEXCOORD_CUBIC_ENV;
-      t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) |
-                           (log2Height << R200_FACE_HEIGHT_1_SHIFT) |
-                           (log2Width << R200_FACE_WIDTH_2_SHIFT) |
-                           (log2Height << R200_FACE_HEIGHT_2_SHIFT) |
-                           (log2Width << R200_FACE_WIDTH_3_SHIFT) |
-                           (log2Height << R200_FACE_HEIGHT_3_SHIFT) |
-                           (log2Width << R200_FACE_WIDTH_4_SHIFT) |
-                           (log2Height << R200_FACE_HEIGHT_4_SHIFT));
-   }
-   else {
-      /* If we don't in fact send enough texture coordinates, q will be 1,
-       * making TEXCOORD_PROJ act like TEXCOORD_NONPROJ (Right?)
-       */
-      t->pp_txformat_x |= R200_TEXCOORD_PROJ;
-   }
-
-   t->pp_txsize = (((tObj->Image[0][t->base.firstLevel]->Width - 1) << 0) |
-                   ((tObj->Image[0][t->base.firstLevel]->Height - 1) << 16));
-
-   /* Only need to round to nearest 32 for textures, but the blitter
-    * requires 64-byte aligned pitches, and we may/may not need the
-    * blitter.   NPOT only!
-    */
-   if ( !t->image_override ) {
-      if (baseImage->IsCompressed)
-         t->pp_txpitch = (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63);
-      else
-         t->pp_txpitch = ((tObj->Image[0][t->base.firstLevel]->Width * texelBytes) + 63) & ~(63);
-      t->pp_txpitch -= 32;
-   }
-
-   t->dirty_state = TEX_ALL;
-
-   /* FYI: r200UploadTexImages( rmesa, t ) used to be called here */
-}
-
-
-
 /* ================================================================
  * Texture combine functions
  */
@@ -981,20 +733,19 @@ void r200SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 {
 	r200ContextPtr rmesa = pDRICtx->driverPrivate;
 	struct gl_texture_object *tObj =
-	    _mesa_lookup_texture(rmesa->glCtx, texname);
-	r200TexObjPtr t;
+	    _mesa_lookup_texture(rmesa->radeon.glCtx, texname);
+	radeonTexObjPtr t = radeon_tex_obj(tObj);
 
 	if (!tObj)
 		return;
 
-	t = (r200TexObjPtr) tObj->DriverData;
-
 	t->image_override = GL_TRUE;
 
 	if (!offset)
 		return;
 
-	t->pp_txoffset = offset;
+	t->bo = NULL;
+	t->override_offset = offset;
 	t->pp_txpitch = pitch - 32;
 
 	switch (depth) {
@@ -1014,6 +765,125 @@ void r200SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 	}
 }
 
+void r200SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format,
+		       __DRIdrawable *dPriv)
+{
+	struct gl_texture_unit *texUnit;
+	struct gl_texture_object *texObj;
+	struct gl_texture_image *texImage;
+	struct radeon_renderbuffer *rb;
+	radeon_texture_image *rImage;
+	radeonContextPtr radeon;
+	r200ContextPtr rmesa;
+	struct radeon_framebuffer *rfb;
+	radeonTexObjPtr t;
+	uint32_t pitch_val;
+	uint32_t internalFormat, type, format;
+
+	type = GL_BGRA;
+	format = GL_UNSIGNED_BYTE;
+	internalFormat = (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT ? 3 : 4);
+
+	radeon = pDRICtx->driverPrivate;
+	rmesa = pDRICtx->driverPrivate;
+
+	rfb = dPriv->driverPrivate;
+        texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit];
+	texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target);
+        texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0);
+
+	rImage = get_radeon_texture_image(texImage);
+	t = radeon_tex_obj(texObj);
+        if (t == NULL) {
+    	    return;
+    	}
+
+	radeon_update_renderbuffers(pDRICtx, dPriv);
+	/* back & depth buffer are useless free them right away */
+	rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer;
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+        rb->bo = NULL;
+	}
+	rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+		rb->bo = NULL;
+	}
+	rb = rfb->color_rb[0];
+	if (rb->bo == NULL) {
+		/* Failed to BO for the buffer */
+		return;
+	}
+	
+	_mesa_lock_texture(radeon->glCtx, texObj);
+	if (t->bo) {
+		radeon_bo_unref(t->bo);
+		t->bo = NULL;
+	}
+	if (rImage->bo) {
+		radeon_bo_unref(rImage->bo);
+		rImage->bo = NULL;
+	}
+	if (t->mt) {
+		radeon_miptree_unreference(t->mt);
+		t->mt = NULL;
+	}
+	if (rImage->mt) {
+		radeon_miptree_unreference(rImage->mt);
+		rImage->mt = NULL;
+	}
+	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
+				   rb->width, rb->height, 1, 0, rb->cpp);
+	texImage->RowStride = rb->pitch / rb->cpp;
+	texImage->TexFormat = radeonChooseTextureFormat(radeon->glCtx,
+							internalFormat,
+							type, format, 0);
+	rImage->bo = rb->bo;
+	radeon_bo_ref(rImage->bo);
+	t->bo = rb->bo;
+	radeon_bo_ref(t->bo);
+	t->tile_bits = 0;
+	t->image_override = GL_TRUE;
+	t->override_offset = 0;
+	t->pp_txpitch &= (1 << 13) -1;
+	pitch_val = rb->pitch;
+	switch (rb->cpp) {
+	case 4:
+		if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT)
+			t->pp_txformat = tx_table_le[MESA_FORMAT_RGB888].format;
+		else
+			t->pp_txformat = tx_table_le[MESA_FORMAT_ARGB8888].format;
+		t->pp_txfilter |= tx_table_le[MESA_FORMAT_ARGB8888].filter;
+		break;
+	case 3:
+	default:
+		t->pp_txformat = tx_table_le[MESA_FORMAT_RGB888].format;
+		t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB888].filter;
+		break;
+	case 2:
+		t->pp_txformat = tx_table_le[MESA_FORMAT_RGB565].format;
+		t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB565].filter;
+		break;
+	}
+        t->pp_txsize = ((rb->width - 1) << RADEON_TEX_USIZE_SHIFT)
+		   | ((rb->height - 1) << RADEON_TEX_VSIZE_SHIFT);
+        t->pp_txformat |= R200_TXFORMAT_NON_POWER2;
+	t->pp_txpitch = pitch_val;
+        t->pp_txpitch -= 32;
+
+	t->validated = GL_TRUE;
+	_mesa_unlock_texture(radeon->glCtx, texObj);
+	return;
+}
+
+
+void r200SetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
+{
+        r200SetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv);
+}
+
+
 #define REF_COLOR 1
 #define REF_ALPHA 2
 
@@ -1207,12 +1077,41 @@ static GLboolean r200UpdateAllTexEnv( GLcontext *ctx )
                                 R200_VOLUME_FILTER_MASK)
 
 
+static void disable_tex_obj_state( r200ContextPtr rmesa, 
+				   int unit )
+{
+   
+   R200_STATECHANGE( rmesa, vtx );
+   rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3));
+
+   if (rmesa->radeon.TclFallback & (R200_TCL_FALLBACK_TEXGEN_0<<unit)) {
+      TCL_FALLBACK( rmesa->radeon.glCtx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE);
+   }
+
+   /* Actually want to keep all units less than max active texture
+    * enabled, right?  Fix this for >2 texunits.
+    */
+
+   {
+      GLuint tmp = rmesa->TexGenEnabled;
+
+      rmesa->TexGenEnabled &= ~(R200_TEXGEN_TEXMAT_0_ENABLE<<unit);
+      rmesa->TexGenEnabled &= ~(R200_TEXMAT_0_ENABLE<<unit);
+      rmesa->TexGenNeedNormals[unit] = GL_FALSE;
+      rmesa->TexGenCompSel &= ~(R200_OUTPUT_TEX_0 << unit);
+
+      if (tmp != rmesa->TexGenEnabled) {
+	 rmesa->recheck_texgen[unit] = GL_TRUE;
+	 rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
+      }
+   }
+}
 static void import_tex_obj_state( r200ContextPtr rmesa,
 				  int unit,
-				  r200TexObjPtr texobj )
+				  radeonTexObjPtr texobj )
 {
 /* do not use RADEON_DB_STATE to avoid stale texture caches */
-   int *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0];
+   GLuint *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0];
 
    R200_STATECHANGE( rmesa, tex[unit] );
 
@@ -1225,36 +1124,21 @@ static void import_tex_obj_state( r200ContextPtr rmesa,
    cmd[TEX_PP_TXSIZE] = texobj->pp_txsize; /* NPOT only! */
    cmd[TEX_PP_TXPITCH] = texobj->pp_txpitch; /* NPOT only! */
    cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color;
-   if (rmesa->r200Screen->drmSupportsFragShader) {
-      cmd[TEX_PP_TXOFFSET_NEWDRM] = texobj->pp_txoffset;
-   }
-   else {
-      cmd[TEX_PP_TXOFFSET_OLDDRM] = texobj->pp_txoffset;
-   }
 
-   if (texobj->base.tObj->Target == GL_TEXTURE_CUBE_MAP) {
-      int *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0];
-      GLuint bytesPerFace = texobj->base.totalSize / 6;
-      ASSERT(texobj->base.totalSize % 6 == 0);
+   if (texobj->base.Target == GL_TEXTURE_CUBE_MAP) {
+      GLuint *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0];
 
       R200_STATECHANGE( rmesa, cube[unit] );
       cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
-      if (rmesa->r200Screen->drmSupportsFragShader) {
+      if (rmesa->radeon.radeonScreen->drmSupportsFragShader) {
 	 /* that value is submitted twice. could change cube atom
 	    to not include that command when new drm is used */
 	 cmd[TEX_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
       }
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_F1] = texobj->pp_txoffset + 1 * bytesPerFace;
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_F2] = texobj->pp_txoffset + 2 * bytesPerFace;
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_F3] = texobj->pp_txoffset + 3 * bytesPerFace;
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_F4] = texobj->pp_txoffset + 4 * bytesPerFace;
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_F5] = texobj->pp_txoffset + 5 * bytesPerFace;
    }
 
-   texobj->dirty_state &= ~(1<<unit);
 }
 
-
 static void set_texgen_matrix( r200ContextPtr rmesa, 
 			       GLuint unit,
 			       const GLfloat *s_plane,
@@ -1377,7 +1261,6 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit )
    } else {
       tgcm |= R200_TEXGEN_COMP_T << (unit * 4);
    }
-
    if (texUnit->TexGenEnabled & R_BIT) {
       if (texUnit->GenR.Mode != mode)
 	 mixed_fallback = GL_TRUE;
@@ -1517,52 +1400,6 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit )
    return GL_TRUE;
 }
 
-
-static void disable_tex( GLcontext *ctx, int unit )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_0_ENABLE<<unit)) {
-      /* Texture unit disabled */
-      if ( rmesa->state.texture.unit[unit].texobj != NULL ) {
-	 /* The old texture is no longer bound to this texture unit.
-	  * Mark it as such.
-	  */
-
-	 rmesa->state.texture.unit[unit].texobj->base.bound &= ~(1UL << unit);
-	 rmesa->state.texture.unit[unit].texobj = NULL;
-      }
-
-      R200_STATECHANGE( rmesa, ctx );
-      rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~(R200_TEX_0_ENABLE << unit);
-	 
-      R200_STATECHANGE( rmesa, vtx );
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3));
-	 
-      if (rmesa->TclFallback & (R200_TCL_FALLBACK_TEXGEN_0<<unit)) {
-	 TCL_FALLBACK( ctx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE);
-      }
-
-      /* Actually want to keep all units less than max active texture
-       * enabled, right?  Fix this for >2 texunits.
-       */
-
-      {
-	 GLuint tmp = rmesa->TexGenEnabled;
-
-	 rmesa->TexGenEnabled &= ~(R200_TEXGEN_TEXMAT_0_ENABLE<<unit);
-	 rmesa->TexGenEnabled &= ~(R200_TEXMAT_0_ENABLE<<unit);
-	 rmesa->TexGenNeedNormals[unit] = GL_FALSE;
-	 rmesa->TexGenCompSel &= ~(R200_OUTPUT_TEX_0 << unit);
-
-	 if (tmp != rmesa->TexGenEnabled) {
-	    rmesa->recheck_texgen[unit] = GL_TRUE;
-	    rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
-	 }
-      }
-   }
-}
-
 void set_re_cntl_d3d( GLcontext *ctx, int unit, GLboolean use_d3d )
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
@@ -1579,237 +1416,169 @@ void set_re_cntl_d3d( GLcontext *ctx, int unit, GLboolean use_d3d )
    }
 }
 
-static GLboolean enable_tex_2d( GLcontext *ctx, int unit )
+/**
+ * Compute the cached hardware register values for the given texture object.
+ *
+ * \param rmesa Context pointer
+ * \param t the r300 texture object
+ */
+static void setup_hardware_state(r200ContextPtr rmesa, radeonTexObj *t)
 {
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData;
-
-   /* Need to load the 2d images associated with this unit.
-    */
-   if (t->pp_txformat & R200_TXFORMAT_NON_POWER2) {
-      t->pp_txformat &= ~R200_TXFORMAT_NON_POWER2;
-      t->base.dirty_images[0] = ~0;
-   }
-
-   ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D);
-
-   if ( t->base.dirty_images[0] ) {
-      R200_FIREVERTICES( rmesa );
-      r200SetTexImages( rmesa, tObj );
-      r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, 0 );
-      if ( !t->base.memBlock && !t->image_override ) 
-	 return GL_FALSE;
+   int firstlevel = t->mt ? t->mt->firstLevel : 0;
+   const struct gl_texture_image *firstImage = t->base.Image[0][firstlevel];
+   GLint log2Width, log2Height, log2Depth, texelBytes;
+   
+   if ( t->bo ) {
+       return;
    }
 
-   set_re_cntl_d3d( ctx, unit, GL_FALSE );
+   log2Width  = firstImage->WidthLog2;
+   log2Height = firstImage->HeightLog2;
+   log2Depth  = firstImage->DepthLog2;
+   texelBytes = firstImage->TexFormat->TexelBytes;
 
-   return GL_TRUE;
-}
 
-#if ENABLE_HW_3D_TEXTURE
-static GLboolean enable_tex_3d( GLcontext *ctx, int unit )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData;
-
-   /* Need to load the 3d images associated with this unit.
-    */
-   if (t->pp_txformat & R200_TXFORMAT_NON_POWER2) {
-      t->pp_txformat &= ~R200_TXFORMAT_NON_POWER2;
-      t->base.dirty_images[0] = ~0;
+   if (!t->image_override) {
+      if (VALID_FORMAT(firstImage->TexFormat->MesaFormat)) {
+	 const struct tx_table *table = _mesa_little_endian() ? tx_table_le :
+	    tx_table_be;
+	 
+	 t->pp_txformat &= ~(R200_TXFORMAT_FORMAT_MASK |
+			     R200_TXFORMAT_ALPHA_IN_MAP);
+	 t->pp_txfilter &= ~R200_YUV_TO_RGB;
+	 
+	 t->pp_txformat |= table[ firstImage->TexFormat->MesaFormat ].format;
+	 t->pp_txfilter |= table[ firstImage->TexFormat->MesaFormat ].filter;
+      } else {
+	 _mesa_problem(NULL, "unexpected texture format in %s",
+		       __FUNCTION__);
+	 return;
+      }
    }
+   
+   t->pp_txfilter &= ~R200_MAX_MIP_LEVEL_MASK;
+   t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << R200_MAX_MIP_LEVEL_SHIFT;
+	
+   t->pp_txformat &= ~(R200_TXFORMAT_WIDTH_MASK |
+		       R200_TXFORMAT_HEIGHT_MASK |
+		       R200_TXFORMAT_CUBIC_MAP_ENABLE |
+		       R200_TXFORMAT_F5_WIDTH_MASK |
+		       R200_TXFORMAT_F5_HEIGHT_MASK);
+   t->pp_txformat |= ((log2Width << R200_TXFORMAT_WIDTH_SHIFT) |
+		      (log2Height << R200_TXFORMAT_HEIGHT_SHIFT));
+   
+   t->tile_bits = 0;
+   
+   t->pp_txformat_x &= ~(R200_DEPTH_LOG2_MASK | R200_TEXCOORD_MASK);
+   if (t->base.Target == GL_TEXTURE_3D) {
+      t->pp_txformat_x |= (log2Depth << R200_DEPTH_LOG2_SHIFT);
+      t->pp_txformat_x |= R200_TEXCOORD_VOLUME;
 
-   ASSERT(tObj->Target == GL_TEXTURE_3D);
-
-   /* R100 & R200 do not support mipmaps for 3D textures.
-    */
-   if ( (tObj->MinFilter != GL_NEAREST) && (tObj->MinFilter != GL_LINEAR) ) {
-      return GL_FALSE;
    }
-
-   if ( t->base.dirty_images[0] ) {
-      R200_FIREVERTICES( rmesa );
-      r200SetTexImages( rmesa, tObj );
-      r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, 0 );
-      if ( !t->base.memBlock ) 
-	 return GL_FALSE;
+   else if (t->base.Target == GL_TEXTURE_CUBE_MAP) {
+      ASSERT(log2Width == log2Height);
+      t->pp_txformat |= ((log2Width << R200_TXFORMAT_F5_WIDTH_SHIFT) |
+			 (log2Height << R200_TXFORMAT_F5_HEIGHT_SHIFT) |
+			 /* don't think we need this bit, if it exists at all - fglrx does not set it */
+			 (R200_TXFORMAT_CUBIC_MAP_ENABLE));
+      t->pp_txformat_x |= R200_TEXCOORD_CUBIC_ENV;
+      t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) |
+                           (log2Height << R200_FACE_HEIGHT_1_SHIFT) |
+                           (log2Width << R200_FACE_WIDTH_2_SHIFT) |
+                           (log2Height << R200_FACE_HEIGHT_2_SHIFT) |
+                           (log2Width << R200_FACE_WIDTH_3_SHIFT) |
+                           (log2Height << R200_FACE_HEIGHT_3_SHIFT) |
+                           (log2Width << R200_FACE_WIDTH_4_SHIFT) |
+                           (log2Height << R200_FACE_HEIGHT_4_SHIFT));
    }
-
-   set_re_cntl_d3d( ctx, unit, GL_TRUE );
-
-   return GL_TRUE;
-}
-#endif
-
-static GLboolean enable_tex_cube( GLcontext *ctx, int unit )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData;
-   GLuint face;
-
-   /* Need to load the 2d images associated with this unit.
-    */
-   if (t->pp_txformat & R200_TXFORMAT_NON_POWER2) {
-      t->pp_txformat &= ~R200_TXFORMAT_NON_POWER2;
-      for (face = 0; face < 6; face++)
-         t->base.dirty_images[face] = ~0;
+   else {
+      /* If we don't in fact send enough texture coordinates, q will be 1,
+       * making TEXCOORD_PROJ act like TEXCOORD_NONPROJ (Right?)
+       */
+      t->pp_txformat_x |= R200_TEXCOORD_PROJ;
    }
 
-   ASSERT(tObj->Target == GL_TEXTURE_CUBE_MAP);
+   t->pp_txsize = (((firstImage->Width - 1) << R200_PP_TX_WIDTHMASK_SHIFT)
+		   | ((firstImage->Height - 1) << R200_PP_TX_HEIGHTMASK_SHIFT));
 
-   if ( t->base.dirty_images[0] || t->base.dirty_images[1] ||
-        t->base.dirty_images[2] || t->base.dirty_images[3] ||
-        t->base.dirty_images[4] || t->base.dirty_images[5] ) {
-      /* flush */
-      R200_FIREVERTICES( rmesa );
-      /* layout memory space, once for all faces */
-      r200SetTexImages( rmesa, tObj );
-   }
-
-   /* upload (per face) */
-   for (face = 0; face < 6; face++) {
-      if (t->base.dirty_images[face]) {
-         r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, face );
-      }
-   }
-      
-   if ( !t->base.memBlock ) {
-      /* texmem alloc failed, use s/w fallback */
-      return GL_FALSE;
+   if ( !t->image_override ) {
+      if (firstImage->IsCompressed)
+         t->pp_txpitch = (firstImage->Width + 63) & ~(63);
+      else
+         t->pp_txpitch = ((firstImage->Width * texelBytes) + 63) & ~(63);
+      t->pp_txpitch -= 32;
    }
 
-   set_re_cntl_d3d( ctx, unit, GL_TRUE );
-
-   return GL_TRUE;
-}
-
-static GLboolean enable_tex_rect( GLcontext *ctx, int unit )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData;
-
-   if (!(t->pp_txformat & R200_TXFORMAT_NON_POWER2)) {
+   if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) {
       t->pp_txformat |= R200_TXFORMAT_NON_POWER2;
-      t->base.dirty_images[0] = ~0;
-   }
-
-   ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV);
-
-   if ( t->base.dirty_images[0] ) {
-      R200_FIREVERTICES( rmesa );
-      r200SetTexImages( rmesa, tObj );
-      r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, 0 );
-      if ( !t->base.memBlock &&
-           !t->image_override &&
-           !rmesa->prefer_gart_client_texturing ) 
-	 return GL_FALSE;
    }
 
-   set_re_cntl_d3d( ctx, unit, GL_FALSE );
-
-   return GL_TRUE;
 }
 
-
-static GLboolean update_tex_common( GLcontext *ctx, int unit )
+static GLboolean r200_validate_texture(GLcontext *ctx, struct gl_texture_object *texObj, int unit)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData;
-
-   /* Fallback if there's a texture border */
-   if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 )
-       return GL_FALSE;
-
-   /* Update state if this is a different texture object to last
-    * time.
-    */
-   if ( rmesa->state.texture.unit[unit].texobj != t ) {
-      if ( rmesa->state.texture.unit[unit].texobj != NULL ) {
-	 /* The old texture is no longer bound to this texture unit.
-	  * Mark it as such.
-	  */
+   radeonTexObj *t = radeon_tex_obj(texObj);
 
-	 rmesa->state.texture.unit[unit].texobj->base.bound &= 
-	     ~(1UL << unit);
-      }
-
-      rmesa->state.texture.unit[unit].texobj = t;
-      t->base.bound |= (1UL << unit);
-      t->dirty_state |= 1<<unit;
-      driUpdateTextureLRU( (driTextureObject *) t ); /* XXX: should be locked! */
-   }
-
-
-   /* Newly enabled?
-    */
-   if ( 1|| !(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_0_ENABLE<<unit))) {
-      R200_STATECHANGE( rmesa, ctx );
-      rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_0_ENABLE << unit;
-
-      R200_STATECHANGE( rmesa, vtx );
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3));
-      rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] |= 4 << (unit * 3);
+   if (!radeon_validate_texture_miptree(ctx, texObj))
+      return GL_FALSE;
 
-      rmesa->recheck_texgen[unit] = GL_TRUE;
-   }
+   r200_validate_texgen(ctx, unit);
+   /* Configure the hardware registers (more precisely, the cached version
+    * of the hardware registers). */
+   setup_hardware_state(rmesa, t);
+
+   if (texObj->Target == GL_TEXTURE_RECTANGLE_NV ||
+       texObj->Target == GL_TEXTURE_2D ||
+       texObj->Target == GL_TEXTURE_1D)
+      set_re_cntl_d3d( ctx, unit, GL_FALSE );
+   else
+      set_re_cntl_d3d( ctx, unit, GL_TRUE );
+   R200_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_0_ENABLE << unit;
+   
+   R200_STATECHANGE( rmesa, vtx );
+   rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3));
+   rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] |= 4 << (unit * 3);
 
-   if (t->dirty_state & (1<<unit)) {
-      import_tex_obj_state( rmesa, unit, t );
-   }
+   rmesa->recheck_texgen[unit] = GL_TRUE;
+   import_tex_obj_state( rmesa, unit, t );
 
    if (rmesa->recheck_texgen[unit]) {
       GLboolean fallback = !r200_validate_texgen( ctx, unit );
       TCL_FALLBACK( ctx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), fallback);
       rmesa->recheck_texgen[unit] = 0;
-      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+      rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
    }
 
-   FALLBACK( rmesa, R200_FALLBACK_BORDER_MODE, t->border_fallback );
-   return !t->border_fallback;
-}
+   t->validated = GL_TRUE;
 
+   FALLBACK( rmesa, RADEON_FALLBACK_BORDER_MODE, t->border_fallback );
 
+   return !t->border_fallback;
+}
 
-static GLboolean r200UpdateTextureUnit( GLcontext *ctx, int unit )
+static GLboolean r200UpdateTextureUnit(GLcontext *ctx, int unit)
 {
    r200ContextPtr rmesa = R200_CONTEXT(ctx);
    GLuint unitneeded = rmesa->state.texture.unit[unit].unitneeded;
 
-   if ( unitneeded & (TEXTURE_RECT_BIT) ) {
-      return (enable_tex_rect( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
-   }
-   else if ( unitneeded & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) {
-      return (enable_tex_2d( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
-   }
-#if ENABLE_HW_3D_TEXTURE
-   else if ( unitneeded & (TEXTURE_3D_BIT) ) {
-      return (enable_tex_3d( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
-   }
-#endif
-   else if ( unitneeded & (TEXTURE_CUBE_BIT) ) {
-      return (enable_tex_cube( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
-   }
-   else if ( unitneeded ) {
-      return GL_FALSE;
-   }
-   else {
-      disable_tex( ctx, unit );
-      return GL_TRUE;
+   if (!unitneeded) {
+      /* disable the unit */
+     disable_tex_obj_state(rmesa, unit);
+     return GL_TRUE;
    }
+
+   if (!r200_validate_texture(ctx, ctx->Texture.Unit[unit]._Current, unit)) {
+    _mesa_warning(ctx,
+		  "failed to validate texture for unit %d.\n",
+		  unit);
+    rmesa->state.texture.unit[unit].texobj = NULL;
+    return GL_FALSE;
+  }
+
+   rmesa->state.texture.unit[unit].texobj = radeon_tex_obj(ctx->Texture.Unit[unit]._Current);
+  return GL_TRUE;
 }
 
 
@@ -1850,11 +1619,11 @@ void r200UpdateTextureState( GLcontext *ctx )
 
    FALLBACK( rmesa, R200_FALLBACK_TEXTURE, !ok );
 
-   if (rmesa->TclFallback)
+   if (rmesa->radeon.TclFallback)
       r200ChooseVertexState( ctx );
 
 
-   if (rmesa->r200Screen->chip_family == CHIP_FAMILY_R200) {
+   if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) {
 
       /*
        * T0 hang workaround -------------
@@ -1867,7 +1636,7 @@ void r200UpdateTextureState( GLcontext *ctx )
 	 R200_STATECHANGE(rmesa, tex[1]);
 	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_1_ENABLE;
 	 if (!(rmesa->hw.cst.cmd[CST_PP_CNTL_X] & R200_PPX_TEX_1_ENABLE))
-	    rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
+	   rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
 	 rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] |= R200_TXFORMAT_LOOKUP_DISABLE;
       }
       else if (!ctx->ATIFragmentShader._Enabled) {
diff --git a/src/mesa/drivers/dri/r200/r200_vertprog.c b/src/mesa/drivers/dri/r200/r200_vertprog.c
index 4ce93b51453..620f29b5c6e 100644
--- a/src/mesa/drivers/dri/r200/r200_vertprog.c
+++ b/src/mesa/drivers/dri/r200/r200_vertprog.c
@@ -1110,9 +1110,9 @@ void r200SetupVertexProg( GLcontext *ctx ) {
    }
    /* could optimize setting up vertex progs away for non-tcl hw */
    fallback = !(vp->native && r200VertexProgUpdateParams(ctx, vp) &&
-      rmesa->r200Screen->drmSupportsVertexProgram);
+      rmesa->radeon.radeonScreen->drmSupportsVertexProgram);
    TCL_FALLBACK(ctx, R200_TCL_FALLBACK_VERTEX_PROGRAM, fallback);
-   if (rmesa->TclFallback) return;
+   if (rmesa->radeon.TclFallback) return;
 
    R200_STATECHANGE( rmesa, vap );
    /* FIXME: fglrx sets R200_VAP_SINGLE_BUF_STATE_ENABLE too. Do we need it?
diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile
index 6ca934204f3..bdb09624be4 100644
--- a/src/mesa/drivers/dri/r300/Makefile
+++ b/src/mesa/drivers/dri/r300/Makefile
@@ -3,6 +3,8 @@
 TOP = ../../../../..
 include $(TOP)/configs/current
 
+CFLAGS += $(RADEON_CFLAGS)
+
 LIBNAME = r300_dri.so
 
 MINIGLX_SOURCES = server/radeon_dri.c
@@ -20,20 +22,26 @@ COMMON_SOURCES = \
 	../common/xmlconfig.c \
 	../common/dri_util.c
 
+RADEON_COMMON_SOURCES = \
+	radeon_texture.c \
+	radeon_common_context.c \
+	radeon_common.c \
+	radeon_dma.c \
+	radeon_lock.c \
+	radeon_bo_legacy.c \
+	radeon_cs_legacy.c \
+	radeon_mipmap_tree.c \
+	radeon_span.c \
+	radeon_fbo.c
+
 DRIVER_SOURCES = \
 		 radeon_screen.c \
-		 radeon_context.c \
-		 radeon_ioctl.c \
-		 radeon_lock.c \
-		 radeon_span.c \
-		 radeon_state.c \
-		 r300_mem.c \
 		 r300_context.c \
+		 r300_draw.c \
 		 r300_ioctl.c \
 		 r300_cmdbuf.c \
 		 r300_state.c \
 		 r300_render.c \
-		 r300_texmem.c \
 		 r300_tex.c \
 		 r300_texstate.c \
 		 radeon_program.c \
@@ -41,6 +49,7 @@ DRIVER_SOURCES = \
 		 radeon_program_pair.c \
 		 radeon_nqssadce.c \
 		 r300_vertprog.c \
+		 r300_fragprog_common.c \
 		 r300_fragprog.c \
 		 r300_fragprog_swizzle.c \
 		 r300_fragprog_emit.c \
@@ -49,12 +58,15 @@ DRIVER_SOURCES = \
 		 r300_shader.c \
 		 r300_emit.c \
 		 r300_swtcl.c \
+		 $(RADEON_COMMON_SOURCES) \
 		 $(EGL_SOURCES)
 
 C_SOURCES = $(COMMON_SOURCES) $(DRIVER_SOURCES)
 
 DRIVER_DEFINES = -DCOMPILE_R300 -DR200_MERGED=0 \
-	-DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R300
+	-DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R300 \
+#	-DRADEON_BO_TRACK \
+	-Wall
 
 SYMLINKS = \
 	server/radeon_dri.c \
@@ -68,7 +80,29 @@ COMMON_SYMLINKS = \
 	radeon_chipset.h \
 	radeon_screen.c \
 	radeon_screen.h \
-	radeon_span.h
+	radeon_span.h \
+	radeon_span.c \
+	radeon_bo_legacy.c \
+	radeon_cs_legacy.c \
+	radeon_bo_legacy.h \
+	radeon_cs_legacy.h \
+	radeon_bocs_wrapper.h \
+	radeon_lock.c \
+	radeon_lock.h \
+	radeon_common.c \
+	radeon_common.h \
+	radeon_common_context.c \
+	radeon_common_context.h \
+	radeon_cmdbuf.h \
+	radeon_dma.c \
+	radeon_dma.h \
+	radeon_mipmap_tree.c \
+	radeon_mipmap_tree.h \
+	radeon_texture.c \
+	radeon_texture.h \
+	radeon_fbo.c
+
+DRI_LIB_DEPS += $(RADEON_LDFLAGS)
 
 ##### TARGETS #####
 
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index f4472756f10..0261a5b1d8a 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -44,245 +44,417 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "drm.h"
 #include "radeon_drm.h"
 
-#include "radeon_ioctl.h"
 #include "r300_context.h"
 #include "r300_ioctl.h"
 #include "radeon_reg.h"
 #include "r300_reg.h"
 #include "r300_cmdbuf.h"
 #include "r300_emit.h"
+#include "radeon_bocs_wrapper.h"
+#include "radeon_mipmap_tree.h"
 #include "r300_state.h"
+#include "radeon_reg.h"
 
-// Set this to 1 for extremely verbose debugging of command buffers
-#define DEBUG_CMDBUF		0
-
-/**
- * Send the current command buffer via ioctl to the hardware.
+/** # of dwords reserved for additional instructions that may need to be written
+ * during flushing.
  */
-int r300FlushCmdBufLocked(r300ContextPtr r300, const char *caller)
+#define SPACE_FOR_FLUSHING	4
+
+static unsigned packet0_count(r300ContextPtr r300, uint32_t *pkt)
 {
-	int ret;
-	int i;
-	drm_radeon_cmd_buffer_t cmd;
-	int start;
-
-	if (r300->radeon.lost_context) {
-		start = 0;
-		r300->radeon.lost_context = GL_FALSE;
-	} else
-		start = r300->cmdbuf.count_reemit;
-
-	if (RADEON_DEBUG & DEBUG_IOCTL) {
-		fprintf(stderr, "%s from %s - %i cliprects\n",
-			__FUNCTION__, caller, r300->radeon.numClipRects);
-
-		if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_VERBOSE)
-			for (i = start; i < r300->cmdbuf.count_used; ++i)
-				fprintf(stderr, "%d: %08x\n", i,
-					r300->cmdbuf.cmd_buf[i]);
-	}
+    if (r300->radeon.radeonScreen->kernel_mm) {
+        return ((((*pkt) >> 16) & 0x3FFF) + 1);
+    } else {
+        drm_r300_cmd_header_t *t = (drm_r300_cmd_header_t*)pkt;
+        return t->packet0.count;
+    }
+}
 
-	cmd.buf = (char *)(r300->cmdbuf.cmd_buf + start);
-	cmd.bufsz = (r300->cmdbuf.count_used - start) * 4;
+#define vpu_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->vpu.count)
+#define r500fp_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->r500fp.count)
 
-	if (r300->radeon.state.scissor.enabled) {
-		cmd.nbox = r300->radeon.state.scissor.numClipRects;
-		cmd.boxes =
-		    (drm_clip_rect_t *) r300->radeon.state.scissor.pClipRects;
-	} else {
-		cmd.nbox = r300->radeon.numClipRects;
-		cmd.boxes = (drm_clip_rect_t *) r300->radeon.pClipRects;
+void emit_vpu(GLcontext *ctx, struct radeon_state_atom * atom)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+	drm_r300_cmd_header_t cmd;
+	uint32_t addr, ndw, i;
+
+	if (!r300->radeon.radeonScreen->kernel_mm) {
+		uint32_t dwords;
+		dwords = (*atom->check) (ctx, atom);
+		BEGIN_BATCH_NO_AUTOSTATE(dwords);
+		OUT_BATCH_TABLE(atom->cmd, dwords);
+		END_BATCH();
+		return;
 	}
 
-	ret = drmCommandWrite(r300->radeon.dri.fd,
-			      DRM_RADEON_CMDBUF, &cmd, sizeof(cmd));
-
-	if (RADEON_DEBUG & DEBUG_SYNC) {
-		fprintf(stderr, "Syncing in %s (from %s)\n\n",
-			__FUNCTION__, caller);
-		radeonWaitForIdleLocked(&r300->radeon);
+	cmd.u = atom->cmd[0];
+	addr = (cmd.vpu.adrhi << 8) | cmd.vpu.adrlo;
+	ndw = cmd.vpu.count * 4;
+	if (ndw) {
+
+		if (r300->vap_flush_needed) {
+			BEGIN_BATCH_NO_AUTOSTATE(15 + ndw);
+
+			/* flush processing vertices */
+			OUT_BATCH_REGVAL(R300_SC_SCREENDOOR, 0);
+			OUT_BATCH_REGVAL(R300_RB3D_DSTCACHE_CTLSTAT, R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
+			OUT_BATCH_REGVAL(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
+			OUT_BATCH_REGVAL(R300_SC_SCREENDOOR, 0xffffff);
+			OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0);
+			r300->vap_flush_needed = GL_FALSE;
+		} else {
+			BEGIN_BATCH_NO_AUTOSTATE(5 + ndw);
+		}
+		OUT_BATCH_REGVAL(R300_VAP_PVS_VECTOR_INDX_REG, addr);
+		OUT_BATCH(CP_PACKET0(R300_VAP_PVS_UPLOAD_DATA, ndw-1) | RADEON_ONE_REG_WR);
+		for (i = 0; i < ndw; i++) {
+			OUT_BATCH(atom->cmd[i+1]);
+		}
+		OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0);
+		END_BATCH();
 	}
-
-	r300->dma.nr_released_bufs = 0;
-	r300->cmdbuf.count_used = 0;
-	r300->cmdbuf.count_reemit = 0;
-
-	return ret;
 }
 
-int r300FlushCmdBuf(r300ContextPtr r300, const char *caller)
+void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom)
 {
-	int ret;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+	drm_r300_cmd_header_t cmd;
+	uint32_t addr, ndw, i, sz;
+	int type, clamp, stride;
+
+	if (!r300->radeon.radeonScreen->kernel_mm) {
+		uint32_t dwords;
+		dwords = (*atom->check) (ctx, atom);
+		BEGIN_BATCH_NO_AUTOSTATE(dwords);
+		OUT_BATCH_TABLE(atom->cmd, dwords);
+		END_BATCH();
+		return;
+	}
 
-	LOCK_HARDWARE(&r300->radeon);
+	cmd.u = atom->cmd[0];
+	sz = cmd.r500fp.count;
+	addr = ((cmd.r500fp.adrhi_flags & 1) << 8) | cmd.r500fp.adrlo;
+	type = !!(cmd.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE);
+	clamp = !!(cmd.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP);
 
-	ret = r300FlushCmdBufLocked(r300, caller);
+	addr |= (type << 16);
+	addr |= (clamp << 17);
 
-	UNLOCK_HARDWARE(&r300->radeon);
+	stride = type ? 4 : 6;
 
-	if (ret) {
-		fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret);
-		_mesa_exit(ret);
-	}
+	ndw = sz * stride;
+	if (ndw) {
 
-	return ret;
+		BEGIN_BATCH_NO_AUTOSTATE(3 + ndw);
+		OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_INDEX, 0));
+		OUT_BATCH(addr);
+		OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_DATA, ndw-1) | RADEON_ONE_REG_WR);
+		for (i = 0; i < ndw; i++) {
+			OUT_BATCH(atom->cmd[i+1]);
+		}
+		END_BATCH();
+	}
 }
 
-static void r300PrintStateAtom(r300ContextPtr r300, struct r300_state_atom *state)
+static void emit_tex_offsets(GLcontext *ctx, struct radeon_state_atom * atom)
 {
-	int i, j, reg;
-	int dwords = (*state->check) (r300, state);
-	drm_r300_cmd_header_t cmd;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+	int numtmus = packet0_count(r300, r300->hw.tex.offset.cmd);
+	int notexture = 0;
 
-	fprintf(stderr, "  emit %s %d/%d\n", state->name, dwords,
-		state->cmd_size);
-
-	if (RADEON_DEBUG & DEBUG_VERBOSE) {
-		for (i = 0; i < dwords;) {
-			cmd = *((drm_r300_cmd_header_t *) &state->cmd[i]);
-			reg = (cmd.packet0.reghi << 8) | cmd.packet0.reglo;
-			fprintf(stderr, "      %s[%d]: cmdpacket0 (first reg=0x%04x, count=%d)\n",
-					state->name, i, reg, cmd.packet0.count);
-			++i;
-			for (j = 0; j < cmd.packet0.count; j++) {
-				fprintf(stderr, "      %s[%d]: 0x%04x = %08x\n",
-					state->name, i, reg, state->cmd[i]);
-				reg += 4;
-				++i;
-			}
+	if (numtmus) {
+		int i;
+
+		for(i = 0; i < numtmus; ++i) {
+		    radeonTexObj *t = r300->hw.textures[i];
+
+		    if (!t)
+			notexture = 1;
+		}
+
+		if (r300->radeon.radeonScreen->kernel_mm && notexture) {
+			return;
+		}
+		for(i = 0; i < numtmus; ++i) {
+		    radeonTexObj *t = r300->hw.textures[i];
+		    if (t && !t->image_override) {
+                BEGIN_BATCH_NO_AUTOSTATE(4);
+                OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1);
+			    OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+					    RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+                END_BATCH();
+		    } else if (!t) {
+                /* Texture unit hasn't a texture bound nothings to do */
+		    } else { /* override cases */
+			    if (t->bo) {
+                    BEGIN_BATCH_NO_AUTOSTATE(4);
+                    OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1);
+				    OUT_BATCH_RELOC(t->tile_bits, t->bo, 0,
+						    RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+                    END_BATCH();
+			    } else if (!r300->radeon.radeonScreen->kernel_mm) {
+                    BEGIN_BATCH_NO_AUTOSTATE(2);
+                    OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1);
+				    OUT_BATCH(t->override_offset);
+                    END_BATCH();
+			    } else {
+                    /* Texture unit hasn't a texture bound nothings to do */
+                }
+		    }
 		}
 	}
 }
 
-/**
- * Emit all atoms with a dirty field equal to dirty.
- *
- * The caller must have ensured that there is enough space in the command
- * buffer.
- */
-static INLINE void r300EmitAtoms(r300ContextPtr r300, GLboolean dirty)
+void r300_emit_scissor(GLcontext *ctx)
 {
-	struct r300_state_atom *atom;
-	uint32_t *dest;
-	int dwords;
-
-	dest = r300->cmdbuf.cmd_buf + r300->cmdbuf.count_used;
-
-	/* Emit WAIT */
-	*dest = cmdwait(R300_WAIT_3D | R300_WAIT_3D_CLEAN);
-	dest++;
-	r300->cmdbuf.count_used++;
-
-	/* Emit cache flush */
-	*dest = cmdpacket0(R300_TX_INVALTAGS, 1);
-	dest++;
-	r300->cmdbuf.count_used++;
-
-	*dest = R300_TX_FLUSH;
-	dest++;
-	r300->cmdbuf.count_used++;
-
-	/* Emit END3D */
-	*dest = cmdpacify();
-	dest++;
-	r300->cmdbuf.count_used++;
-
-	/* Emit actual atoms */
-
-	foreach(atom, &r300->hw.atomlist) {
-		if ((atom->dirty || r300->hw.all_dirty) == dirty) {
-			dwords = (*atom->check) (r300, atom);
-			if (dwords) {
-				if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) {
-					r300PrintStateAtom(r300, atom);
-				}
-				memcpy(dest, atom->cmd, dwords * 4);
-				dest += dwords;
-				r300->cmdbuf.count_used += dwords;
-				atom->dirty = GL_FALSE;
-			} else {
-				if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) {
-					fprintf(stderr, "  skip state %s\n",
-						atom->name);
-				}
-			}
-		}
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+    unsigned x1, y1, x2, y2;
+	struct radeon_renderbuffer *rrb;
+
+    if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) {
+        return;
+    }
+	rrb = radeon_get_colorbuffer(&r300->radeon);
+	if (!rrb || !rrb->bo) {
+		fprintf(stderr, "no rrb\n");
+		return;
 	}
+    if (r300->radeon.state.scissor.enabled) {
+        x1 = r300->radeon.state.scissor.rect.x1;
+        y1 = r300->radeon.state.scissor.rect.y1;
+        x2 = r300->radeon.state.scissor.rect.x2 - 1;
+        y2 = r300->radeon.state.scissor.rect.y2 - 1;
+    } else {
+        x1 = 0;
+        y1 = 0;
+        x2 = rrb->width - 1;
+        y2 = rrb->height - 1;
+    }
+    if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) {
+        x1 += R300_SCISSORS_OFFSET;
+        y1 += R300_SCISSORS_OFFSET;
+        x2 += R300_SCISSORS_OFFSET;
+        y2 += R300_SCISSORS_OFFSET;
+    }
+    BEGIN_BATCH_NO_AUTOSTATE(3);
+    OUT_BATCH_REGSEQ(R300_SC_SCISSORS_TL, 2);
+    OUT_BATCH((x1 << R300_SCISSORS_X_SHIFT)|(y1 << R300_SCISSORS_Y_SHIFT));
+    OUT_BATCH((x2 << R300_SCISSORS_X_SHIFT)|(y2 << R300_SCISSORS_Y_SHIFT));
+    END_BATCH();
 }
 
-/**
- * Copy dirty hardware state atoms into the command buffer.
- *
- * We also copy out clean state if we're at the start of a buffer. That makes
- * it easy to recover from lost contexts.
- */
-void r300EmitState(r300ContextPtr r300)
+static void emit_cb_offset(GLcontext *ctx, struct radeon_state_atom * atom)
 {
-	if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_PRIMS))
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	if (r300->cmdbuf.count_used && !r300->hw.is_dirty
-	    && !r300->hw.all_dirty)
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+	struct radeon_renderbuffer *rrb;
+	uint32_t cbpitch;
+	uint32_t offset = r300->radeon.state.color.draw_offset;
+	uint32_t dw = 6;
+    int i;
+
+	rrb = radeon_get_colorbuffer(&r300->radeon);
+	if (!rrb || !rrb->bo) {
+		fprintf(stderr, "no rrb\n");
 		return;
+	}
 
-	/* To avoid going across the entire set of states multiple times, just check
-	 * for enough space for the case of emitting all state, and inline the
-	 * r300AllocCmdBuf code here without all the checks.
-	 */
-	r300EnsureCmdBufSpace(r300, r300->hw.max_state_size, __FUNCTION__);
+	cbpitch = (rrb->pitch / rrb->cpp);
+	if (rrb->cpp == 4)
+		cbpitch |= R300_COLOR_FORMAT_ARGB8888;
+	else
+		cbpitch |= R300_COLOR_FORMAT_RGB565;
+
+	if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
+		cbpitch |= R300_COLOR_TILE_ENABLE;
+
+    	if (r300->radeon.radeonScreen->kernel_mm)
+		dw += 2;
+	BEGIN_BATCH_NO_AUTOSTATE(dw);
+	OUT_BATCH_REGSEQ(R300_RB3D_COLOROFFSET0, 1);
+	OUT_BATCH_RELOC(offset, rrb->bo, offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+	OUT_BATCH_REGSEQ(R300_RB3D_COLORPITCH0, 1);
+    	if (!r300->radeon.radeonScreen->kernel_mm)
+		OUT_BATCH(cbpitch);
+	else
+		OUT_BATCH_RELOC(cbpitch, rrb->bo, cbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+	END_BATCH();
+    if (r300->radeon.radeonScreen->driScreen->dri2.enabled) {
+        if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+            BEGIN_BATCH_NO_AUTOSTATE(3);
+            OUT_BATCH_REGSEQ(R300_SC_SCISSORS_TL, 2);
+            OUT_BATCH(0);
+            OUT_BATCH(((rrb->width - 1) << R300_SCISSORS_X_SHIFT) |
+                    ((rrb->height - 1) << R300_SCISSORS_Y_SHIFT));
+            END_BATCH();
+            BEGIN_BATCH_NO_AUTOSTATE(16);
+            for (i = 0; i < 4; i++) {
+                OUT_BATCH_REGSEQ(R300_SC_CLIPRECT_TL_0 + (i * 8), 2);
+                OUT_BATCH((0 << R300_CLIPRECT_X_SHIFT) | (0 << R300_CLIPRECT_Y_SHIFT));
+                OUT_BATCH(((rrb->width - 1) << R300_CLIPRECT_X_SHIFT) | ((rrb->height - 1) << R300_CLIPRECT_Y_SHIFT));
+            }
+            OUT_BATCH_REGSEQ(R300_SC_CLIP_RULE, 1);
+            OUT_BATCH(0xAAAA);
+            OUT_BATCH_REGSEQ(R300_SC_SCREENDOOR, 1);
+            OUT_BATCH(0xffffff);
+            END_BATCH();
+        } else {
+            BEGIN_BATCH_NO_AUTOSTATE(3);
+            OUT_BATCH_REGSEQ(R300_SC_SCISSORS_TL, 2);
+            OUT_BATCH((R300_SCISSORS_OFFSET << R300_SCISSORS_X_SHIFT) |
+                    (R300_SCISSORS_OFFSET << R300_SCISSORS_Y_SHIFT));
+            OUT_BATCH(((rrb->width + R300_SCISSORS_OFFSET - 1) << R300_SCISSORS_X_SHIFT) |
+                    ((rrb->height + R300_SCISSORS_OFFSET - 1) << R300_SCISSORS_Y_SHIFT));
+            END_BATCH();
+            BEGIN_BATCH_NO_AUTOSTATE(16);
+            for (i = 0; i < 4; i++) {
+                OUT_BATCH_REGSEQ(R300_SC_CLIPRECT_TL_0 + (i * 8), 2);
+                OUT_BATCH((R300_SCISSORS_OFFSET << R300_CLIPRECT_X_SHIFT) | (R300_SCISSORS_OFFSET << R300_CLIPRECT_Y_SHIFT));
+                OUT_BATCH(((R300_SCISSORS_OFFSET + rrb->width - 1) << R300_CLIPRECT_X_SHIFT) |
+                          ((R300_SCISSORS_OFFSET + rrb->height - 1) << R300_CLIPRECT_Y_SHIFT));
+            }
+            OUT_BATCH_REGSEQ(R300_SC_CLIP_RULE, 1);
+            OUT_BATCH(0xAAAA);
+            OUT_BATCH_REGSEQ(R300_SC_SCREENDOOR, 1);
+            OUT_BATCH(0xffffff);
+            END_BATCH();
+        }
+    }
+}
 
-	if (!r300->cmdbuf.count_used) {
-		if (RADEON_DEBUG & DEBUG_STATE)
-			fprintf(stderr, "Begin reemit state\n");
+static void emit_zb_offset(GLcontext *ctx, struct radeon_state_atom * atom)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+	struct radeon_renderbuffer *rrb;
+	uint32_t zbpitch;
 
-		r300EmitAtoms(r300, GL_FALSE);
-		r300->cmdbuf.count_reemit = r300->cmdbuf.count_used;
+	rrb = radeon_get_depthbuffer(&r300->radeon);
+	if (!rrb)
+		return;
+
+	zbpitch = (rrb->pitch / rrb->cpp);
+	if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) {
+		zbpitch |= R300_DEPTHMACROTILE_ENABLE;
+	}
+	if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE){
+		zbpitch |= R300_DEPTHMICROTILE_TILED;
 	}
 
-	if (RADEON_DEBUG & DEBUG_STATE)
-		fprintf(stderr, "Begin dirty state\n");
+	BEGIN_BATCH_NO_AUTOSTATE(6);
+	OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1);
+	OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+	OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, zbpitch);
+	END_BATCH();
+}
 
-	r300EmitAtoms(r300, GL_TRUE);
+static void emit_gb_misc(GLcontext *ctx, struct radeon_state_atom * atom)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+    if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) {
+        BEGIN_BATCH_NO_AUTOSTATE(4);
+        OUT_BATCH(atom->cmd[0]);
+        OUT_BATCH(atom->cmd[1]);
+        OUT_BATCH(atom->cmd[2]);
+        OUT_BATCH(atom->cmd[3]);
+        END_BATCH();
+    }
+}
 
-	assert(r300->cmdbuf.count_used < r300->cmdbuf.size);
+static void emit_threshold_misc(GLcontext *ctx, struct radeon_state_atom * atom)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+    if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+        BEGIN_BATCH_NO_AUTOSTATE(3);
+        OUT_BATCH(atom->cmd[0]);
+        OUT_BATCH(atom->cmd[1]);
+        OUT_BATCH(atom->cmd[2]);
+        END_BATCH();
+    }
+}
 
-	r300->hw.is_dirty = GL_FALSE;
-	r300->hw.all_dirty = GL_FALSE;
+static void emit_shade_misc(GLcontext *ctx, struct radeon_state_atom * atom)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+
+    if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) {
+        BEGIN_BATCH_NO_AUTOSTATE(2);
+        OUT_BATCH(atom->cmd[0]);
+        OUT_BATCH(atom->cmd[1]);
+        END_BATCH();
+    }
 }
 
-#define packet0_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->packet0.count)
-#define vpu_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->vpu.count)
-#define r500fp_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->r500fp.count)
+static void emit_zstencil_format(GLcontext *ctx, struct radeon_state_atom * atom)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&r300->radeon);
+	struct radeon_renderbuffer *rrb;
+	uint32_t format = 0;
+
+	rrb = radeon_get_depthbuffer(&r300->radeon);
+	if (!rrb)
+	  format = 0;
+	else {
+	  if (rrb->cpp == 2)
+	    format = R300_DEPTHFORMAT_16BIT_INT_Z;
+	  else if (rrb->cpp == 4)
+	    format = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
+	}
+
+	OUT_BATCH(atom->cmd[0]);
+	atom->cmd[1] &= ~0xf;
+	atom->cmd[1] |= format;
+	OUT_BATCH(atom->cmd[1]);
+	OUT_BATCH(atom->cmd[2]);
+	OUT_BATCH(atom->cmd[3]);
+	OUT_BATCH(atom->cmd[4]);
+}
 
-static int check_always(r300ContextPtr r300, struct r300_state_atom *atom)
+static int check_always(GLcontext *ctx, struct radeon_state_atom *atom)
 {
 	return atom->cmd_size;
 }
 
-static int check_variable(r300ContextPtr r300, struct r300_state_atom *atom)
+static int check_variable(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	int cnt;
-	cnt = packet0_count(atom->cmd);
+	if (atom->cmd[0] == CP_PACKET2) {
+		return 0;
+	}
+	cnt = packet0_count(r300, atom->cmd);
 	return cnt ? cnt + 1 : 0;
 }
 
-static int check_vpu(r300ContextPtr r300, struct r300_state_atom *atom)
+int check_vpu(GLcontext *ctx, struct radeon_state_atom *atom)
 {
 	int cnt;
+
 	cnt = vpu_count(atom->cmd);
 	return cnt ? (cnt * 4) + 1 : 0;
 }
 
-static int check_r500fp(r300ContextPtr r300, struct r300_state_atom *atom)
+int check_r500fp(GLcontext *ctx, struct radeon_state_atom *atom)
 {
 	int cnt;
+
 	cnt = r500fp_count(atom->cmd);
 	return cnt ? (cnt * 6) + 1 : 0;
 }
 
-static int check_r500fp_const(r300ContextPtr r300, struct r300_state_atom *atom)
+int check_r500fp_const(GLcontext *ctx, struct radeon_state_atom *atom)
 {
 	int cnt;
+
 	cnt = r500fp_count(atom->cmd);
 	return cnt ? (cnt * 4) + 1 : 0;
 }
@@ -295,8 +467,8 @@ static int check_r500fp_const(r300ContextPtr r300, struct r300_state_atom *atom)
       r300->hw.ATOM.idx = (IDX);					\
       r300->hw.ATOM.check = check_##CHK;				\
       r300->hw.ATOM.dirty = GL_FALSE;					\
-      r300->hw.max_state_size += (SZ);					\
-      insert_at_tail(&r300->hw.atomlist, &r300->hw.ATOM);		\
+      r300->radeon.hw.max_state_size += (SZ);					\
+      insert_at_tail(&r300->radeon.hw.atomlist, &r300->hw.ATOM);		\
    } while (0)
 /**
  * Allocate memory for the command buffer and initialize the state atom
@@ -304,18 +476,16 @@ static int check_r500fp_const(r300ContextPtr r300, struct r300_state_atom *atom)
  */
 void r300InitCmdBuf(r300ContextPtr r300)
 {
-	int size, mtu;
-	int has_tcl = 1;
+	int mtu;
+	int has_tcl;
 	int is_r500 = 0;
-	int i;
 
-	if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
-		has_tcl = 0;
+	has_tcl = r300->options.hw_tcl_enabled;
 
 	if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
 		is_r500 = 1;
 
-	r300->hw.max_state_size = 2 + 2;	/* reserve extra space for WAIT_IDLE and tex cache flush */
+	r300->radeon.hw.max_state_size = 2 + 2;	/* reserve extra space for WAIT_IDLE and tex cache flush */
 
 	mtu = r300->radeon.glCtx->Const.MaxTextureUnits;
 	if (RADEON_DEBUG & DEBUG_TEXTURE) {
@@ -323,232 +493,248 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	}
 
 	/* Setup the atom linked list */
-	make_empty_list(&r300->hw.atomlist);
-	r300->hw.atomlist.name = "atom-list";
+	make_empty_list(&r300->radeon.hw.atomlist);
+	r300->radeon.hw.atomlist.name = "atom-list";
 
 	/* Initialize state atoms */
 	ALLOC_STATE(vpt, always, R300_VPT_CMDSIZE, 0);
-	r300->hw.vpt.cmd[R300_VPT_CMD_0] = cmdpacket0(R300_SE_VPORT_XSCALE, 6);
+	r300->hw.vpt.cmd[R300_VPT_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_SE_VPORT_XSCALE, 6);
 	ALLOC_STATE(vap_cntl, always, R300_VAP_CNTL_SIZE, 0);
-	r300->hw.vap_cntl.cmd[R300_VAP_CNTL_FLUSH] = cmdpacket0(R300_VAP_PVS_STATE_FLUSH_REG, 1);
+	r300->hw.vap_cntl.cmd[R300_VAP_CNTL_FLUSH] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PVS_STATE_FLUSH_REG, 1);
 	r300->hw.vap_cntl.cmd[R300_VAP_CNTL_FLUSH_1] = 0;
-	r300->hw.vap_cntl.cmd[R300_VAP_CNTL_CMD] = cmdpacket0(R300_VAP_CNTL, 1);
+	r300->hw.vap_cntl.cmd[R300_VAP_CNTL_CMD] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_CNTL, 1);
 	if (is_r500) {
 	    ALLOC_STATE(vap_index_offset, always, 2, 0);
-	    r300->hw.vap_index_offset.cmd[0] = cmdpacket0(R500_VAP_INDEX_OFFSET, 1);
+	    r300->hw.vap_index_offset.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_VAP_INDEX_OFFSET, 1);
 	    r300->hw.vap_index_offset.cmd[1] = 0;
 	}
 	ALLOC_STATE(vte, always, 3, 0);
-	r300->hw.vte.cmd[0] = cmdpacket0(R300_SE_VTE_CNTL, 2);
+	r300->hw.vte.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SE_VTE_CNTL, 2);
 	ALLOC_STATE(vap_vf_max_vtx_indx, always, 3, 0);
-	r300->hw.vap_vf_max_vtx_indx.cmd[0] = cmdpacket0(R300_VAP_VF_MAX_VTX_INDX, 2);
+	r300->hw.vap_vf_max_vtx_indx.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_VF_MAX_VTX_INDX, 2);
 	ALLOC_STATE(vap_cntl_status, always, 2, 0);
-	r300->hw.vap_cntl_status.cmd[0] = cmdpacket0(R300_VAP_CNTL_STATUS, 1);
+	r300->hw.vap_cntl_status.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_CNTL_STATUS, 1);
 	ALLOC_STATE(vir[0], variable, R300_VIR_CMDSIZE, 0);
 	r300->hw.vir[0].cmd[R300_VIR_CMD_0] =
-	    cmdpacket0(R300_VAP_PROG_STREAM_CNTL_0, 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PROG_STREAM_CNTL_0, 1);
 	ALLOC_STATE(vir[1], variable, R300_VIR_CMDSIZE, 1);
 	r300->hw.vir[1].cmd[R300_VIR_CMD_0] =
-	    cmdpacket0(R300_VAP_PROG_STREAM_CNTL_EXT_0, 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PROG_STREAM_CNTL_EXT_0, 1);
 	ALLOC_STATE(vic, always, R300_VIC_CMDSIZE, 0);
-	r300->hw.vic.cmd[R300_VIC_CMD_0] = cmdpacket0(R300_VAP_VTX_STATE_CNTL, 2);
+	r300->hw.vic.cmd[R300_VIC_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_VTX_STATE_CNTL, 2);
 	ALLOC_STATE(vap_psc_sgn_norm_cntl, always, 2, 0);
-	r300->hw.vap_psc_sgn_norm_cntl.cmd[0] = cmdpacket0(R300_VAP_PSC_SGN_NORM_CNTL, SGN_NORM_ZERO_CLAMP_MINUS_ONE);
+	r300->hw.vap_psc_sgn_norm_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PSC_SGN_NORM_CNTL, SGN_NORM_ZERO_CLAMP_MINUS_ONE);
 
 	if (has_tcl) {
 		ALLOC_STATE(vap_clip_cntl, always, 2, 0);
-		r300->hw.vap_clip_cntl.cmd[0] = cmdpacket0(R300_VAP_CLIP_CNTL, 1);
+		r300->hw.vap_clip_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_CLIP_CNTL, 1);
 		ALLOC_STATE(vap_clip, always, 5, 0);
-		r300->hw.vap_clip.cmd[0] = cmdpacket0(R300_VAP_GB_VERT_CLIP_ADJ, 4);
+		r300->hw.vap_clip.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_GB_VERT_CLIP_ADJ, 4);
 		ALLOC_STATE(vap_pvs_vtx_timeout_reg, always, 2, 0);
-		r300->hw.vap_pvs_vtx_timeout_reg.cmd[0] = cmdpacket0(VAP_PVS_VTX_TIMEOUT_REG, 1);
+		r300->hw.vap_pvs_vtx_timeout_reg.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, VAP_PVS_VTX_TIMEOUT_REG, 1);
 	}
 
 	ALLOC_STATE(vof, always, R300_VOF_CMDSIZE, 0);
 	r300->hw.vof.cmd[R300_VOF_CMD_0] =
-	    cmdpacket0(R300_VAP_OUTPUT_VTX_FMT_0, 2);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_VAP_OUTPUT_VTX_FMT_0, 2);
 
 	if (has_tcl) {
 		ALLOC_STATE(pvs, always, R300_PVS_CMDSIZE, 0);
 		r300->hw.pvs.cmd[R300_PVS_CMD_0] =
-		    cmdpacket0(R300_VAP_PVS_CODE_CNTL_0, 3);
+		    cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PVS_CODE_CNTL_0, 3);
 	}
 
 	ALLOC_STATE(gb_enable, always, 2, 0);
-	r300->hw.gb_enable.cmd[0] = cmdpacket0(R300_GB_ENABLE, 1);
+	r300->hw.gb_enable.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GB_ENABLE, 1);
 	ALLOC_STATE(gb_misc, always, R300_GB_MISC_CMDSIZE, 0);
-	r300->hw.gb_misc.cmd[0] = cmdpacket0(R300_GB_MSPOS0, 5);
+	r300->hw.gb_misc.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GB_MSPOS0, 3);
+	r300->hw.gb_misc.emit = emit_gb_misc;
+	ALLOC_STATE(gb_misc2, always, R300_GB_MISC2_CMDSIZE, 0);
+    r300->hw.gb_misc2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x401C, 2);
 	ALLOC_STATE(txe, always, R300_TXE_CMDSIZE, 0);
-	r300->hw.txe.cmd[R300_TXE_CMD_0] = cmdpacket0(R300_TX_ENABLE, 1);
+	r300->hw.txe.cmd[R300_TXE_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_ENABLE, 1);
 	ALLOC_STATE(ga_point_s0, always, 5, 0);
-	r300->hw.ga_point_s0.cmd[0] = cmdpacket0(R300_GA_POINT_S0, 4);
+	r300->hw.ga_point_s0.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POINT_S0, 4);
 	ALLOC_STATE(ga_triangle_stipple, always, 2, 0);
-	r300->hw.ga_triangle_stipple.cmd[0] = cmdpacket0(R300_GA_TRIANGLE_STIPPLE, 1);
+	r300->hw.ga_triangle_stipple.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_TRIANGLE_STIPPLE, 1);
 	ALLOC_STATE(ps, always, R300_PS_CMDSIZE, 0);
-	r300->hw.ps.cmd[0] = cmdpacket0(R300_GA_POINT_SIZE, 1);
+	r300->hw.ps.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POINT_SIZE, 1);
 	ALLOC_STATE(ga_point_minmax, always, 4, 0);
-	r300->hw.ga_point_minmax.cmd[0] = cmdpacket0(R300_GA_POINT_MINMAX, 3);
+	r300->hw.ga_point_minmax.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POINT_MINMAX, 3);
 	ALLOC_STATE(lcntl, always, 2, 0);
-	r300->hw.lcntl.cmd[0] = cmdpacket0(R300_GA_LINE_CNTL, 1);
+	r300->hw.lcntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_LINE_CNTL, 1);
 	ALLOC_STATE(ga_line_stipple, always, 4, 0);
-	r300->hw.ga_line_stipple.cmd[0] = cmdpacket0(R300_GA_LINE_STIPPLE_VALUE, 3);
-	ALLOC_STATE(shade, always, 5, 0);
-	r300->hw.shade.cmd[0] = cmdpacket0(R300_GA_ENHANCE, 4);
+	r300->hw.ga_line_stipple.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_LINE_STIPPLE_VALUE, 3);
+	ALLOC_STATE(shade, always, 2, 0);
+	r300->hw.shade.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_ENHANCE, 1);
+	r300->hw.shade.emit = emit_shade_misc;
+	ALLOC_STATE(shade2, always, 4, 0);
+	r300->hw.shade2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x4278, 3);
 	ALLOC_STATE(polygon_mode, always, 4, 0);
-	r300->hw.polygon_mode.cmd[0] = cmdpacket0(R300_GA_POLY_MODE, 3);
+	r300->hw.polygon_mode.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POLY_MODE, 3);
 	ALLOC_STATE(fogp, always, 3, 0);
-	r300->hw.fogp.cmd[0] = cmdpacket0(R300_GA_FOG_SCALE, 2);
+	r300->hw.fogp.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_FOG_SCALE, 2);
 	ALLOC_STATE(zbias_cntl, always, 2, 0);
-	r300->hw.zbias_cntl.cmd[0] = cmdpacket0(R300_SU_TEX_WRAP, 1);
+	r300->hw.zbias_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_TEX_WRAP, 1);
 	ALLOC_STATE(zbs, always, R300_ZBS_CMDSIZE, 0);
 	r300->hw.zbs.cmd[R300_ZBS_CMD_0] =
-	    cmdpacket0(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
 	ALLOC_STATE(occlusion_cntl, always, 2, 0);
-	r300->hw.occlusion_cntl.cmd[0] = cmdpacket0(R300_SU_POLY_OFFSET_ENABLE, 1);
+	r300->hw.occlusion_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_POLY_OFFSET_ENABLE, 1);
 	ALLOC_STATE(cul, always, R300_CUL_CMDSIZE, 0);
-	r300->hw.cul.cmd[R300_CUL_CMD_0] = cmdpacket0(R300_SU_CULL_MODE, 1);
+	r300->hw.cul.cmd[R300_CUL_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_CULL_MODE, 1);
 	ALLOC_STATE(su_depth_scale, always, 3, 0);
-	r300->hw.su_depth_scale.cmd[0] = cmdpacket0(R300_SU_DEPTH_SCALE, 2);
+	r300->hw.su_depth_scale.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_DEPTH_SCALE, 2);
 	ALLOC_STATE(rc, always, R300_RC_CMDSIZE, 0);
-	r300->hw.rc.cmd[R300_RC_CMD_0] = cmdpacket0(R300_RS_COUNT, 2);
+	r300->hw.rc.cmd[R300_RC_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_COUNT, 2);
 	if (is_r500) {
-		ALLOC_STATE(ri, always, R500_RI_CMDSIZE, 0);
-		r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(R500_RS_IP_0, 16);
-		for (i = 0; i < 8; i++) {
-			r300->hw.ri.cmd[R300_RI_CMD_0 + i +1] =
-			  (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) |
-                          (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) |
-                          (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) |
-                          (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT);
-		}
+		ALLOC_STATE(ri, variable, R500_RI_CMDSIZE, 0);
+		r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_IP_0, 16);
 		ALLOC_STATE(rr, variable, R300_RR_CMDSIZE, 0);
-		r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R500_RS_INST_0, 1);
+		r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_INST_0, 1);
 	} else {
-		ALLOC_STATE(ri, always, R300_RI_CMDSIZE, 0);
-		r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(R300_RS_IP_0, 8);
+		ALLOC_STATE(ri, variable, R300_RI_CMDSIZE, 0);
+		r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_IP_0, 8);
 		ALLOC_STATE(rr, variable, R300_RR_CMDSIZE, 0);
-		r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R300_RS_INST_0, 1);
+		r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_INST_0, 1);
 	}
 	ALLOC_STATE(sc_hyperz, always, 3, 0);
-	r300->hw.sc_hyperz.cmd[0] = cmdpacket0(R300_SC_HYPERZ, 2);
+	r300->hw.sc_hyperz.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SC_HYPERZ, 2);
 	ALLOC_STATE(sc_screendoor, always, 2, 0);
-	r300->hw.sc_screendoor.cmd[0] = cmdpacket0(R300_SC_SCREENDOOR, 1);
+	r300->hw.sc_screendoor.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SC_SCREENDOOR, 1);
 	ALLOC_STATE(us_out_fmt, always, 6, 0);
-	r300->hw.us_out_fmt.cmd[0] = cmdpacket0(R300_US_OUT_FMT, 5);
+	r300->hw.us_out_fmt.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_OUT_FMT, 5);
 
 	if (is_r500) {
 		ALLOC_STATE(fp, always, R500_FP_CMDSIZE, 0);
-		r300->hw.fp.cmd[R500_FP_CMD_0] = cmdpacket0(R500_US_CONFIG, 2);
+		r300->hw.fp.cmd[R500_FP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_US_CONFIG, 2);
 		r300->hw.fp.cmd[R500_FP_CNTL] = R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO;
-		r300->hw.fp.cmd[R500_FP_CMD_1] = cmdpacket0(R500_US_CODE_ADDR, 3);
-		r300->hw.fp.cmd[R500_FP_CMD_2] = cmdpacket0(R500_US_FC_CTRL, 1);
+		r300->hw.fp.cmd[R500_FP_CMD_1] = cmdpacket0(r300->radeon.radeonScreen, R500_US_CODE_ADDR, 3);
+		r300->hw.fp.cmd[R500_FP_CMD_2] = cmdpacket0(r300->radeon.radeonScreen, R500_US_FC_CTRL, 1);
 		r300->hw.fp.cmd[R500_FP_FC_CNTL] = 0; /* FIXME when we add flow control */
 
 		ALLOC_STATE(r500fp, r500fp, R500_FPI_CMDSIZE, 0);
-		r300->hw.r500fp.cmd[R300_FPI_CMD_0] = cmdr500fp(0, 0, 0, 0);
+		r300->hw.r500fp.cmd[R300_FPI_CMD_0] =
+			cmdr500fp(r300->radeon.radeonScreen, 0, 0, 0, 0);
+		r300->hw.r500fp.emit = emit_r500fp;
 		ALLOC_STATE(r500fp_const, r500fp_const, R500_FPP_CMDSIZE, 0);
-		r300->hw.r500fp_const.cmd[R300_FPI_CMD_0] = cmdr500fp(0, 0, 1, 0);
+		r300->hw.r500fp_const.cmd[R300_FPI_CMD_0] =
+			cmdr500fp(r300->radeon.radeonScreen, 0, 0, 1, 0);
+		r300->hw.r500fp_const.emit = emit_r500fp;
 	} else {
 		ALLOC_STATE(fp, always, R300_FP_CMDSIZE, 0);
-		r300->hw.fp.cmd[R300_FP_CMD_0] = cmdpacket0(R300_US_CONFIG, 3);
-		r300->hw.fp.cmd[R300_FP_CMD_1] = cmdpacket0(R300_US_CODE_ADDR_0, 4);
+		r300->hw.fp.cmd[R300_FP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_CONFIG, 3);
+		r300->hw.fp.cmd[R300_FP_CMD_1] = cmdpacket0(r300->radeon.radeonScreen, R300_US_CODE_ADDR_0, 4);
+
 		ALLOC_STATE(fpt, variable, R300_FPT_CMDSIZE, 0);
-		r300->hw.fpt.cmd[R300_FPT_CMD_0] = cmdpacket0(R300_US_TEX_INST_0, 0);
+		r300->hw.fpt.cmd[R300_FPT_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_TEX_INST_0, 0);
 
 		ALLOC_STATE(fpi[0], variable, R300_FPI_CMDSIZE, 0);
-		r300->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_INST_0, 1);
+		r300->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_RGB_INST_0, 1);
 		ALLOC_STATE(fpi[1], variable, R300_FPI_CMDSIZE, 1);
-		r300->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_ADDR_0, 1);
+		r300->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_RGB_ADDR_0, 1);
 		ALLOC_STATE(fpi[2], variable, R300_FPI_CMDSIZE, 2);
-		r300->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_INST_0, 1);
+		r300->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_ALPHA_INST_0, 1);
 		ALLOC_STATE(fpi[3], variable, R300_FPI_CMDSIZE, 3);
-		r300->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_ADDR_0, 1);
+		r300->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_ALPHA_ADDR_0, 1);
 		ALLOC_STATE(fpp, variable, R300_FPP_CMDSIZE, 0);
-		r300->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(R300_PFS_PARAM_0_X, 0);
+		r300->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_PFS_PARAM_0_X, 0);
 	}
 	ALLOC_STATE(fogs, always, R300_FOGS_CMDSIZE, 0);
-	r300->hw.fogs.cmd[R300_FOGS_CMD_0] = cmdpacket0(R300_FG_FOG_BLEND, 1);
+	r300->hw.fogs.cmd[R300_FOGS_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_FOG_BLEND, 1);
 	ALLOC_STATE(fogc, always, R300_FOGC_CMDSIZE, 0);
-	r300->hw.fogc.cmd[R300_FOGC_CMD_0] = cmdpacket0(R300_FG_FOG_COLOR_R, 3);
+	r300->hw.fogc.cmd[R300_FOGC_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_FOG_COLOR_R, 3);
 	ALLOC_STATE(at, always, R300_AT_CMDSIZE, 0);
-	r300->hw.at.cmd[R300_AT_CMD_0] = cmdpacket0(R300_FG_ALPHA_FUNC, 2);
+	r300->hw.at.cmd[R300_AT_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_ALPHA_FUNC, 2);
 	ALLOC_STATE(fg_depth_src, always, 2, 0);
-	r300->hw.fg_depth_src.cmd[0] = cmdpacket0(R300_FG_DEPTH_SRC, 1);
+	r300->hw.fg_depth_src.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_DEPTH_SRC, 1);
 	ALLOC_STATE(rb3d_cctl, always, 2, 0);
-	r300->hw.rb3d_cctl.cmd[0] = cmdpacket0(R300_RB3D_CCTL, 1);
+	r300->hw.rb3d_cctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_CCTL, 1);
 	ALLOC_STATE(bld, always, R300_BLD_CMDSIZE, 0);
-	r300->hw.bld.cmd[R300_BLD_CMD_0] = cmdpacket0(R300_RB3D_CBLEND, 2);
+	r300->hw.bld.cmd[R300_BLD_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_CBLEND, 2);
 	ALLOC_STATE(cmk, always, R300_CMK_CMDSIZE, 0);
-	r300->hw.cmk.cmd[R300_CMK_CMD_0] = cmdpacket0(RB3D_COLOR_CHANNEL_MASK, 1);
+	r300->hw.cmk.cmd[R300_CMK_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, RB3D_COLOR_CHANNEL_MASK, 1);
 	if (is_r500) {
 		ALLOC_STATE(blend_color, always, 3, 0);
-		r300->hw.blend_color.cmd[0] = cmdpacket0(R500_RB3D_CONSTANT_COLOR_AR, 2);
+		r300->hw.blend_color.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_RB3D_CONSTANT_COLOR_AR, 2);
 	} else {
 		ALLOC_STATE(blend_color, always, 2, 0);
-		r300->hw.blend_color.cmd[0] = cmdpacket0(R300_RB3D_BLEND_COLOR, 1);
+		r300->hw.blend_color.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_BLEND_COLOR, 1);
 	}
 	ALLOC_STATE(rop, always, 2, 0);
-	r300->hw.rop.cmd[0] = cmdpacket0(R300_RB3D_ROPCNTL, 1);
+	r300->hw.rop.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_ROPCNTL, 1);
 	ALLOC_STATE(cb, always, R300_CB_CMDSIZE, 0);
-	r300->hw.cb.cmd[R300_CB_CMD_0] = cmdpacket0(R300_RB3D_COLOROFFSET0, 1);
-	r300->hw.cb.cmd[R300_CB_CMD_1] = cmdpacket0(R300_RB3D_COLORPITCH0, 1);
+	r300->hw.cb.emit = &emit_cb_offset;
 	ALLOC_STATE(rb3d_dither_ctl, always, 10, 0);
-	r300->hw.rb3d_dither_ctl.cmd[0] = cmdpacket0(R300_RB3D_DITHER_CTL, 9);
+	r300->hw.rb3d_dither_ctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_DITHER_CTL, 9);
 	ALLOC_STATE(rb3d_aaresolve_ctl, always, 2, 0);
-	r300->hw.rb3d_aaresolve_ctl.cmd[0] = cmdpacket0(R300_RB3D_AARESOLVE_CTL, 1);
-	ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, always, 3, 0);
-	r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[0] = cmdpacket0(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 2);
+	r300->hw.rb3d_aaresolve_ctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_AARESOLVE_CTL, 1);
+    ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, always, 3, 0);
+    r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 2);
+	r300->hw.rb3d_discard_src_pixel_lte_threshold.emit = emit_threshold_misc;
 	ALLOC_STATE(zs, always, R300_ZS_CMDSIZE, 0);
 	r300->hw.zs.cmd[R300_ZS_CMD_0] =
-	    cmdpacket0(R300_ZB_CNTL, 3);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_ZB_CNTL, 3);
+
 	ALLOC_STATE(zstencil_format, always, 5, 0);
 	r300->hw.zstencil_format.cmd[0] =
-	    cmdpacket0(R300_ZB_FORMAT, 4);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_ZB_FORMAT, 4);
+	r300->hw.zstencil_format.emit = emit_zstencil_format;
+
 	ALLOC_STATE(zb, always, R300_ZB_CMDSIZE, 0);
-	r300->hw.zb.cmd[R300_ZB_CMD_0] = cmdpacket0(R300_ZB_DEPTHOFFSET, 2);
+	r300->hw.zb.emit = emit_zb_offset;
 	ALLOC_STATE(zb_depthclearvalue, always, 2, 0);
-	r300->hw.zb_depthclearvalue.cmd[0] = cmdpacket0(R300_ZB_DEPTHCLEARVALUE, 1);
-	ALLOC_STATE(unk4F30, always, 3, 0);
-	r300->hw.unk4F30.cmd[0] = cmdpacket0(0x4F30, 2);
+	r300->hw.zb_depthclearvalue.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_DEPTHCLEARVALUE, 1);
+	ALLOC_STATE(zb_zmask, always, 3, 0);
+	r300->hw.zb_zmask.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_ZMASK_OFFSET, 2);
 	ALLOC_STATE(zb_hiz_offset, always, 2, 0);
-	r300->hw.zb_hiz_offset.cmd[0] = cmdpacket0(R300_ZB_HIZ_OFFSET, 1);
+	r300->hw.zb_hiz_offset.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_HIZ_OFFSET, 1);
 	ALLOC_STATE(zb_hiz_pitch, always, 2, 0);
-	r300->hw.zb_hiz_pitch.cmd[0] = cmdpacket0(R300_ZB_HIZ_PITCH, 1);
+	r300->hw.zb_hiz_pitch.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_HIZ_PITCH, 1);
 
 	/* VPU only on TCL */
 	if (has_tcl) {
    	        int i;
 		ALLOC_STATE(vpi, vpu, R300_VPI_CMDSIZE, 0);
-		r300->hw.vpi.cmd[R300_VPI_CMD_0] =
-		    cmdvpu(R300_PVS_CODE_START, 0);
+		r300->hw.vpi.cmd[0] =
+		    cmdvpu(r300->radeon.radeonScreen, R300_PVS_CODE_START, 0);
+		r300->hw.vpi.emit = emit_vpu;
 
 		if (is_r500) {
 		    ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0);
-		    r300->hw.vpp.cmd[R300_VPP_CMD_0] =
-			cmdvpu(R500_PVS_CONST_START, 0);
+		    r300->hw.vpp.cmd[0] =
+			cmdvpu(r300->radeon.radeonScreen, R500_PVS_CONST_START, 0);
+		    r300->hw.vpp.emit = emit_vpu;
 
 		    ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0);
-		    r300->hw.vps.cmd[R300_VPS_CMD_0] =
-			cmdvpu(R500_POINT_VPORT_SCALE_OFFSET, 1);
+		    r300->hw.vps.cmd[0] =
+			cmdvpu(r300->radeon.radeonScreen, R500_POINT_VPORT_SCALE_OFFSET, 1);
+		    r300->hw.vps.emit = emit_vpu;
 
 			for (i = 0; i < 6; i++) {
-				ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0);
-				r300->hw.vpucp[i].cmd[R300_VPUCP_CMD_0] =
-					cmdvpu(R500_PVS_UCP_START + i, 1);
+			  ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0);
+			  r300->hw.vpucp[i].cmd[0] =
+				  cmdvpu(r300->radeon.radeonScreen,
+                           R500_PVS_UCP_START + i, 1);
+				r300->hw.vpucp[i].emit = emit_vpu;
 			}
 		} else {
 		    ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0);
-		    r300->hw.vpp.cmd[R300_VPP_CMD_0] =
-			cmdvpu(R300_PVS_CONST_START, 0);
+		    r300->hw.vpp.cmd[0] =
+			cmdvpu(r300->radeon.radeonScreen, R300_PVS_CONST_START, 0);
+		    r300->hw.vpp.emit = emit_vpu;
 
 		    ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0);
-		    r300->hw.vps.cmd[R300_VPS_CMD_0] =
-			cmdvpu(R300_POINT_VPORT_SCALE_OFFSET, 1);
+		    r300->hw.vps.cmd[0] =
+			cmdvpu(r300->radeon.radeonScreen, R300_POINT_VPORT_SCALE_OFFSET, 1);
+		    r300->hw.vps.emit = emit_vpu;
 
 			for (i = 0; i < 6; i++) {
 				ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0);
-				r300->hw.vpucp[i].cmd[R300_VPUCP_CMD_0] =
-					cmdvpu(R300_PVS_UCP_START + i, 1);
+				r300->hw.vpucp[i].cmd[0] =
+					cmdvpu(r300->radeon.radeonScreen,
+					       R300_PVS_UCP_START + i, 1);
+				r300->hw.vpucp[i].emit = emit_vpu;
 			}
 		}
 	}
@@ -556,130 +742,37 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	/* Textures */
 	ALLOC_STATE(tex.filter, variable, mtu + 1, 0);
 	r300->hw.tex.filter.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FILTER0_0, 0);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, 0);
 
 	ALLOC_STATE(tex.filter_1, variable, mtu + 1, 0);
 	r300->hw.tex.filter_1.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FILTER1_0, 0);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER1_0, 0);
 
 	ALLOC_STATE(tex.size, variable, mtu + 1, 0);
-	r300->hw.tex.size.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_SIZE_0, 0);
+	r300->hw.tex.size.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_SIZE_0, 0);
 
 	ALLOC_STATE(tex.format, variable, mtu + 1, 0);
 	r300->hw.tex.format.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FORMAT_0, 0);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT_0, 0);
 
 	ALLOC_STATE(tex.pitch, variable, mtu + 1, 0);
-	r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_FORMAT2_0, 0);
+	r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT2_0, 0);
 
-	ALLOC_STATE(tex.offset, variable, mtu + 1, 0);
+	ALLOC_STATE(tex.offset, variable, 1, 0);
 	r300->hw.tex.offset.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_OFFSET_0, 0);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_OFFSET_0, 0);
+	r300->hw.tex.offset.emit = &emit_tex_offsets;
 
 	ALLOC_STATE(tex.chroma_key, variable, mtu + 1, 0);
 	r300->hw.tex.chroma_key.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_CHROMA_KEY_0, 0);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_CHROMA_KEY_0, 0);
 
 	ALLOC_STATE(tex.border_color, variable, mtu + 1, 0);
 	r300->hw.tex.border_color.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_BORDER_COLOR_0, 0);
-
-	r300->hw.is_dirty = GL_TRUE;
-	r300->hw.all_dirty = GL_TRUE;
-
-	/* Initialize command buffer */
-	size =
-	    256 * driQueryOptioni(&r300->radeon.optionCache,
-				  "command_buffer_size");
-	if (size < 2 * r300->hw.max_state_size) {
-		size = 2 * r300->hw.max_state_size + 65535;
-	}
-	if (size > 64 * 256)
-		size = 64 * 256;
-
-	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) {
-		fprintf(stderr, "sizeof(drm_r300_cmd_header_t)=%zd\n",
-			sizeof(drm_r300_cmd_header_t));
-		fprintf(stderr, "sizeof(drm_radeon_cmd_buffer_t)=%zd\n",
-			sizeof(drm_radeon_cmd_buffer_t));
-		fprintf(stderr,
-			"Allocating %d bytes command buffer (max state is %d bytes)\n",
-			size * 4, r300->hw.max_state_size * 4);
-	}
-
-	r300->cmdbuf.size = size;
-	r300->cmdbuf.cmd_buf = (uint32_t *) CALLOC(size * 4);
-	r300->cmdbuf.count_used = 0;
-	r300->cmdbuf.count_reemit = 0;
-}
-
-/**
- * Destroy the command buffer and state atoms.
- */
-void r300DestroyCmdBuf(r300ContextPtr r300)
-{
-	struct r300_state_atom *atom;
-
-	FREE(r300->cmdbuf.cmd_buf);
-
-	foreach(atom, &r300->hw.atomlist) {
-		FREE(atom->cmd);
-	}
-}
-
-void r300EmitBlit(r300ContextPtr rmesa,
-		  GLuint color_fmt,
-		  GLuint src_pitch,
-		  GLuint src_offset,
-		  GLuint dst_pitch,
-		  GLuint dst_offset,
-		  GLint srcx, GLint srcy,
-		  GLint dstx, GLint dsty, GLuint w, GLuint h)
-{
-	drm_r300_cmd_header_t *cmd;
-
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr,
-			"%s src %x/%x %d,%d dst: %x/%x %d,%d sz: %dx%d\n",
-			__FUNCTION__, src_pitch, src_offset, srcx, srcy,
-			dst_pitch, dst_offset, dstx, dsty, w, h);
-
-	assert((src_pitch & 63) == 0);
-	assert((dst_pitch & 63) == 0);
-	assert((src_offset & 1023) == 0);
-	assert((dst_offset & 1023) == 0);
-	assert(w < (1 << 16));
-	assert(h < (1 << 16));
-
-	cmd = (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, 8, __FUNCTION__);
-
-	cmd[0].header.cmd_type = R300_CMD_PACKET3;
-	cmd[0].header.pad0 = R300_CMD_PACKET3_RAW;
-	cmd[1].u = R300_CP_CMD_BITBLT_MULTI | (5 << 16);
-	cmd[2].u = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
-		    RADEON_GMC_DST_PITCH_OFFSET_CNTL |
-		    RADEON_GMC_BRUSH_NONE |
-		    (color_fmt << 8) |
-		    RADEON_GMC_SRC_DATATYPE_COLOR |
-		    RADEON_ROP3_S |
-		    RADEON_DP_SRC_SOURCE_MEMORY |
-		    RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
-
-	cmd[3].u = ((src_pitch / 64) << 22) | (src_offset >> 10);
-	cmd[4].u = ((dst_pitch / 64) << 22) | (dst_offset >> 10);
-	cmd[5].u = (srcx << 16) | srcy;
-	cmd[6].u = (dstx << 16) | dsty;	/* dst */
-	cmd[7].u = (w << 16) | h;
-}
-
-void r300EmitWait(r300ContextPtr rmesa, GLuint flags)
-{
-	drm_r300_cmd_header_t *cmd;
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_BORDER_COLOR_0, 0);
 
-	assert(!(flags & ~(R300_WAIT_2D | R300_WAIT_3D)));
+	r300->radeon.hw.is_dirty = GL_TRUE;
+	r300->radeon.hw.all_dirty = GL_TRUE;
 
-	cmd = (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__);
-	cmd[0].u = 0;
-	cmd[0].wait.cmd_type = R300_CMD_WAIT;
-	cmd[0].wait.flags = flags;
+	rcommonInitCmdBuf(&r300->radeon);
 }
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.h b/src/mesa/drivers/dri/r300/r300_cmdbuf.h
index a8eaa580bd9..53bcc0eeb49 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.h
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.h
@@ -38,79 +38,14 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "r300_context.h"
 
-extern int r300FlushCmdBufLocked(r300ContextPtr r300, const char *caller);
-extern int r300FlushCmdBuf(r300ContextPtr r300, const char *caller);
-
-extern void r300EmitState(r300ContextPtr r300);
-
 extern void r300InitCmdBuf(r300ContextPtr r300);
-extern void r300DestroyCmdBuf(r300ContextPtr r300);
-
-/**
- * Make sure that enough space is available in the command buffer
- * by flushing if necessary.
- *
- * \param dwords The number of dwords we need to be free on the command buffer
- */
-static INLINE void r300EnsureCmdBufSpace(r300ContextPtr r300,
-					     int dwords, const char *caller)
-{
-	assert(dwords < r300->cmdbuf.size);
-
-	if (r300->cmdbuf.count_used + dwords > r300->cmdbuf.size)
-		r300FlushCmdBuf(r300, caller);
-}
-
-/**
- * Allocate the given number of dwords in the command buffer and return
- * a pointer to the allocated area.
- * When necessary, these functions cause a flush. r300AllocCmdBuf() also
- * causes state reemission after a flush. This is necessary to ensure
- * correct hardware state after an unlock.
- */
-static INLINE uint32_t *r300RawAllocCmdBuf(r300ContextPtr r300,
-					       int dwords, const char *caller)
-{
-	uint32_t *ptr;
-
-	r300EnsureCmdBufSpace(r300, dwords, caller);
-
-	ptr = &r300->cmdbuf.cmd_buf[r300->cmdbuf.count_used];
-	r300->cmdbuf.count_used += dwords;
-	return ptr;
-}
-
-static INLINE uint32_t *r300AllocCmdBuf(r300ContextPtr r300,
-					    int dwords, const char *caller)
-{
-	uint32_t *ptr;
-
-	r300EnsureCmdBufSpace(r300, dwords, caller);
-
-	if (!r300->cmdbuf.count_used) {
-		if (RADEON_DEBUG & DEBUG_IOCTL)
-			fprintf(stderr,
-				"Reemit state after flush (from %s)\n", caller);
-		r300EmitState(r300);
-	}
-
-	ptr = &r300->cmdbuf.cmd_buf[r300->cmdbuf.count_used];
-	r300->cmdbuf.count_used += dwords;
-	return ptr;
-}
+void r300_emit_scissor(GLcontext *ctx);
 
-extern void r300EmitBlit(r300ContextPtr rmesa,
-			 GLuint color_fmt,
-			 GLuint src_pitch,
-			 GLuint src_offset,
-			 GLuint dst_pitch,
-			 GLuint dst_offset,
-			 GLint srcx, GLint srcy,
-			 GLint dstx, GLint dsty, GLuint w, GLuint h);
+void emit_vpu(GLcontext *ctx, struct radeon_state_atom * atom);
+int check_vpu(GLcontext *ctx, struct radeon_state_atom *atom);
 
-extern void r300EmitWait(r300ContextPtr rmesa, GLuint flags);
-extern void r300EmitLOAD_VBPNTR(r300ContextPtr rmesa, int start);
-extern void r300EmitVertexShader(r300ContextPtr rmesa);
-extern void r300EmitPixelShader(r300ContextPtr rmesa);
+void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom);
+int check_r500fp(GLcontext *ctx, struct radeon_state_atom *atom);
+int check_r500fp_const(GLcontext *ctx, struct radeon_state_atom *atom);
 
 #endif				/* __R300_CMDBUF_H__ */
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index 8f0effd83e2..9004a3ffab5 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -43,8 +43,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/matrix.h"
 #include "main/extensions.h"
 #include "main/state.h"
-#include "main/texobj.h"
 #include "main/bufferobj.h"
+#include "main/texobj.h"
 
 #include "swrast/swrast.h"
 #include "swrast_setup/swrast_setup.h"
@@ -56,42 +56,40 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "drivers/common/driverfuncs.h"
 
-#include "radeon_ioctl.h"
-#include "radeon_span.h"
 #include "r300_context.h"
+#include "radeon_context.h"
+#include "radeon_span.h"
 #include "r300_cmdbuf.h"
 #include "r300_state.h"
 #include "r300_ioctl.h"
 #include "r300_tex.h"
 #include "r300_emit.h"
+#include "r300_render.h"
 #include "r300_swtcl.h"
+#include "radeon_bocs_wrapper.h"
 
-#ifdef USER_BUFFERS
-#include "r300_mem.h"
-#endif
 
 #include "vblank.h"
 #include "utils.h"
 #include "xmlpool.h"		/* for symbolic values of enum-type options */
 
-/* hw_tcl_on derives from future_hw_tcl_on when its safe to change it. */
-int future_hw_tcl_on = 1;
-int hw_tcl_on = 1;
-
 #define need_GL_VERSION_2_0
 #define need_GL_ARB_point_parameters
 #define need_GL_ARB_vertex_program
 #define need_GL_EXT_blend_equation_separate
 #define need_GL_EXT_blend_func_separate
 #define need_GL_EXT_blend_minmax
+#define need_GL_EXT_framebuffer_object
 #define need_GL_EXT_fog_coord
 #define need_GL_EXT_gpu_program_parameters
 #define need_GL_EXT_secondary_color
 #define need_GL_EXT_stencil_two_side
 #define need_GL_ATI_separate_stencil
 #define need_GL_NV_vertex_program
+
 #include "extension_helper.h"
 
+
 const struct dri_extension card_extensions[] = {
   /* *INDENT-OFF* */
   {"GL_ARB_depth_texture",		NULL},
@@ -112,6 +110,7 @@ const struct dri_extension card_extensions[] = {
   {"GL_EXT_blend_func_separate",	GL_EXT_blend_func_separate_functions},
   {"GL_EXT_blend_minmax",		GL_EXT_blend_minmax_functions},
   {"GL_EXT_blend_subtract",		NULL},
+  {"GL_EXT_packed_depth_stencil",	NULL},
   {"GL_EXT_fog_coord",			GL_EXT_fog_coord_functions },
   {"GL_EXT_gpu_program_parameters",     GL_EXT_gpu_program_parameters_functions},
   {"GL_EXT_secondary_color", 		GL_EXT_secondary_color_functions},
@@ -125,6 +124,7 @@ const struct dri_extension card_extensions[] = {
   {"GL_EXT_texture_lod_bias",		NULL},
   {"GL_EXT_texture_mirror_clamp",	NULL},
   {"GL_EXT_texture_rectangle",		NULL},
+  {"GL_EXT_vertex_array_bgra",		NULL},
   {"GL_ATI_separate_stencil",		GL_ATI_separate_stencil_functions},
   {"GL_ATI_texture_env_combine3",	NULL},
   {"GL_ATI_texture_mirror_once",	NULL},
@@ -139,6 +139,11 @@ const struct dri_extension card_extensions[] = {
 };
 
 
+const struct dri_extension mm_extensions[] = {
+  { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions },
+  { NULL, NULL }
+};
+
 /**
  * The GL 2.0 functions are needed to make display lists work with
  * functions added by GL_ATI_separate_stencil.
@@ -147,16 +152,8 @@ const struct dri_extension gl_20_extension[] = {
   {"GL_VERSION_2_0",			GL_VERSION_2_0_functions },
 };
 
-
-extern struct tnl_pipeline_stage _r300_render_stage;
-extern const struct tnl_pipeline_stage _r300_tcl_stage;
-
 static const struct tnl_pipeline_stage *r300_pipeline[] = {
 
-	/* Try and go straight to t&l
-	 */
-	&_r300_tcl_stage,
-
 	/* Catch any t&l fallbacks
 	 */
 	&_tnl_vertex_transform_stage,
@@ -165,6 +162,7 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = {
 	&_tnl_fog_coordinate_stage,
 	&_tnl_texgen_stage,
 	&_tnl_texture_transform_stage,
+	&_tnl_point_attenuation_stage,
 	&_tnl_vertex_program_stage,
 
 	/* Try again to go to tcl?
@@ -184,6 +182,189 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = {
 	0,
 };
 
+static void r300_get_lock(radeonContextPtr rmesa)
+{
+	drm_radeon_sarea_t *sarea = rmesa->sarea;
+
+	if (sarea->ctx_owner != rmesa->dri.hwContext) {
+		sarea->ctx_owner = rmesa->dri.hwContext;
+		if (!rmesa->radeonScreen->kernel_mm)
+			radeon_bo_legacy_texture_age(rmesa->radeonScreen->bom);
+	}
+}
+
+static void r300_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmesa)
+{
+    /* please flush pipe do all pending work */
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_SC_SCREENDOOR, 1));
+    radeon_cs_write_dword(cs, 0x0);
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_SC_SCREENDOOR, 1));
+    radeon_cs_write_dword(cs, 0x00FFFFFF);
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_SC_HYPERZ, 1));
+    radeon_cs_write_dword(cs, 0x0);
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_US_CONFIG, 1));
+    radeon_cs_write_dword(cs, 0x0);
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_ZB_CNTL, 1));
+    radeon_cs_write_dword(cs, 0x0);
+    radeon_cs_write_dword(cs, cmdwait(rmesa->radeonScreen, R300_WAIT_3D));
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_RB3D_DSTCACHE_CTLSTAT, 1));
+    radeon_cs_write_dword(cs, R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
+    radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen,
+                                  R300_ZB_ZCACHE_CTLSTAT, 1));
+    radeon_cs_write_dword(cs, R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
+    radeon_cs_write_dword(cs, cmdwait(rmesa->radeonScreen,
+                               R300_WAIT_3D | R300_WAIT_3D_CLEAN));
+}
+
+static void r300_vtbl_pre_emit_atoms(radeonContextPtr radeon)
+{
+	r300ContextPtr r300 = (r300ContextPtr)radeon;
+	BATCH_LOCALS(radeon);
+
+	r300->vap_flush_needed = GL_TRUE;
+
+	cp_wait(radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
+	BEGIN_BATCH_NO_AUTOSTATE(2);
+	OUT_BATCH_REGVAL(R300_TX_INVALTAGS, R300_TX_FLUSH);
+	END_BATCH();
+	end_3d(radeon);
+}
+
+static void r300_fallback(GLcontext *ctx, GLuint bit, GLboolean mode)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	if (mode)
+		r300->radeon.Fallback |= bit;
+	else
+		r300->radeon.Fallback &= ~bit;
+}
+
+static void r300_init_vtbl(radeonContextPtr radeon)
+{
+	radeon->vtbl.get_lock = r300_get_lock;
+	radeon->vtbl.update_viewport_offset = r300UpdateViewportOffset;
+	radeon->vtbl.emit_cs_header = r300_vtbl_emit_cs_header;
+	radeon->vtbl.swtcl_flush = r300_swtcl_flush;
+	radeon->vtbl.pre_emit_atoms = r300_vtbl_pre_emit_atoms;
+	radeon->vtbl.fallback = r300_fallback;
+}
+
+static void r300InitConstValues(GLcontext *ctx, radeonScreenPtr screen)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+
+	ctx->Const.MaxTextureImageUnits =
+	    driQueryOptioni(&r300->radeon.optionCache, "texture_image_units");
+	ctx->Const.MaxTextureCoordUnits =
+	    driQueryOptioni(&r300->radeon.optionCache, "texture_coord_units");
+	ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureImageUnits,
+		 ctx->Const.MaxTextureCoordUnits);
+    /* FIXME: When no memory manager is available we should set this 
+     * to some reasonable value based on texture memory pool size */
+    /* FIXME: r5xx limit is 4096 */
+    ctx->Const.MaxTextureLevels = 12;
+	ctx->Const.MaxTextureMaxAnisotropy = 16.0;
+	ctx->Const.MaxTextureLodBias = 16.0;
+
+	if (screen->chip_family >= CHIP_FAMILY_RV515)
+		ctx->Const.MaxTextureLevels = 13;
+	else
+		ctx->Const.MaxTextureLevels = 12;
+
+	ctx->Const.MinPointSize = 1.0;
+	ctx->Const.MinPointSizeAA = 1.0;
+	ctx->Const.MaxPointSize = R300_POINTSIZE_MAX;
+	ctx->Const.MaxPointSizeAA = R300_POINTSIZE_MAX;
+
+	ctx->Const.MinLineWidth = 1.0;
+	ctx->Const.MinLineWidthAA = 1.0;
+	ctx->Const.MaxLineWidth = R300_LINESIZE_MAX;
+	ctx->Const.MaxLineWidthAA = R300_LINESIZE_MAX;
+
+	ctx->Const.MaxDrawBuffers = 1;
+
+	/* currently bogus data */
+	if (r300->options.hw_tcl_enabled) {
+		ctx->Const.VertexProgram.MaxInstructions = VSF_MAX_FRAGMENT_LENGTH / 4;
+		ctx->Const.VertexProgram.MaxNativeInstructions =
+		  VSF_MAX_FRAGMENT_LENGTH / 4;
+		ctx->Const.VertexProgram.MaxNativeAttribs = 16;	/* r420 */
+		ctx->Const.VertexProgram.MaxTemps = 32;
+		ctx->Const.VertexProgram.MaxNativeTemps =
+		  /*VSF_MAX_FRAGMENT_TEMPS */ 32;
+		ctx->Const.VertexProgram.MaxNativeParameters = 256;	/* r420 */
+		ctx->Const.VertexProgram.MaxNativeAddressRegs = 1;
+	}
+
+	if (screen->chip_family >= CHIP_FAMILY_RV515) {
+		ctx->Const.FragmentProgram.MaxNativeTemps = R500_PFS_NUM_TEMP_REGS;
+		ctx->Const.FragmentProgram.MaxNativeAttribs = 11;	/* copy i915... */
+		ctx->Const.FragmentProgram.MaxNativeParameters = R500_PFS_NUM_CONST_REGS;
+		ctx->Const.FragmentProgram.MaxNativeAluInstructions = R500_PFS_MAX_INST;
+		ctx->Const.FragmentProgram.MaxNativeTexInstructions = R500_PFS_MAX_INST;
+		ctx->Const.FragmentProgram.MaxNativeInstructions = R500_PFS_MAX_INST;
+		ctx->Const.FragmentProgram.MaxNativeTexIndirections = R500_PFS_MAX_INST;
+		ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;
+	} else {
+		ctx->Const.FragmentProgram.MaxNativeTemps = R300_PFS_NUM_TEMP_REGS;
+		ctx->Const.FragmentProgram.MaxNativeAttribs = 11;	/* copy i915... */
+		ctx->Const.FragmentProgram.MaxNativeParameters = R300_PFS_NUM_CONST_REGS;
+		ctx->Const.FragmentProgram.MaxNativeAluInstructions = R300_PFS_MAX_ALU_INST;
+		ctx->Const.FragmentProgram.MaxNativeTexInstructions = R300_PFS_MAX_TEX_INST;
+		ctx->Const.FragmentProgram.MaxNativeInstructions = R300_PFS_MAX_ALU_INST + R300_PFS_MAX_TEX_INST;
+		ctx->Const.FragmentProgram.MaxNativeTexIndirections = R300_PFS_MAX_TEX_INDIRECT;
+		ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;
+	}
+}
+
+static void r300ParseOptions(r300ContextPtr r300, radeonScreenPtr screen)
+{
+	struct r300_options options = { 0 };
+
+	driParseConfigFiles(&r300->radeon.optionCache, &screen->optionCache,
+			    screen->driScreen->myNum, "r300");
+
+	r300->radeon.initialMaxAnisotropy = driQueryOptionf(&r300->radeon.optionCache, "def_max_anisotropy");
+
+	options.stencil_two_side_disabled = driQueryOptionb(&r300->radeon.optionCache, "disable_stencil_two_side");
+	options.s3tc_force_enabled = driQueryOptionb(&r300->radeon.optionCache, "force_s3tc_enable");
+	options.s3tc_force_disabled = driQueryOptionb(&r300->radeon.optionCache, "disable_s3tc");
+
+	if (!(screen->chip_flags & RADEON_CHIPSET_TCL) || driQueryOptioni(&r300->radeon.optionCache, "tcl_mode") == DRI_CONF_TCL_SW)
+		options.hw_tcl_enabled = 0;
+	else
+		options.hw_tcl_enabled = 1;
+
+	options.conformance_mode = !driQueryOptionb(&r300->radeon.optionCache, "disable_lowimpact_fallback");
+
+	r300->options = options;
+}
+
+static void r300InitGLExtensions(GLcontext *ctx)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+
+	driInitExtensions(ctx, card_extensions, GL_TRUE);
+	if (r300->radeon.radeonScreen->kernel_mm)
+		driInitExtensions(ctx, mm_extensions, GL_FALSE);
+
+	if (r300->options.stencil_two_side_disabled)
+		_mesa_disable_extension(ctx, "GL_EXT_stencil_two_side");
+
+	if (ctx->Mesa_DXTn && !r300->options.s3tc_force_enabled) {
+		_mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
+		_mesa_enable_extension(ctx, "GL_S3_s3tc");
+	} else if (r300->options.s3tc_force_disabled) {
+		_mesa_disable_extension(ctx, "GL_EXT_texture_compression_s3tc");
+	}
+}
+
 /* Create the device specific rendering context.
  */
 GLboolean r300CreateContext(const __GLcontextModes * glVisual,
@@ -195,42 +376,25 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	struct dd_function_table functions;
 	r300ContextPtr r300;
 	GLcontext *ctx;
-	int tcl_mode, i;
 
 	assert(glVisual);
 	assert(driContextPriv);
 	assert(screen);
 
-	/* Allocate the R300 context */
 	r300 = (r300ContextPtr) CALLOC(sizeof(*r300));
 	if (!r300)
 		return GL_FALSE;
 
-	if (!(screen->chip_flags & RADEON_CHIPSET_TCL))
-		hw_tcl_on = future_hw_tcl_on = 0;
+	r300ParseOptions(r300, screen);
 
-	/* Parse configuration files.
-	 * Do this here so that initialMaxAnisotropy is set before we create
-	 * the default textures.
-	 */
-	driParseConfigFiles(&r300->radeon.optionCache, &screen->optionCache,
-			    screen->driScreen->myNum, "r300");
-	r300->initialMaxAnisotropy = driQueryOptionf(&r300->radeon.optionCache,
-						     "def_max_anisotropy");
+	r300_init_vtbl(&r300->radeon);
 
-	/* Init default driver functions then plug in our R300-specific functions
-	 * (the texture functions are especially important)
-	 */
 	_mesa_init_driver_functions(&functions);
 	r300InitIoctlFuncs(&functions);
 	r300InitStateFuncs(&functions);
 	r300InitTextureFuncs(&functions);
 	r300InitShaderFuncs(&functions);
 
-#ifdef USER_BUFFERS
-	r300_mem_init(r300);
-#endif
-
 	if (!radeonInitContext(&r300->radeon, &functions,
 			       glVisual, driContextPriv,
 			       sharedContextPrivate)) {
@@ -238,94 +402,15 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 		return GL_FALSE;
 	}
 
-	/* Init r300 context data */
-	r300->dma.buf0_address =
-	    r300->radeon.radeonScreen->buffers->list[0].address;
-
-	(void)memset(r300->texture_heaps, 0, sizeof(r300->texture_heaps));
-	make_empty_list(&r300->swapped);
-
-	r300->nr_heaps = 1 /* screen->numTexHeaps */ ;
-	assert(r300->nr_heaps < RADEON_NR_TEX_HEAPS);
-	for (i = 0; i < r300->nr_heaps; i++) {
-		/* *INDENT-OFF* */
-		r300->texture_heaps[i] = driCreateTextureHeap(i, r300,
-							       screen->
-							       texSize[i], 12,
-							       RADEON_NR_TEX_REGIONS,
-							       (drmTextureRegionPtr)
-							       r300->radeon.sarea->
-							       tex_list[i],
-							       &r300->radeon.sarea->
-							       tex_age[i],
-							       &r300->swapped,
-							       sizeof
-							       (r300TexObj),
-							       (destroy_texture_object_t
-								*)
-							       r300DestroyTexObj);
-		/* *INDENT-ON* */
-	}
-	r300->texture_depth = driQueryOptioni(&r300->radeon.optionCache,
-					      "texture_depth");
-	if (r300->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
-		r300->texture_depth = (screen->cpp == 4) ?
-		    DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16;
-
-	/* Set the maximum texture size small enough that we can guarentee that
-	 * all texture units can bind a maximal texture and have them both in
-	 * texturable memory at once.
-	 */
-
 	ctx = r300->radeon.glCtx;
 
-	ctx->Const.MaxTextureImageUnits =
-	    driQueryOptioni(&r300->radeon.optionCache, "texture_image_units");
-	ctx->Const.MaxTextureCoordUnits =
-	    driQueryOptioni(&r300->radeon.optionCache, "texture_coord_units");
-	ctx->Const.MaxTextureUnits =
-	    MIN2(ctx->Const.MaxTextureImageUnits,
-		 ctx->Const.MaxTextureCoordUnits);
-	ctx->Const.MaxTextureMaxAnisotropy = 16.0;
-	ctx->Const.MaxTextureLodBias = 16.0;
-
-	if (screen->chip_family >= CHIP_FAMILY_RV515)
-	    ctx->Const.MaxTextureLevels = 13;
-	else
-	    ctx->Const.MaxTextureLevels = 12;
-
-        driCalculateMaxTextureLevels( r300->texture_heaps,
-                                      r300->nr_heaps,
-                                      & ctx->Const,
-                                      4,
-                                      ctx->Const.MaxTextureLevels - 1,
-                                      MIN2(ctx->Const.MaxTextureLevels,
-                                           MAX_3D_TEXTURE_LEVELS) - 1,
-                                      ctx->Const.MaxTextureLevels - 1,
-                                      ctx->Const.MaxTextureLevels - 1,
-                                      ctx->Const.MaxTextureLevels - 1,
-                                      GL_FALSE,
-                                      2 );
+	r300->fallback = 0;
+	if (r300->options.hw_tcl_enabled)
+		ctx->VertexProgram._MaintainTnlProgram = GL_TRUE;
 
-	ctx->Const.MinPointSize = 1.0;
-	ctx->Const.MinPointSizeAA = 1.0;
-	ctx->Const.MaxPointSize = R300_POINTSIZE_MAX;
-	ctx->Const.MaxPointSizeAA = R300_POINTSIZE_MAX;
-
-	ctx->Const.MinLineWidth = 1.0;
-	ctx->Const.MinLineWidthAA = 1.0;
-	ctx->Const.MaxLineWidth = R300_LINESIZE_MAX;
-	ctx->Const.MaxLineWidthAA = R300_LINESIZE_MAX;
-
-#ifdef USER_BUFFERS
-	/* Needs further modifications */
-#if 0
-	ctx->Const.MaxArrayLockSize =
-	    ( /*512 */ RADEON_BUFFER_SIZE * 16 * 1024) / (4 * 4);
-#endif
-#endif
+	ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE;
 
-	ctx->Const.MaxDrawBuffers = 1;
+	r300InitConstValues(ctx, screen);
 
 	/* Initialize the software rasterizer and helper modules.
 	 */
@@ -334,16 +419,12 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	_tnl_CreateContext(ctx);
 	_swsetup_CreateContext(ctx);
 	_swsetup_Wakeup(ctx);
-	_ae_create_context(ctx);
 
 	/* Install the customized pipeline:
 	 */
 	_tnl_destroy_pipeline(ctx);
 	_tnl_install_pipeline(ctx, r300_pipeline);
-
-	/* Try and keep materials and vertices separate:
-	 */
-/* 	_tnl_isolate_materials(ctx, GL_TRUE); */
+	TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
 
 	/* Configure swrast and TNL to match hardware characteristics:
 	 */
@@ -352,226 +433,25 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	_tnl_allow_pixel_fog(ctx, GL_FALSE);
 	_tnl_allow_vertex_fog(ctx, GL_TRUE);
 
-	/* currently bogus data */
-	if (screen->chip_flags & RADEON_CHIPSET_TCL) {
-	        ctx->Const.VertexProgram.MaxInstructions = VSF_MAX_FRAGMENT_LENGTH / 4;
-		ctx->Const.VertexProgram.MaxNativeInstructions =
-		  VSF_MAX_FRAGMENT_LENGTH / 4;
-		ctx->Const.VertexProgram.MaxNativeAttribs = 16;	/* r420 */
-		ctx->Const.VertexProgram.MaxTemps = 32;
-		ctx->Const.VertexProgram.MaxNativeTemps =
-		  /*VSF_MAX_FRAGMENT_TEMPS */ 32;
-		ctx->Const.VertexProgram.MaxNativeParameters = 256;	/* r420 */
-		ctx->Const.VertexProgram.MaxNativeAddressRegs = 1;
+	if (r300->options.hw_tcl_enabled) {
+		r300InitDraw(ctx);
+	} else {
+		r300InitSwtcl(ctx);
 	}
 
-	ctx->Const.FragmentProgram.MaxNativeTemps = PFS_NUM_TEMP_REGS;
-	ctx->Const.FragmentProgram.MaxNativeAttribs = 11;	/* copy i915... */
-	ctx->Const.FragmentProgram.MaxNativeParameters = PFS_NUM_CONST_REGS;
-	ctx->Const.FragmentProgram.MaxNativeAluInstructions = PFS_MAX_ALU_INST;
-	ctx->Const.FragmentProgram.MaxNativeTexInstructions = PFS_MAX_TEX_INST;
-	ctx->Const.FragmentProgram.MaxNativeInstructions =
-	    PFS_MAX_ALU_INST + PFS_MAX_TEX_INST;
-	ctx->Const.FragmentProgram.MaxNativeTexIndirections =
-	    PFS_MAX_TEX_INDIRECT;
-	ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0;	/* and these are?? */
-	ctx->VertexProgram._MaintainTnlProgram = GL_TRUE;
-	ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE;
-
-	driInitExtensions(ctx, card_extensions, GL_TRUE);
-
-	if (driQueryOptionb
-	    (&r300->radeon.optionCache, "disable_stencil_two_side"))
-		_mesa_disable_extension(ctx, "GL_EXT_stencil_two_side");
-
-	if (r300->radeon.glCtx->Mesa_DXTn
-	    && !driQueryOptionb(&r300->radeon.optionCache, "disable_s3tc")) {
-		_mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
-		_mesa_enable_extension(ctx, "GL_S3_s3tc");
-	} else
-	    if (driQueryOptionb(&r300->radeon.optionCache, "force_s3tc_enable"))
-	{
-		_mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
-	}
-
-	r300->disable_lowimpact_fallback =
-	    driQueryOptionb(&r300->radeon.optionCache,
-			    "disable_lowimpact_fallback");
-
-	radeonInitSpanFuncs(ctx);
+	radeon_fbo_init(&r300->radeon);
+	radeonInitSpanFuncs( ctx );
 	r300InitCmdBuf(r300);
 	r300InitState(r300);
-	if (!(screen->chip_flags & RADEON_CHIPSET_TCL))
-	        r300InitSwtcl(ctx);
-
-	TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
+	r300InitShaderFunctions(r300);
 
-	tcl_mode = driQueryOptioni(&r300->radeon.optionCache, "tcl_mode");
-	if (driQueryOptionb(&r300->radeon.optionCache, "no_rast")) {
-		fprintf(stderr, "disabling 3D acceleration\n");
-#if R200_MERGED
-		FALLBACK(&r300->radeon, RADEON_FALLBACK_DISABLE, 1);
-#endif
-	}
-	if (tcl_mode == DRI_CONF_TCL_SW ||
-	    !(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
-		if (r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
-			r300->radeon.radeonScreen->chip_flags &=
-			    ~RADEON_CHIPSET_TCL;
-			fprintf(stderr, "Disabling HW TCL support\n");
-		}
-		TCL_FALLBACK(r300->radeon.glCtx,
-			     RADEON_TCL_FALLBACK_TCL_DISABLE, 1);
+	if (screen->chip_family == CHIP_FAMILY_RS600 ||	screen->chip_family == CHIP_FAMILY_RS690 ||
+		screen->chip_family == CHIP_FAMILY_RS740) {
+		r300->radeon.texture_row_align = 64;
 	}
 
-	return GL_TRUE;
-}
-
-static void r300FreeGartAllocations(r300ContextPtr r300)
-{
-	int i, ret, tries = 0, done_age, in_use = 0;
-	drm_radeon_mem_free_t memfree;
+	r300InitGLExtensions(ctx);
 
-	memfree.region = RADEON_MEM_REGION_GART;
-
-#ifdef USER_BUFFERS
-	for (i = r300->rmm->u_last; i > 0; i--) {
-		if (r300->rmm->u_list[i].ptr == NULL) {
-			continue;
-		}
-
-		/* check whether this buffer is still in use */
-		if (r300->rmm->u_list[i].pending) {
-			in_use++;
-		}
-	}
-	/* Cannot flush/lock if no context exists. */
-	if (in_use)
-		r300FlushCmdBuf(r300, __FUNCTION__);
-
-	done_age = radeonGetAge((radeonContextPtr) r300);
-
-	for (i = r300->rmm->u_last; i > 0; i--) {
-		if (r300->rmm->u_list[i].ptr == NULL) {
-			continue;
-		}
-
-		/* check whether this buffer is still in use */
-		if (!r300->rmm->u_list[i].pending) {
-			continue;
-		}
-
-		assert(r300->rmm->u_list[i].h_pending == 0);
-
-		tries = 0;
-		while (r300->rmm->u_list[i].age > done_age && tries++ < 1000) {
-			usleep(10);
-			done_age = radeonGetAge((radeonContextPtr) r300);
-		}
-		if (tries >= 1000) {
-			WARN_ONCE("Failed to idle region!");
-		}
-
-		memfree.region_offset = (char *)r300->rmm->u_list[i].ptr -
-		    (char *)r300->radeon.radeonScreen->gartTextures.map;
-
-		ret = drmCommandWrite(r300->radeon.radeonScreen->driScreen->fd,
-				      DRM_RADEON_FREE, &memfree,
-				      sizeof(memfree));
-		if (ret) {
-			fprintf(stderr, "Failed to free at %p\nret = %s\n",
-				r300->rmm->u_list[i].ptr, strerror(-ret));
-		} else {
-			if (i == r300->rmm->u_last)
-				r300->rmm->u_last--;
-
-			r300->rmm->u_list[i].pending = 0;
-			r300->rmm->u_list[i].ptr = NULL;
-		}
-	}
-	r300->rmm->u_head = i;
-#endif				/* USER_BUFFERS */
+	return GL_TRUE;
 }
 
-/* Destroy the device specific context.
- */
-void r300DestroyContext(__DRIcontextPrivate * driContextPriv)
-{
-	GET_CURRENT_CONTEXT(ctx);
-	r300ContextPtr r300 = (r300ContextPtr) driContextPriv->driverPrivate;
-	radeonContextPtr radeon = (radeonContextPtr) r300;
-	radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL;
-	int i;
-
-	if (RADEON_DEBUG & DEBUG_DRI) {
-		fprintf(stderr, "Destroying context !\n");
-	}
-
-	/* check if we're deleting the currently bound context */
-	if (&r300->radeon == current) {
-		radeonFlush(r300->radeon.glCtx);
-		_mesa_make_current(NULL, NULL, NULL);
-	}
-
-	/* Free r300 context resources */
-	assert(r300);		/* should never be null */
-
-	if (r300) {
-		GLboolean release_texture_heaps;
-
-		release_texture_heaps =
-		    (r300->radeon.glCtx->Shared->RefCount == 1);
-		_swsetup_DestroyContext(r300->radeon.glCtx);
-		_tnl_DestroyContext(r300->radeon.glCtx);
-		_vbo_DestroyContext(r300->radeon.glCtx);
-		_swrast_DestroyContext(r300->radeon.glCtx);
-
-		if (r300->dma.current.buf) {
-			r300ReleaseDmaRegion(r300, &r300->dma.current,
-					     __FUNCTION__);
-#ifndef USER_BUFFERS
-			r300FlushCmdBuf(r300, __FUNCTION__);
-#endif
-		}
-		r300FreeGartAllocations(r300);
-		r300DestroyCmdBuf(r300);
-
-		if (radeon->state.scissor.pClipRects) {
-			FREE(radeon->state.scissor.pClipRects);
-			radeon->state.scissor.pClipRects = NULL;
-		}
-
-		if (release_texture_heaps) {
-			/* This share group is about to go away, free our private
-			 * texture object data.
-			 */
-			int i;
-
-			for (i = 0; i < r300->nr_heaps; i++) {
-				driDestroyTextureHeap(r300->texture_heaps[i]);
-				r300->texture_heaps[i] = NULL;
-			}
-
-			assert(is_empty_list(&r300->swapped));
-		}
-
-                /* Drop texture object references from current hardware state */
-		for (i = 0; i < 8; i++) {
-			_mesa_reference_texobj(&r300->state.texture.unit[i].texobj, NULL);
-		}
-
-		radeonCleanupContext(&r300->radeon);
-
-#ifdef USER_BUFFERS
-		/* the memory manager might be accessed when Mesa frees the shared
-		 * state, so don't destroy it earlier
-		 */
-		r300_mem_destroy(r300);
-#endif
-
-		/* free the option cache */
-		driDestroyOptionCache(&r300->radeon.optionCache);
-
-		FREE(r300);
-	}
-}
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index 96a3205f1a3..026c33c67cb 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -37,26 +37,20 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifndef __R300_CONTEXT_H__
 #define __R300_CONTEXT_H__
 
-#include "tnl/t_vertex.h"
 #include "drm.h"
 #include "radeon_drm.h"
 #include "dri_util.h"
-#include "texmem.h"
+#include "radeon_common.h"
 
-#include "main/macros.h"
 #include "main/mtypes.h"
-#include "main/colormac.h"
-
-#define USER_BUFFERS
+#include "shader/prog_instruction.h"
 
 struct r300_context;
 typedef struct r300_context r300ContextRec;
 typedef struct r300_context *r300ContextPtr;
 
-#include "radeon_lock.h"
-#include "main/mm.h"
 
-/* From http://gcc.gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html .
+/* From http://gcc. gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html .
    I suppose we could inline this and use macro to fetch out __LINE__ and stuff in case we run into trouble
    with other compilers ... GLUE!
 */
@@ -73,180 +67,14 @@ typedef struct r300_context *r300ContextPtr;
 	}
 
 #include "r300_vertprog.h"
-#include "r500_fragprog.h"
-
-/**
- * This function takes a float and packs it into a uint32_t
- */
-static INLINE uint32_t r300PackFloat32(float fl)
-{
-	union {
-		float fl;
-		uint32_t u;
-	} u;
-
-	u.fl = fl;
-	return u.u;
-}
-
-/* This is probably wrong for some values, I need to test this
- * some more.  Range checking would be a good idea also..
- *
- * But it works for most things.  I'll fix it later if someone
- * else with a better clue doesn't
- */
-static INLINE uint32_t r300PackFloat24(float f)
-{
-	float mantissa;
-	int exponent;
-	uint32_t float24 = 0;
-
-	if (f == 0.0)
-		return 0;
-
-	mantissa = frexpf(f, &exponent);
-
-	/* Handle -ve */
-	if (mantissa < 0) {
-		float24 |= (1 << 23);
-		mantissa = mantissa * -1.0;
-	}
-	/* Handle exponent, bias of 63 */
-	exponent += 62;
-	float24 |= (exponent << 16);
-	/* Kill 7 LSB of mantissa */
-	float24 |= (r300PackFloat32(mantissa) & 0x7FFFFF) >> 7;
-
-	return float24;
-}
-
-/************ DMA BUFFERS **************/
-
-/* Need refcounting on dma buffers:
- */
-struct r300_dma_buffer {
-	int refcount;		/**< the number of retained regions in buf */
-	drmBufPtr buf;
-	int id;
-};
-#undef GET_START
-#ifdef USER_BUFFERS
-#define GET_START(rvb) (r300GartOffsetFromVirtual(rmesa, (rvb)->address+(rvb)->start))
-#else
-#define GET_START(rvb) (rmesa->radeon.radeonScreen->gart_buffer_offset +		\
-			(rvb)->address - rmesa->dma.buf0_address +	\
-			(rvb)->start)
-#endif
-/* A retained region, eg vertices for indexed vertices.
- */
-struct r300_dma_region {
-	struct r300_dma_buffer *buf;
-	char *address;		/* == buf->address */
-	int start, end, ptr;	/* offsets from start of buf */
-
-	int aos_offset;		/* address in GART memory */
-	int aos_stride;		/* distance between elements, in dwords */
-	int aos_size;		/* number of components (1-4) */
-};
-
-struct r300_dma {
-	/* Active dma region.  Allocations for vertices and retained
-	 * regions come from here.  Also used for emitting random vertices,
-	 * these may be flushed by calling flush_current();
-	 */
-	struct r300_dma_region current;
 
-	void (*flush) (r300ContextPtr);
-
-	char *buf0_address;	/* start of buf[0], for index calcs */
-
-	/* Number of "in-flight" DMA buffers, i.e. the number of buffers
-	 * for which a DISCARD command is currently queued in the command buffer.
-	 */
-	GLuint nr_released_bufs;
-};
-
-       /* Texture related */
-
-typedef struct r300_tex_obj r300TexObj, *r300TexObjPtr;
-
-/* Maximum number of mipmap levels supported by any supported GPU
- */
-#define R300_MAX_TEXTURE_LEVELS 13
-
-/* Texture object in locally shared texture space.
- */
-struct r300_tex_obj {
-	driTextureObject base;
-
-	GLuint bufAddr;		/* Offset to start of locally
-				   shared texture block */
-
-	drm_radeon_tex_image_t image[6][R300_MAX_TEXTURE_LEVELS];
-	/* Six, for the cube faces */
-
-	GLboolean image_override;	/* Image overridden by GLX_EXT_tfp */
-
-	GLuint pitch;		/* this isn't sent to hardware just used in calculations */
-	/* hardware register values */
-	/* Note that R200 has 8 registers per texture and R300 only 7 */
-	GLuint filter;
-	GLuint filter_1;
-	GLuint pitch_reg;
-	GLuint size;		/* npot only */
-	GLuint format;
-	GLuint offset;		/* Image location in the card's address space.
-				   All cube faces follow. */
-	GLuint unknown4;
-	GLuint unknown5;
-	/* end hardware registers */
-
-	/* registers computed by r200 code - keep them here to
-	   compare against what is actually written.
-
-	   to be removed later.. */
-	GLuint pp_border_color;
-	GLuint pp_cubic_faces;	/* cube face 1,2,3,4 log2 sizes */
-	GLuint format_x;
-
-	GLboolean border_fallback;
-
-	GLuint tile_bits;	/* hw texture tile bits used on this texture */
-};
-
-struct r300_texture_env_state {
-	struct gl_texture_object *texobj;
-	GLenum format;
-	GLenum envMode;
-};
 
 /* The blit width for texture uploads
  */
 #define R300_BLIT_WIDTH_BYTES 1024
 #define R300_MAX_TEXTURE_UNITS 8
 
-struct r300_texture_state {
-	struct r300_texture_env_state unit[R300_MAX_TEXTURE_UNITS];
-	int tc_count;		/* number of incoming texture coordinates from VAP */
-};
 
-/**
- * A block of hardware state.
- *
- * When check returns non-zero, the returned number of dwords must be
- * copied verbatim into the command buffer in order to update a state atom
- * when it is dirty.
- */
-struct r300_state_atom {
-	struct r300_state_atom *next, *prev;
-	const char *name;	/* for debug */
-	int cmd_size;		/* maximum size in dwords */
-	GLuint idx;		/* index in an array (e.g. textures) */
-	uint32_t *cmd;
-	GLboolean dirty;
-
-	int (*check) (r300ContextPtr, struct r300_state_atom * atom);
-};
 
 #define R300_VPT_CMD_0		0
 #define R300_VPT_XSCALE		1
@@ -288,9 +116,11 @@ struct r300_state_atom {
 #define R300_GB_MISC_MSPOS_0		1
 #define R300_GB_MISC_MSPOS_1		2
 #define R300_GB_MISC_TILE_CONFIG	3
-#define R300_GB_MISC_SELECT		4
-#define R300_GB_MISC_AA_CONFIG		5
-#define R300_GB_MISC_CMDSIZE		6
+#define R300_GB_MISC_CMDSIZE		4
+#define R300_GB_MISC2_CMD_0		    0
+#define R300_GB_MISC2_SELECT		1
+#define R300_GB_MISC2_AA_CONFIG		2
+#define R300_GB_MISC2_CMDSIZE		3
 
 #define R300_TXE_CMD_0		0
 #define R300_TXE_ENABLE		1
@@ -463,124 +293,100 @@ struct r300_state_atom {
  * Cache for hardware register state.
  */
 struct r300_hw_state {
-	struct r300_state_atom atomlist;
-
-	GLboolean is_dirty;
-	GLboolean all_dirty;
-	int max_state_size;	/* in dwords */
-
-	struct r300_state_atom vpt;	/* viewport (1D98) */
-	struct r300_state_atom vap_cntl;
-        struct r300_state_atom vap_index_offset; /* 0x208c r5xx only */
-	struct r300_state_atom vof;	/* VAP output format register 0x2090 */
-	struct r300_state_atom vte;	/* (20B0) */
-	struct r300_state_atom vap_vf_max_vtx_indx;	/* Maximum Vertex Indx Clamp (2134) */
-	struct r300_state_atom vap_cntl_status;
-	struct r300_state_atom vir[2];	/* vap input route (2150/21E0) */
-	struct r300_state_atom vic;	/* vap input control (2180) */
-	struct r300_state_atom vap_psc_sgn_norm_cntl; /* Programmable Stream Control Signed Normalize Control (21DC) */
-	struct r300_state_atom vap_clip_cntl;
-	struct r300_state_atom vap_clip;
-	struct r300_state_atom vap_pvs_vtx_timeout_reg;	/* Vertex timeout register (2288) */
-	struct r300_state_atom pvs;	/* pvs_cntl (22D0) */
-	struct r300_state_atom gb_enable;	/* (4008) */
-	struct r300_state_atom gb_misc;	/* Multisampling position shifts ? (4010) */
-	struct r300_state_atom ga_point_s0;	/* S Texture Coordinate of Vertex 0 for Point texture stuffing (LLC) (4200) */
-	struct r300_state_atom ga_triangle_stipple;	/* (4214) */
-	struct r300_state_atom ps;	/* pointsize (421C) */
-	struct r300_state_atom ga_point_minmax;	/* (4230) */
-	struct r300_state_atom lcntl;	/* line control */
-	struct r300_state_atom ga_line_stipple;	/* (4260) */
-	struct r300_state_atom shade;
-	struct r300_state_atom polygon_mode;
-	struct r300_state_atom fogp;	/* fog parameters (4294) */
-	struct r300_state_atom ga_soft_reset;	/* (429C) */
-	struct r300_state_atom zbias_cntl;
-	struct r300_state_atom zbs;	/* zbias (42A4) */
-	struct r300_state_atom occlusion_cntl;
-	struct r300_state_atom cul;	/* cull cntl (42B8) */
-	struct r300_state_atom su_depth_scale;	/* (42C0) */
-	struct r300_state_atom rc;	/* rs control (4300) */
-	struct r300_state_atom ri;	/* rs interpolators (4310) */
-	struct r300_state_atom rr;	/* rs route (4330) */
-	struct r300_state_atom sc_hyperz;	/* (43A4) */
-	struct r300_state_atom sc_screendoor;	/* (43E8) */
-	struct r300_state_atom fp;	/* fragment program cntl + nodes (4600) */
-	struct r300_state_atom fpt;	/* texi - (4620) */
-	struct r300_state_atom us_out_fmt;	/* (46A4) */
-	struct r300_state_atom r500fp;	/* r500 fp instructions */
-	struct r300_state_atom r500fp_const;	/* r500 fp constants */
-	struct r300_state_atom fpi[4];	/* fp instructions (46C0/47C0/48C0/49C0) */
-	struct r300_state_atom fogs;	/* fog state (4BC0) */
-	struct r300_state_atom fogc;	/* fog color (4BC8) */
-	struct r300_state_atom at;	/* alpha test (4BD4) */
-	struct r300_state_atom fg_depth_src;	/* (4BD8) */
-	struct r300_state_atom fpp;	/* 0x4C00 and following */
-	struct r300_state_atom rb3d_cctl;	/* (4E00) */
-	struct r300_state_atom bld;	/* blending (4E04) */
-	struct r300_state_atom cmk;	/* colormask (4E0C) */
-	struct r300_state_atom blend_color;	/* constant blend color */
-	struct r300_state_atom rop;	/* ropcntl */
-	struct r300_state_atom cb;	/* colorbuffer (4E28) */
-	struct r300_state_atom rb3d_dither_ctl;	/* (4E50) */
-	struct r300_state_atom rb3d_aaresolve_ctl;	/* (4E88) */
-	struct r300_state_atom rb3d_discard_src_pixel_lte_threshold;	/* (4E88) I saw it only written on RV350 hardware..  */
-	struct r300_state_atom zs;	/* zstencil control (4F00) */
-	struct r300_state_atom zstencil_format;
-	struct r300_state_atom zb;	/* z buffer (4F20) */
-	struct r300_state_atom zb_depthclearvalue;	/* (4F28) */
-	struct r300_state_atom unk4F30;	/* (4F30) */
-	struct r300_state_atom zb_hiz_offset;	/* (4F44) */
-	struct r300_state_atom zb_hiz_pitch;	/* (4F54) */
-
-	struct r300_state_atom vpi;	/* vp instructions */
-	struct r300_state_atom vpp;	/* vp parameters */
-	struct r300_state_atom vps;	/* vertex point size (?) */
-	struct r300_state_atom vpucp[6];	/* vp user clip plane - 6 */
+	struct radeon_state_atom vpt;	/* viewport (1D98) */
+	struct radeon_state_atom vap_cntl;
+	struct radeon_state_atom vap_index_offset; /* 0x208c r5xx only */
+	struct radeon_state_atom vof;	/* VAP output format register 0x2090 */
+	struct radeon_state_atom vte;	/* (20B0) */
+	struct radeon_state_atom vap_vf_max_vtx_indx;	/* Maximum Vertex Indx Clamp (2134) */
+	struct radeon_state_atom vap_cntl_status;
+	struct radeon_state_atom vir[2];	/* vap input route (2150/21E0) */
+	struct radeon_state_atom vic;	/* vap input control (2180) */
+	struct radeon_state_atom vap_psc_sgn_norm_cntl; /* Programmable Stream Control Signed Normalize Control (21DC) */
+	struct radeon_state_atom vap_clip_cntl;
+	struct radeon_state_atom vap_clip;
+	struct radeon_state_atom vap_pvs_vtx_timeout_reg;	/* Vertex timeout register (2288) */
+	struct radeon_state_atom pvs;	/* pvs_cntl (22D0) */
+	struct radeon_state_atom gb_enable;	/* (4008) */
+	struct radeon_state_atom gb_misc;	/* Multisampling position shifts ? (4010) */
+	struct radeon_state_atom gb_misc2;	/* Multisampling position shifts ? (4010) */
+	struct radeon_state_atom ga_point_s0;	/* S Texture Coordinate of Vertex 0 for Point texture stuffing (LLC) (4200) */
+	struct radeon_state_atom ga_triangle_stipple;	/* (4214) */
+	struct radeon_state_atom ps;	/* pointsize (421C) */
+	struct radeon_state_atom ga_point_minmax;	/* (4230) */
+	struct radeon_state_atom lcntl;	/* line control */
+	struct radeon_state_atom ga_line_stipple;	/* (4260) */
+	struct radeon_state_atom shade;
+	struct radeon_state_atom shade2;
+	struct radeon_state_atom polygon_mode;
+	struct radeon_state_atom fogp;	/* fog parameters (4294) */
+	struct radeon_state_atom ga_soft_reset;	/* (429C) */
+	struct radeon_state_atom zbias_cntl;
+	struct radeon_state_atom zbs;	/* zbias (42A4) */
+	struct radeon_state_atom occlusion_cntl;
+	struct radeon_state_atom cul;	/* cull cntl (42B8) */
+	struct radeon_state_atom su_depth_scale;	/* (42C0) */
+	struct radeon_state_atom rc;	/* rs control (4300) */
+	struct radeon_state_atom ri;	/* rs interpolators (4310) */
+	struct radeon_state_atom rr;	/* rs route (4330) */
+	struct radeon_state_atom sc_hyperz;	/* (43A4) */
+	struct radeon_state_atom sc_screendoor;	/* (43E8) */
+	struct radeon_state_atom fp;	/* fragment program cntl + nodes (4600) */
+	struct radeon_state_atom fpt;	/* texi - (4620) */
+	struct radeon_state_atom us_out_fmt;	/* (46A4) */
+	struct radeon_state_atom r500fp;	/* r500 fp instructions */
+	struct radeon_state_atom r500fp_const;	/* r500 fp constants */
+	struct radeon_state_atom fpi[4];	/* fp instructions (46C0/47C0/48C0/49C0) */
+	struct radeon_state_atom fogs;	/* fog state (4BC0) */
+	struct radeon_state_atom fogc;	/* fog color (4BC8) */
+	struct radeon_state_atom at;	/* alpha test (4BD4) */
+	struct radeon_state_atom fg_depth_src;	/* (4BD8) */
+	struct radeon_state_atom fpp;	/* 0x4C00 and following */
+	struct radeon_state_atom rb3d_cctl;	/* (4E00) */
+	struct radeon_state_atom bld;	/* blending (4E04) */
+	struct radeon_state_atom cmk;	/* colormask (4E0C) */
+	struct radeon_state_atom blend_color;	/* constant blend color */
+	struct radeon_state_atom rop;	/* ropcntl */
+	struct radeon_state_atom cb;	/* colorbuffer (4E28) */
+	struct radeon_state_atom rb3d_dither_ctl;	/* (4E50) */
+	struct radeon_state_atom rb3d_aaresolve_ctl;	/* (4E88) */
+	struct radeon_state_atom rb3d_discard_src_pixel_lte_threshold;	/* (4E88) I saw it only written on RV350 hardware..  */
+	struct radeon_state_atom zs;	/* zstencil control (4F00) */
+	struct radeon_state_atom zstencil_format;
+	struct radeon_state_atom zb;	/* z buffer (4F20) */
+	struct radeon_state_atom zb_depthclearvalue;	/* (4F28) */
+	struct radeon_state_atom zb_zmask;	/* (4F30) */
+	struct radeon_state_atom zb_hiz_offset;	/* (4F44) */
+	struct radeon_state_atom zb_hiz_pitch;	/* (4F54) */
+
+	struct radeon_state_atom vpi;	/* vp instructions */
+	struct radeon_state_atom vpp;	/* vp parameters */
+	struct radeon_state_atom vps;	/* vertex point size (?) */
+	struct radeon_state_atom vpucp[6];	/* vp user clip plane - 6 */
 	/* 8 texture units */
 	/* the state is grouped by function and not by
 	   texture unit. This makes single unit updates
 	   really awkward - we are much better off
 	   updating the whole thing at once */
 	struct {
-		struct r300_state_atom filter;
-		struct r300_state_atom filter_1;
-		struct r300_state_atom size;
-		struct r300_state_atom format;
-		struct r300_state_atom pitch;
-		struct r300_state_atom offset;
-		struct r300_state_atom chroma_key;
-		struct r300_state_atom border_color;
+		struct radeon_state_atom filter;
+		struct radeon_state_atom filter_1;
+		struct radeon_state_atom size;
+		struct radeon_state_atom format;
+		struct radeon_state_atom pitch;
+		struct radeon_state_atom offset;
+		struct radeon_state_atom chroma_key;
+		struct radeon_state_atom border_color;
 	} tex;
-	struct r300_state_atom txe;	/* tex enable (4104) */
-};
+	struct radeon_state_atom txe;	/* tex enable (4104) */
 
-/**
- * This structure holds the command buffer while it is being constructed.
- *
- * The first batch of commands in the buffer is always the state that needs
- * to be re-emitted when the context is lost. This batch can be skipped
- * otherwise.
- */
-struct r300_cmdbuf {
-	int size;		/* DWORDs allocated for buffer */
-	uint32_t *cmd_buf;
-	int count_used;		/* DWORDs filled so far */
-	int count_reemit;	/* size of re-emission batch */
+	radeonTexObj *textures[R300_MAX_TEXTURE_UNITS];
 };
 
 /**
  * State cache
  */
 
-struct r300_depthbuffer_state {
-	GLfloat scale;
-};
-
-struct r300_stencilbuffer_state {
-	GLboolean hw_stencil;
-};
-
 /* Vertex shader state */
 
 /* Perhaps more if we store programs in vmem? */
@@ -593,73 +399,55 @@ struct r300_stencilbuffer_state {
 #define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0)
 #define STATE_R300_TEXRECT_FACTOR (STATE_INTERNAL_DRIVER+1)
 
-struct r300_vertex_shader_fragment {
-	int length;
-	union {
-		GLuint d[VSF_MAX_FRAGMENT_LENGTH];
-		float f[VSF_MAX_FRAGMENT_LENGTH];
-		GLuint i[VSF_MAX_FRAGMENT_LENGTH];
-	} body;
-};
-
-struct r300_vertex_shader_state {
-	struct r300_vertex_shader_fragment program;
-};
-
-extern int hw_tcl_on;
-
 #define COLOR_IS_RGBA
 #define TAG(x) r300##x
 #include "tnl_dd/t_dd_vertex.h"
 #undef TAG
 
-//#define CURRENT_VERTEX_SHADER(ctx) (ctx->VertexProgram._Current)
-#define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->selected_vp)
-
-/* Should but doesnt work */
-//#define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->curr_vp)
-
-/* r300_vertex_shader_state and r300_vertex_program should probably be merged together someday.
- * Keeping them them seperate for now should ensure fixed pipeline keeps functioning properly.
- */
-
-struct r300_vertex_program_key {
-	GLuint InputsRead;
-	GLuint OutputsWritten;
-	GLuint OutputsAdded;
-};
-
 struct r300_vertex_program {
 	struct r300_vertex_program *next;
-	struct r300_vertex_program_key key;
-	int translated;
 
-	struct r300_vertex_shader_fragment program;
+	struct r300_vertex_program_key {
+		GLuint InputsRead;
+		GLuint OutputsWritten;
+		GLuint OutputsAdded;
+	} key;
+	
+	struct r300_vertex_shader_hw_code {
+		int length;
+		union {
+			GLuint d[VSF_MAX_FRAGMENT_LENGTH];
+			float f[VSF_MAX_FRAGMENT_LENGTH];
+		} body;
+	} hw_code;
+
+	GLboolean translated;
+	GLboolean error;
 
 	int pos_end;
 	int num_temporaries;	/* Number of temp vars used by program */
 	int wpos_idx;
 	int inputs[VERT_ATTRIB_MAX];
 	int outputs[VERT_RESULT_MAX];
-	int native;
-	int ref_count;
-	int use_ref_count;
 };
 
 struct r300_vertex_program_cont {
 	struct gl_vertex_program mesa_program;	/* Must be first */
-	struct r300_vertex_shader_fragment params;
 	struct r300_vertex_program *progs;
 };
 
-#define PFS_MAX_ALU_INST	64
-#define PFS_MAX_TEX_INST	64
-#define PFS_MAX_TEX_INDIRECT 4
-#define PFS_NUM_TEMP_REGS	32
-#define PFS_NUM_CONST_REGS	16
+#define R300_PFS_MAX_ALU_INST	64
+#define R300_PFS_MAX_TEX_INST	32
+#define R300_PFS_MAX_TEX_INDIRECT 4
+#define R300_PFS_NUM_TEMP_REGS	32
+#define R300_PFS_NUM_CONST_REGS	32
 
-struct r300_pfs_compile_state;
+#define R500_PFS_MAX_INST 512
+#define R500_PFS_NUM_TEMP_REGS 128
+#define R500_PFS_NUM_CONST_REGS 256
 
+struct r300_pfs_compile_state;
+struct r500_pfs_compile_state;
 
 /**
  * Stores state that influences the compilation of a fragment program.
@@ -702,7 +490,7 @@ struct r300_fragment_program_node {
 struct r300_fragment_program_code {
 	struct {
 		int length; /**< total # of texture instructions used */
-		GLuint inst[PFS_MAX_TEX_INST];
+		GLuint inst[R300_PFS_MAX_TEX_INST];
 	} tex;
 
 	struct {
@@ -712,7 +500,7 @@ struct r300_fragment_program_code {
 			GLuint inst1;
 			GLuint inst2;
 			GLuint inst3;
-		} inst[PFS_MAX_ALU_INST];
+		} inst[R300_PFS_MAX_ALU_INST];
 	} alu;
 
 	struct r300_fragment_program_node node[4];
@@ -723,53 +511,12 @@ struct r300_fragment_program_code {
 	 * Remember which program register a given hardware constant
 	 * belongs to.
 	 */
-	struct prog_src_register constant[PFS_NUM_CONST_REGS];
+	struct prog_src_register constant[R300_PFS_NUM_CONST_REGS];
 	int const_nr;
 
 	int max_temp_idx;
 };
 
-/**
- * Store everything about a fragment program that is needed
- * to render with that program.
- */
-struct r300_fragment_program {
-	struct gl_fragment_program mesa_program;
-
-	GLboolean translated;
-	GLboolean error;
-
-	struct r300_fragment_program_external_state state;
-	struct r300_fragment_program_code code;
-
-	GLboolean WritesDepth;
-	GLuint optimization;
-};
-
-struct r500_pfs_compile_state;
-
-struct r500_fragment_program_external_state {
-	struct {
-		/**
-		 * If the sampler is used as a shadow sampler,
-		 * this field is:
-		 *  0 - GL_LUMINANCE
-		 *  1 - GL_INTENSITY
-		 *  2 - GL_ALPHA
-		 * depending on the depth texture mode.
-		 */
-		GLuint depth_texture_mode : 2;
-
-		/**
-		 * If the sampler is used as a shadow sampler,
-		 * this field is (texture_compare_func - GL_NEVER).
-		 * [e.g. if compare function is GL_LEQUAL, this field is 3]
-		 *
-		 * Otherwise, this field is 0.
-		 */
-		GLuint texture_compare_func : 3;
-	} unit[16];
-};
 
 struct r500_fragment_program_code {
 	struct {
@@ -779,7 +526,7 @@ struct r500_fragment_program_code {
 		GLuint inst3;
 		GLuint inst4;
 		GLuint inst5;
-	} inst[512];
+	} inst[R500_PFS_MAX_INST];
 
 	int inst_offset;
 	int inst_end;
@@ -788,94 +535,46 @@ struct r500_fragment_program_code {
 	 * Remember which program register a given hardware constant
 	 * belongs to.
 	 */
-	struct prog_src_register constant[PFS_NUM_CONST_REGS];
+	struct prog_src_register constant[R500_PFS_NUM_CONST_REGS];
 	int const_nr;
 
 	int max_temp_idx;
 };
 
-struct r500_fragment_program {
-	struct gl_fragment_program mesa_program;
+/**
+* Store everything about a fragment program that is needed
+* to render with that program.
+*/
+struct r300_fragment_program {
+	struct gl_fragment_program Base;
 
-	GLcontext *ctx;
 	GLboolean translated;
 	GLboolean error;
 
-	struct r500_fragment_program_external_state state;
-	struct r500_fragment_program_code code;
+	struct r300_fragment_program_external_state state;
+	union rX00_fragment_program_code {
+		struct r300_fragment_program_code r300;
+		struct r500_fragment_program_code r500;
+	} code;
 
 	GLboolean writes_depth;
-
 	GLuint optimization;
 };
 
-#define R300_MAX_AOS_ARRAYS		16
-
-#define REG_COORDS	0
-#define REG_COLOR0	1
-#define REG_TEX0	2
-
-struct r300_state {
-	struct r300_depthbuffer_state depth;
-	struct r300_texture_state texture;
-	int sw_tcl_inputs[VERT_ATTRIB_MAX];
-	struct r300_vertex_shader_state vertex_shader;
-	struct r300_dma_region aos[R300_MAX_AOS_ARRAYS];
-	int aos_count;
-
-	GLuint *Elts;
-	struct r300_dma_region elt_dma;
-
-	struct r300_dma_region swtcl_dma;
-	DECLARE_RENDERINPUTS(render_inputs_bitset);	/* actual render inputs that R300 was configured for.
-							   They are the same as tnl->render_inputs for fixed pipeline */
-
-	struct r300_stencilbuffer_state stencil;
-
+struct r300_fragment_program_compiler {
+	r300ContextPtr r300;
+	struct r300_fragment_program *fp;
+	union rX00_fragment_program_code *code;
+	struct gl_program *program;
 };
 
-#define R300_FALLBACK_NONE 0
-#define R300_FALLBACK_TCL 1
-#define R300_FALLBACK_RAST 2
+#define R300_MAX_AOS_ARRAYS		16
+
 
 /* r300_swtcl.c
  */
 struct r300_swtcl_info {
-   GLuint RenderIndex;
-
-   /**
-    * Size of a hardware vertex.  This is calculated when \c ::vertex_attrs is
-    * installed in the Mesa state vector.
-    */
-   GLuint vertex_size;
-
-   /**
-    * Attributes instructing the Mesa TCL pipeline where / how to put vertex
-    * data in the hardware buffer.
-    */
-   struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
-
-   /**
-    * Number of elements of \c ::vertex_attrs that are actually used.
-    */
-   GLuint vertex_attr_count;
-
-   /**
-    * Cached pointer to the buffer where Mesa will store vertex data.
-    */
-   GLubyte *verts;
-
-   /* Fallback rasterization functions
-    */
-  //   r200_point_func draw_point;
-  //   r200_line_func draw_line;
-  //   r200_tri_func draw_tri;
-
-   GLuint hw_primitive;
-   GLenum render_primitive;
-   GLuint numverts;
-
-   /**
+  /*
     * Offset of the 4UB color data within a hardware (swtcl) vertex.
     */
    GLuint coloroffset;
@@ -884,15 +583,44 @@ struct r300_swtcl_info {
     * Offset of the 3UB specular color data within a hardware (swtcl) vertex.
     */
    GLuint specoffset;
+};
 
-   /**
-    * Should Mesa project vertex data or will the hardware do it?
-    */
-   GLboolean needproj;
+struct r300_vtable {
+	void (* SetupRSUnit)(GLcontext *ctx);
+	void (* SetupFragmentShaderTextures)(GLcontext *ctx, int *tmu_mappings);
+	GLboolean (* BuildFragmentProgramHwCode)(struct r300_fragment_program_compiler *compiler);
+	void (* FragmentProgramDump)(union rX00_fragment_program_code *code);
+	void (* SetupPixelShader)(GLcontext *ctx);
+};
 
-   struct r300_dma_region indexed_verts;
+struct r300_vertex_buffer {
+	struct vertex_attribute {
+		/* generic */
+		GLubyte element;
+		GLvoid *data;
+		GLboolean free_needed;
+		GLuint stride;
+		GLuint dwords;
+		GLubyte size; /* number of components */
+
+		/* hw specific */
+		uint32_t data_type:4;
+		uint32_t dst_loc:5;
+		uint32_t _signed:1;
+		uint32_t normalize:1;
+		uint32_t swizzle:12;
+		uint32_t write_mask:4;
+	} attribs[VERT_ATTRIB_MAX];
+
+	GLubyte num_attribs;
 };
 
+struct r300_index_buffer {
+	GLvoid *ptr;
+	GLboolean is_32bit;
+	GLboolean free_needed;
+	GLuint count;
+};
 
 /**
  * \brief R300 context structure.
@@ -900,46 +628,33 @@ struct r300_swtcl_info {
 struct r300_context {
 	struct radeon_context radeon;	/* parent class, must be first */
 
+	struct r300_vtable vtbl;
+
 	struct r300_hw_state hw;
-	struct r300_cmdbuf cmdbuf;
-	struct r300_state state;
-	struct gl_vertex_program *curr_vp;
+
 	struct r300_vertex_program *selected_vp;
 
 	/* Vertex buffers
 	 */
-	struct r300_dma dma;
-	GLboolean save_on_next_unlock;
-	GLuint NewGLState;
-
-	/* Texture object bookkeeping
-	 */
-	unsigned nr_heaps;
-	driTexHeap *texture_heaps[RADEON_NR_TEX_HEAPS];
-	driTextureObject swapped;
-	int texture_depth;
-	float initialMaxAnisotropy;
-
-	/* Clientdata textures;
-	 */
-	GLuint prefer_gart_client_texturing;
-
-#ifdef USER_BUFFERS
-	struct r300_memory_manager *rmm;
-#endif
-
 	GLvector4f dummy_attrib[_TNL_ATTRIB_MAX];
 	GLvector4f *temp_attrib[_TNL_ATTRIB_MAX];
 
-	GLboolean disable_lowimpact_fallback;
-
-	DECLARE_RENDERINPUTS(tnl_index_bitset);	/* index of bits for last tnl_install_attrs */
+	struct r300_options {
+		uint32_t conformance_mode:1;
+		uint32_t hw_tcl_enabled:1;
+		uint32_t s3tc_force_enabled:1;
+		uint32_t s3tc_force_disabled:1;
+		uint32_t stencil_two_side_disabled:1;
+	} options;
+	
 	struct r300_swtcl_info swtcl;
-};
+	struct r300_vertex_buffer vbuf;
+	struct r300_index_buffer ind_buf;
+	GLboolean vap_flush_needed;
+
+	uint32_t fallback;
 
-struct r300_buffer_object {
-	struct gl_buffer_object mesa_obj;
-	int id;
+	DECLARE_RENDERINPUTS(render_inputs_bitset);
 };
 
 #define R300_CONTEXT(ctx)		((r300ContextPtr)(ctx->DriverCtx))
@@ -955,9 +670,11 @@ extern int r300VertexProgUpdateParams(GLcontext * ctx,
 				      struct r300_vertex_program_cont *vp,
 				      float *dst);
 
-#define RADEON_D_CAPTURE 0
-#define RADEON_D_PLAYBACK 1
-#define RADEON_D_PLAYBACK_RAW 2
-#define RADEON_D_T 3
+extern void r300InitShaderFunctions(r300ContextPtr r300);
+
+extern void r300InitDraw(GLcontext *ctx);
+
+#define r300PackFloat32 radeonPackFloat32
+#define r300PackFloat24 radeonPackFloat24
 
 #endif				/* __R300_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
new file mode 100644
index 00000000000..cc5650fb7c8
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -0,0 +1,475 @@
+/**************************************************************************
+ *
+ * Copyright 2009 Maciej Cencora
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHOR(S) AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include <stdlib.h>
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/state.h"
+#include "main/api_validate.h"
+#include "main/enums.h"
+
+#include "r300_reg.h"
+#include "r300_context.h"
+#include "r300_emit.h"
+#include "r300_render.h"
+#include "r300_state.h"
+#include "r300_tex.h"
+
+#include "tnl/tnl.h"
+#include "tnl/t_vp_build.h"
+#include "vbo/vbo_context.h"
+#include "swrast/swrast.h"
+#include "swrast_setup/swrast_setup.h"
+
+static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf, struct gl_buffer_object **bo, GLuint *nr_bo)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_index_buffer *ind_buf = &r300->ind_buf;
+	GLvoid *src_ptr;
+
+	if (!mesa_ind_buf) {
+		ind_buf->ptr = NULL;
+		return;
+	}
+
+	ind_buf->count = mesa_ind_buf->count;
+	if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) {
+		bo[*nr_bo] = mesa_ind_buf->obj;
+		(*nr_bo)++;
+		ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj);
+		assert(mesa_ind_buf->obj->Pointer != NULL);
+	}
+	src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr);
+
+	if (mesa_ind_buf->type == GL_UNSIGNED_BYTE) {
+		GLubyte *in = (GLubyte *)src_ptr;
+		GLuint *out = _mesa_malloc(sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1));
+		int i;
+
+		ind_buf->ptr = out;
+
+		for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
+			*out++ = in[i] | in[i + 1] << 16;
+		}
+
+		if (i < mesa_ind_buf->count) {
+			*out++ = in[i];
+		}
+
+		ind_buf->free_needed = GL_TRUE;
+		ind_buf->is_32bit = GL_FALSE;
+	} else if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) {
+#if MESA_BIG_ENDIAN
+		GLushort *in = (GLushort *)src_ptr;
+		GLuint *out = _mesa_malloc(sizeof(GLushort) *
+					   ((mesa_ind_buf->count + 1) & ~1));
+		int i;
+
+		ind_buf->ptr = out;
+
+		for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
+			*out++ = in[i] | in[i + 1] << 16;
+		}
+
+		if (i < mesa_ind_buf->count) {
+			*out++ = in[i];
+		}
+
+		ind_buf->free_needed = GL_TRUE;
+#else
+		ind_buf->ptr = src_ptr;
+		ind_buf->free_needed = GL_FALSE;
+#endif
+		ind_buf->is_32bit = GL_FALSE;
+	} else {
+		ind_buf->ptr = src_ptr;
+		ind_buf->free_needed = GL_FALSE;
+		ind_buf->is_32bit = GL_TRUE;
+	}
+}
+
+static int getTypeSize(GLenum type)
+{
+	switch (type) {
+		case GL_DOUBLE:
+			return sizeof(GLdouble);
+		case GL_FLOAT:
+			return sizeof(GLfloat);
+		case GL_INT:
+			return sizeof(GLint);
+		case GL_UNSIGNED_INT:
+			return sizeof(GLuint);
+		case GL_SHORT:
+			return sizeof(GLshort);
+		case GL_UNSIGNED_SHORT:
+			return sizeof(GLushort);
+		case GL_BYTE:
+			return sizeof(GLbyte);
+		case GL_UNSIGNED_BYTE:
+			return sizeof(GLubyte);
+		default:
+			assert(0);
+			return 0;
+	}
+}
+
+#define CONVERT( TYPE, MACRO ) do {		\
+	GLuint i, j, sz;				\
+	sz = input->Size;				\
+	if (input->Normalized) {			\
+		for (i = 0; i < count; i++) {		\
+			const TYPE *in = (TYPE *)src_ptr;		\
+			for (j = 0; j < sz; j++) {		\
+				*dst_ptr++ = MACRO(*in);		\
+				in++;				\
+			}					\
+			src_ptr += stride;			\
+		}						\
+	} else {					\
+		for (i = 0; i < count; i++) {		\
+			const TYPE *in = (TYPE *)src_ptr;		\
+			for (j = 0; j < sz; j++) {		\
+				*dst_ptr++ = (GLfloat)(*in);		\
+				in++;				\
+			}					\
+			src_ptr += stride;			\
+		}						\
+	}						\
+} while (0)
+
+static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const struct gl_client_array *input, struct gl_buffer_object **bo, GLuint *nr_bo)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_vertex_buffer *vbuf = &r300->vbuf;
+	struct vertex_attribute r300_attr;
+	const void *src_ptr;
+	GLenum type;
+	GLuint stride;
+
+	if (input->BufferObj->Name) {
+		if (!input->BufferObj->Pointer) {
+			bo[*nr_bo] = input->BufferObj;
+			(*nr_bo)++;
+			ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj);
+			assert(input->BufferObj->Pointer != NULL);
+		}
+
+		src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr);
+	} else
+		src_ptr = input->Ptr;
+
+	stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB;
+
+	if (input->Type == GL_DOUBLE || input->Type == GL_UNSIGNED_INT || input->Type == GL_INT ||
+#if MESA_BIG_ENDIAN
+	    getTypeSize(input->Type) != 4 ||
+#endif
+	    stride < 4) {
+		if (RADEON_DEBUG & DEBUG_FALLBACKS) {
+			fprintf(stderr, "%s: Converting vertex attributes, attribute data format %x,", __FUNCTION__, input->Type);
+			fprintf(stderr, "stride %d, components %d\n", stride, input->Size);
+		}
+
+		GLfloat *dst_ptr, *tmp;
+		tmp = dst_ptr = _mesa_malloc(sizeof(GLfloat) * input->Size * count);
+
+		switch (input->Type) {
+			case GL_DOUBLE:
+				CONVERT(GLdouble, (GLfloat));
+				break;
+			case GL_UNSIGNED_INT:
+				CONVERT(GLuint, UINT_TO_FLOAT);
+				break;
+			case GL_INT:
+				CONVERT(GLint, INT_TO_FLOAT);
+				break;
+			case GL_UNSIGNED_SHORT:
+				CONVERT(GLushort, USHORT_TO_FLOAT);
+				break;
+			case GL_SHORT:
+				CONVERT(GLshort, SHORT_TO_FLOAT);
+				break;
+			case GL_UNSIGNED_BYTE:
+				assert(input->Format != GL_BGRA);
+				CONVERT(GLubyte, UBYTE_TO_FLOAT);
+				break;
+			case GL_BYTE:
+				CONVERT(GLbyte, BYTE_TO_FLOAT);
+				break;
+			default:
+				assert(0);
+				break;
+		}
+
+		type = GL_FLOAT;
+		r300_attr.free_needed = GL_TRUE;
+		r300_attr.data = tmp;
+		r300_attr.stride = sizeof(GLfloat) * input->Size;
+		r300_attr.dwords = input->Size;
+	} else {
+		type = input->Type;
+		r300_attr.free_needed = GL_FALSE;
+		r300_attr.data = (GLvoid *)src_ptr;
+		r300_attr.stride = stride;
+		r300_attr.dwords = (getTypeSize(type) * input->Size  + 3)/ 4;
+	}
+
+	r300_attr.size = input->Size;
+	r300_attr.element = attr;
+	r300_attr.dst_loc = vbuf->num_attribs;
+
+	switch (type) {
+		case GL_FLOAT:
+			switch (input->Size) {
+				case 1: r300_attr.data_type = R300_DATA_TYPE_FLOAT_1; break;
+				case 2: r300_attr.data_type = R300_DATA_TYPE_FLOAT_2; break;
+				case 3: r300_attr.data_type = R300_DATA_TYPE_FLOAT_3; break;
+				case 4: r300_attr.data_type = R300_DATA_TYPE_FLOAT_4; break;
+			}
+			r300_attr._signed = 0;
+			r300_attr.normalize = 0;
+			break;
+		case GL_SHORT:
+			r300_attr._signed = 1;
+			r300_attr.normalize = input->Normalized;
+			switch (input->Size) {
+				case 1:
+				case 2:
+					r300_attr.data_type = R300_DATA_TYPE_SHORT_2;
+					break;
+				case 3:
+				case 4:
+					r300_attr.data_type = R300_DATA_TYPE_SHORT_4;
+					break;
+			}
+			break;
+		case GL_BYTE:
+			r300_attr._signed = 1;
+			r300_attr.normalize = input->Normalized;
+			r300_attr.data_type = R300_DATA_TYPE_BYTE;
+			break;
+		case GL_UNSIGNED_SHORT:
+			r300_attr._signed = 0;
+			r300_attr.normalize = input->Normalized;
+			switch (input->Size) {
+				case 1:
+				case 2:
+					r300_attr.data_type = R300_DATA_TYPE_SHORT_2;
+					break;
+				case 3:
+				case 4:
+					r300_attr.data_type = R300_DATA_TYPE_SHORT_4;
+					break;
+			}
+			break;
+		case GL_UNSIGNED_BYTE:
+			r300_attr._signed = 0;
+			r300_attr.normalize = input->Normalized;
+			if (input->Format == GL_BGRA)
+				r300_attr.data_type = R300_DATA_TYPE_D3DCOLOR;
+			else
+				r300_attr.data_type = R300_DATA_TYPE_BYTE;
+			break;
+
+		default:
+		case GL_DOUBLE:
+		case GL_INT:
+		case GL_UNSIGNED_INT:
+			assert(0);
+			break;
+	}
+
+	switch (input->Size) {
+		case 4:
+			r300_attr.swizzle = SWIZZLE_XYZW;
+			break;
+		case 3:
+			r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
+			break;
+		case 2:
+			r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ONE);
+			break;
+		case 1:
+			r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE);
+			break;
+	}
+
+	r300_attr.write_mask = MASK_XYZW;
+
+	vbuf->attribs[vbuf->num_attribs] = r300_attr;
+	++vbuf->num_attribs;
+}
+
+static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], int count, struct gl_buffer_object **bo, GLuint *nr_bo)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_vertex_buffer *vbuf = &r300->vbuf;
+
+	{
+		int i, tmp;
+
+		tmp = r300->selected_vp->key.InputsRead;
+		i = 0;
+		vbuf->num_attribs = 0;
+		while (tmp) {
+			/* find first enabled bit */
+			while (!(tmp & 1)) {
+				tmp >>= 1;
+				++i;
+			}
+
+			r300TranslateAttrib(ctx, i, count, arrays[i], bo, nr_bo);
+
+			tmp >>= 1;
+			++i;
+		}
+	}
+
+	r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS);
+	if (r300->fallback)
+		return;
+
+	{
+		int i;
+
+		for (i = 0; i < vbuf->num_attribs; i++) {
+			rcommon_emit_vector(ctx, &r300->radeon.tcl.aos[i],
+						vbuf->attribs[i].data, vbuf->attribs[i].dwords,
+						vbuf->attribs[i].stride, count);
+		}
+
+		r300->radeon.tcl.aos_count = vbuf->num_attribs;
+	}
+}
+
+static void r300FreeData(GLcontext *ctx, struct gl_buffer_object **bo, GLuint nr_bo)
+{
+	{
+		struct r300_vertex_buffer *vbuf = &R300_CONTEXT(ctx)->vbuf;
+		int i;
+
+		for (i = 0; i < vbuf->num_attribs; i++) {
+			if (vbuf->attribs[i].free_needed)
+				_mesa_free(vbuf->attribs[i].data);
+		}
+	}
+
+	{
+		struct r300_index_buffer *ind_buf = &R300_CONTEXT(ctx)->ind_buf;
+		if (ind_buf->free_needed)
+			_mesa_free(ind_buf->ptr);
+	}
+
+	{
+		int i;
+
+		for (i = 0; i < nr_bo; ++i) {
+			ctx->Driver.UnmapBuffer(ctx, 0, bo[i]);
+		}
+	}
+}
+
+static GLboolean r300TryDrawPrims(GLcontext *ctx,
+					 const struct gl_client_array *arrays[],
+					 const struct _mesa_prim *prim,
+					 GLuint nr_prims,
+					 const struct _mesa_index_buffer *ib,
+					 GLuint min_index,
+					 GLuint max_index )
+{
+	struct r300_context *r300 = R300_CONTEXT(ctx);
+	struct gl_buffer_object *bo[VERT_ATTRIB_MAX+1];
+	GLuint i, nr_bo = 0;
+
+	if (ctx->NewState)
+		_mesa_update_state( ctx );
+
+	if (r300->options.hw_tcl_enabled)
+		_tnl_UpdateFixedFunctionProgram(ctx);
+
+	r300UpdateShaders(r300);
+
+	r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx));
+
+	r300FixupIndexBuffer(ctx, ib, bo, &nr_bo);
+
+	r300SetVertexFormat(ctx, arrays, max_index + 1, bo, &nr_bo);
+
+	if (r300->fallback)
+		return GL_FALSE;
+
+	r300SetupVAP(ctx, r300->selected_vp->key.InputsRead, r300->selected_vp->key.OutputsWritten);
+
+	r300UpdateShaderStates(r300);
+
+	r300EmitCacheFlush(r300);
+	radeonEmitState(&r300->radeon);
+
+	for (i = 0; i < nr_prims; ++i) {
+		r300RunRenderPrimitive(ctx, prim[i].start, prim[i].start + prim[i].count, prim[i].mode);
+	}
+
+	r300EmitCacheFlush(r300);
+
+	radeonReleaseArrays(ctx, ~0);
+
+	r300FreeData(ctx, bo, nr_bo);
+
+	return GL_TRUE;
+}
+
+/* TODO: rebase if number of indices in any of primitives is > 8192 for 32bit indices or 16384 for 16bit indices */
+
+static void r300DrawPrims(GLcontext *ctx,
+			 const struct gl_client_array *arrays[],
+			 const struct _mesa_prim *prim,
+			 GLuint nr_prims,
+			 const struct _mesa_index_buffer *ib,
+			 GLuint min_index,
+			 GLuint max_index)
+{
+	GLboolean retval;
+
+	if (min_index) {
+		vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims );
+		return;
+	}
+
+	/* Make an attempt at drawing */
+	retval = r300TryDrawPrims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+
+	/* If failed run tnl pipeline - it should take care of fallbacks */
+	if (!retval)
+		_tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+}
+
+void r300InitDraw(GLcontext *ctx)
+{
+	struct vbo_context *vbo = vbo_context(ctx);
+
+	vbo->draw_prims = r300DrawPrims;
+}
diff --git a/src/mesa/drivers/dri/r300/r300_emit.c b/src/mesa/drivers/dri/r300/r300_emit.c
index 28c3157427c..c3817721dc4 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.c
+++ b/src/mesa/drivers/dri/r300/r300_emit.c
@@ -31,6 +31,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * \file
  *
  * \author Keith Whitwell <[email protected]>
+ * \author Maciej Cencora <[email protected]>
  */
 
 #include "main/glheader.h"
@@ -46,222 +47,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/t_context.h"
 
 #include "r300_context.h"
-#include "radeon_ioctl.h"
 #include "r300_state.h"
 #include "r300_emit.h"
 #include "r300_ioctl.h"
-
-#ifdef USER_BUFFERS
-#include "r300_mem.h"
-#endif
-
-#if SWIZZLE_X != R300_INPUT_ROUTE_SELECT_X || \
-    SWIZZLE_Y != R300_INPUT_ROUTE_SELECT_Y || \
-    SWIZZLE_Z != R300_INPUT_ROUTE_SELECT_Z || \
-    SWIZZLE_W != R300_INPUT_ROUTE_SELECT_W || \
-    SWIZZLE_ZERO != R300_INPUT_ROUTE_SELECT_ZERO || \
-    SWIZZLE_ONE != R300_INPUT_ROUTE_SELECT_ONE
-#error Cannot change these!
-#endif
-
-#define DEBUG_ALL DEBUG_VERTS
-
-#if defined(USE_X86_ASM)
-#define COPY_DWORDS( dst, src, nr )					\
-do {									\
-	int __tmp;							\
-	__asm__ __volatile__( "rep ; movsl"				\
-			      : "=%c" (__tmp), "=D" (dst), "=S" (__tmp)	\
-			      : "0" (nr),				\
-			        "D" ((long)dst),			\
-			        "S" ((long)src) );			\
-} while (0)
-#else
-#define COPY_DWORDS( dst, src, nr )		\
-do {						\
-   int j;					\
-   for ( j = 0 ; j < nr ; j++ )			\
-      dst[j] = ((int *)src)[j];			\
-   dst += nr;					\
-} while (0)
-#endif
-
-static void r300EmitVec4(GLcontext * ctx, struct r300_dma_region *rvb,
-			 GLvoid * data, int stride, int count)
-{
-	int i;
-	int *out = (int *)(rvb->address + rvb->start);
-
-	if (RADEON_DEBUG & DEBUG_VERTS)
-		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
-			__FUNCTION__, count, stride, (void *)out, (void *)data);
-
-	if (stride == 4)
-		COPY_DWORDS(out, data, count);
-	else
-		for (i = 0; i < count; i++) {
-			out[0] = *(int *)data;
-			out++;
-			data += stride;
-		}
-}
-
-static void r300EmitVec8(GLcontext * ctx, struct r300_dma_region *rvb,
-			 GLvoid * data, int stride, int count)
-{
-	int i;
-	int *out = (int *)(rvb->address + rvb->start);
-
-	if (RADEON_DEBUG & DEBUG_VERTS)
-		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
-			__FUNCTION__, count, stride, (void *)out, (void *)data);
-
-	if (stride == 8)
-		COPY_DWORDS(out, data, count * 2);
-	else
-		for (i = 0; i < count; i++) {
-			out[0] = *(int *)data;
-			out[1] = *(int *)(data + 4);
-			out += 2;
-			data += stride;
-		}
-}
-
-static void r300EmitVec12(GLcontext * ctx, struct r300_dma_region *rvb,
-			  GLvoid * data, int stride, int count)
-{
-	int i;
-	int *out = (int *)(rvb->address + rvb->start);
-
-	if (RADEON_DEBUG & DEBUG_VERTS)
-		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
-			__FUNCTION__, count, stride, (void *)out, (void *)data);
-
-	if (stride == 12)
-		COPY_DWORDS(out, data, count * 3);
-	else
-		for (i = 0; i < count; i++) {
-			out[0] = *(int *)data;
-			out[1] = *(int *)(data + 4);
-			out[2] = *(int *)(data + 8);
-			out += 3;
-			data += stride;
-		}
-}
-
-static void r300EmitVec16(GLcontext * ctx, struct r300_dma_region *rvb,
-			  GLvoid * data, int stride, int count)
-{
-	int i;
-	int *out = (int *)(rvb->address + rvb->start);
-
-	if (RADEON_DEBUG & DEBUG_VERTS)
-		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
-			__FUNCTION__, count, stride, (void *)out, (void *)data);
-
-	if (stride == 16)
-		COPY_DWORDS(out, data, count * 4);
-	else
-		for (i = 0; i < count; i++) {
-			out[0] = *(int *)data;
-			out[1] = *(int *)(data + 4);
-			out[2] = *(int *)(data + 8);
-			out[3] = *(int *)(data + 12);
-			out += 4;
-			data += stride;
-		}
-}
-
-static void r300EmitVec(GLcontext * ctx, struct r300_dma_region *rvb,
-			GLvoid * data, int size, int stride, int count)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-
-	if (stride == 0) {
-		r300AllocDmaRegion(rmesa, rvb, size * 4, 4);
-		count = 1;
-		rvb->aos_offset = GET_START(rvb);
-		rvb->aos_stride = 0;
-	} else {
-		r300AllocDmaRegion(rmesa, rvb, size * count * 4, 4);
-		rvb->aos_offset = GET_START(rvb);
-		rvb->aos_stride = size;
-	}
-
-	switch (size) {
-	case 1:
-		r300EmitVec4(ctx, rvb, data, stride, count);
-		break;
-	case 2:
-		r300EmitVec8(ctx, rvb, data, stride, count);
-		break;
-	case 3:
-		r300EmitVec12(ctx, rvb, data, stride, count);
-		break;
-	case 4:
-		r300EmitVec16(ctx, rvb, data, stride, count);
-		break;
-	default:
-		assert(0);
-		break;
-	}
-}
-
-#define DW_SIZE(x) ((inputs[tab[(x)]] << R300_DST_VEC_LOC_SHIFT) |	\
-		    (attribptr[tab[(x)]]->size - 1) << R300_DATA_TYPE_0_SHIFT)
-
-GLuint r300VAPInputRoute0(uint32_t * dst, GLvector4f ** attribptr,
-				 int *inputs, GLint * tab, GLuint nr)
-{
-	GLuint i, dw;
-
-	/* type, inputs, stop bit, size */
-	for (i = 0; i < nr; i += 2) {
-		/* make sure input is valid, would lockup the gpu */
-		assert(inputs[tab[i]] != -1);
-		dw = (R300_SIGNED | DW_SIZE(i));
-		if (i + 1 == nr) {
-			dw |= R300_LAST_VEC << R300_DATA_TYPE_0_SHIFT;
-		} else {
-			assert(inputs[tab[i + 1]] != -1);
-			dw |= (R300_SIGNED |
-			       DW_SIZE(i + 1)) << R300_DATA_TYPE_1_SHIFT;
-			if (i + 2 == nr) {
-				dw |= R300_LAST_VEC << R300_DATA_TYPE_1_SHIFT;
-			}
-		}
-		dst[i >> 1] = dw;
-	}
-
-	return (nr + 1) >> 1;
-}
-
-static GLuint r300VAPInputRoute1Swizzle(int swizzle[4])
-{
-	return (swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
-	    (swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
-	    (swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
-	    (swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT);
-}
-
-GLuint r300VAPInputRoute1(uint32_t * dst, int swizzle[][4], GLuint nr)
-{
-	GLuint i, dw;
-
-	for (i = 0; i < nr; i += 2) {
-		dw = (r300VAPInputRoute1Swizzle(swizzle[i]) |
-		      ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y |
-			R300_WRITE_ENA_Z | R300_WRITE_ENA_W) << R300_WRITE_ENA_SHIFT)) << R300_SWIZZLE0_SHIFT;
-		if (i + 1 < nr) {
-			dw |= (r300VAPInputRoute1Swizzle(swizzle[i + 1]) |
-			       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y |
-				 R300_WRITE_ENA_Z | R300_WRITE_ENA_W) << R300_WRITE_ENA_SHIFT)) << R300_SWIZZLE1_SHIFT;
-		}
-		dst[i >> 1] = dw;
-	}
-
-	return (nr + 1) >> 1;
-}
+#include "r300_render.h"
+#include "r300_swtcl.h"
 
 GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead)
 {
@@ -272,7 +62,6 @@ GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead)
 
 GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead)
 {
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	GLuint i, vic_1 = 0;
 
 	if (InputsRead & (1 << VERT_ATTRIB_POS))
@@ -284,281 +73,112 @@ GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead)
 	if (InputsRead & (1 << VERT_ATTRIB_COLOR0))
 		vic_1 |= R300_INPUT_CNTL_COLOR;
 
-	rmesa->state.texture.tc_count = 0;
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++)
 		if (InputsRead & (1 << (VERT_ATTRIB_TEX0 + i))) {
-			rmesa->state.texture.tc_count++;
 			vic_1 |= R300_INPUT_CNTL_TC0 << i;
 		}
 
 	return vic_1;
 }
 
-GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten)
+GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
 {
 	GLuint ret = 0;
 
-	if (OutputsWritten & (1 << VERT_RESULT_HPOS))
+	if (vp_writes & (1 << VERT_RESULT_HPOS))
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 
-	if (OutputsWritten & (1 << VERT_RESULT_COL0))
+	if (vp_writes & (1 << VERT_RESULT_COL0) && fp_reads & FRAG_BIT_COL0)
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT;
 
-	if (OutputsWritten & (1 << VERT_RESULT_COL1))
+	if (vp_writes & (1 << VERT_RESULT_COL1) && fp_reads & FRAG_BIT_COL1)
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT;
 
-	if (OutputsWritten & (1 << VERT_RESULT_BFC0)
-	    || OutputsWritten & (1 << VERT_RESULT_BFC1))
-		ret |=
-		    R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT |
-		    R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT |
-		    R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT;
+	/* Two sided lighting works only if all 4 colors are written */
+	if (vp_writes & (1 << VERT_RESULT_BFC0) || vp_writes & (1 << VERT_RESULT_BFC1))
+		ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT | R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT |
+			   R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT | R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT;
 
-	if (OutputsWritten & (1 << VERT_RESULT_PSIZ))
+	if (vp_writes & (1 << VERT_RESULT_PSIZ))
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 
 	return ret;
 }
 
-GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint OutputsWritten)
+GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
 {
 	GLuint i, ret = 0, first_free_texcoord = 0;
 
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-		if (OutputsWritten & (1 << (VERT_RESULT_TEX0 + i))) {
-			ret |= (4 << (3 * i));
+		if (vp_writes & (1 << (VERT_RESULT_TEX0 + i)) && fp_reads & FRAG_BIT_TEX(i)) {
+			ret |= (4 << (3 * first_free_texcoord));
 			++first_free_texcoord;
 		}
 	}
 
-	if (OutputsWritten & (1 << VERT_RESULT_FOGC)) {
-		if (first_free_texcoord > 8) {
-			fprintf(stderr, "\tout of free texcoords to write fog coord\n");
-			_mesa_exit(-1);
-		}
-		ret |= 4 << (3 * first_free_texcoord);
-	}
-
-	return ret;
-}
-
-/* Emit vertex data to GART memory
- * Route inputs to the vertex processor
- * This function should never return R300_FALLBACK_TCL when using software tcl.
- */
-int r300EmitArrays(GLcontext * ctx)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *vb = &tnl->vb;
-	GLuint nr;
-	GLuint count = vb->Count;
-	GLuint i;
-	GLuint InputsRead = 0, OutputsWritten = 0;
-	int *inputs = NULL;
-	int vir_inputs[VERT_ATTRIB_MAX];
-	GLint tab[VERT_ATTRIB_MAX];
-	int swizzle[VERT_ATTRIB_MAX][4];
-	struct r300_vertex_program *prog =
-	    (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
-
-	if (hw_tcl_on) {
-		inputs = prog->inputs;
-		InputsRead = prog->key.InputsRead;
-		OutputsWritten = prog->key.OutputsWritten;
-	} else {
-		inputs = rmesa->state.sw_tcl_inputs;
-
-		DECLARE_RENDERINPUTS(render_inputs_bitset);
-		RENDERINPUTS_COPY(render_inputs_bitset, tnl->render_inputs_bitset);
-
-		vb->AttribPtr[VERT_ATTRIB_POS] = vb->ClipPtr;
-
-		assert(RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_POS));
-		assert(RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_NORMAL) == 0);
-		//assert(RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_COLOR0));
-
-		if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_POS)) {
-			InputsRead |= 1 << VERT_ATTRIB_POS;
-			OutputsWritten |= 1 << VERT_RESULT_HPOS;
-		}
-
-		if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_COLOR0)) {
-			InputsRead |= 1 << VERT_ATTRIB_COLOR0;
-			OutputsWritten |= 1 << VERT_RESULT_COL0;
-		}
-
-		if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_COLOR1)) {
-			InputsRead |= 1 << VERT_ATTRIB_COLOR1;
-			OutputsWritten |= 1 << VERT_RESULT_COL1;
-		}
-
-		for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-			if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_TEX(i))) {
-				InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i);
-				OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i);
-			}
-		}
-
-		for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) {
-			if (InputsRead & (1 << i)) {
-				inputs[i] = nr++;
-			} else {
-				inputs[i] = -1;
-			}
-		}
-
-		/* Fixed, apply to vir0 only */
-		memcpy(vir_inputs, inputs, VERT_ATTRIB_MAX * sizeof(int));
-		inputs = vir_inputs;
-		if (InputsRead & VERT_ATTRIB_POS)
-			inputs[VERT_ATTRIB_POS] = 0;
-		if (InputsRead & (1 << VERT_ATTRIB_COLOR0))
-			inputs[VERT_ATTRIB_COLOR0] = 2;
-		if (InputsRead & (1 << VERT_ATTRIB_COLOR1))
-			inputs[VERT_ATTRIB_COLOR1] = 3;
-		for (i = VERT_ATTRIB_TEX0; i <= VERT_ATTRIB_TEX7; i++)
-			if (InputsRead & (1 << i))
-				inputs[i] = 6 + (i - VERT_ATTRIB_TEX0);
-
-		RENDERINPUTS_COPY(rmesa->state.render_inputs_bitset, render_inputs_bitset);
+	if (fp_reads & FRAG_BIT_WPOS) {
+		ret |= (4 << (3 * first_free_texcoord));
+		++first_free_texcoord;
 	}
 
-	assert(InputsRead);
-	assert(OutputsWritten);
-
-	for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) {
-		if (InputsRead & (1 << i)) {
-			tab[nr++] = i;
-		}
+	if (vp_writes & (1 << VERT_RESULT_FOGC) && fp_reads & FRAG_BIT_FOGC) {
+		ret |= 4 << (3 * first_free_texcoord);
 	}
 
-	if (nr > R300_MAX_AOS_ARRAYS) {
-		return R300_FALLBACK_TCL;
+	if (first_free_texcoord > 8) {
+		fprintf(stderr, "\tout of free texcoords\n");
+		_mesa_exit(-1);
 	}
 
-	for (i = 0; i < nr; i++) {
-		int ci, fix, found = 0;
-
-		swizzle[i][0] = SWIZZLE_ZERO;
-		swizzle[i][1] = SWIZZLE_ZERO;
-		swizzle[i][2] = SWIZZLE_ZERO;
-		swizzle[i][3] = SWIZZLE_ONE;
+	return ret;
+}
 
-		for (ci = 0; ci < vb->AttribPtr[tab[i]]->size; ci++) {
-			swizzle[i][ci] = ci;
-		}
+GLboolean r300EmitArrays(GLcontext * ctx)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_vertex_buffer *vbuf = &r300->vbuf;
+	GLuint InputsRead, OutputsWritten;
 
-		if (r300IsGartMemory(rmesa, vb->AttribPtr[tab[i]]->data, 4)) {
-			if (vb->AttribPtr[tab[i]]->stride % 4) {
-				return R300_FALLBACK_TCL;
-			}
-			rmesa->state.aos[i].address = (void *)(vb->AttribPtr[tab[i]]->data);
-			rmesa->state.aos[i].start = 0;
-			rmesa->state.aos[i].aos_offset = r300GartOffsetFromVirtual(rmesa, vb->AttribPtr[tab[i]]->data);
-			rmesa->state.aos[i].aos_stride = vb->AttribPtr[tab[i]]->stride / 4;
-			rmesa->state.aos[i].aos_size = vb->AttribPtr[tab[i]]->size;
-		} else {
-			r300EmitVec(ctx, &rmesa->state.aos[i],
-				    vb->AttribPtr[tab[i]]->data,
-				    vb->AttribPtr[tab[i]]->size,
-				    vb->AttribPtr[tab[i]]->stride, count);
-		}
+	r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten);
 
-		rmesa->state.aos[i].aos_size = vb->AttribPtr[tab[i]]->size;
+	r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS);
+	if (r300->fallback & R300_RASTER_FALLBACK_MASK)
+		return GL_FALSE;
 
-		for (fix = 0; fix <= 4 - vb->AttribPtr[tab[i]]->size; fix++) {
-			if ((rmesa->state.aos[i].aos_offset - _mesa_sizeof_type(GL_FLOAT) * fix) % 4) {
-				continue;
-			}
-			found = 1;
-			break;
-		}
+	{
+		struct vertex_buffer *mesa_vb = &TNL_CONTEXT(ctx)->vb;
+		GLuint attr, i;
 
-		if (found) {
-			if (fix > 0) {
-				WARN_ONCE("Feeling lucky?\n");
-			}
-			rmesa->state.aos[i].aos_offset -= _mesa_sizeof_type(GL_FLOAT) * fix;
-			for (ci = 0; ci < vb->AttribPtr[tab[i]]->size; ci++) {
-				swizzle[i][ci] += fix;
-			}
-		} else {
-			WARN_ONCE
-			    ("Cannot handle offset %x with stride %d, comp %d\n",
-			     rmesa->state.aos[i].aos_offset,
-			     rmesa->state.aos[i].aos_stride,
-			     vb->AttribPtr[tab[i]]->size);
-			return R300_FALLBACK_TCL;
+		for (i = 0; i < vbuf->num_attribs; i++) {
+			attr = vbuf->attribs[i].element;
+			rcommon_emit_vector(ctx, &r300->radeon.tcl.aos[i], mesa_vb->AttribPtr[attr]->data,
+					mesa_vb->AttribPtr[attr]->size, mesa_vb->AttribPtr[attr]->stride, mesa_vb->Count);
 		}
-	}
 
-	/* Setup INPUT_ROUTE. */
-	R300_STATECHANGE(rmesa, vir[0]);
-	((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count =
-	    r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0],
-			       vb->AttribPtr, inputs, tab, nr);
-	R300_STATECHANGE(rmesa, vir[1]);
-	((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count =
-	    r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle,
-			       nr);
-
-	/* Setup INPUT_CNTL. */
-	R300_STATECHANGE(rmesa, vic);
-	rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
-	rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
-
-	/* Setup OUTPUT_VTX_FMT. */
-	R300_STATECHANGE(rmesa, vof);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_0] =
-	    r300VAPOutputCntl0(ctx, OutputsWritten);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_1] =
-	    r300VAPOutputCntl1(ctx, OutputsWritten);
-
-	rmesa->state.aos_count = nr;
-
-	return R300_FALLBACK_NONE;
-}
+		r300->radeon.tcl.aos_count = vbuf->num_attribs;
 
-#ifdef USER_BUFFERS
-void r300UseArrays(GLcontext * ctx)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	int i;
-
-	if (rmesa->state.elt_dma.buf)
-		r300_mem_use(rmesa, rmesa->state.elt_dma.buf->id);
-
-	for (i = 0; i < rmesa->state.aos_count; i++) {
-		if (rmesa->state.aos[i].buf)
-			r300_mem_use(rmesa, rmesa->state.aos[i].buf->id);
+		/* Fill index buffer info */
+		r300->ind_buf.ptr = mesa_vb->Elts;
+		r300->ind_buf.is_32bit = GL_TRUE;
+		r300->ind_buf.free_needed = GL_FALSE;
 	}
-}
-#endif
 
-void r300ReleaseArrays(GLcontext * ctx)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	int i;
+	r300SetupVAP(ctx, InputsRead, OutputsWritten);
 
-	r300ReleaseDmaRegion(rmesa, &rmesa->state.elt_dma, __FUNCTION__);
-	for (i = 0; i < rmesa->state.aos_count; i++) {
-		r300ReleaseDmaRegion(rmesa, &rmesa->state.aos[i], __FUNCTION__);
-	}
+	return GL_TRUE;
 }
 
 void r300EmitCacheFlush(r300ContextPtr rmesa)
 {
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-
-	drm_radeon_cmd_header_t *cmd = NULL;
-
-	reg_start(R300_RB3D_DSTCACHE_CTLSTAT, 0);
-	e32(R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
-	    R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
-
-	reg_start(R300_ZB_ZCACHE_CTLSTAT, 0);
-	e32(R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
-	    R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
+	BATCH_LOCALS(&rmesa->radeon);
+
+	BEGIN_BATCH_NO_AUTOSTATE(4);
+	OUT_BATCH_REGVAL(R300_RB3D_DSTCACHE_CTLSTAT,
+		R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
+		R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
+	OUT_BATCH_REGVAL(R300_ZB_ZCACHE_CTLSTAT,
+		R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
+		R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
+	END_BATCH();
+	COMMIT_BATCH();
 }
diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h
index 89d738339f8..2fb8b82d3a1 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.h
+++ b/src/mesa/drivers/dri/r300/r300_emit.h
@@ -44,28 +44,31 @@
 #include "r300_cmdbuf.h"
 #include "radeon_reg.h"
 
-/* TODO: move these defines (and the ones from DRM) into r300_reg.h and sync up
- * with DRM */
-#define CP_PACKET0(reg, n)	(RADEON_CP_PACKET0 | ((n)<<16) | ((reg)>>2))
-#define CP_PACKET3( pkt, n )						\
-	(RADEON_CP_PACKET3 | (pkt) | ((n) << 16))
-
-static INLINE uint32_t cmdpacket0(int reg, int count)
+static INLINE uint32_t cmdpacket0(struct radeon_screen *rscrn,
+                                  int reg, int count)
 {
-	drm_r300_cmd_header_t cmd;
-
-	cmd.packet0.cmd_type = R300_CMD_PACKET0;
-	cmd.packet0.count = count;
-	cmd.packet0.reghi = ((unsigned int)reg & 0xFF00) >> 8;
-	cmd.packet0.reglo = ((unsigned int)reg & 0x00FF);
-
-	return cmd.u;
+    if (!rscrn->kernel_mm) {
+	    drm_r300_cmd_header_t cmd;
+
+	cmd.u = 0;
+    	cmd.packet0.cmd_type = R300_CMD_PACKET0;
+	    cmd.packet0.count = count;
+    	cmd.packet0.reghi = ((unsigned int)reg & 0xFF00) >> 8;
+	    cmd.packet0.reglo = ((unsigned int)reg & 0x00FF);
+
+    	return cmd.u;
+    }
+    if (count) {
+        return CP_PACKET0(reg, count - 1);
+    }
+    return CP_PACKET2;
 }
 
-static INLINE uint32_t cmdvpu(int addr, int count)
+static INLINE uint32_t cmdvpu(struct radeon_screen *rscrn, int addr, int count)
 {
 	drm_r300_cmd_header_t cmd;
 
+	cmd.u = 0;
 	cmd.vpu.cmd_type = R300_CMD_VPU;
 	cmd.vpu.count = count;
 	cmd.vpu.adrhi = ((unsigned int)addr & 0xFF00) >> 8;
@@ -74,10 +77,12 @@ static INLINE uint32_t cmdvpu(int addr, int count)
 	return cmd.u;
 }
 
-static INLINE uint32_t cmdr500fp(int addr, int count, int type, int clamp)
+static INLINE uint32_t cmdr500fp(struct radeon_screen *rscrn,
+                                 int addr, int count, int type, int clamp)
 {
 	drm_r300_cmd_header_t cmd;
 
+	cmd.u = 0;
 	cmd.r500fp.cmd_type = R300_CMD_R500FP;
 	cmd.r500fp.count = count;
 	cmd.r500fp.adrhi_flags = ((unsigned int)addr & 0x100) >> 8;
@@ -88,181 +93,139 @@ static INLINE uint32_t cmdr500fp(int addr, int count, int type, int clamp)
 	return cmd.u;
 }
 
-static INLINE uint32_t cmdpacket3(int packet)
+static INLINE uint32_t cmdpacket3(struct radeon_screen *rscrn, int packet)
 {
 	drm_r300_cmd_header_t cmd;
 
+	cmd.u = 0;
 	cmd.packet3.cmd_type = R300_CMD_PACKET3;
 	cmd.packet3.packet = packet;
 
 	return cmd.u;
 }
 
-static INLINE uint32_t cmdcpdelay(unsigned short count)
+static INLINE uint32_t cmdcpdelay(struct radeon_screen *rscrn,  
+                                  unsigned short count)
 {
 	drm_r300_cmd_header_t cmd;
 
+	cmd.u = 0;
+
 	cmd.delay.cmd_type = R300_CMD_CP_DELAY;
 	cmd.delay.count = count;
 
 	return cmd.u;
 }
 
-static INLINE uint32_t cmdwait(unsigned char flags)
+static INLINE uint32_t cmdwait(struct radeon_screen *rscrn,
+                               unsigned char flags)
 {
 	drm_r300_cmd_header_t cmd;
 
+	cmd.u = 0;
 	cmd.wait.cmd_type = R300_CMD_WAIT;
 	cmd.wait.flags = flags;
 
 	return cmd.u;
 }
 
-static INLINE uint32_t cmdpacify(void)
+static INLINE uint32_t cmdpacify(struct radeon_screen *rscrn)
 {
 	drm_r300_cmd_header_t cmd;
 
+	cmd.u = 0;
 	cmd.header.cmd_type = R300_CMD_END3D;
 
 	return cmd.u;
 }
 
 /**
- * Prepare to write a register value to register at address reg.
- * If num_extra > 0 then the following extra values are written
- * to registers with address +4, +8 and so on..
- */
-#define reg_start(reg, num_extra)					\
-	do {								\
-		int _n;							\
-		_n=(num_extra);						\
-		cmd = (drm_radeon_cmd_header_t*)			\
-			r300AllocCmdBuf(rmesa,				\
-					(_n+2),				\
-					__FUNCTION__);			\
-		cmd_reserved=_n+2;					\
-		cmd_written=1;						\
-		cmd[0].i=cmdpacket0((reg), _n+1);			\
-	} while (0);
-
-/**
- * Emit GLuint freestyle
+ * Write the header of a packet3 to the command buffer.
+ * Outputs 2 dwords and expects (num_extra+1) additional dwords afterwards.
  */
-#define e32(dword)							\
-	do {								\
-		if(cmd_written<cmd_reserved) {				\
-			cmd[cmd_written].i=(dword);			\
-			cmd_written++;					\
-		} else {						\
-			fprintf(stderr,					\
-				"e32 but no previous packet "		\
-				"declaration.\n"			\
-				"Aborting! in %s::%s at line %d, "	\
-				"cmd_written=%d cmd_reserved=%d\n",	\
-				__FILE__, __FUNCTION__, __LINE__,	\
-				cmd_written, cmd_reserved);		\
-			_mesa_exit(-1);					\
-		}							\
+#define OUT_BATCH_PACKET3(packet, num_extra) do {\
+    if (!b_l_rmesa->radeonScreen->kernel_mm) {		\
+    	OUT_BATCH(cmdpacket3(b_l_rmesa->radeonScreen,\
+                  R300_CMD_PACKET3_RAW)); \
+    } else b_l_rmesa->cmdbuf.cs->section_cdw++;\
+	OUT_BATCH(CP_PACKET3((packet), (num_extra))); \
 	} while(0)
 
-#define	efloat(f) e32(r300PackFloat32(f))
-
-#define vsf_start_fragment(dest, length)				\
-	do {								\
-		int _n;							\
-		_n = (length);						\
-		cmd = (drm_radeon_cmd_header_t*)			\
-			r300AllocCmdBuf(rmesa,				\
-					(_n+1),				\
-					__FUNCTION__);			\
-		cmd_reserved = _n+2;					\
-		cmd_written =1;						\
-		cmd[0].i = cmdvpu((dest), _n/4);			\
-	} while (0);
-
-#define r500fp_start_fragment(dest, length)				\
-	do {								\
-		int _n;							\
-		_n = (length);						\
-		cmd = (drm_radeon_cmd_header_t*)			\
-			r300AllocCmdBuf(rmesa,				\
-					(_n+1),				\
-					__FUNCTION__);			\
-		cmd_reserved = _n+1;					\
-		cmd_written =1;						\
-		cmd[0].i = cmdr500fp((dest), _n/6, 0, 0);		\
-	} while (0);
-
-#define start_packet3(packet, count)					\
-	{								\
-		int _n;							\
-		GLuint _p;						\
-		_n = (count);						\
-		_p = (packet);						\
-		cmd = (drm_radeon_cmd_header_t*)			\
-			r300AllocCmdBuf(rmesa,				\
-					(_n+3),				\
-					__FUNCTION__);			\
-		cmd_reserved = _n+3;					\
-		cmd_written = 2;					\
-		if(_n > 0x3fff) {					\
-			fprintf(stderr,"Too big packet3 %08x: cannot "	\
-				"store %d dwords\n",			\
-				_p, _n);				\
-			_mesa_exit(-1);					\
-		}							\
-		cmd[0].i = cmdpacket3(R300_CMD_PACKET3_RAW);		\
-		cmd[1].i = _p | ((_n & 0x3fff)<<16);			\
-	}
-
 /**
  * Must be sent to switch to 2d commands
  */
-void static INLINE end_3d(r300ContextPtr rmesa)
+void static INLINE end_3d(radeonContextPtr radeon)
 {
-	drm_radeon_cmd_header_t *cmd = NULL;
+	BATCH_LOCALS(radeon);
 
-	cmd =
-	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__);
-	cmd[0].header.cmd_type = R300_CMD_END3D;
+	if (!radeon->radeonScreen->kernel_mm) {
+		BEGIN_BATCH_NO_AUTOSTATE(1);
+		OUT_BATCH(cmdpacify(radeon->radeonScreen));
+		END_BATCH();
+	}
 }
 
 void static INLINE cp_delay(r300ContextPtr rmesa, unsigned short count)
 {
-	drm_radeon_cmd_header_t *cmd = NULL;
+	BATCH_LOCALS(&rmesa->radeon);
 
-	cmd =
-	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__);
-	cmd[0].i = cmdcpdelay(count);
+	if (!rmesa->radeon.radeonScreen->kernel_mm) {
+		BEGIN_BATCH_NO_AUTOSTATE(1);
+		OUT_BATCH(cmdcpdelay(rmesa->radeon.radeonScreen, count));
+		END_BATCH();
+	}
 }
 
-void static INLINE cp_wait(r300ContextPtr rmesa, unsigned char flags)
+void static INLINE cp_wait(radeonContextPtr radeon, unsigned char flags)
 {
-	drm_radeon_cmd_header_t *cmd = NULL;
-
-	cmd =
-	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__);
-	cmd[0].i = cmdwait(flags);
+	BATCH_LOCALS(radeon);
+	uint32_t wait_until;
+
+	if (!radeon->radeonScreen->kernel_mm) {
+		BEGIN_BATCH_NO_AUTOSTATE(1);
+		OUT_BATCH(cmdwait(radeon->radeonScreen, flags));
+		END_BATCH();
+	} else {
+		switch(flags) {
+		case R300_WAIT_2D:
+			wait_until = (1 << 14);
+			break;
+		case R300_WAIT_3D:
+			wait_until = (1 << 15);
+			break;
+		case R300_NEW_WAIT_2D_3D:
+			wait_until = (1 << 14) | (1 << 15);
+			break;
+		case R300_NEW_WAIT_2D_2D_CLEAN:
+			wait_until = (1 << 14) | (1 << 16) | (1 << 18);
+			break;
+		case R300_NEW_WAIT_3D_3D_CLEAN:
+			wait_until = (1 << 15) | (1 << 17) | (1 << 18);
+			break;
+		case R300_NEW_WAIT_2D_2D_CLEAN_3D_3D_CLEAN:
+			wait_until  = (1 << 14) | (1 << 16) | (1 << 18);
+			wait_until |= (1 << 15) | (1 << 17) | (1 << 18);
+			break;
+		default:
+			return;
+		}
+		BEGIN_BATCH_NO_AUTOSTATE(2);
+		OUT_BATCH(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+		OUT_BATCH(wait_until);
+		END_BATCH();
+	}
 }
 
-extern int r300EmitArrays(GLcontext * ctx);
-
-#ifdef USER_BUFFERS
-void r300UseArrays(GLcontext * ctx);
-#endif
+extern GLboolean r300EmitArrays(GLcontext * ctx);
 
-extern void r300ReleaseArrays(GLcontext * ctx);
 extern int r300PrimitiveType(r300ContextPtr rmesa, int prim);
 extern int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim);
 
 extern void r300EmitCacheFlush(r300ContextPtr rmesa);
 
-extern GLuint r300VAPInputRoute0(uint32_t * dst, GLvector4f ** attribptr,
-				 int *inputs, GLint * tab, GLuint nr);
-extern GLuint r300VAPInputRoute1(uint32_t * dst, int swizzle[][4], GLuint nr);
 extern GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead);
 extern GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead);
-extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten);
-extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint OutputsWritten);
+extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads);
+extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c
index 2f45429cf27..55c1cfe6317 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.c
@@ -25,32 +25,12 @@
  *
  */
 
-/**
- * \file
- *
- * Fragment program compiler. Perform transformations on the intermediate
- * representation until the program is in a form where we can translate
- * it more or less directly into machine-readable form.
- *
- * \author Ben Skeggs <[email protected]>
- * \author Jerome Glisse <[email protected]>
- */
+#include "r300_fragprog.h"
 
-#include "main/glheader.h"
-#include "main/macros.h"
-#include "main/enums.h"
-#include "shader/prog_instruction.h"
 #include "shader/prog_parameter.h"
-#include "shader/prog_print.h"
 
 #include "r300_context.h"
-#include "r300_fragprog.h"
 #include "r300_fragprog_swizzle.h"
-#include "r300_state.h"
-
-#include "radeon_nqssadce.h"
-#include "radeon_program_alu.h"
-
 
 static void reset_srcreg(struct prog_src_register* reg)
 {
@@ -81,7 +61,7 @@ static struct prog_src_register shadow_ambient(struct gl_program *program, int t
  * \todo If/when r5xx uses the radeon_program architecture, this can probably
  * be reused.
  */
-static GLboolean transform_TEX(
+GLboolean r300_transform_TEX(
 	struct radeon_transform_context *t,
 	struct prog_instruction* orig_inst, void* data)
 {
@@ -160,6 +140,8 @@ static GLboolean transform_TEX(
 			inst.DstReg.Index = tempreg;
 			inst.DstReg.WriteMask = WRITEMASK_XYZW;
 			destredirect = GL_TRUE;
+		} else if (inst.SaturateMode) {
+			destredirect = GL_TRUE;
 		}
 	}
 
@@ -175,7 +157,7 @@ static GLboolean transform_TEX(
 		inst.SrcReg[0].File = PROGRAM_TEMPORARY;
 		inst.SrcReg[0].Index = tmpreg;
 	}
-	
+
 	tgt = radeonAppendInstructions(t->Program, 1);
 	_mesa_copy_instructions(tgt, &inst, 1);
 
@@ -239,6 +221,7 @@ static GLboolean transform_TEX(
 
 		tgt->Opcode = OPCODE_MOV;
 		tgt->DstReg = orig_inst->DstReg;
+		tgt->SaturateMode = inst.SaturateMode;
 		tgt->SrcReg[0].File = PROGRAM_TEMPORARY;
 		tgt->SrcReg[0].Index = inst.DstReg.Index;
 	}
@@ -246,242 +229,10 @@ static GLboolean transform_TEX(
 	return GL_TRUE;
 }
 
-
-static void update_params(r300ContextPtr r300, struct r300_fragment_program *fp)
-{
-	struct gl_fragment_program *mp = &fp->mesa_program;
-
-	/* Ask Mesa nicely to fill in ParameterValues for us */
-	if (mp->Base.Parameters)
-		_mesa_load_state_parameters(r300->radeon.glCtx, mp->Base.Parameters);
-}
-
-
-/**
- * Transform the program to support fragment.position.
- *
- * Introduce a small fragment at the start of the program that will be
- * the only code that directly reads the FRAG_ATTRIB_WPOS input.
- * All other code pieces that reference that input will be rewritten
- * to read from a newly allocated temporary.
- *
- * \todo if/when r5xx supports the radeon_program architecture, this is a
- * likely candidate for code sharing.
- */
-static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
-{
-	GLuint InputsRead = compiler->fp->mesa_program.Base.InputsRead;
-
-	if (!(InputsRead & FRAG_BIT_WPOS))
-		return;
-
-	static gl_state_index tokens[STATE_LENGTH] = {
-		STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
-	};
-	struct prog_instruction *fpi;
-	GLuint window_index;
-	int i = 0;
-	GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY);
-
-	_mesa_insert_instructions(compiler->program, 0, 3);
-	fpi = compiler->program->Instructions;
-
-	/* perspective divide */
-	fpi[i].Opcode = OPCODE_RCP;
-
-	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
-	fpi[i].DstReg.Index = tempregi;
-	fpi[i].DstReg.WriteMask = WRITEMASK_W;
-	fpi[i].DstReg.CondMask = COND_TR;
-
-	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
-	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
-	fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
-	i++;
-
-	fpi[i].Opcode = OPCODE_MUL;
-
-	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
-	fpi[i].DstReg.Index = tempregi;
-	fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
-	fpi[i].DstReg.CondMask = COND_TR;
-
-	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
-	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
-	fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-
-	fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
-	fpi[i].SrcReg[1].Index = tempregi;
-	fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
-	i++;
-
-	/* viewport transformation */
-	window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens);
-
-	fpi[i].Opcode = OPCODE_MAD;
-
-	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
-	fpi[i].DstReg.Index = tempregi;
-	fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
-	fpi[i].DstReg.CondMask = COND_TR;
-
-	fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
-	fpi[i].SrcReg[0].Index = tempregi;
-	fpi[i].SrcReg[0].Swizzle =
-	    MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-
-	fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
-	fpi[i].SrcReg[1].Index = window_index;
-	fpi[i].SrcReg[1].Swizzle =
-	    MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-
-	fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
-	fpi[i].SrcReg[2].Index = window_index;
-	fpi[i].SrcReg[2].Swizzle =
-	    MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-	i++;
-
-	for (; i < compiler->program->NumInstructions; ++i) {
-		int reg;
-		for (reg = 0; reg < 3; reg++) {
-			if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT &&
-			    fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) {
-				fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY;
-				fpi[i].SrcReg[reg].Index = tempregi;
-			}
-		}
-	}
-}
-
-
-static void nqssadce_init(struct nqssadce_state* s)
-{
-	s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW;
-	s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W;
-}
-
-
-static GLuint build_dtm(GLuint depthmode)
-{
-	switch(depthmode) {
-	default:
-	case GL_LUMINANCE: return 0;
-	case GL_INTENSITY: return 1;
-	case GL_ALPHA: return 2;
-	}
-}
-
-static GLuint build_func(GLuint comparefunc)
-{
-	return comparefunc - GL_NEVER;
-}
-
-
-/**
- * Collect all external state that is relevant for compiling the given
- * fragment program.
- */
-static void build_state(
-	r300ContextPtr r300,
-	struct r300_fragment_program *fp,
-	struct r300_fragment_program_external_state *state)
-{
-	int unit;
-
-	_mesa_bzero(state, sizeof(*state));
-
-	for(unit = 0; unit < 16; ++unit) {
-		if (fp->mesa_program.Base.ShadowSamplers & (1 << unit)) {
-			struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
-
-			state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode);
-			state->unit[unit].texture_compare_func = build_func(tex->CompareFunc);
-		}
-	}
-}
-
-
-void r300TranslateFragmentShader(r300ContextPtr r300,
-				 struct r300_fragment_program *fp)
-{
-	struct r300_fragment_program_external_state state;
-
-	build_state(r300, fp, &state);
-	if (_mesa_memcmp(&fp->state, &state, sizeof(state))) {
-		/* TODO: cache compiled programs */
-		fp->translated = GL_FALSE;
-		_mesa_memcpy(&fp->state, &state, sizeof(state));
-	}
-
-	if (!fp->translated) {
-		struct r300_fragment_program_compiler compiler;
-
-		compiler.r300 = r300;
-		compiler.fp = fp;
-		compiler.code = &fp->code;
-		compiler.program = _mesa_clone_program(r300->radeon.glCtx, &fp->mesa_program.Base);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Fragment Program: Initial program:\n");
-			_mesa_print_program(compiler.program);
-		}
-
-		insert_WPOS_trailer(&compiler);
-
-		struct radeon_program_transformation transformations[] = {
-			{ &transform_TEX, &compiler },
-			{ &radeonTransformALU, 0 },
-			{ &radeonTransformTrigSimple, 0 }
-		};
-		radeonLocalTransform(
-			r300->radeon.glCtx,
-			compiler.program,
-			3, transformations);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Fragment Program: After native rewrite:\n");
-			_mesa_print_program(compiler.program);
-		}
-
-		struct radeon_nqssadce_descr nqssadce = {
-			.Init = &nqssadce_init,
-			.IsNativeSwizzle = &r300FPIsNativeSwizzle,
-			.BuildSwizzle = &r300FPBuildSwizzle,
-			.RewriteDepthOut = GL_TRUE
-		};
-		radeonNqssaDce(r300->radeon.glCtx, compiler.program, &nqssadce);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Compiler: after NqSSA-DCE:\n");
-			_mesa_print_program(compiler.program);
-		}
-
-		if (!r300FragmentProgramEmit(&compiler))
-			fp->error = GL_TRUE;
-
-		/* Subtle: Rescue any parameters that have been added during transformations */
-		_mesa_free_parameter_list(fp->mesa_program.Base.Parameters);
-		fp->mesa_program.Base.Parameters = compiler.program->Parameters;
-		compiler.program->Parameters = 0;
-
-		_mesa_reference_program(r300->radeon.glCtx, &compiler.program, NULL);
-
-		if (!fp->error)
-			fp->translated = GL_TRUE;
-		if (fp->error || (RADEON_DEBUG & DEBUG_PIXEL))
-			r300FragmentProgramDump(fp, &fp->code);
-		r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM |
-                                          _NEW_PROGRAM_CONSTANTS);
-	}
-
-	update_params(r300, fp);
-}
-
 /* just some random things... */
-void r300FragmentProgramDump(
-	struct r300_fragment_program *fp,
-	struct r300_fragment_program_code *code)
+void r300FragmentProgramDump(union rX00_fragment_program_code *c)
 {
+	struct r300_fragment_program_code *code = &c->r300;
 	int n, i, j;
 	static int pc = 0;
 
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.h b/src/mesa/drivers/dri/r300/r300_fragprog.h
index 94fb554fb37..5ce6f33cee7 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.h
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.h
@@ -33,9 +33,6 @@
 #ifndef __R300_FRAGPROG_H_
 #define __R300_FRAGPROG_H_
 
-#include "main/glheader.h"
-#include "main/macros.h"
-#include "main/enums.h"
 #include "shader/program.h"
 #include "shader/prog_instruction.h"
 
@@ -105,28 +102,10 @@
 
 #endif
 
-struct r300_fragment_program;
+extern GLboolean r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler);
 
-extern void r300TranslateFragmentShader(r300ContextPtr r300,
-					struct r300_fragment_program *fp);
+extern void r300FragmentProgramDump(union rX00_fragment_program_code *c);
 
-
-/**
- * Used internally by the r300 fragment program code to store compile-time
- * only data.
- */
-struct r300_fragment_program_compiler {
-	r300ContextPtr r300;
-	struct r300_fragment_program *fp;
-	struct r300_fragment_program_code *code;
-	struct gl_program *program;
-};
-
-extern GLboolean r300FragmentProgramEmit(struct r300_fragment_program_compiler *compiler);
-
-
-extern void r300FragmentProgramDump(
-	struct r300_fragment_program *fp,
-	struct r300_fragment_program_code *code);
+extern GLboolean r300_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c
new file mode 100644
index 00000000000..abc8757ba1f
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c
@@ -0,0 +1,291 @@
+/*
+ * Copyright (C) 2009 Maciej Cencora <[email protected]>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+/**
+ * \file
+ *
+ * Fragment program compiler. Perform transformations on the intermediate
+ * representation until the program is in a form where we can translate
+ * it more or less directly into machine-readable form.
+ *
+ * \author Ben Skeggs <[email protected]>
+ * \author Jerome Glisse <[email protected]>
+ */
+
+#include "r300_fragprog_common.h"
+
+#include "shader/program.h"
+#include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
+
+#include "r300_state.h"
+#include "r300_fragprog.h"
+#include "r300_fragprog_swizzle.h"
+#include "r500_fragprog.h"
+
+#include "radeon_program.h"
+#include "radeon_program_alu.h"
+
+static void update_params(GLcontext *ctx, struct gl_fragment_program *fp)
+{
+	/* Ask Mesa nicely to fill in ParameterValues for us */
+	if (fp->Base.Parameters)
+		_mesa_load_state_parameters(ctx, fp->Base.Parameters);
+}
+
+static void nqssadce_init(struct nqssadce_state* s)
+{
+	s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW;
+	s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W;
+}
+
+/**
+ * Transform the program to support fragment.position.
+ *
+ * Introduce a small fragment at the start of the program that will be
+ * the only code that directly reads the FRAG_ATTRIB_WPOS input.
+ * All other code pieces that reference that input will be rewritten
+ * to read from a newly allocated temporary.
+ *
+ */
+static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
+{
+	GLuint InputsRead = compiler->fp->Base.Base.InputsRead;
+
+	if (!(InputsRead & FRAG_BIT_WPOS))
+		return;
+
+	static gl_state_index tokens[STATE_LENGTH] = {
+		STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
+	};
+	struct prog_instruction *fpi;
+	GLuint window_index;
+	int i = 0;
+	GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY);
+
+	_mesa_insert_instructions(compiler->program, 0, 3);
+	fpi = compiler->program->Instructions;
+
+	/* perspective divide */
+	fpi[i].Opcode = OPCODE_RCP;
+
+	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
+	fpi[i].DstReg.Index = tempregi;
+	fpi[i].DstReg.WriteMask = WRITEMASK_W;
+	fpi[i].DstReg.CondMask = COND_TR;
+
+	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
+	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
+	fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
+	i++;
+
+	fpi[i].Opcode = OPCODE_MUL;
+
+	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
+	fpi[i].DstReg.Index = tempregi;
+	fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
+	fpi[i].DstReg.CondMask = COND_TR;
+
+	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
+	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
+	fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+	fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
+	fpi[i].SrcReg[1].Index = tempregi;
+	fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
+	i++;
+
+	/* viewport transformation */
+	window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens);
+
+	fpi[i].Opcode = OPCODE_MAD;
+
+	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
+	fpi[i].DstReg.Index = tempregi;
+	fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
+	fpi[i].DstReg.CondMask = COND_TR;
+
+	fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
+	fpi[i].SrcReg[0].Index = tempregi;
+	fpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+	fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
+	fpi[i].SrcReg[1].Index = window_index;
+	fpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+	fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
+	fpi[i].SrcReg[2].Index = window_index;
+	fpi[i].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+	i++;
+
+	for (; i < compiler->program->NumInstructions; ++i) {
+		int reg;
+		for (reg = 0; reg < 3; reg++) {
+			if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT &&
+			    fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) {
+				fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY;
+				fpi[i].SrcReg[reg].Index = tempregi;
+			}
+		}
+	}
+}
+
+static GLuint build_dtm(GLuint depthmode)
+{
+	switch(depthmode) {
+	default:
+	case GL_LUMINANCE: return 0;
+	case GL_INTENSITY: return 1;
+	case GL_ALPHA: return 2;
+	}
+}
+
+static GLuint build_func(GLuint comparefunc)
+{
+	return comparefunc - GL_NEVER;
+}
+
+/**
+ * Collect all external state that is relevant for compiling the given
+ * fragment program.
+ */
+static void build_state(
+	r300ContextPtr r300,
+	struct r300_fragment_program *fp,
+	struct r300_fragment_program_external_state *state)
+{
+	int unit;
+
+	_mesa_bzero(state, sizeof(*state));
+
+	for(unit = 0; unit < 16; ++unit) {
+		if (fp->Base.Base.ShadowSamplers & (1 << unit)) {
+			struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
+
+			state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode);
+			state->unit[unit].texture_compare_func = build_func(tex->CompareFunc);
+		}
+	}
+}
+
+void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)fp;
+	struct r300_fragment_program_external_state state;
+
+	build_state(r300, r300_fp, &state);
+	if (_mesa_memcmp(&r300_fp->state, &state, sizeof(state))) {
+		/* TODO: cache compiled programs */
+		r300_fp->translated = GL_FALSE;
+		_mesa_memcpy(&r300_fp->state, &state, sizeof(state));
+	}
+
+	if (!r300_fp->translated) {
+		struct r300_fragment_program_compiler compiler;
+
+		compiler.r300 = r300;
+		compiler.fp = r300_fp;
+		compiler.code = &r300_fp->code;
+		compiler.program = _mesa_clone_program(ctx, &fp->Base);
+
+		if (RADEON_DEBUG & DEBUG_PIXEL) {
+			fflush(stdout);
+			_mesa_printf("Fragment Program: Initial program:\n");
+			_mesa_print_program(compiler.program);
+			fflush(stdout);
+		}
+
+		insert_WPOS_trailer(&compiler);
+
+		if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+			struct radeon_program_transformation transformations[] = {
+				{ &r500_transform_TEX, &compiler },
+				{ &radeonTransformALU, 0 },
+				{ &radeonTransformDeriv, 0 },
+				{ &radeonTransformTrigScale, 0 }
+			};
+			radeonLocalTransform(ctx, compiler.program, 4, transformations);
+		} else {
+			struct radeon_program_transformation transformations[] = {
+				{ &r300_transform_TEX, &compiler },
+				{ &radeonTransformALU, 0 },
+				{ &radeonTransformTrigSimple, 0 }
+			};
+			radeonLocalTransform(ctx, compiler.program, 3, transformations);
+		}
+
+		if (RADEON_DEBUG & DEBUG_PIXEL) {
+			_mesa_printf("Fragment Program: After native rewrite:\n");
+			_mesa_print_program(compiler.program);
+			fflush(stdout);
+		}
+
+		if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+			struct radeon_nqssadce_descr nqssadce = {
+				.Init = &nqssadce_init,
+				.IsNativeSwizzle = &r500FPIsNativeSwizzle,
+				.BuildSwizzle = &r500FPBuildSwizzle,
+				.RewriteDepthOut = GL_TRUE
+			};
+			radeonNqssaDce(ctx, compiler.program, &nqssadce);
+		} else {
+			struct radeon_nqssadce_descr nqssadce = {
+				.Init = &nqssadce_init,
+				.IsNativeSwizzle = &r300FPIsNativeSwizzle,
+				.BuildSwizzle = &r300FPBuildSwizzle,
+				.RewriteDepthOut = GL_TRUE
+			};
+			radeonNqssaDce(ctx, compiler.program, &nqssadce);
+		}
+
+		if (RADEON_DEBUG & DEBUG_PIXEL) {
+			_mesa_printf("Compiler: after NqSSA-DCE:\n");
+			_mesa_print_program(compiler.program);
+			fflush(stdout);
+		}
+
+		if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler))
+			r300_fp->error = GL_TRUE;
+
+		/* Subtle: Rescue any parameters that have been added during transformations */
+		_mesa_free_parameter_list(fp->Base.Parameters);
+		fp->Base.Parameters = compiler.program->Parameters;
+		compiler.program->Parameters = 0;
+
+		_mesa_reference_program(ctx, &compiler.program, NULL);
+
+		r300_fp->translated = GL_TRUE;
+
+		r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+
+		if (r300_fp->error || (RADEON_DEBUG & DEBUG_PIXEL))
+			r300->vtbl.FragmentProgramDump(&r300_fp->code);
+	}
+
+	update_params(ctx, fp);
+}
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.h b/src/mesa/drivers/dri/r300/r300_fragprog_common.h
new file mode 100644
index 00000000000..85ea86fecb1
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (C) 2009 Maciej Cencora <[email protected]>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __R300_FRAGPROG_COMMON_H_
+#define __R300_FRAGPROG_COMMON_H_
+
+#include "main/mtypes.h"
+
+extern void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp);
+
+#endif
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
index 9f0b7e35349..b75656e7ee1 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
@@ -47,7 +47,7 @@
 
 #define PROG_CODE \
 	struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \
-	struct r300_fragment_program_code *code = c->code
+	struct r300_fragment_program_code *code = &c->code->r300
 
 #define error(fmt, args...) do {			\
 		fprintf(stderr, "%s::%s(): " fmt "\n",	\
@@ -66,7 +66,7 @@ static GLboolean emit_const(void* data, GLuint file, GLuint index, GLuint *hwind
 	}
 
 	if (*hwindex >= code->const_nr) {
-		if (*hwindex >= PFS_NUM_CONST_REGS) {
+		if (*hwindex >= R300_PFS_NUM_CONST_REGS) {
 			error("Out of hw constants!\n");
 			return GL_FALSE;
 		}
@@ -138,7 +138,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst)
 {
 	PROG_CODE;
 
-	if (code->alu.length >= PFS_MAX_ALU_INST) {
+	if (code->alu.length >= R300_PFS_MAX_ALU_INST) {
 		error("Too many ALU instructions");
 		return GL_FALSE;
 	}
@@ -201,7 +201,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst)
 	if (inst->Alpha.DepthWriteMask) {
 		code->alu.inst[ip].inst3 |= R300_ALU_DSTA_DEPTH;
 		code->node[code->cur_node].flags |= R300_W_OUT;
-		c->fp->WritesDepth = GL_TRUE;
+		c->fp->writes_depth = GL_TRUE;
 	}
 
 	return GL_TRUE;
@@ -213,7 +213,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst)
  */
 static GLboolean finish_node(struct r300_fragment_program_compiler *c)
 {
-	struct r300_fragment_program_code *code = c->code;
+	struct r300_fragment_program_code *code = &c->code->r300;
 	struct r300_fragment_program_node *node = &code->node[code->cur_node];
 
 	if (node->alu_end < 0) {
@@ -275,7 +275,7 @@ static GLboolean emit_tex(void* data, struct prog_instruction* inst)
 {
 	PROG_CODE;
 
-	if (code->tex.length >= PFS_MAX_TEX_INST) {
+	if (code->tex.length >= R300_PFS_MAX_TEX_INST) {
 		error("Too many TEX instructions");
 		return GL_FALSE;
 	}
@@ -318,16 +318,16 @@ static const struct radeon_pair_handler pair_handler = {
 	.EmitPaired = &emit_alu,
 	.EmitTex = &emit_tex,
 	.BeginTexBlock = &begin_tex,
-	.MaxHwTemps = PFS_NUM_TEMP_REGS
+	.MaxHwTemps = R300_PFS_NUM_TEMP_REGS
 };
 
 /**
  * Final compilation step: Turn the intermediate radeon_program into
  * machine-readable instructions.
  */
-GLboolean r300FragmentProgramEmit(struct r300_fragment_program_compiler *compiler)
+GLboolean r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler)
 {
-	struct r300_fragment_program_code *code = compiler->code;
+	struct r300_fragment_program_code *code = &compiler->code->r300;
 
 	_mesa_bzero(code, sizeof(struct r300_fragment_program_code));
 	code->node[0].alu_end = -1;
diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c
index ee85e229f0b..104079b4dbe 100644
--- a/src/mesa/drivers/dri/r300/r300_ioctl.c
+++ b/src/mesa/drivers/dri/r300/r300_ioctl.c
@@ -46,8 +46,9 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/context.h"
 #include "swrast/swrast.h"
 
+#include "radeon_common.h"
+#include "radeon_lock.h"
 #include "r300_context.h"
-#include "radeon_ioctl.h"
 #include "r300_ioctl.h"
 #include "r300_cmdbuf.h"
 #include "r300_state.h"
@@ -55,71 +56,90 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_reg.h"
 #include "r300_emit.h"
 #include "r300_fragprog.h"
+#include "r300_context.h"
 
 #include "vblank.h"
 
+#define R200_3D_DRAW_IMMD_2      0xC0003500
+
 #define CLEARBUFFER_COLOR	0x1
 #define CLEARBUFFER_DEPTH	0x2
 #define CLEARBUFFER_STENCIL	0x4
 
-static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
+static void r300EmitClearState(GLcontext * ctx);
+
+static void r300UserClear(GLcontext *ctx, GLuint mask)
 {
+	radeon_clear_tris(ctx, mask);
+}
+
+static void r300ClearBuffer(r300ContextPtr r300, int flags,
+			    struct radeon_renderbuffer *rrb,
+			    struct radeon_renderbuffer *rrbd)
+{
+	BATCH_LOCALS(&r300->radeon);
 	GLcontext *ctx = r300->radeon.glCtx;
-	__DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable;
-	GLuint cboffset, cbpitch;
-	drm_r300_cmd_header_t *cmd2;
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-	drm_radeon_cmd_header_t *cmd = NULL;
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon);
+	GLuint cbpitch = 0;
 	r300ContextPtr rmesa = r300;
 
 	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s: %s buffer (%i,%i %ix%i)\n",
-			__FUNCTION__, buffer ? "back" : "front",
-			dPriv->x, dPriv->y, dPriv->w, dPriv->h);
-
-	if (buffer) {
-		cboffset = r300->radeon.radeonScreen->backOffset;
-		cbpitch = r300->radeon.radeonScreen->backPitch;
-	} else {
-		cboffset = r300->radeon.radeonScreen->frontOffset;
-		cbpitch = r300->radeon.radeonScreen->frontPitch;
+		fprintf(stderr, "%s: buffer %p (%i,%i %ix%i)\n",
+			__FUNCTION__, rrb, dPriv->x, dPriv->y,
+			dPriv->w, dPriv->h);
+
+	if (rrb) {
+		cbpitch = (rrb->pitch / rrb->cpp);
+		if (rrb->cpp == 4)
+			cbpitch |= R300_COLOR_FORMAT_ARGB8888;
+		else
+			cbpitch |= R300_COLOR_FORMAT_RGB565;
+
+		if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE){
+			cbpitch |= R300_COLOR_TILE_ENABLE;
+        }
 	}
 
-	cboffset += r300->radeon.radeonScreen->fbLocation;
-
-	cp_wait(r300, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
-	end_3d(rmesa);
-
-	R300_STATECHANGE(r300, cb);
-	reg_start(R300_RB3D_COLOROFFSET0, 0);
-	e32(cboffset);
-
-	if (r300->radeon.radeonScreen->cpp == 4)
-		cbpitch |= R300_COLOR_FORMAT_ARGB8888;
-	else
-		cbpitch |= R300_COLOR_FORMAT_RGB565;
-
-	if (r300->radeon.sarea->tiling_enabled)
-		cbpitch |= R300_COLOR_TILE_ENABLE;
-
-	reg_start(R300_RB3D_COLORPITCH0, 0);
-	e32(cbpitch);
-
-	R300_STATECHANGE(r300, cmk);
-	reg_start(RB3D_COLOR_CHANNEL_MASK, 0);
+	/* TODO in bufmgr */
+	cp_wait(&r300->radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
+	end_3d(&rmesa->radeon);
 
 	if (flags & CLEARBUFFER_COLOR) {
-		e32((ctx->Color.ColorMask[BCOMP] ? RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0 : 0) |
-		    (ctx->Color.ColorMask[GCOMP] ? RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0 : 0) |
-		    (ctx->Color.ColorMask[RCOMP] ? RB3D_COLOR_CHANNEL_MASK_RED_MASK0 : 0) |
-		    (ctx->Color.ColorMask[ACOMP] ? RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0 : 0));
+		assert(rrb != 0);
+		BEGIN_BATCH_NO_AUTOSTATE(6);
+		OUT_BATCH_REGSEQ(R300_RB3D_COLOROFFSET0, 1);
+		OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+		OUT_BATCH_REGVAL(R300_RB3D_COLORPITCH0, cbpitch);
+		END_BATCH();
+	}
+#if 1
+	if (flags & (CLEARBUFFER_DEPTH | CLEARBUFFER_STENCIL)) {
+		assert(rrbd != 0);
+		cbpitch = (rrbd->pitch / rrbd->cpp);
+		if (rrbd->bo->flags & RADEON_BO_FLAGS_MACRO_TILE){
+			cbpitch |= R300_DEPTHMACROTILE_ENABLE;
+        }
+		if (rrbd->bo->flags & RADEON_BO_FLAGS_MICRO_TILE){
+            cbpitch |= R300_DEPTHMICROTILE_TILED;
+        }
+		BEGIN_BATCH_NO_AUTOSTATE(6);
+		OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1);
+		OUT_BATCH_RELOC(0, rrbd->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+		OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, cbpitch);
+		END_BATCH();
+	}
+#endif
+	BEGIN_BATCH_NO_AUTOSTATE(6);
+	OUT_BATCH_REGSEQ(RB3D_COLOR_CHANNEL_MASK, 1);
+	if (flags & CLEARBUFFER_COLOR) {
+		OUT_BATCH((ctx->Color.ColorMask[BCOMP] ? RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0 : 0) |
+			  (ctx->Color.ColorMask[GCOMP] ? RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0 : 0) |
+			  (ctx->Color.ColorMask[RCOMP] ? RB3D_COLOR_CHANNEL_MASK_RED_MASK0 : 0) |
+			  (ctx->Color.ColorMask[ACOMP] ? RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0 : 0));
 	} else {
-		e32(0x0);
+		OUT_BATCH(0);
 	}
 
-	R300_STATECHANGE(r300, zs);
-	reg_start(R300_ZB_CNTL, 2);
 
 	{
 		uint32_t t1, t2;
@@ -146,73 +166,92 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 			     R300_S_FRONT_ZFAIL_OP_SHIFT);
 		}
 
-		e32(t1);
-		e32(t2);
-		e32(((ctx->Stencil.WriteMask[0] & R300_STENCILREF_MASK) << R300_STENCILWRITEMASK_SHIFT) |
-		    (ctx->Stencil.Clear & R300_STENCILREF_MASK));
+		OUT_BATCH_REGSEQ(R300_ZB_CNTL, 3);
+		OUT_BATCH(t1);
+		OUT_BATCH(t2);
+		OUT_BATCH(((ctx->Stencil.WriteMask[0] & R300_STENCILREF_MASK) <<
+                   R300_STENCILWRITEMASK_SHIFT) |
+			  (ctx->Stencil.Clear & R300_STENCILREF_MASK));
+		END_BATCH();
 	}
 
-	cmd2 = (drm_r300_cmd_header_t *) r300AllocCmdBuf(r300, 9, __FUNCTION__);
-	cmd2[0].packet3.cmd_type = R300_CMD_PACKET3;
-	cmd2[0].packet3.packet = R300_CMD_PACKET3_CLEAR;
-	cmd2[1].u = r300PackFloat32(dPriv->w / 2.0);
-	cmd2[2].u = r300PackFloat32(dPriv->h / 2.0);
-	cmd2[3].u = r300PackFloat32(ctx->Depth.Clear);
-	cmd2[4].u = r300PackFloat32(1.0);
-	cmd2[5].u = r300PackFloat32(ctx->Color.ClearColor[0]);
-	cmd2[6].u = r300PackFloat32(ctx->Color.ClearColor[1]);
-	cmd2[7].u = r300PackFloat32(ctx->Color.ClearColor[2]);
-	cmd2[8].u = r300PackFloat32(ctx->Color.ClearColor[3]);
+	if (!rmesa->radeon.radeonScreen->kernel_mm) {
+		BEGIN_BATCH_NO_AUTOSTATE(9);
+		OUT_BATCH(cmdpacket3(r300->radeon.radeonScreen, R300_CMD_PACKET3_CLEAR));
+		OUT_BATCH_FLOAT32(dPriv->w / 2.0);
+		OUT_BATCH_FLOAT32(dPriv->h / 2.0);
+		OUT_BATCH_FLOAT32(ctx->Depth.Clear);
+		OUT_BATCH_FLOAT32(1.0);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[0]);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[1]);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[2]);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[3]);
+		END_BATCH();
+	} else {
+		OUT_BATCH(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8));
+		OUT_BATCH(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
+			  (1 << R300_PRIM_NUM_VERTICES_SHIFT));
+		OUT_BATCH_FLOAT32(dPriv->w / 2.0);
+		OUT_BATCH_FLOAT32(dPriv->h / 2.0);
+		OUT_BATCH_FLOAT32(ctx->Depth.Clear);
+		OUT_BATCH_FLOAT32(1.0);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[0]);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[1]);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[2]);
+		OUT_BATCH_FLOAT32(ctx->Color.ClearColor[3]);
+	}
 
 	r300EmitCacheFlush(rmesa);
-	cp_wait(rmesa, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
+	cp_wait(&r300->radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
+
+	R300_STATECHANGE(r300, cb);
+	R300_STATECHANGE(r300, cmk);
+	R300_STATECHANGE(r300, zs);
 }
 
 static void r300EmitClearState(GLcontext * ctx)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
-	r300ContextPtr rmesa = r300;
-	__DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable;
+	BATCH_LOCALS(&r300->radeon);
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon);
 	int i;
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-	drm_radeon_cmd_header_t *cmd = NULL;
-	int has_tcl = 1;
+	int has_tcl;
 	int is_r500 = 0;
 	GLuint vap_cntl;
 
-	if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
-		has_tcl = 0;
-
-        if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
-                is_r500 = 1;
+	has_tcl = r300->options.hw_tcl_enabled;
 
+	if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
+		is_r500 = 1;
 
-	/* FIXME: the values written to R300_VAP_INPUT_ROUTE_0_0 and
-	 * R300_VAP_INPUT_ROUTE_0_1 are in fact known, however, the values are
-	 * quite complex; see the functions in r300_emit.c.
+	/* State atom dirty tracking is a little subtle here.
+	 *
+	 * On the one hand, we need to make sure base state is emitted
+	 * here if we start with an empty batch buffer, otherwise clear
+	 * works incorrectly with multiple processes. Therefore, the first
+	 * BEGIN_BATCH cannot be a BEGIN_BATCH_NO_AUTOSTATE.
 	 *
-	 * I believe it would be a good idea to extend the functions in
-	 * r300_emit.c so that they can be used to setup the default values for
-	 * these registers, as well as the actual values used for rendering.
+	 * On the other hand, implicit state emission clears the state atom
+	 * dirty bits, so we have to call R300_STATECHANGE later than the
+	 * first BEGIN_BATCH.
+	 *
+	 * The final trickiness is that, because we change state, we need
+	 * to ensure that any stored swtcl primitives are flushed properly
+	 * before we start changing state. See the R300_NEWPRIM in r300Clear
+	 * for this.
 	 */
-	R300_STATECHANGE(r300, vir[0]);
-	reg_start(R300_VAP_PROG_STREAM_CNTL_0, 0);
+	BEGIN_BATCH(31);
+	OUT_BATCH_REGSEQ(R300_VAP_PROG_STREAM_CNTL_0, 1);
 	if (!has_tcl)
-	    e32(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) |
+		OUT_BATCH(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) |
 		 ((R300_LAST_VEC | (2 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT)));
 	else
-	    e32(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) |
+		OUT_BATCH(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) |
 		 ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT)));
 
-	/* disable fog */
-	R300_STATECHANGE(r300, fogs);
-	reg_start(R300_FG_FOG_BLEND, 0);
-	e32(0x0);
-
-	R300_STATECHANGE(r300, vir[1]);
-	reg_start(R300_VAP_PROG_STREAM_CNTL_EXT_0, 0);
-	e32(((((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) |
+	OUT_BATCH_REGVAL(R300_FG_FOG_BLEND, 0);
+	OUT_BATCH_REGVAL(R300_VAP_PROG_STREAM_CNTL_EXT_0,
+	   ((((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) |
 	       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_SHIFT) |
 	       (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_SHIFT) |
 	       (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_SHIFT) |
@@ -226,619 +265,402 @@ static void r300EmitClearState(GLcontext * ctx)
 	      << R300_SWIZZLE1_SHIFT)));
 
 	/* R300_VAP_INPUT_CNTL_0, R300_VAP_INPUT_CNTL_1 */
-	R300_STATECHANGE(r300, vic);
-	reg_start(R300_VAP_VTX_STATE_CNTL, 1);
-	e32((R300_SEL_USER_COLOR_0 << R300_COLOR_0_ASSEMBLY_SHIFT));
-	e32(R300_INPUT_CNTL_POS | R300_INPUT_CNTL_COLOR | R300_INPUT_CNTL_TC0);
+	OUT_BATCH_REGSEQ(R300_VAP_VTX_STATE_CNTL, 2);
+	OUT_BATCH((R300_SEL_USER_COLOR_0 << R300_COLOR_0_ASSEMBLY_SHIFT));
+	OUT_BATCH(R300_INPUT_CNTL_POS | R300_INPUT_CNTL_COLOR | R300_INPUT_CNTL_TC0);
 
-	R300_STATECHANGE(r300, vte);
 	/* comes from fglrx startup of clear */
-	reg_start(R300_SE_VTE_CNTL, 1);
-	e32(R300_VTX_W0_FMT | R300_VPORT_X_SCALE_ENA |
-	    R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA |
-	    R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA |
-	    R300_VPORT_Z_OFFSET_ENA);
-	e32(0x8);
+	OUT_BATCH_REGSEQ(R300_SE_VTE_CNTL, 2);
+	OUT_BATCH(R300_VTX_W0_FMT | R300_VPORT_X_SCALE_ENA |
+		  R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA |
+		  R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA |
+		  R300_VPORT_Z_OFFSET_ENA);
+	OUT_BATCH(0x8);
 
-	reg_start(R300_VAP_PSC_SGN_NORM_CNTL, 0);
-	e32(0xaaaaaaaa);
+	OUT_BATCH_REGVAL(R300_VAP_PSC_SGN_NORM_CNTL, 0xaaaaaaaa);
 
-	R300_STATECHANGE(r300, vof);
-	reg_start(R300_VAP_OUTPUT_VTX_FMT_0, 1);
-	e32(R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT |
-	    R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT);
-	e32(0x0);		/* no textures */
+	OUT_BATCH_REGSEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
+	OUT_BATCH(R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT |
+		  R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT);
+	OUT_BATCH(0); /* no textures */
 
-	R300_STATECHANGE(r300, txe);
-	reg_start(R300_TX_ENABLE, 0);
-	e32(0x0);
+	OUT_BATCH_REGVAL(R300_TX_ENABLE, 0);
 
-	R300_STATECHANGE(r300, vpt);
-	reg_start(R300_SE_VPORT_XSCALE, 5);
-	efloat(1.0);
-	efloat(dPriv->x);
-	efloat(1.0);
-	efloat(dPriv->y);
-	efloat(1.0);
-	efloat(0.0);
+	OUT_BATCH_REGSEQ(R300_SE_VPORT_XSCALE, 6);
+	OUT_BATCH_FLOAT32(1.0);
+	OUT_BATCH_FLOAT32(dPriv->x);
+	OUT_BATCH_FLOAT32(1.0);
+	OUT_BATCH_FLOAT32(dPriv->y);
+	OUT_BATCH_FLOAT32(1.0);
+	OUT_BATCH_FLOAT32(0.0);
 
-	R300_STATECHANGE(r300, at);
-	reg_start(R300_FG_ALPHA_FUNC, 0);
-	e32(0x0);
+	OUT_BATCH_REGVAL(R300_FG_ALPHA_FUNC, 0);
+
+	OUT_BATCH_REGSEQ(R300_RB3D_CBLEND, 2);
+	OUT_BATCH(0x0);
+	OUT_BATCH(0x0);
+	END_BATCH();
 
+	R300_STATECHANGE(r300, vir[0]);
+	R300_STATECHANGE(r300, fogs);
+	R300_STATECHANGE(r300, vir[1]);
+	R300_STATECHANGE(r300, vic);
+	R300_STATECHANGE(r300, vte);
+	R300_STATECHANGE(r300, vof);
+	R300_STATECHANGE(r300, txe);
+	R300_STATECHANGE(r300, vpt);
+	R300_STATECHANGE(r300, at);
 	R300_STATECHANGE(r300, bld);
-	reg_start(R300_RB3D_CBLEND, 1);
-	e32(0x0);
-	e32(0x0);
+	R300_STATECHANGE(r300, ps);
 
 	if (has_tcl) {
-	    R300_STATECHANGE(r300, vap_clip_cntl);
-	    reg_start(R300_VAP_CLIP_CNTL, 0);
-	    e32(R300_PS_UCP_MODE_CLIP_AS_TRIFAN | R300_CLIP_DISABLE);
+		R300_STATECHANGE(r300, vap_clip_cntl);
+
+		BEGIN_BATCH_NO_AUTOSTATE(2);
+		OUT_BATCH_REGVAL(R300_VAP_CLIP_CNTL, R300_PS_UCP_MODE_CLIP_AS_TRIFAN | R300_CLIP_DISABLE);
+		END_BATCH();
         }
 
-	R300_STATECHANGE(r300, ps);
-	reg_start(R300_GA_POINT_SIZE, 0);
-	e32(((dPriv->w * 6) << R300_POINTSIZE_X_SHIFT) |
-	    ((dPriv->h * 6) << R300_POINTSIZE_Y_SHIFT));
+	BEGIN_BATCH_NO_AUTOSTATE(2);
+	OUT_BATCH_REGVAL(R300_GA_POINT_SIZE,
+		((dPriv->w * 6) << R300_POINTSIZE_X_SHIFT) |
+		((dPriv->h * 6) << R300_POINTSIZE_Y_SHIFT));
+	END_BATCH();
 
 	if (!is_r500) {
 		R300_STATECHANGE(r300, ri);
-		reg_start(R300_RS_IP_0, 7);
-		for (i = 0; i < 8; ++i) {
-			e32(R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3));
-		}
-
 		R300_STATECHANGE(r300, rc);
-		/* The second constant is needed to get glxgears display anything .. */
-		reg_start(R300_RS_COUNT, 1);
-		e32((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN);
-		e32(0x0);
-
 		R300_STATECHANGE(r300, rr);
-		reg_start(R300_RS_INST_0, 0);
-		e32(R300_RS_INST_COL_CN_WRITE);
+
+		BEGIN_BATCH(14);
+		OUT_BATCH_REGSEQ(R300_RS_IP_0, 8);
+		for (i = 0; i < 8; ++i)
+			OUT_BATCH(R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3));
+
+		OUT_BATCH_REGSEQ(R300_RS_COUNT, 2);
+		OUT_BATCH((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN);
+		OUT_BATCH(0x0);
+
+		OUT_BATCH_REGVAL(R300_RS_INST_0, R300_RS_INST_COL_CN_WRITE);
+		END_BATCH();
 	} else {
 		R300_STATECHANGE(r300, ri);
-		reg_start(R500_RS_IP_0, 7);
+		R300_STATECHANGE(r300, rc);
+		R300_STATECHANGE(r300, rr);
+
+		BEGIN_BATCH(14);
+		OUT_BATCH_REGSEQ(R500_RS_IP_0, 8);
 		for (i = 0; i < 8; ++i) {
-			e32((R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) |
-			    (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) |
-			    (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) |
-			    (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT));
+			OUT_BATCH((R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) |
+				  (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) |
+				  (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) |
+				  (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT));
 		}
 
-		R300_STATECHANGE(r300, rc);
-		/* The second constant is needed to get glxgears display anything .. */
-		reg_start(R300_RS_COUNT, 1);
-		e32((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN);
-		e32(0x0);
-
-		R300_STATECHANGE(r300, rr);
-		reg_start(R500_RS_INST_0, 0);
-		e32(R500_RS_INST_COL_CN_WRITE);
+		OUT_BATCH_REGSEQ(R300_RS_COUNT, 2);
+		OUT_BATCH((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN);
+		OUT_BATCH(0x0);
 
+		OUT_BATCH_REGVAL(R500_RS_INST_0, R500_RS_INST_COL_CN_WRITE);
+		END_BATCH();
 	}
 
 	if (!is_r500) {
 		R300_STATECHANGE(r300, fp);
-		reg_start(R300_US_CONFIG, 2);
-		e32(0x0);
-		e32(0x0);
-		e32(0x0);
-		reg_start(R300_US_CODE_ADDR_0, 3);
-		e32(0x0);
-		e32(0x0);
-		e32(0x0);
-		e32(R300_RGBA_OUT);
-
 		R300_STATECHANGE(r300, fpi[0]);
 		R300_STATECHANGE(r300, fpi[1]);
 		R300_STATECHANGE(r300, fpi[2]);
 		R300_STATECHANGE(r300, fpi[3]);
 
-		reg_start(R300_US_ALU_RGB_INST_0, 0);
-		e32(FP_INSTRC(MAD, FP_ARGC(SRC0C_XYZ), FP_ARGC(ONE), FP_ARGC(ZERO)));
-
-		reg_start(R300_US_ALU_RGB_ADDR_0, 0);
-		e32(FP_SELC(0, NO, XYZ, FP_TMP(0), 0, 0));
-
-		reg_start(R300_US_ALU_ALPHA_INST_0, 0);
-		e32(FP_INSTRA(MAD, FP_ARGA(SRC0A), FP_ARGA(ONE), FP_ARGA(ZERO)));
-
-		reg_start(R300_US_ALU_ALPHA_ADDR_0, 0);
-		e32(FP_SELA(0, NO, W, FP_TMP(0), 0, 0));
+		BEGIN_BATCH(17);
+		OUT_BATCH_REGSEQ(R300_US_CONFIG, 3);
+		OUT_BATCH(0x0);
+		OUT_BATCH(0x0);
+		OUT_BATCH(0x0);
+		OUT_BATCH_REGSEQ(R300_US_CODE_ADDR_0, 4);
+		OUT_BATCH(0x0);
+		OUT_BATCH(0x0);
+		OUT_BATCH(0x0);
+		OUT_BATCH(R300_RGBA_OUT);
+
+		OUT_BATCH_REGVAL(R300_US_ALU_RGB_INST_0,
+			FP_INSTRC(MAD, FP_ARGC(SRC0C_XYZ), FP_ARGC(ONE), FP_ARGC(ZERO)));
+		OUT_BATCH_REGVAL(R300_US_ALU_RGB_ADDR_0,
+			FP_SELC(0, NO, XYZ, FP_TMP(0), 0, 0));
+		OUT_BATCH_REGVAL(R300_US_ALU_ALPHA_INST_0,
+			FP_INSTRA(MAD, FP_ARGA(SRC0A), FP_ARGA(ONE), FP_ARGA(ZERO)));
+		OUT_BATCH_REGVAL(R300_US_ALU_ALPHA_ADDR_0,
+			FP_SELA(0, NO, W, FP_TMP(0), 0, 0));
+		END_BATCH();
 	} else {
- 		R300_STATECHANGE(r300, fp);
- 		reg_start(R500_US_CONFIG, 1);
- 		e32(R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
- 		e32(0x0);
- 		reg_start(R500_US_CODE_ADDR, 2);
- 		e32(R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(1));
- 		e32(R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(1));
- 		e32(R500_US_CODE_OFFSET_ADDR(0));
+		struct radeon_state_atom r500fp;
+		uint32_t _cmd[10];
 
+		R300_STATECHANGE(r300, fp);
 		R300_STATECHANGE(r300, r500fp);
-		r500fp_start_fragment(0, 6);
-
-		e32(R500_INST_TYPE_OUT |
-		    R500_INST_TEX_SEM_WAIT |
-		    R500_INST_LAST |
-		    R500_INST_RGB_OMASK_R |
-		    R500_INST_RGB_OMASK_G |
-		    R500_INST_RGB_OMASK_B |
-		    R500_INST_ALPHA_OMASK |
-		    R500_INST_RGB_CLAMP |
-		    R500_INST_ALPHA_CLAMP);
-
-		e32(R500_RGB_ADDR0(0) |
-		    R500_RGB_ADDR1(0) |
-		    R500_RGB_ADDR1_CONST |
-		    R500_RGB_ADDR2(0) |
-		    R500_RGB_ADDR2_CONST);
-
-		e32(R500_ALPHA_ADDR0(0) |
-		    R500_ALPHA_ADDR1(0) |
-		    R500_ALPHA_ADDR1_CONST |
-		    R500_ALPHA_ADDR2(0) |
-		    R500_ALPHA_ADDR2_CONST);
-
-		e32(R500_ALU_RGB_SEL_A_SRC0 |
-		    R500_ALU_RGB_R_SWIZ_A_R |
-		    R500_ALU_RGB_G_SWIZ_A_G |
-		    R500_ALU_RGB_B_SWIZ_A_B |
-		    R500_ALU_RGB_SEL_B_SRC0 |
-		    R500_ALU_RGB_R_SWIZ_B_R |
-		    R500_ALU_RGB_B_SWIZ_B_G |
-		    R500_ALU_RGB_G_SWIZ_B_B);
-
-		e32(R500_ALPHA_OP_CMP |
-		    R500_ALPHA_SWIZ_A_A |
-		    R500_ALPHA_SWIZ_B_A);
-
-		e32(R500_ALU_RGBA_OP_CMP |
-		    R500_ALU_RGBA_R_SWIZ_0 |
-		    R500_ALU_RGBA_G_SWIZ_0 |
-		    R500_ALU_RGBA_B_SWIZ_0 |
-		    R500_ALU_RGBA_A_SWIZ_0);
+
+		BEGIN_BATCH(7);
+		OUT_BATCH_REGSEQ(R500_US_CONFIG, 2);
+		OUT_BATCH(R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
+		OUT_BATCH(0x0);
+		OUT_BATCH_REGSEQ(R500_US_CODE_ADDR, 3);
+		OUT_BATCH(R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(1));
+		OUT_BATCH(R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(1));
+		OUT_BATCH(R500_US_CODE_OFFSET_ADDR(0));
+		END_BATCH();
+
+		r500fp.check = check_r500fp;
+		r500fp.cmd = _cmd;
+		r500fp.cmd[0] = cmdr500fp(r300->radeon.radeonScreen, 0, 1, 0, 0);
+		r500fp.cmd[1] = R500_INST_TYPE_OUT |
+			R500_INST_TEX_SEM_WAIT |
+			R500_INST_LAST |
+			R500_INST_RGB_OMASK_R |
+			R500_INST_RGB_OMASK_G |
+			R500_INST_RGB_OMASK_B |
+			R500_INST_ALPHA_OMASK |
+			R500_INST_RGB_CLAMP |
+			R500_INST_ALPHA_CLAMP;
+		r500fp.cmd[2] = R500_RGB_ADDR0(0) |
+			R500_RGB_ADDR1(0) |
+			R500_RGB_ADDR1_CONST |
+			R500_RGB_ADDR2(0) |
+			R500_RGB_ADDR2_CONST;
+		r500fp.cmd[3] = R500_ALPHA_ADDR0(0) |
+			R500_ALPHA_ADDR1(0) |
+			R500_ALPHA_ADDR1_CONST |
+			R500_ALPHA_ADDR2(0) |
+			R500_ALPHA_ADDR2_CONST;
+		r500fp.cmd[4] = R500_ALU_RGB_SEL_A_SRC0 |
+			R500_ALU_RGB_R_SWIZ_A_R |
+			R500_ALU_RGB_G_SWIZ_A_G |
+			R500_ALU_RGB_B_SWIZ_A_B |
+			R500_ALU_RGB_SEL_B_SRC0 |
+			R500_ALU_RGB_R_SWIZ_B_R |
+			R500_ALU_RGB_B_SWIZ_B_G |
+			R500_ALU_RGB_G_SWIZ_B_B;
+		r500fp.cmd[5] = R500_ALPHA_OP_CMP |
+			R500_ALPHA_SWIZ_A_A |
+			R500_ALPHA_SWIZ_B_A;
+		r500fp.cmd[6] = R500_ALU_RGBA_OP_CMP |
+			R500_ALU_RGBA_R_SWIZ_0 |
+			R500_ALU_RGBA_G_SWIZ_0 |
+			R500_ALU_RGBA_B_SWIZ_0 |
+			R500_ALU_RGBA_A_SWIZ_0;
+
+		r500fp.cmd[7] = 0;
+		emit_r500fp(ctx, &r500fp);
 	}
 
-	reg_start(R300_VAP_PVS_STATE_FLUSH_REG, 0);
-	e32(0x00000000);
+	BEGIN_BATCH(2);
+	OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0);
+	END_BATCH();
+
 	if (has_tcl) {
-	    vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
+		vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
 			(5 << R300_PVS_NUM_CNTLRS_SHIFT) |
 			(12 << R300_VF_MAX_VTX_NUM_SHIFT));
-	    if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
-		vap_cntl |= R500_TCL_STATE_OPTIMIZATION;
-	} else
-	    vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
+		if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
+			vap_cntl |= R500_TCL_STATE_OPTIMIZATION;
+	} else {
+		vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
 			(5 << R300_PVS_NUM_CNTLRS_SHIFT) |
 			(5 << R300_VF_MAX_VTX_NUM_SHIFT));
+	}
 
 	if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV515)
-	    vap_cntl |= (2 << R300_PVS_NUM_FPUS_SHIFT);
+		vap_cntl |= (2 << R300_PVS_NUM_FPUS_SHIFT);
 	else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) ||
 		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV560) ||
 		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV570))
-	    vap_cntl |= (5 << R300_PVS_NUM_FPUS_SHIFT);
+		vap_cntl |= (5 << R300_PVS_NUM_FPUS_SHIFT);
 	else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV410) ||
 		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R420))
-	    vap_cntl |= (6 << R300_PVS_NUM_FPUS_SHIFT);
+		vap_cntl |= (6 << R300_PVS_NUM_FPUS_SHIFT);
 	else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R520) ||
 		 (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R580))
-	    vap_cntl |= (8 << R300_PVS_NUM_FPUS_SHIFT);
+		vap_cntl |= (8 << R300_PVS_NUM_FPUS_SHIFT);
 	else
-	    vap_cntl |= (4 << R300_PVS_NUM_FPUS_SHIFT);
+		vap_cntl |= (4 << R300_PVS_NUM_FPUS_SHIFT);
+
+	R300_STATECHANGE(r300, vap_cntl);
 
-	R300_STATECHANGE(rmesa, vap_cntl);
-	reg_start(R300_VAP_CNTL, 0);
-	e32(vap_cntl);
+	BEGIN_BATCH(2);
+	OUT_BATCH_REGVAL(R300_VAP_CNTL, vap_cntl);
+	END_BATCH();
 
 	if (has_tcl) {
+        struct radeon_state_atom vpu;
+        uint32_t _cmd[10];
 		R300_STATECHANGE(r300, pvs);
-		reg_start(R300_VAP_PVS_CODE_CNTL_0, 2);
-
-		e32((0 << R300_PVS_FIRST_INST_SHIFT) |
-		    (0 << R300_PVS_XYZW_VALID_INST_SHIFT) |
-		    (1 << R300_PVS_LAST_INST_SHIFT));
-		e32((0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) |
-		    (0 << R300_PVS_MAX_CONST_ADDR_SHIFT));
-		e32(1 << R300_PVS_LAST_VTX_SRC_INST_SHIFT);
-
 		R300_STATECHANGE(r300, vpi);
-		vsf_start_fragment(0x0, 8);
 
-		e32(PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 0, 0xf, PVS_DST_REG_OUT));
-		e32(PVS_SRC_OPERAND(0, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, VSF_FLAG_NONE));
-		e32(PVS_SRC_OPERAND(0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_REG_INPUT, VSF_FLAG_NONE));
-		e32(0x0);
-
-		e32(PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 1, 0xf, PVS_DST_REG_OUT));
-		e32(PVS_SRC_OPERAND(1, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, VSF_FLAG_NONE));
-		e32(PVS_SRC_OPERAND(1, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_REG_INPUT, VSF_FLAG_NONE));
-		e32(0x0);
+		BEGIN_BATCH(4);
+		OUT_BATCH_REGSEQ(R300_VAP_PVS_CODE_CNTL_0, 3);
+		OUT_BATCH((0 << R300_PVS_FIRST_INST_SHIFT) |
+			  (0 << R300_PVS_XYZW_VALID_INST_SHIFT) |
+			  (1 << R300_PVS_LAST_INST_SHIFT));
+		OUT_BATCH((0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) |
+			  (0 << R300_PVS_MAX_CONST_ADDR_SHIFT));
+		OUT_BATCH(1 << R300_PVS_LAST_VTX_SRC_INST_SHIFT);
+		END_BATCH();
+
+		vpu.check = check_vpu;
+		vpu.cmd = _cmd;
+		vpu.cmd[0] = cmdvpu(r300->radeon.radeonScreen, 0, 2);
+
+		vpu.cmd[1] = PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE,
+                                         0, 0xf, PVS_DST_REG_OUT);
+		vpu.cmd[2] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y,
+                                      PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W,
+                                      PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
+		vpu.cmd[3] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
+		vpu.cmd[4] = 0x0;
+
+		vpu.cmd[5] = PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 1, 0xf,
+                                         PVS_DST_REG_OUT);
+		vpu.cmd[6] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_X,
+                                      PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z,
+                                      PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT,
+
+                                      VSF_FLAG_NONE);
+		vpu.cmd[7] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_SELECT_FORCE_0,
+                                      PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
+		vpu.cmd[8] = 0x0;
+
+		r300->vap_flush_needed = GL_TRUE;
+		emit_vpu(ctx, &vpu);
 	}
 }
 
-/**
- * Buffer clear
- */
-static void r300Clear(GLcontext * ctx, GLbitfield mask)
+static void r300KernelClear(GLcontext *ctx, GLuint flags)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
-	__DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable;
-	int flags = 0;
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon);
+	struct radeon_framebuffer *rfb = dPriv->driverPrivate;
+	struct radeon_renderbuffer *rrb;
+	struct radeon_renderbuffer *rrbd;
 	int bits = 0;
-	int swapped;
-
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "r300Clear\n");
-
-	{
-		LOCK_HARDWARE(&r300->radeon);
-		UNLOCK_HARDWARE(&r300->radeon);
-		if (dPriv->numClipRects == 0)
-			return;
-	}
 
-	if (mask & BUFFER_BIT_FRONT_LEFT) {
-		flags |= BUFFER_BIT_FRONT_LEFT;
-		mask &= ~BUFFER_BIT_FRONT_LEFT;
-	}
-
-	if (mask & BUFFER_BIT_BACK_LEFT) {
-		flags |= BUFFER_BIT_BACK_LEFT;
-		mask &= ~BUFFER_BIT_BACK_LEFT;
-	}
-
-	if (mask & BUFFER_BIT_DEPTH) {
+	/* Make sure it fits there. */
+	rcommonEnsureCmdBufSpace(&r300->radeon, 421 * 3, __FUNCTION__);
+	if (flags || bits)
+		r300EmitClearState(ctx);
+	rrbd = radeon_get_renderbuffer(&rfb->base, BUFFER_DEPTH);
+	if (rrbd && (flags & BUFFER_BIT_DEPTH))
 		bits |= CLEARBUFFER_DEPTH;
-		mask &= ~BUFFER_BIT_DEPTH;
-	}
 
-	if ((mask & BUFFER_BIT_STENCIL) && r300->state.stencil.hw_stencil) {
+	if (rrbd && (flags & BUFFER_BIT_STENCIL))
 		bits |= CLEARBUFFER_STENCIL;
-		mask &= ~BUFFER_BIT_STENCIL;
-	}
 
-	if (mask) {
-		if (RADEON_DEBUG & DEBUG_FALLBACKS)
-			fprintf(stderr, "%s: swrast clear, mask: %x\n",
-				__FUNCTION__, mask);
-		_swrast_Clear(ctx, mask);
+	if (flags & BUFFER_BIT_COLOR0) {
+		rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_COLOR0);
+		r300ClearBuffer(r300, CLEARBUFFER_COLOR, rrb, NULL);
+		bits = 0;
 	}
 
-	swapped = r300->radeon.sarea->pfCurrentPage == 1;
-
-	/* Make sure it fits there. */
-	r300EnsureCmdBufSpace(r300, 421 * 3, __FUNCTION__);
-	if (flags || bits)
-		r300EmitClearState(ctx);
-
 	if (flags & BUFFER_BIT_FRONT_LEFT) {
-		r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, swapped);
+		rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_FRONT_LEFT);
+		r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, rrb, rrbd);
 		bits = 0;
 	}
 
 	if (flags & BUFFER_BIT_BACK_LEFT) {
-		r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, swapped ^ 1);
+		rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_BACK_LEFT);
+		r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, rrb, rrbd);
 		bits = 0;
 	}
 
 	if (bits)
-		r300ClearBuffer(r300, bits, 0);
+		r300ClearBuffer(r300, bits, NULL, rrbd);
 
+	COMMIT_BATCH();
 }
 
-void r300Flush(GLcontext * ctx)
+/**
+ * Buffer clear
+ */
+static void r300Clear(GLcontext * ctx, GLbitfield mask)
 {
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon);
+	const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask);
+	GLbitfield swrast_mask = 0, tri_mask = 0;
+	int i;
+	struct gl_framebuffer *fb = ctx->DrawBuffer;
 
 	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	if (rmesa->dma.flush)
-		rmesa->dma.flush( rmesa );
-
-	if (rmesa->cmdbuf.count_used > rmesa->cmdbuf.count_reemit)
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-}
-
-#ifdef USER_BUFFERS
-#include "r300_mem.h"
-
-void r300RefillCurrentDmaRegion(r300ContextPtr rmesa, int size)
-{
-	struct r300_dma_buffer *dmabuf;
-	size = MAX2(size, RADEON_BUFFER_SIZE * 16);
-
-	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	if (rmesa->dma.flush) {
-		rmesa->dma.flush(rmesa);
-	}
+		fprintf(stderr, "r300Clear\n");
 
-	if (rmesa->dma.current.buf) {
-#ifdef USER_BUFFERS
-		r300_mem_use(rmesa, rmesa->dma.current.buf->id);
-#endif
-		r300ReleaseDmaRegion(rmesa, &rmesa->dma.current, __FUNCTION__);
+	if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) {
+		LOCK_HARDWARE(&r300->radeon);
+		UNLOCK_HARDWARE(&r300->radeon);
+		if (dPriv->numClipRects == 0)
+			return;
 	}
-	if (rmesa->dma.nr_released_bufs > 4)
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-
-	dmabuf = CALLOC_STRUCT(r300_dma_buffer);
-	dmabuf->buf = (void *)1;	/* hack */
-	dmabuf->refcount = 1;
 
-	dmabuf->id = r300_mem_alloc(rmesa, 4, size);
-	if (dmabuf->id == 0) {
-		LOCK_HARDWARE(&rmesa->radeon);	/* no need to validate */
-
-		r300FlushCmdBufLocked(rmesa, __FUNCTION__);
-		radeonWaitForIdleLocked(&rmesa->radeon);
+	/* Flush swtcl vertices if necessary, because we will change hardware
+	 * state during clear. See also the state-related comment in
+	 * r300EmitClearState.
+	 */
+	R300_NEWPRIM(r300);
 
-		dmabuf->id = r300_mem_alloc(rmesa, 4, size);
+	if (colorMask == ~0)
+	  tri_mask |= (mask & BUFFER_BITS_COLOR);
 
-		UNLOCK_HARDWARE(&rmesa->radeon);
 
-		if (dmabuf->id == 0) {
-			fprintf(stderr,
-				"Error: Could not get dma buffer... exiting\n");
-			_mesa_exit(-1);
-		}
+	/* HW stencil */
+	if (mask & BUFFER_BIT_STENCIL) {
+		tri_mask |= BUFFER_BIT_STENCIL;
 	}
 
-	rmesa->dma.current.buf = dmabuf;
-	rmesa->dma.current.address = r300_mem_ptr(rmesa, dmabuf->id);
-	rmesa->dma.current.end = size;
-	rmesa->dma.current.start = 0;
-	rmesa->dma.current.ptr = 0;
-}
-
-void r300ReleaseDmaRegion(r300ContextPtr rmesa,
-			  struct r300_dma_region *region, const char *caller)
-{
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s from %s\n", __FUNCTION__, caller);
-
-	if (!region->buf)
-		return;
-
-	if (rmesa->dma.flush)
-		rmesa->dma.flush(rmesa);
-
-	if (--region->buf->refcount == 0) {
-		r300_mem_free(rmesa, region->buf->id);
-		FREE(region->buf);
-		rmesa->dma.nr_released_bufs++;
+	/* HW depth */
+	if (mask & BUFFER_BIT_DEPTH) {
+    	        tri_mask |= BUFFER_BIT_DEPTH;
 	}
 
-	region->buf = 0;
-	region->start = 0;
-}
-
-/* Allocates a region from rmesa->dma.current.  If there isn't enough
- * space in current, grab a new buffer (and discard what was left of current)
- */
-void r300AllocDmaRegion(r300ContextPtr rmesa,
-			struct r300_dma_region *region,
-			int bytes, int alignment)
-{
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
-
-	if (rmesa->dma.flush)
-		rmesa->dma.flush(rmesa);
-
-	if (region->buf)
-		r300ReleaseDmaRegion(rmesa, region, __FUNCTION__);
-
-	alignment--;
-	rmesa->dma.current.start = rmesa->dma.current.ptr =
-	    (rmesa->dma.current.ptr + alignment) & ~alignment;
-
-	if (rmesa->dma.current.ptr + bytes > rmesa->dma.current.end)
-		r300RefillCurrentDmaRegion(rmesa, (bytes + 0x7) & ~0x7);
-
-	region->start = rmesa->dma.current.start;
-	region->ptr = rmesa->dma.current.start;
-	region->end = rmesa->dma.current.start + bytes;
-	region->address = rmesa->dma.current.address;
-	region->buf = rmesa->dma.current.buf;
-	region->buf->refcount++;
-
-	rmesa->dma.current.ptr += bytes;	/* bug - if alignment > 7 */
-	rmesa->dma.current.start =
-	    rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;
-
-	assert(rmesa->dma.current.ptr <= rmesa->dma.current.end);
-}
+	/* If we're doing a tri pass for depth/stencil, include a likely color
+	 * buffer with it.
+	 */
 
-#else
-static void r300RefillCurrentDmaRegion(r300ContextPtr rmesa)
-{
-	struct r300_dma_buffer *dmabuf;
-	int fd = rmesa->radeon.dri.fd;
-	int index = 0;
-	int size = 0;
-	drmDMAReq dma;
-	int ret;
-
-	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	if (rmesa->dma.flush) {
-		rmesa->dma.flush(rmesa);
+	for (i = 0; i < BUFFER_COUNT; i++) {
+	  GLuint bufBit = 1 << i;
+	  if ((tri_mask) & bufBit) {
+	    if (!fb->Attachment[i].Renderbuffer->ClassID) {
+	      tri_mask &= ~bufBit;
+	      swrast_mask |= bufBit;
+	    }
+	  }
 	}
 
-	if (rmesa->dma.current.buf)
-		r300ReleaseDmaRegion(rmesa, &rmesa->dma.current, __FUNCTION__);
-
-	if (rmesa->dma.nr_released_bufs > 4)
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-
-	dma.context = rmesa->radeon.dri.hwContext;
-	dma.send_count = 0;
-	dma.send_list = NULL;
-	dma.send_sizes = NULL;
-	dma.flags = 0;
-	dma.request_count = 1;
-	dma.request_size = RADEON_BUFFER_SIZE;
-	dma.request_list = &index;
-	dma.request_sizes = &size;
-	dma.granted_count = 0;
-
-	LOCK_HARDWARE(&rmesa->radeon);	/* no need to validate */
-
-	ret = drmDMA(fd, &dma);
-
-	if (ret != 0) {
-		/* Try to release some buffers and wait until we can't get any more */
-		if (rmesa->dma.nr_released_bufs) {
-			r300FlushCmdBufLocked(rmesa, __FUNCTION__);
-		}
-
-		if (RADEON_DEBUG & DEBUG_DMA)
-			fprintf(stderr, "Waiting for buffers\n");
-
-		radeonWaitForIdleLocked(&rmesa->radeon);
-		ret = drmDMA(fd, &dma);
+	/* SW fallback clearing */
+	swrast_mask = mask & ~tri_mask;
 
-		if (ret != 0) {
-			UNLOCK_HARDWARE(&rmesa->radeon);
-			fprintf(stderr,
-				"Error: Could not get dma buffer... exiting\n");
-			_mesa_exit(-1);
-		}
+	if (tri_mask) {
+		if (r300->radeon.radeonScreen->kernel_mm)
+			r300UserClear(ctx, tri_mask);
+		else
+			r300KernelClear(ctx, tri_mask);
 	}
-
-	UNLOCK_HARDWARE(&rmesa->radeon);
-
-	if (RADEON_DEBUG & DEBUG_DMA)
-		fprintf(stderr, "Allocated buffer %d\n", index);
-
-	dmabuf = CALLOC_STRUCT(r300_dma_buffer);
-	dmabuf->buf = &rmesa->radeon.radeonScreen->buffers->list[index];
-	dmabuf->refcount = 1;
-
-	rmesa->dma.current.buf = dmabuf;
-	rmesa->dma.current.address = dmabuf->buf->address;
-	rmesa->dma.current.end = dmabuf->buf->total;
-	rmesa->dma.current.start = 0;
-	rmesa->dma.current.ptr = 0;
-}
-
-void r300ReleaseDmaRegion(r300ContextPtr rmesa,
-			  struct r300_dma_region *region, const char *caller)
-{
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s from %s\n", __FUNCTION__, caller);
-
-	if (!region->buf)
-		return;
-
-	if (rmesa->dma.flush)
-		rmesa->dma.flush(rmesa);
-
-	if (--region->buf->refcount == 0) {
-		drm_radeon_cmd_header_t *cmd;
-
-		if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
-			fprintf(stderr, "%s -- DISCARD BUF %d\n",
-				__FUNCTION__, region->buf->buf->idx);
-		cmd =
-		    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa,
-								sizeof
-								(*cmd) / 4,
-								__FUNCTION__);
-		cmd->dma.cmd_type = R300_CMD_DMA_DISCARD;
-		cmd->dma.buf_idx = region->buf->buf->idx;
-
-		FREE(region->buf);
-		rmesa->dma.nr_released_bufs++;
+	if (swrast_mask) {
+		if (RADEON_DEBUG & DEBUG_FALLBACKS)
+			fprintf(stderr, "%s: swrast clear, mask: %x\n",
+				__FUNCTION__, swrast_mask);
+		_swrast_Clear(ctx, swrast_mask);
 	}
-
-	region->buf = 0;
-	region->start = 0;
-}
-
-/* Allocates a region from rmesa->dma.current.  If there isn't enough
- * space in current, grab a new buffer (and discard what was left of current)
- */
-void r300AllocDmaRegion(r300ContextPtr rmesa,
-			struct r300_dma_region *region,
-			int bytes, int alignment)
-{
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
-
-	if (rmesa->dma.flush)
-		rmesa->dma.flush(rmesa);
-
-	if (region->buf)
-		r300ReleaseDmaRegion(rmesa, region, __FUNCTION__);
-
-	alignment--;
-	rmesa->dma.current.start = rmesa->dma.current.ptr =
-	    (rmesa->dma.current.ptr + alignment) & ~alignment;
-
-	if (rmesa->dma.current.ptr + bytes > rmesa->dma.current.end)
-		r300RefillCurrentDmaRegion(rmesa);
-
-	region->start = rmesa->dma.current.start;
-	region->ptr = rmesa->dma.current.start;
-	region->end = rmesa->dma.current.start + bytes;
-	region->address = rmesa->dma.current.address;
-	region->buf = rmesa->dma.current.buf;
-	region->buf->refcount++;
-
-	rmesa->dma.current.ptr += bytes;	/* bug - if alignment > 7 */
-	rmesa->dma.current.start =
-	    rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;
-
-	assert(rmesa->dma.current.ptr <= rmesa->dma.current.end);
 }
 
-#endif
-
-GLboolean r300IsGartMemory(r300ContextPtr rmesa, const GLvoid * pointer,
-			   GLint size)
-{
-	int offset =
-	    (char *)pointer -
-	    (char *)rmesa->radeon.radeonScreen->gartTextures.map;
-	int valid = (size >= 0 && offset >= 0
-		     && offset + size <
-		     rmesa->radeon.radeonScreen->gartTextures.size);
-
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "r300IsGartMemory( %p ) : %d\n", pointer,
-			valid);
-
-	return valid;
-}
-
-GLuint r300GartOffsetFromVirtual(r300ContextPtr rmesa, const GLvoid * pointer)
-{
-	int offset =
-	    (char *)pointer -
-	    (char *)rmesa->radeon.radeonScreen->gartTextures.map;
-
-	//fprintf(stderr, "offset=%08x\n", offset);
-
-	if (offset < 0
-	    || offset > rmesa->radeon.radeonScreen->gartTextures.size)
-		return ~0;
-	else
-		return rmesa->radeon.radeonScreen->gart_texture_offset + offset;
-}
 
 void r300InitIoctlFuncs(struct dd_function_table *functions)
 {
 	functions->Clear = r300Clear;
 	functions->Finish = radeonFinish;
-	functions->Flush = r300Flush;
+	functions->Flush = radeonFlush;
 }
diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.h b/src/mesa/drivers/dri/r300/r300_ioctl.h
index e1143fb6c34..3abfa71a6e8 100644
--- a/src/mesa/drivers/dri/r300/r300_ioctl.h
+++ b/src/mesa/drivers/dri/r300/r300_ioctl.h
@@ -39,22 +39,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_context.h"
 #include "radeon_drm.h"
 
-extern GLboolean r300IsGartMemory(r300ContextPtr rmesa,
-				  const GLvoid * pointer, GLint size);
-
-extern GLuint r300GartOffsetFromVirtual(r300ContextPtr rmesa,
-					const GLvoid * pointer);
-
-extern void r300Flush(GLcontext * ctx);
-
-extern void r300ReleaseDmaRegion(r300ContextPtr rmesa,
-				 struct r300_dma_region *region,
-				 const char *caller);
-extern void r300AllocDmaRegion(r300ContextPtr rmesa,
-			       struct r300_dma_region *region, int bytes,
-			       int alignment);
-
 extern void r300InitIoctlFuncs(struct dd_function_table *functions);
 
-extern void r300RefillCurrentDmaRegion(r300ContextPtr rmesa, int size);
 #endif				/* __R300_IOCTL_H__ */
diff --git a/src/mesa/drivers/dri/r300/r300_mem.c b/src/mesa/drivers/dri/r300/r300_mem.c
deleted file mode 100644
index f8f9d4fcdf1..00000000000
--- a/src/mesa/drivers/dri/r300/r300_mem.c
+++ /dev/null
@@ -1,385 +0,0 @@
-/*
- * Copyright (C) 2005 Aapo Tahkola.
- *
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining
- * a copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sublicense, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial
- * portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
- * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
- * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
- * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
- * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- */
-
-/**
- * \file
- *
- * \author Aapo Tahkola <[email protected]>
- */
-
-#include <unistd.h>
-
-#include "r300_context.h"
-#include "r300_cmdbuf.h"
-#include "r300_ioctl.h"
-#include "r300_mem.h"
-#include "radeon_ioctl.h"
-
-#ifdef USER_BUFFERS
-
-static void resize_u_list(r300ContextPtr rmesa)
-{
-	void *temp;
-	int nsize;
-
-	temp = rmesa->rmm->u_list;
-	nsize = rmesa->rmm->u_size * 2;
-
-	rmesa->rmm->u_list = _mesa_malloc(nsize * sizeof(*rmesa->rmm->u_list));
-	_mesa_memset(rmesa->rmm->u_list, 0,
-		     nsize * sizeof(*rmesa->rmm->u_list));
-
-	if (temp) {
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-
-		_mesa_memcpy(rmesa->rmm->u_list, temp,
-			     rmesa->rmm->u_size * sizeof(*rmesa->rmm->u_list));
-		_mesa_free(temp);
-	}
-
-	rmesa->rmm->u_size = nsize;
-}
-
-void r300_mem_init(r300ContextPtr rmesa)
-{
-	rmesa->rmm = malloc(sizeof(struct r300_memory_manager));
-	memset(rmesa->rmm, 0, sizeof(struct r300_memory_manager));
-
-	rmesa->rmm->u_size = 128;
-	resize_u_list(rmesa);
-}
-
-void r300_mem_destroy(r300ContextPtr rmesa)
-{
-	_mesa_free(rmesa->rmm->u_list);
-	rmesa->rmm->u_list = NULL;
-
-	_mesa_free(rmesa->rmm);
-	rmesa->rmm = NULL;
-}
-
-void *r300_mem_ptr(r300ContextPtr rmesa, int id)
-{
-	assert(id <= rmesa->rmm->u_last);
-	return rmesa->rmm->u_list[id].ptr;
-}
-
-int r300_mem_find(r300ContextPtr rmesa, void *ptr)
-{
-	int i;
-
-	for (i = 1; i < rmesa->rmm->u_size + 1; i++)
-		if (rmesa->rmm->u_list[i].ptr &&
-		    ptr >= rmesa->rmm->u_list[i].ptr &&
-		    ptr <
-		    rmesa->rmm->u_list[i].ptr + rmesa->rmm->u_list[i].size)
-			break;
-
-	if (i < rmesa->rmm->u_size + 1)
-		return i;
-
-	fprintf(stderr, "%p failed\n", ptr);
-	return 0;
-}
-
-//#define MM_DEBUG
-int r300_mem_alloc(r300ContextPtr rmesa, int alignment, int size)
-{
-	drm_radeon_mem_alloc_t alloc;
-	int offset = 0, ret;
-	int i, free = -1;
-	int done_age;
-	drm_radeon_mem_free_t memfree;
-	int tries = 0;
-	static int bytes_wasted = 0, allocated = 0;
-
-	if (size < 4096)
-		bytes_wasted += 4096 - size;
-
-	allocated += size;
-
-#if 0
-	static int t = 0;
-	if (t != time(NULL)) {
-		t = time(NULL);
-		fprintf(stderr, "slots used %d, wasted %d kb, allocated %d\n",
-			rmesa->rmm->u_last, bytes_wasted / 1024,
-			allocated / 1024);
-	}
-#endif
-
-	memfree.region = RADEON_MEM_REGION_GART;
-
-      again:
-
-	done_age = radeonGetAge((radeonContextPtr) rmesa);
-
-	if (rmesa->rmm->u_last + 1 >= rmesa->rmm->u_size)
-		resize_u_list(rmesa);
-
-	for (i = rmesa->rmm->u_last + 1; i > 0; i--) {
-		if (rmesa->rmm->u_list[i].ptr == NULL) {
-			free = i;
-			continue;
-		}
-
-		if (rmesa->rmm->u_list[i].h_pending == 0 &&
-		    rmesa->rmm->u_list[i].pending
-		    && rmesa->rmm->u_list[i].age <= done_age) {
-			memfree.region_offset =
-			    (char *)rmesa->rmm->u_list[i].ptr -
-			    (char *)rmesa->radeon.radeonScreen->gartTextures.
-			    map;
-
-			ret =
-			    drmCommandWrite(rmesa->radeon.radeonScreen->
-					    driScreen->fd, DRM_RADEON_FREE,
-					    &memfree, sizeof(memfree));
-
-			if (ret) {
-				fprintf(stderr, "Failed to free at %p\n",
-					rmesa->rmm->u_list[i].ptr);
-				fprintf(stderr, "ret = %s\n", strerror(-ret));
-				exit(1);
-			} else {
-#ifdef MM_DEBUG
-				fprintf(stderr, "really freed %d at age %x\n",
-					i,
-					radeonGetAge((radeonContextPtr) rmesa));
-#endif
-				if (i == rmesa->rmm->u_last)
-					rmesa->rmm->u_last--;
-
-				if (rmesa->rmm->u_list[i].size < 4096)
-					bytes_wasted -=
-					    4096 - rmesa->rmm->u_list[i].size;
-
-				allocated -= rmesa->rmm->u_list[i].size;
-				rmesa->rmm->u_list[i].pending = 0;
-				rmesa->rmm->u_list[i].ptr = NULL;
-				free = i;
-			}
-		}
-	}
-	rmesa->rmm->u_head = i;
-
-	if (free == -1) {
-		WARN_ONCE("Ran out of slots!\n");
-		//usleep(100);
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-		tries++;
-		if (tries > 100) {
-			WARN_ONCE("Ran out of slots!\n");
-			exit(1);
-		}
-		goto again;
-	}
-
-	alloc.region = RADEON_MEM_REGION_GART;
-	alloc.alignment = alignment;
-	alloc.size = size;
-	alloc.region_offset = &offset;
-
-	ret =
-	    drmCommandWriteRead(rmesa->radeon.dri.fd, DRM_RADEON_ALLOC, &alloc,
-				sizeof(alloc));
-	if (ret) {
-#if 0
-		WARN_ONCE("Ran out of mem!\n");
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-		//usleep(100);
-		tries2++;
-		tries = 0;
-		if (tries2 > 100) {
-			WARN_ONCE("Ran out of GART memory!\n");
-			exit(1);
-		}
-		goto again;
-#else
-		WARN_ONCE
-		    ("Ran out of GART memory (for %d)!\nPlease consider adjusting GARTSize option.\n",
-		     size);
-		return 0;
-#endif
-	}
-
-	i = free;
-
-	if (i > rmesa->rmm->u_last)
-		rmesa->rmm->u_last = i;
-
-	rmesa->rmm->u_list[i].ptr =
-	    ((GLubyte *) rmesa->radeon.radeonScreen->gartTextures.map) + offset;
-	rmesa->rmm->u_list[i].size = size;
-	rmesa->rmm->u_list[i].age = 0;
-	//fprintf(stderr, "alloc %p at id %d\n", rmesa->rmm->u_list[i].ptr, i);
-
-#ifdef MM_DEBUG
-	fprintf(stderr, "allocated %d at age %x\n", i,
-		radeonGetAge((radeonContextPtr) rmesa));
-#endif
-
-	return i;
-}
-
-void r300_mem_use(r300ContextPtr rmesa, int id)
-{
-	uint64_t ull;
-#ifdef MM_DEBUG
-	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id,
-		radeonGetAge((radeonContextPtr) rmesa));
-#endif
-	drm_r300_cmd_header_t *cmd;
-
-	assert(id <= rmesa->rmm->u_last);
-
-	if (id == 0)
-		return;
-
-	cmd =
-	    (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa,
-						      2 + sizeof(ull) / 4,
-						      __FUNCTION__);
-	cmd[0].scratch.cmd_type = R300_CMD_SCRATCH;
-	cmd[0].scratch.reg = R300_MEM_SCRATCH;
-	cmd[0].scratch.n_bufs = 1;
-	cmd[0].scratch.flags = 0;
-	cmd++;
-
-	ull = (uint64_t) (intptr_t) & rmesa->rmm->u_list[id].age;
-	_mesa_memcpy(cmd, &ull, sizeof(ull));
-	cmd += sizeof(ull) / 4;
-
-	cmd[0].u = /*id */ 0;
-
-	LOCK_HARDWARE(&rmesa->radeon);	/* Protect from DRM. */
-	rmesa->rmm->u_list[id].h_pending++;
-	UNLOCK_HARDWARE(&rmesa->radeon);
-}
-
-unsigned long r300_mem_offset(r300ContextPtr rmesa, int id)
-{
-	unsigned long offset;
-
-	assert(id <= rmesa->rmm->u_last);
-
-	offset = (char *)rmesa->rmm->u_list[id].ptr -
-	    (char *)rmesa->radeon.radeonScreen->gartTextures.map;
-	offset += rmesa->radeon.radeonScreen->gart_texture_offset;
-
-	return offset;
-}
-
-void *r300_mem_map(r300ContextPtr rmesa, int id, int access)
-{
-#ifdef MM_DEBUG
-	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id,
-		radeonGetAge((radeonContextPtr) rmesa));
-#endif
-	void *ptr;
-	int tries = 0;
-
-	assert(id <= rmesa->rmm->u_last);
-
-	if (access == R300_MEM_R) {
-
-		if (rmesa->rmm->u_list[id].mapped == 1)
-			WARN_ONCE("buffer %d already mapped\n", id);
-
-		rmesa->rmm->u_list[id].mapped = 1;
-		ptr = r300_mem_ptr(rmesa, id);
-
-		return ptr;
-	}
-
-	if (rmesa->rmm->u_list[id].h_pending)
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
-
-	if (rmesa->rmm->u_list[id].h_pending) {
-		return NULL;
-	}
-
-	while (rmesa->rmm->u_list[id].age >
-	       radeonGetAge((radeonContextPtr) rmesa) && tries++ < 1000)
-		usleep(10);
-
-	if (tries >= 1000) {
-		fprintf(stderr, "Idling failed (%x vs %x)\n",
-			rmesa->rmm->u_list[id].age,
-			radeonGetAge((radeonContextPtr) rmesa));
-		return NULL;
-	}
-
-	if (rmesa->rmm->u_list[id].mapped == 1)
-		WARN_ONCE("buffer %d already mapped\n", id);
-
-	rmesa->rmm->u_list[id].mapped = 1;
-	ptr = r300_mem_ptr(rmesa, id);
-
-	return ptr;
-}
-
-void r300_mem_unmap(r300ContextPtr rmesa, int id)
-{
-#ifdef MM_DEBUG
-	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id,
-		radeonGetAge((radeonContextPtr) rmesa));
-#endif
-
-	assert(id <= rmesa->rmm->u_last);
-
-	if (rmesa->rmm->u_list[id].mapped == 0)
-		WARN_ONCE("buffer %d not mapped\n", id);
-
-	rmesa->rmm->u_list[id].mapped = 0;
-}
-
-void r300_mem_free(r300ContextPtr rmesa, int id)
-{
-#ifdef MM_DEBUG
-	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id,
-		radeonGetAge((radeonContextPtr) rmesa));
-#endif
-
-	assert(id <= rmesa->rmm->u_last);
-
-	if (id == 0)
-		return;
-
-	if (rmesa->rmm->u_list[id].ptr == NULL) {
-		WARN_ONCE("Not allocated!\n");
-		return;
-	}
-
-	if (rmesa->rmm->u_list[id].pending) {
-		WARN_ONCE("%p already pended!\n", rmesa->rmm->u_list[id].ptr);
-		return;
-	}
-
-	rmesa->rmm->u_list[id].pending = 1;
-}
-#endif
diff --git a/src/mesa/drivers/dri/r300/r300_mem.h b/src/mesa/drivers/dri/r300/r300_mem.h
deleted file mode 100644
index 625a7f6d8d5..00000000000
--- a/src/mesa/drivers/dri/r300/r300_mem.h
+++ /dev/null
@@ -1,37 +0,0 @@
-#ifndef __R300_MEM_H__
-#define __R300_MEM_H__
-
-//#define R300_MEM_PDL 0
-#define R300_MEM_UL 1
-
-#define R300_MEM_R 1
-#define R300_MEM_W 2
-#define R300_MEM_RW (R300_MEM_R | R300_MEM_W)
-
-#define R300_MEM_SCRATCH 2
-
-struct r300_memory_manager {
-	struct {
-		void *ptr;
-		uint32_t size;
-		uint32_t age;
-		uint32_t h_pending;
-		int pending;
-		int mapped;
-	} *u_list;
-	int u_head, u_size, u_last;
-
-};
-
-extern void r300_mem_init(r300ContextPtr rmesa);
-extern void r300_mem_destroy(r300ContextPtr rmesa);
-extern void *r300_mem_ptr(r300ContextPtr rmesa, int id);
-extern int r300_mem_find(r300ContextPtr rmesa, void *ptr);
-extern int r300_mem_alloc(r300ContextPtr rmesa, int alignment, int size);
-extern void r300_mem_use(r300ContextPtr rmesa, int id);
-extern unsigned long r300_mem_offset(r300ContextPtr rmesa, int id);
-extern void *r300_mem_map(r300ContextPtr rmesa, int id, int access);
-extern void r300_mem_unmap(r300ContextPtr rmesa, int id);
-extern void r300_mem_free(r300ContextPtr rmesa, int id);
-
-#endif
diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h
index 8f1a6630d5c..c22616b95f6 100644
--- a/src/mesa/drivers/dri/r300/r300_reg.h
+++ b/src/mesa/drivers/dri/r300/r300_reg.h
@@ -1531,6 +1531,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #	define R500_SEL_FILTER4_TC3		 (3 << 18)
 
 #define R300_TX_OFFSET_0                    0x4540
+#define R300_TX_OFFSET_1                    0x4544
+#define R300_TX_OFFSET_2                    0x4548
+#define R300_TX_OFFSET_3                    0x454C
+#define R300_TX_OFFSET_4                    0x4550
+#define R300_TX_OFFSET_5                    0x4554
+#define R300_TX_OFFSET_6                    0x4558
+#define R300_TX_OFFSET_7                    0x455C
 	/* BEGIN: Guess from R200 */
 #       define R300_TXO_ENDIAN_NO_SWAP           (0 << 0)
 #       define R300_TXO_ENDIAN_BYTE_SWAP         (1 << 0)
@@ -2425,6 +2432,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* Z Buffer Clear Value */
 #define R300_ZB_DEPTHCLEARVALUE                  0x4f28
 
+#define R300_ZB_ZMASK_OFFSET                     0x4f30
+#define R300_ZB_ZMASK_PITCH                      0x4f34
+#define R300_ZB_ZMASK_WRINDEX                    0x4f38
+#define R300_ZB_ZMASK_DWORD                      0x4f3c
+#define R300_ZB_ZMASK_RDINDEX                    0x4f40
+
 /* Hierarchical Z Memory Offset */
 #define R300_ZB_HIZ_OFFSET                       0x4f44
 
@@ -3165,6 +3178,9 @@ enum {
 #   define R300_W_SRC_RAS				(1 << 2)
 
 
+/* Packet0 field ordering to write all values to the same reg */
+#define RADEON_ONE_REG_WR        (1 << 15)
+
 /* Draw a primitive from vertex data in arrays loaded via 3D_LOAD_VBPNTR.
  * Two parameter dwords:
  * 0. VAP_VTX_FMT: The first parameter is not written to hardware
diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index 16ce4a11999..1356305a212 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -50,6 +50,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  * no bugs...
  */
 
+#include "r300_render.h"
+
 #include "main/glheader.h"
 #include "main/state.h"
 #include "main/imports.h"
@@ -66,16 +68,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/t_vp_build.h"
 #include "radeon_reg.h"
 #include "radeon_macros.h"
-#include "radeon_ioctl.h"
-#include "radeon_state.h"
 #include "r300_context.h"
 #include "r300_ioctl.h"
 #include "r300_state.h"
 #include "r300_reg.h"
 #include "r300_tex.h"
 #include "r300_emit.h"
-#include "r300_fragprog.h"
-extern int future_hw_tcl_on;
+#include "r300_fragprog_common.h"
+#include "r300_swtcl.h"
 
 /**
  * \brief Convert a OpenGL primitive type into a R300 primitive type.
@@ -172,96 +172,187 @@ int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
 	return num_verts - verts_off;
 }
 
-static void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts)
+static void r300EmitElts(GLcontext * ctx, unsigned long n_elts)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct r300_dma_region *rvb = &rmesa->state.elt_dma;
 	void *out;
+	GLuint size;
 
-	if (r300IsGartMemory(rmesa, elts, n_elts * 4)) {
-		rvb->address = rmesa->radeon.radeonScreen->gartTextures.map;
-		rvb->start = ((char *)elts) - rvb->address;
-		rvb->aos_offset =
-		    rmesa->radeon.radeonScreen->gart_texture_offset +
-		    rvb->start;
-		return;
-	} else if (r300IsGartMemory(rmesa, elts, 1)) {
-		WARN_ONCE("Pointer not within GART memory!\n");
-		_mesa_exit(-1);
-	}
+	size = ((rmesa->ind_buf.is_32bit ? 4 : 2) * n_elts + 3) & ~3;
 
-	r300AllocDmaRegion(rmesa, rvb, n_elts * 4, 4);
-	rvb->aos_offset = GET_START(rvb);
-
-	out = rvb->address + rvb->start;
-	memcpy(out, elts, n_elts * 4);
+	radeonAllocDmaRegion(&rmesa->radeon, &rmesa->radeon.tcl.elt_dma_bo,
+			     &rmesa->radeon.tcl.elt_dma_offset, size, 4);
+	radeon_bo_map(rmesa->radeon.tcl.elt_dma_bo, 1);
+	out = rmesa->radeon.tcl.elt_dma_bo->ptr + rmesa->radeon.tcl.elt_dma_offset;
+	memcpy(out, rmesa->ind_buf.ptr, size);
+	radeon_bo_unmap(rmesa->radeon.tcl.elt_dma_bo);
 }
 
-static void r300FireEB(r300ContextPtr rmesa, unsigned long addr,
-		       int vertex_count, int type)
+static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
 {
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-	drm_radeon_cmd_header_t *cmd = NULL;
-
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0), 0);
-	e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count << 16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
-
-	start_packet3(CP_PACKET3(R300_PACKET3_INDX_BUFFER, 2), 2);
-	e32(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
-	    (R300_VAP_PORT_IDX0 >> 2));
-	e32(addr);
-	e32(vertex_count);
+	BATCH_LOCALS(&rmesa->radeon);
+
+    r300_emit_scissor(rmesa->radeon.glCtx);
+	if (vertex_count > 0) {
+		int size;
+
+		BEGIN_BATCH(10);
+		OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0);
+		if (rmesa->ind_buf.is_32bit) {
+			size = vertex_count;
+			OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
+			  ((vertex_count + 0) << 16) | type |
+			  R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+		} else {
+			size = (vertex_count + 1) >> 1;
+			OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
+			   ((vertex_count + 0) << 16) | type);
+		}
+
+		if (!rmesa->radeon.radeonScreen->kernel_mm) {
+			OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
+			OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
+	    			 (R300_VAP_PORT_IDX0 >> 2));
+			OUT_BATCH_RELOC(rmesa->radeon.tcl.elt_dma_offset,
+					rmesa->radeon.tcl.elt_dma_bo,
+					rmesa->radeon.tcl.elt_dma_offset,
+					RADEON_GEM_DOMAIN_GTT, 0, 0);
+			OUT_BATCH(size);
+		} else {
+			OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2);
+			OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) |
+	    			 (R300_VAP_PORT_IDX0 >> 2));
+			OUT_BATCH(rmesa->radeon.tcl.elt_dma_offset);
+			OUT_BATCH(size);
+			radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+					      rmesa->radeon.tcl.elt_dma_bo,
+					      RADEON_GEM_DOMAIN_GTT, 0, 0);
+		}
+		END_BATCH();
+	}
 }
 
 static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
 {
+	BATCH_LOCALS(&rmesa->radeon);
+	uint32_t voffset;
 	int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2;
 	int i;
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-	drm_radeon_cmd_header_t *cmd = NULL;
 
 	if (RADEON_DEBUG & DEBUG_VERTS)
 		fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__, nr,
 			offset);
 
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1), sz - 1);
-	e32(nr);
 
-	for (i = 0; i + 1 < nr; i += 2) {
-		e32((rmesa->state.aos[i].aos_size << 0) |
-		    (rmesa->state.aos[i].aos_stride << 8) |
-		    (rmesa->state.aos[i + 1].aos_size << 16) |
-		    (rmesa->state.aos[i + 1].aos_stride << 24));
+	if (!rmesa->radeon.radeonScreen->kernel_mm) {
+		BEGIN_BATCH(sz+2+(nr * 2));
+		OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1);
+		OUT_BATCH(nr);
+
+		for (i = 0; i + 1 < nr; i += 2) {
+			OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) |
+				  (rmesa->radeon.tcl.aos[i].stride << 8) |
+				  (rmesa->radeon.tcl.aos[i + 1].components << 16) |
+				  (rmesa->radeon.tcl.aos[i + 1].stride << 24));
+
+			voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+			OUT_BATCH_RELOC(voffset,
+					rmesa->radeon.tcl.aos[i].bo,
+					voffset,
+					RADEON_GEM_DOMAIN_GTT,
+					0, 0);
+			voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+			  offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+			OUT_BATCH_RELOC(voffset,
+					rmesa->radeon.tcl.aos[i+1].bo,
+					voffset,
+					RADEON_GEM_DOMAIN_GTT,
+					0, 0);
+		}
 
-		e32(rmesa->state.aos[i].aos_offset + offset * 4 * rmesa->state.aos[i].aos_stride);
-		e32(rmesa->state.aos[i + 1].aos_offset + offset * 4 * rmesa->state.aos[i + 1].aos_stride);
-	}
+		if (nr & 1) {
+			OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) |
+				  (rmesa->radeon.tcl.aos[nr - 1].stride << 8));
+			voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+			OUT_BATCH_RELOC(voffset,
+					rmesa->radeon.tcl.aos[nr - 1].bo,
+					voffset,
+					RADEON_GEM_DOMAIN_GTT,
+					0, 0);
+		}
+		END_BATCH();
+	} else {
+
+		BEGIN_BATCH(sz+2+(nr * 2));
+		OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1);
+		OUT_BATCH(nr);
+
+		for (i = 0; i + 1 < nr; i += 2) {
+			OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) |
+				  (rmesa->radeon.tcl.aos[i].stride << 8) |
+				  (rmesa->radeon.tcl.aos[i + 1].components << 16) |
+				  (rmesa->radeon.tcl.aos[i + 1].stride << 24));
+
+			voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+			OUT_BATCH(voffset);
+			voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+			OUT_BATCH(voffset);
+		}
 
-	if (nr & 1) {
-		e32((rmesa->state.aos[nr - 1].aos_size << 0) |
-		    (rmesa->state.aos[nr - 1].aos_stride << 8));
-		e32(rmesa->state.aos[nr - 1].aos_offset + offset * 4 * rmesa->state.aos[nr - 1].aos_stride);
+		if (nr & 1) {
+			OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) |
+			  (rmesa->radeon.tcl.aos[nr - 1].stride << 8));
+			voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+			OUT_BATCH(voffset);
+		}
+		for (i = 0; i + 1 < nr; i += 2) {
+			voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+			radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+					      rmesa->radeon.tcl.aos[i+0].bo,
+					      RADEON_GEM_DOMAIN_GTT,
+					      0, 0);
+			voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+			radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+					      rmesa->radeon.tcl.aos[i+1].bo,
+					      RADEON_GEM_DOMAIN_GTT,
+					      0, 0);
+		}
+		if (nr & 1) {
+			voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+				offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+			radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+					      rmesa->radeon.tcl.aos[nr-1].bo,
+					      RADEON_GEM_DOMAIN_GTT,
+					      0, 0);
+		}
+		END_BATCH();
 	}
+
 }
 
 static void r300FireAOS(r300ContextPtr rmesa, int vertex_count, int type)
 {
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-	drm_radeon_cmd_header_t *cmd = NULL;
+	BATCH_LOCALS(&rmesa->radeon);
 
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0);
-	e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type);
+    r300_emit_scissor(rmesa->radeon.glCtx);
+	BEGIN_BATCH(3);
+	OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
+	OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type);
+	END_BATCH();
 }
 
-static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
-				   int start, int end, int prim)
+void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim)
 {
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	BATCH_LOCALS(&rmesa->radeon);
 	int type, num_verts;
-	TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *vb = &tnl->vb;
 
 	type = r300PrimitiveType(rmesa, prim);
 	num_verts = r300NumVerts(rmesa, end - start, prim);
@@ -269,7 +360,13 @@ static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
 	if (type < 0 || num_verts <= 0)
 		return;
 
-	if (vb->Elts) {
+	/* Make space for at least 64 dwords.
+	 * This is supposed to ensure that we can get all rendering
+	 * commands into a single command buffer.
+	 */
+	rcommonEnsureCmdBufSpace(&rmesa->radeon, 128, __FUNCTION__);
+
+	if (rmesa->ind_buf.ptr) {
 		if (num_verts > 65535) {
 			/* not implemented yet */
 			WARN_ONCE("Too many elts\n");
@@ -286,114 +383,142 @@ static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
 		 * allocating the index array might actually evict the vertex
 		 * arrays. *sigh*
 		 */
-		r300EmitElts(ctx, vb->Elts, num_verts);
-		r300EmitAOS(rmesa, rmesa->state.aos_count, start);
-		r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset, num_verts, type);
+		r300EmitElts(ctx, num_verts);
+		r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, start);
+		if (rmesa->radeon.radeonScreen->kernel_mm) {
+			BEGIN_BATCH_NO_AUTOSTATE(2);
+			OUT_BATCH_REGSEQ(R300_VAP_VF_MAX_VTX_INDX, 1);
+			OUT_BATCH(rmesa->radeon.tcl.aos[0].count);
+			END_BATCH();
+		}
+		r300FireEB(rmesa, num_verts, type);
 	} else {
-		r300EmitAOS(rmesa, rmesa->state.aos_count, start);
+		r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, start);
 		r300FireAOS(rmesa, num_verts, type);
 	}
+	COMMIT_BATCH();
 }
 
-static GLboolean r300RunRender(GLcontext * ctx,
-			       struct tnl_pipeline_stage *stage)
+static void r300RunRender(GLcontext * ctx, struct tnl_pipeline_stage *stage)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	int i;
 	TNLcontext *tnl = TNL_CONTEXT(ctx);
 	struct vertex_buffer *vb = &tnl->vb;
 
-
 	if (RADEON_DEBUG & DEBUG_PRIMS)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
 	r300UpdateShaders(rmesa);
-	if (r300EmitArrays(ctx))
-		return GL_TRUE;
+	r300EmitArrays(ctx);
 
 	r300UpdateShaderStates(rmesa);
 
 	r300EmitCacheFlush(rmesa);
-	r300EmitState(rmesa);
+	radeonEmitState(&rmesa->radeon);
 
 	for (i = 0; i < vb->PrimitiveCount; i++) {
 		GLuint prim = _tnl_translate_prim(&vb->Primitive[i]);
 		GLuint start = vb->Primitive[i].start;
 		GLuint end = vb->Primitive[i].start + vb->Primitive[i].count;
-		r300RunRenderPrimitive(rmesa, ctx, start, end, prim);
+		r300RunRenderPrimitive(ctx, start, end, prim);
 	}
 
 	r300EmitCacheFlush(rmesa);
 
-#ifdef USER_BUFFERS
-	r300UseArrays(ctx);
-#endif
+	radeonReleaseArrays(ctx, ~0);
+}
 
-	r300ReleaseArrays(ctx);
 
-	return GL_FALSE;
+static const char *getFallbackString(uint32_t bit)
+{
+	switch (bit) {
+		case R300_FALLBACK_VERTEX_PROGRAM :
+			return "vertex program";
+		case R300_FALLBACK_LINE_SMOOTH:
+			return "smooth lines";
+		case R300_FALLBACK_POINT_SMOOTH:
+			return "smooth points";
+		case R300_FALLBACK_POLYGON_SMOOTH:
+			return "smooth polygons";
+		case R300_FALLBACK_LINE_STIPPLE:
+			return "line stipple";
+		case R300_FALLBACK_POLYGON_STIPPLE:
+			return "polygon stipple";
+		case R300_FALLBACK_STENCIL_TWOSIDE:
+			return "two-sided stencil";
+		case R300_FALLBACK_RENDER_MODE:
+			return "render mode != GL_RENDER";
+		case R300_FALLBACK_FRAGMENT_PROGRAM:
+			return "fragment program";
+		case R300_FALLBACK_AOS_LIMIT:
+			return "aos limit";
+		case R300_FALLBACK_INVALID_BUFFERS:
+			return "invalid buffers";
+		default:
+			return "unknown";
+	}
 }
 
-#define FALLBACK_IF(expr)						\
-	do {								\
-		if (expr) {						\
-			if (1 || RADEON_DEBUG & DEBUG_FALLBACKS)	\
-				WARN_ONCE("Software fallback:%s\n",	\
-					  #expr);			\
-			return R300_FALLBACK_RAST;			\
-		}							\
-	} while(0)
-
-static int r300Fallback(GLcontext * ctx)
+void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode)
 {
-	r300ContextPtr r300 = R300_CONTEXT(ctx);
-	const unsigned back = ctx->Stencil._BackFace;
-
-	/* Do we need to use new-style shaders?
-	 * Also is there a better way to do this? */
-	if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
-		struct r500_fragment_program *fp = (struct r500_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
-		if (fp) {
-			if (!fp->translated) {
-				r500TranslateFragmentShader(r300, fp);
-				FALLBACK_IF(!fp->translated);
+	TNLcontext *tnl = TNL_CONTEXT(ctx);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	uint32_t old_fallback = rmesa->fallback;
+	static uint32_t fallback_warn = 0;
+	
+	if (mode) {
+		if ((fallback_warn & bit) == 0) {
+			_mesa_fprintf(stderr, "WARNING! Falling back to software for %s\n", getFallbackString(bit));
+			fallback_warn |= bit;
+		}
+		rmesa->fallback |= bit;
+
+		/* update only if we change from no tcl fallbacks to some tcl fallbacks */
+		if (rmesa->options.hw_tcl_enabled) {
+			if (((old_fallback & R300_TCL_FALLBACK_MASK) == 0) &&
+				((bit & R300_TCL_FALLBACK_MASK) > 0)) {
+				R300_STATECHANGE(rmesa, vap_cntl_status);
+				rmesa->hw.vap_cntl_status.cmd[1] |= R300_VAP_TCL_BYPASS;
 			}
 		}
+
+		/* update only if we change from no raster fallbacks to some raster fallbacks */
+		if (((old_fallback & R300_RASTER_FALLBACK_MASK) == 0) &&
+			((bit & R300_RASTER_FALLBACK_MASK) > 0)) {
+			
+			radeon_firevertices(&rmesa->radeon);
+			rmesa->radeon.swtcl.RenderIndex = ~0;
+			_swsetup_Wakeup( ctx );
+		}
 	} else {
-		struct r300_fragment_program *fp = (struct r300_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
-		if (fp) {
-			if (!fp->translated) {
-				r300TranslateFragmentShader(r300, fp);
-				FALLBACK_IF(!fp->translated);
+		rmesa->fallback &= ~bit;
+
+		/* update only if we have disabled all tcl fallbacks */
+		if (rmesa->options.hw_tcl_enabled) {
+			if ((old_fallback & R300_RASTER_FALLBACK_MASK) == bit) {
+				R300_STATECHANGE(rmesa, vap_cntl_status);
+				rmesa->hw.vap_cntl_status.cmd[1] &= ~R300_VAP_TCL_BYPASS;
 			}
 		}
-	}
 
-	FALLBACK_IF(ctx->RenderMode != GL_RENDER);
-
-	/* If GL_EXT_stencil_two_side is disabled, this fallback check can
-	 * be removed.
-	 */
-	FALLBACK_IF(ctx->Stencil.Ref[0] != ctx->Stencil.Ref[back]
-		    || ctx->Stencil.ValueMask[0] !=
-		    ctx->Stencil.ValueMask[back]
-		    || ctx->Stencil.WriteMask[0] !=
-		    ctx->Stencil.WriteMask[back]);
-
-	if (ctx->Extensions.NV_point_sprite || ctx->Extensions.ARB_point_sprite)
-		FALLBACK_IF(ctx->Point.PointSprite);
-
-	if (!r300->disable_lowimpact_fallback) {
-		FALLBACK_IF(ctx->Polygon.StippleFlag);
-		FALLBACK_IF(ctx->Multisample._Enabled);
-		FALLBACK_IF(ctx->Line.StippleFlag);
-		FALLBACK_IF(ctx->Line.SmoothFlag);
-		FALLBACK_IF(ctx->Point.SmoothFlag);
+		/* update only if we have disabled all raster fallbacks */
+		if ((old_fallback & R300_RASTER_FALLBACK_MASK) == bit) {
+			_swrast_flush( ctx );
+			
+			tnl->Driver.Render.Start = r300RenderStart;
+			tnl->Driver.Render.Finish = r300RenderFinish;
+			tnl->Driver.Render.PrimitiveNotify = r300RenderPrimitive;
+			tnl->Driver.Render.ResetLineStipple = r300ResetLineStipple;
+			tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
+			tnl->Driver.Render.CopyPV = _tnl_copy_pv;
+			tnl->Driver.Render.Interp = _tnl_interp;
+			
+			_tnl_invalidate_vertex_state( ctx, ~0 );
+			_tnl_invalidate_vertices( ctx, ~0 );
+		}
 	}
-
-	return R300_FALLBACK_NONE;
+	
 }
 
 static GLboolean r300RunNonTCLRender(GLcontext * ctx,
@@ -404,43 +529,15 @@ static GLboolean r300RunNonTCLRender(GLcontext * ctx,
 	if (RADEON_DEBUG & DEBUG_PRIMS)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
-	if (r300Fallback(ctx) >= R300_FALLBACK_RAST)
-		return GL_TRUE;
-
-	if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
- 	        return GL_TRUE;
-
-	return r300RunRender(ctx, stage);
-}
-
-static GLboolean r300RunTCLRender(GLcontext * ctx,
-				  struct tnl_pipeline_stage *stage)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct r300_vertex_program *vp;
-
-	hw_tcl_on = future_hw_tcl_on;
-
-	if (RADEON_DEBUG & DEBUG_PRIMS)
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	if (hw_tcl_on == GL_FALSE)
+	if (rmesa->fallback & R300_RASTER_FALLBACK_MASK)
 		return GL_TRUE;
 
-	if (r300Fallback(ctx) >= R300_FALLBACK_TCL) {
-		hw_tcl_on = GL_FALSE;
+	if (rmesa->options.hw_tcl_enabled == GL_FALSE)
 		return GL_TRUE;
-	}
-
-	r300UpdateShaders(rmesa);
 
-	vp = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
-	if (vp->native == GL_FALSE) {
-		hw_tcl_on = GL_FALSE;
-		return GL_TRUE;
-	}
+	r300RunRender(ctx, stage);
 
-	return r300RunRender(ctx, stage);
+	return GL_FALSE;
 }
 
 const struct tnl_pipeline_stage _r300_render_stage = {
@@ -451,12 +548,3 @@ const struct tnl_pipeline_stage _r300_render_stage = {
 	NULL,
 	r300RunNonTCLRender
 };
-
-const struct tnl_pipeline_stage _r300_tcl_stage = {
-	"r300 Hardware Transform, Clipping and Lighting",
-	NULL,
-	NULL,
-	NULL,
-	NULL,
-	r300RunTCLRender
-};
diff --git a/src/mesa/drivers/dri/r300/r300_render.h b/src/mesa/drivers/dri/r300/r300_render.h
new file mode 100644
index 00000000000..ec785474a67
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_render.h
@@ -0,0 +1,69 @@
+/*
+ * Copyright 2009 Maciej Cencora <[email protected]>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __R300_RENDER_H__
+#define __R300_RENDER_H__
+
+#include "main/mtypes.h"
+
+#define R300_FALLBACK_VERTEX_PROGRAM    (1 << 0)
+#define R300_TCL_FALLBACK_MASK           0x0000ffff
+
+#define R300_FALLBACK_LINE_SMOOTH       (1 << 16)
+#define R300_FALLBACK_POINT_SMOOTH      (1 << 17)
+#define R300_FALLBACK_POLYGON_SMOOTH    (1 << 18)
+#define R300_FALLBACK_LINE_STIPPLE      (1 << 19)
+#define R300_FALLBACK_POLYGON_STIPPLE   (1 << 20)
+#define R300_FALLBACK_STENCIL_TWOSIDE   (1 << 21)
+#define R300_FALLBACK_RENDER_MODE       (1 << 22)
+#define R300_FALLBACK_FRAGMENT_PROGRAM  (1 << 23)
+#define R300_FALLBACK_AOS_LIMIT         (1 << 30)
+#define R300_FALLBACK_INVALID_BUFFERS   (1 << 31)
+#define R300_RASTER_FALLBACK_MASK        0xffff0000
+
+#define MASK_XYZW (R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
+#define MASK_X R300_WRITE_ENA_X
+#define MASK_Y R300_WRITE_ENA_Y
+#define MASK_Z R300_WRITE_ENA_Z
+#define MASK_W R300_WRITE_ENA_W
+
+#if SWIZZLE_X != R300_INPUT_ROUTE_SELECT_X || \
+    SWIZZLE_Y != R300_INPUT_ROUTE_SELECT_Y || \
+    SWIZZLE_Z != R300_INPUT_ROUTE_SELECT_Z || \
+    SWIZZLE_W != R300_INPUT_ROUTE_SELECT_W || \
+    SWIZZLE_ZERO != R300_INPUT_ROUTE_SELECT_ZERO || \
+    SWIZZLE_ONE != R300_INPUT_ROUTE_SELECT_ONE
+#error Cannot change these!
+#endif
+
+extern const struct tnl_pipeline_stage _r300_render_stage;
+
+extern void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode);
+
+extern void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim);
+
+#endif
diff --git a/src/mesa/drivers/dri/r300/r300_shader.c b/src/mesa/drivers/dri/r300/r300_shader.c
index f30fd986e0f..0133b837966 100644
--- a/src/mesa/drivers/dri/r300/r300_shader.c
+++ b/src/mesa/drivers/dri/r300/r300_shader.c
@@ -1,18 +1,42 @@
+/*
+ * Copyright 2009 Maciej Cencora <[email protected]>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
 
 #include "main/glheader.h"
 
 #include "shader/program.h"
 #include "tnl/tnl.h"
 #include "r300_context.h"
-#include "r300_fragprog.h"
+#include "r300_fragprog_common.h"
 
 static struct gl_program *r300NewProgram(GLcontext * ctx, GLenum target,
 					 GLuint id)
 {
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	struct r300_vertex_program_cont *vp;
-	struct r300_fragment_program *r300_fp;
-	struct r500_fragment_program *r500_fp;
+	struct r300_fragment_program *fp;
 
 	switch (target) {
 	case GL_VERTEX_STATE_PROGRAM_NV:
@@ -20,28 +44,12 @@ static struct gl_program *r300NewProgram(GLcontext * ctx, GLenum target,
 		vp = CALLOC_STRUCT(r300_vertex_program_cont);
 		return _mesa_init_vertex_program(ctx, &vp->mesa_program,
 						 target, id);
-	case GL_FRAGMENT_PROGRAM_ARB:
-		if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
-			r500_fp = CALLOC_STRUCT(r500_fragment_program);
-			r500_fp->ctx = ctx;
-			return _mesa_init_fragment_program(ctx, &r500_fp->mesa_program,
-							   target, id);
-		} else {
-			r300_fp = CALLOC_STRUCT(r300_fragment_program);
-			return _mesa_init_fragment_program(ctx, &r300_fp->mesa_program,
-							   target, id);
-		}
 
 	case GL_FRAGMENT_PROGRAM_NV:
-		if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
-			r500_fp = CALLOC_STRUCT(r500_fragment_program);
-			return _mesa_init_fragment_program(ctx, &r500_fp->mesa_program,
-							   target, id);
-		} else {
-			r300_fp = CALLOC_STRUCT(r300_fragment_program);
-			return _mesa_init_fragment_program(ctx, &r300_fp->mesa_program,
-							   target, id);
-		}
+	case GL_FRAGMENT_PROGRAM_ARB:
+		fp = CALLOC_STRUCT(r300_fragment_program);
+		return _mesa_init_fragment_program(ctx, &fp->Base, target, id);
+
 	default:
 		_mesa_problem(ctx, "Bad target in r300NewProgram");
 	}
@@ -57,20 +65,15 @@ static void r300DeleteProgram(GLcontext * ctx, struct gl_program *prog)
 static void
 r300ProgramStringNotify(GLcontext * ctx, GLenum target, struct gl_program *prog)
 {
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	struct r300_vertex_program_cont *vp = (void *)prog;
 	struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)prog;
-	struct r500_fragment_program *r500_fp = (struct r500_fragment_program *)prog;
 
 	switch (target) {
 	case GL_VERTEX_PROGRAM_ARB:
 		vp->progs = NULL;
 		break;
 	case GL_FRAGMENT_PROGRAM_ARB:
-		if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
-			r500_fp->translated = GL_FALSE;
-		else
-			r300_fp->translated = GL_FALSE;
+		r300_fp->translated = GL_FALSE;
 		break;
 	}
 
@@ -81,7 +84,14 @@ r300ProgramStringNotify(GLcontext * ctx, GLenum target, struct gl_program *prog)
 static GLboolean
 r300IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
 {
-	return GL_TRUE;
+	if (target == GL_FRAGMENT_PROGRAM_ARB) {
+		struct r300_fragment_program *fp = (struct r300_fragment_program *)prog;
+		if (!fp->translated)
+			r300TranslateFragmentShader(ctx, &fp->Base);
+
+		return !fp->error;
+	} else
+		return GL_TRUE;
 }
 
 void r300InitShaderFuncs(struct dd_function_table *functions)
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index 07299f3b36f..c0eda977db8 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -42,6 +42,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/macros.h"
 #include "main/context.h"
 #include "main/dd.h"
+#include "main/framebuffer.h"
 #include "main/simple_list.h"
 #include "main/api_arrayelt.h"
 #include "main/texformat.h"
@@ -52,22 +53,22 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "shader/prog_statevars.h"
 #include "vbo/vbo.h"
 #include "tnl/tnl.h"
+#include "tnl/t_vp_build.h"
 
-#include "radeon_ioctl.h"
-#include "radeon_state.h"
 #include "r300_context.h"
 #include "r300_ioctl.h"
 #include "r300_state.h"
 #include "r300_reg.h"
 #include "r300_emit.h"
-#include "r300_fragprog.h"
 #include "r300_tex.h"
+#include "r300_fragprog_common.h"
+#include "r300_fragprog.h"
+#include "r500_fragprog.h"
+#include "r300_render.h"
+#include "r300_vertprog.h"
 
 #include "drirenderbuffer.h"
 
-extern int future_hw_tcl_on;
-extern void _tnl_UpdateFixedFunctionProgram(GLcontext * ctx);
-
 static void r300BlendColor(GLcontext * ctx, const GLfloat cf[4])
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
@@ -366,7 +367,7 @@ static void r300ClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq )
 	GLint *ip;
 
 	/* no VAP UCP on non-TCL chipsets */
-	if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
+	if (!rmesa->options.hw_tcl_enabled)
 			return;
 
 	p = (GLint) plane - (GLint) GL_CLIP_PLANE0;
@@ -385,7 +386,7 @@ static void r300SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state)
 	GLuint p;
 
 	/* no VAP UCP on non-TCL chipsets */
-	if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
+	if (!r300->options.hw_tcl_enabled)
 		return;
 
 	p = cap - GL_CLIP_PLANE0;
@@ -451,24 +452,16 @@ static void r300SetPolygonOffsetState(GLcontext * ctx, GLboolean state)
 
 static GLboolean current_fragment_program_writes_depth(GLcontext* ctx)
 {
-	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
 
-	if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) {
-		struct r300_fragment_program *fp = (struct r300_fragment_program *)
-			(char *)ctx->FragmentProgram._Current;
-		return (fp && fp->WritesDepth);
-	} else {
-		struct r500_fragment_program* fp =
-			(struct r500_fragment_program*)(char*)
-			ctx->FragmentProgram._Current;
-		return (fp && fp->writes_depth);
-	}
+	return (fp && fp->writes_depth);
 }
 
 static void r300SetEarlyZState(GLcontext * ctx)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	GLuint topZ = R300_ZTOP_ENABLE;
+	GLuint w_fmt, fgdepthsrc;
 
 	if (ctx->Color.AlphaEnabled && ctx->Color.AlphaFunc != GL_ALWAYS)
 		topZ = R300_ZTOP_DISABLE;
@@ -485,6 +478,26 @@ static void r300SetEarlyZState(GLcontext * ctx)
 		R300_STATECHANGE(r300, zstencil_format);
 		r300->hw.zstencil_format.cmd[2] = topZ;
 	}
+
+	/* w_fmt value is set to get best performance
+	* see p.130 R5xx 3D acceleration guide v1.3 */
+	if (current_fragment_program_writes_depth(ctx)) {
+		fgdepthsrc = R300_FG_DEPTH_SRC_SHADER;
+		w_fmt = R300_W_FMT_W24 | R300_W_SRC_US;
+	} else {
+		fgdepthsrc = R300_FG_DEPTH_SRC_SCAN;
+		w_fmt = R300_W_FMT_W0 | R300_W_SRC_US;
+	}
+
+	if (w_fmt != r300->hw.us_out_fmt.cmd[5]) {
+		R300_STATECHANGE(r300, us_out_fmt);
+		r300->hw.us_out_fmt.cmd[5] = w_fmt;
+	}
+
+	if (fgdepthsrc != r300->hw.fg_depth_src.cmd[1]) {
+		R300_STATECHANGE(r300, fg_depth_src);
+		r300->hw.fg_depth_src.cmd[1] = fgdepthsrc;
+	}
 }
 
 static void r300SetAlphaState(GLcontext * ctx)
@@ -535,8 +548,6 @@ static void r300SetAlphaState(GLcontext * ctx)
 	R300_STATECHANGE(r300, at);
 	r300->hw.at.cmd[R300_AT_ALPHA_TEST] = pp_misc;
 	r300->hw.at.cmd[R300_AT_UNKNOWN] = 0;
-
-	r300SetEarlyZState(ctx);
 }
 
 static void r300AlphaFunc(GLcontext * ctx, GLenum func, GLfloat ref)
@@ -584,15 +595,35 @@ static void r300SetDepthState(GLcontext * ctx)
 		r300->hw.zs.cmd[R300_ZS_CNTL_1] |=
 		    translate_func(ctx->Depth.Func) << R300_Z_FUNC_SHIFT;
 	}
+}
 
-	r300SetEarlyZState(ctx);
+static void r300CatchStencilFallback(GLcontext *ctx)
+{
+	const unsigned back = ctx->Stencil._BackFace;
+
+	if (ctx->Stencil._Enabled && (ctx->Stencil.Ref[0] != ctx->Stencil.Ref[back]
+		|| ctx->Stencil.ValueMask[0] != ctx->Stencil.ValueMask[back]
+		|| ctx->Stencil.WriteMask[0] != ctx->Stencil.WriteMask[back])) {
+		r300SwitchFallback(ctx, R300_FALLBACK_STENCIL_TWOSIDE, GL_TRUE);
+	} else {
+		r300SwitchFallback(ctx, R300_FALLBACK_STENCIL_TWOSIDE, GL_FALSE);
+	}
 }
 
 static void r300SetStencilState(GLcontext * ctx, GLboolean state)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	GLboolean hw_stencil = GL_FALSE;
 
-	if (r300->state.stencil.hw_stencil) {
+	r300CatchStencilFallback(ctx);
+
+	if (ctx->DrawBuffer) {
+		struct radeon_renderbuffer *rrbStencil
+			= radeon_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL);
+		hw_stencil = (rrbStencil && rrbStencil->bo);
+	}
+
+	if (hw_stencil) {
 		R300_STATECHANGE(r300, zs);
 		if (state) {
 			r300->hw.zs.cmd[R300_ZS_CNTL_0] |=
@@ -601,10 +632,6 @@ static void r300SetStencilState(GLcontext * ctx, GLboolean state)
 			r300->hw.zs.cmd[R300_ZS_CNTL_0] &=
 			    ~R300_STENCIL_ENABLE;
 		}
-	} else {
-#if R200_MERGED
-		FALLBACK(&r300->radeon, RADEON_FALLBACK_STENCIL, state);
-#endif
 	}
 }
 
@@ -737,7 +764,12 @@ static void r300ColorMask(GLcontext * ctx,
 static void r300PointSize(GLcontext * ctx, GLfloat size)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
-        /* same size limits for AA, non-AA points */
+
+	/* We need to clamp to user defined range here, because
+	 * the HW clamping happens only for per vertex point size. */
+	size = CLAMP(size, ctx->Point.MinSize, ctx->Point.MaxSize);
+
+	/* same size limits for AA, non-AA points */
 	size = CLAMP(size, ctx->Const.MinPointSize, ctx->Const.MaxPointSize);
 
 	R300_STATECHANGE(r300, ps);
@@ -830,29 +862,33 @@ static void r300ShadeModel(GLcontext * ctx, GLenum mode)
 
 	R300_STATECHANGE(rmesa, shade);
 	rmesa->hw.shade.cmd[1] = 0x00000002;
+	R300_STATECHANGE(rmesa, shade2);
 	switch (mode) {
 	case GL_FLAT:
-		rmesa->hw.shade.cmd[2] = R300_RE_SHADE_MODEL_FLAT;
+		rmesa->hw.shade2.cmd[1] = R300_RE_SHADE_MODEL_FLAT;
 		break;
 	case GL_SMOOTH:
-		rmesa->hw.shade.cmd[2] = R300_RE_SHADE_MODEL_SMOOTH;
+		rmesa->hw.shade2.cmd[1] = R300_RE_SHADE_MODEL_SMOOTH;
 		break;
 	default:
 		return;
 	}
-	rmesa->hw.shade.cmd[3] = 0x00000000;
-	rmesa->hw.shade.cmd[4] = 0x00000000;
+	rmesa->hw.shade2.cmd[2] = 0x00000000;
+	rmesa->hw.shade2.cmd[3] = 0x00000000;
 }
 
 static void r300StencilFuncSeparate(GLcontext * ctx, GLenum face,
 				    GLenum func, GLint ref, GLuint mask)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	GLuint refmask =
-	    ((ctx->Stencil.Ref[0] & 0xff) << R300_STENCILREF_SHIFT)
-	     | ((ctx->Stencil.ValueMask[0] & 0xff) << R300_STENCILMASK_SHIFT);
-	const unsigned back = ctx->Stencil._BackFace;
+	GLuint refmask;
 	GLuint flag;
+	const unsigned back = ctx->Stencil._BackFace;
+
+	r300CatchStencilFallback(ctx);
+
+	refmask = ((ctx->Stencil.Ref[0] & 0xff) << R300_STENCILREF_SHIFT)
+	     | ((ctx->Stencil.ValueMask[0] & 0xff) << R300_STENCILMASK_SHIFT);
 
 	R300_STATECHANGE(rmesa, zs);
 	rmesa->hw.zs.cmd[R300_ZS_CNTL_0] |= R300_STENCIL_FRONT_BACK;
@@ -880,6 +916,8 @@ static void r300StencilMaskSeparate(GLcontext * ctx, GLenum face, GLuint mask)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 
+	r300CatchStencilFallback(ctx);
+
 	R300_STATECHANGE(rmesa, zs);
 	rmesa->hw.zs.cmd[R300_ZS_CNTL_2] &=
 	    ~(R300_STENCILREF_MASK <<
@@ -896,6 +934,8 @@ static void r300StencilOpSeparate(GLcontext * ctx, GLenum face,
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	const unsigned back = ctx->Stencil._BackFace;
 
+	r300CatchStencilFallback(ctx);
+
 	R300_STATECHANGE(rmesa, zs);
 	/* It is easier to mask what's left.. */
 	rmesa->hw.zs.cmd[R300_ZS_CNTL_1] &=
@@ -924,28 +964,32 @@ static void r300StencilOpSeparate(GLcontext * ctx, GLenum face,
  * Window position and viewport transformation
  */
 
-/*
- * To correctly position primitives:
- */
-#define SUBPIXEL_X 0.125
-#define SUBPIXEL_Y 0.125
-
 static void r300UpdateWindow(GLcontext * ctx)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	__DRIdrawablePrivate *dPriv = rmesa->radeon.dri.drawable;
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
 	GLfloat xoffset = dPriv ? (GLfloat) dPriv->x : 0;
 	GLfloat yoffset = dPriv ? (GLfloat) dPriv->y + dPriv->h : 0;
 	const GLfloat *v = ctx->Viewport._WindowMap.m;
+	const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
+	const GLboolean render_to_fbo = (ctx->DrawBuffer->Name != 0);
+	GLfloat y_scale, y_bias;
+
+	if (render_to_fbo) {
+		y_scale = 1.0;
+		y_bias = 0;
+	} else {
+		y_scale = -1.0;
+		y_bias = yoffset;
+	}
 
 	GLfloat sx = v[MAT_SX];
-	GLfloat tx = v[MAT_TX] + xoffset + SUBPIXEL_X;
-	GLfloat sy = -v[MAT_SY];
-	GLfloat ty = (-v[MAT_TY]) + yoffset + SUBPIXEL_Y;
-	GLfloat sz = v[MAT_SZ] * rmesa->state.depth.scale;
-	GLfloat tz = v[MAT_TZ] * rmesa->state.depth.scale;
+	GLfloat tx = v[MAT_TX] + xoffset;
+	GLfloat sy = v[MAT_SY] * y_scale;
+	GLfloat ty = (v[MAT_TY] * y_scale) + y_bias;
+	GLfloat sz = v[MAT_SZ] * depthScale;
+	GLfloat tz = v[MAT_TZ] * depthScale;
 
-	R300_FIREVERTICES(rmesa);
 	R300_STATECHANGE(rmesa, vpt);
 
 	rmesa->hw.vpt.cmd[R300_VPT_XSCALE] = r300PackFloat32(sx);
@@ -964,6 +1008,8 @@ static void r300Viewport(GLcontext * ctx, GLint x, GLint y,
 	 * values, or keep the originals hanging around.
 	 */
 	r300UpdateWindow(ctx);
+
+	radeon_viewport(ctx, x, y, width, height);
 }
 
 static void r300DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval)
@@ -974,13 +1020,13 @@ static void r300DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval)
 void r300UpdateViewportOffset(GLcontext * ctx)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	__DRIdrawablePrivate *dPriv = ((radeonContextPtr) rmesa)->dri.drawable;
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
 	GLfloat xoffset = (GLfloat) dPriv->x;
 	GLfloat yoffset = (GLfloat) dPriv->y + dPriv->h;
 	const GLfloat *v = ctx->Viewport._WindowMap.m;
 
-	GLfloat tx = v[MAT_TX] + xoffset + SUBPIXEL_X;
-	GLfloat ty = (-v[MAT_TY]) + yoffset + SUBPIXEL_Y;
+	GLfloat tx = v[MAT_TX] + xoffset;
+	GLfloat ty = (-v[MAT_TY]) + yoffset;
 
 	if (rmesa->hw.vpt.cmd[R300_VPT_XOFFSET] != r300PackFloat32(tx) ||
 	    rmesa->hw.vpt.cmd[R300_VPT_YOFFSET] != r300PackFloat32(ty)) {
@@ -996,64 +1042,6 @@ void r300UpdateViewportOffset(GLcontext * ctx)
 	radeonUpdateScissor(ctx);
 }
 
-/**
- * Tell the card where to render (offset, pitch).
- * Effected by glDrawBuffer, etc
- */
-void r300UpdateDrawBuffer(GLcontext * ctx)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	r300ContextPtr r300 = rmesa;
-	struct gl_framebuffer *fb = ctx->DrawBuffer;
-	driRenderbuffer *drb;
-
-	if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) {
-		/* draw to front */
-		drb =
-		    (driRenderbuffer *) fb->Attachment[BUFFER_FRONT_LEFT].
-		    Renderbuffer;
-	} else if (fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) {
-		/* draw to back */
-		drb =
-		    (driRenderbuffer *) fb->Attachment[BUFFER_BACK_LEFT].
-		    Renderbuffer;
-	} else {
-		/* drawing to multiple buffers, or none */
-		return;
-	}
-
-	assert(drb);
-	assert(drb->flippedPitch);
-
-	R300_STATECHANGE(rmesa, cb);
-
-	r300->hw.cb.cmd[R300_CB_OFFSET] = drb->flippedOffset +	//r300->radeon.state.color.drawOffset +
-	    r300->radeon.radeonScreen->fbLocation;
-	r300->hw.cb.cmd[R300_CB_PITCH] = drb->flippedPitch;	//r300->radeon.state.color.drawPitch;
-
-	if (r300->radeon.radeonScreen->cpp == 4)
-		r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_ARGB8888;
-	else
-		r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_RGB565;
-
-	if (r300->radeon.sarea->tiling_enabled)
-		r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_TILE_ENABLE;
-#if 0
-	R200_STATECHANGE(rmesa, ctx);
-
-	/* Note: we used the (possibly) page-flipped values */
-	rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET]
-	    = ((drb->flippedOffset + rmesa->r200Screen->fbLocation)
-	       & R200_COLOROFFSET_MASK);
-	rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = drb->flippedPitch;
-
-	if (rmesa->sarea->tiling_enabled) {
-		rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |=
-		    R200_COLOR_TILE_ENABLE;
-	}
-#endif
-}
-
 static void
 r300FetchStateParameter(GLcontext * ctx,
 			const gl_state_index state[STATE_LENGTH],
@@ -1064,12 +1052,14 @@ r300FetchStateParameter(GLcontext * ctx,
 	switch (state[0]) {
 	case STATE_INTERNAL:
 		switch (state[1]) {
-		case STATE_R300_WINDOW_DIMENSION:
-			value[0] = r300->radeon.dri.drawable->w * 0.5f;	/* width*0.5 */
-			value[1] = r300->radeon.dri.drawable->h * 0.5f;	/* height*0.5 */
-			value[2] = 0.5F;	/* for moving range [-1 1] -> [0 1] */
-			value[3] = 1.0F;	/* not used */
-			break;
+		case STATE_R300_WINDOW_DIMENSION: {
+				__DRIdrawablePrivate * drawable = radeon_get_drawable(&r300->radeon);
+				value[0] = drawable->w * 0.5f;	/* width*0.5 */
+				value[1] = drawable->h * 0.5f;	/* height*0.5 */
+				value[2] = 0.5F;	/* for moving range [-1 1] -> [0 1] */
+				value[3] = 1.0F;	/* not used */
+				break;
+			}
 
 		case STATE_R300_TEXRECT_FACTOR:{
 				struct gl_texture_object *t =
@@ -1116,7 +1106,7 @@ void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state)
 	if (!fp)
 		return;
 
-	paramList = fp->mesa_program.Base.Parameters;
+	paramList = fp->Base.Base.Parameters;
 
 	if (!paramList)
 		return;
@@ -1235,9 +1225,8 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	int i;
-	struct r300_fragment_program *fp = (struct r300_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
-	struct r300_fragment_program_code *code = &fp->code;
+	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
+	struct r300_fragment_program_code *code = &fp->code.r300;
 
 	R300_STATECHANGE(r300, fpt);
 
@@ -1271,15 +1260,15 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 	}
 
 	r300->hw.fpt.cmd[R300_FPT_CMD_0] =
-		cmdpacket0(R300_US_TEX_INST_0, code->tex.length);
+		cmdpacket0(r300->radeon.radeonScreen,
+                   R300_US_TEX_INST_0, code->tex.length);
 }
 
 static void r500SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 {
 	int i;
-	struct r500_fragment_program *fp = (struct r500_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
-	struct r500_fragment_program_code *code = &fp->code;
+	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
+	struct r500_fragment_program_code *code = &fp->code.r500;
 
 	/* find all the texture instructions and relocate the texture units */
 	for (i = 0; i < code->inst_end + 1; i++) {
@@ -1322,7 +1311,7 @@ static GLuint translate_lod_bias(GLfloat bias)
 static void r300SetupTextures(GLcontext * ctx)
 {
 	int i, mtu;
-	struct r300_tex_obj *t;
+	struct radeon_tex_obj *t;
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	int hw_tmu = 0;
 	int last_hw_tmu = -1;	/* -1 translates into no setup costs for fields */
@@ -1356,21 +1345,16 @@ static void r300SetupTextures(GLcontext * ctx)
 	/* We cannot let disabled tmu offsets pass DRM */
 	for (i = 0; i < mtu; i++) {
 		if (ctx->Texture.Unit[i]._ReallyEnabled) {
-
-#if 0				/* Enables old behaviour */
-			hw_tmu = i;
-#endif
 			tmu_mappings[i] = hw_tmu;
 
-			t = (r300TexObjPtr) r300->state.texture.unit[i].texobj->DriverData;
-			/* XXX questionable fix for bug 9170: */
+			t = radeon_tex_obj(ctx->Texture.Unit[i]._Current);
 			if (!t)
 				continue;
 
-			if ((t->format & 0xffffff00) == 0xffffff00) {
+			if ((t->pp_txformat & 0xffffff00) == 0xffffff00) {
 				WARN_ONCE
 				    ("unknown texture format (entry %x) encountered. Help me !\n",
-				     t->format & 0xff);
+				     t->pp_txformat & 0xff);
 			}
 
 			if (RADEON_DEBUG & DEBUG_STATE)
@@ -1381,29 +1365,28 @@ static void r300SetupTextures(GLcontext * ctx)
 
 			r300->hw.tex.filter.cmd[R300_TEX_VALUE_0 +
 						hw_tmu] =
-			    gen_fixed_filter(t->filter) | (hw_tmu << 28);
+			    gen_fixed_filter(t->pp_txfilter) | (hw_tmu << 28);
 			/* Note: There is a LOD bias per texture unit and a LOD bias
 			 * per texture object. We add them here to get the correct behaviour.
 			 * (The per-texture object LOD bias was introduced in OpenGL 1.4
 			 * and is not present in the EXT_texture_object extension).
 			 */
 			r300->hw.tex.filter_1.cmd[R300_TEX_VALUE_0 + hw_tmu] =
-				t->filter_1 |
-				translate_lod_bias(ctx->Texture.Unit[i].LodBias + t->base.tObj->LodBias);
+				t->pp_txfilter_1 |
+				translate_lod_bias(ctx->Texture.Unit[i].LodBias + t->base.LodBias);
 			r300->hw.tex.size.cmd[R300_TEX_VALUE_0 + hw_tmu] =
-			    t->size;
+			    t->pp_txsize;
 			r300->hw.tex.format.cmd[R300_TEX_VALUE_0 +
-						hw_tmu] = t->format;
+						hw_tmu] = t->pp_txformat;
 			r300->hw.tex.pitch.cmd[R300_TEX_VALUE_0 + hw_tmu] =
-			    t->pitch_reg;
-			r300->hw.tex.offset.cmd[R300_TEX_VALUE_0 +
-						hw_tmu] = t->offset;
+			  t->pp_txpitch;
+			r300->hw.textures[hw_tmu] = t;
 
-			if (t->offset & R300_TXO_MACRO_TILE) {
+			if (t->tile_bits & R300_TXO_MACRO_TILE) {
 				WARN_ONCE("macro tiling enabled!\n");
 			}
 
-			if (t->offset & R300_TXO_MICRO_TILE) {
+			if (t->tile_bits & R300_TXO_MICRO_TILE) {
 				WARN_ONCE("micro tiling enabled!\n");
 			}
 
@@ -1420,37 +1403,33 @@ static void r300SetupTextures(GLcontext * ctx)
 	}
 
 	r300->hw.tex.filter.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FILTER0_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, last_hw_tmu + 1);
 	r300->hw.tex.filter_1.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FILTER1_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER1_0, last_hw_tmu + 1);
 	r300->hw.tex.size.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_SIZE_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_SIZE_0, last_hw_tmu + 1);
 	r300->hw.tex.format.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FORMAT_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT_0, last_hw_tmu + 1);
 	r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_FORMAT2_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT2_0, last_hw_tmu + 1);
 	r300->hw.tex.offset.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_OFFSET_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_OFFSET_0, last_hw_tmu + 1);
 	r300->hw.tex.chroma_key.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_CHROMA_KEY_0, last_hw_tmu + 1);
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_CHROMA_KEY_0, last_hw_tmu + 1);
 	r300->hw.tex.border_color.cmd[R300_TEX_CMD_0] =
-	    cmdpacket0(R300_TX_BORDER_COLOR_0, last_hw_tmu + 1);
-
-	if (!fp)		/* should only happenen once, just after context is created */
-		return;
+	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_BORDER_COLOR_0, last_hw_tmu + 1);
 
 	if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) {
-		if (fp->mesa_program.UsesKill && last_hw_tmu < 0) {
+		if (fp->Base.UsesKill && last_hw_tmu < 0) {
 			// The KILL operation requires the first texture unit
 			// to be enabled.
 			r300->hw.txe.cmd[R300_TXE_ENABLE] |= 1;
 			r300->hw.tex.filter.cmd[R300_TEX_VALUE_0] = 0;
 			r300->hw.tex.filter.cmd[R300_TEX_CMD_0] =
-				cmdpacket0(R300_TX_FILTER0_0, 1);
+				cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, 1);
 		}
-		r300SetupFragmentShaderTextures(ctx, tmu_mappings);
-	} else
-		r500SetupFragmentShaderTextures(ctx, tmu_mappings);
+	}
+	r300->vtbl.SetupFragmentShaderTextures(ctx, tmu_mappings);
 
 	if (RADEON_DEBUG & DEBUG_STATE)
 		fprintf(stderr, "TX_ENABLE: %08x  last_hw_tmu=%d\n",
@@ -1469,26 +1448,21 @@ union r300_outputs_written {
 static void r300SetupRSUnit(GLcontext * ctx)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
-        TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *VB = &tnl->vb;
 	union r300_outputs_written OutputsWritten;
 	GLuint InputsRead;
 	int fp_reg, high_rr;
 	int col_ip, tex_ip;
 	int rs_tex_count = 0;
-	int i, count, col_fmt;
+	int i, col_fmt, hw_tcl_on;
+
+	hw_tcl_on = r300->options.hw_tcl_enabled;
 
 	if (hw_tcl_on)
-		OutputsWritten.vp_outputs = CURRENT_VERTEX_SHADER(ctx)->key.OutputsWritten;
+		OutputsWritten.vp_outputs = r300->selected_vp->key.OutputsWritten;
 	else
-		RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->state.render_inputs_bitset);
+		RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset);
 
-	if (ctx->FragmentProgram._Current)
-		InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
-	else {
-		fprintf(stderr, "No ctx->FragmentProgram._Current!!\n");
-		return;		/* This should only ever happen once.. */
-	}
+	InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
 
 	R300_STATECHANGE(r300, ri);
 	R300_STATECHANGE(r300, rc);
@@ -1507,15 +1481,7 @@ static void r300SetupRSUnit(GLcontext * ctx)
 
 	if (InputsRead & FRAG_BIT_COL0) {
 		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL0, _TNL_ATTRIB_COLOR0)) {
-			count = VB->AttribPtr[_TNL_ATTRIB_COLOR0]->size;
-			if (count == 4)
-			    col_fmt = R300_RS_COL_FMT_RGBA;
-			else if (count == 3)
-			    col_fmt = R300_RS_COL_FMT_RGB1;
-			else
-			    col_fmt = R300_RS_COL_FMT_0001;
-
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(col_fmt);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 			r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R300_RS_INST_COL_ID(col_ip) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_reg);
 			InputsRead &= ~FRAG_BIT_COL0;
 			++col_ip;
@@ -1527,15 +1493,7 @@ static void r300SetupRSUnit(GLcontext * ctx)
 
 	if (InputsRead & FRAG_BIT_COL1) {
 		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL1, _TNL_ATTRIB_COLOR1)) {
-			count = VB->AttribPtr[_TNL_ATTRIB_COLOR1]->size;
-			if (count == 4)
-			    col_fmt = R300_RS_COL_FMT_RGBA;
-			else if (count == 3)
-			    col_fmt = R300_RS_COL_FMT_RGB1;
-			else
-			    col_fmt = R300_RS_COL_FMT_0001;
-
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(col_fmt);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 			r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R300_RS_INST_COL_ID(col_ip) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_reg);
 			InputsRead &= ~FRAG_BIT_COL1;
 			++col_ip;
@@ -1545,6 +1503,7 @@ static void r300SetupRSUnit(GLcontext * ctx)
 		}
 	}
 
+	/* We always route 4 texcoord components */
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
 		if (! ( InputsRead & FRAG_BIT_TEX(i) ) )
 		    continue;
@@ -1554,33 +1513,27 @@ static void r300SetupRSUnit(GLcontext * ctx)
 		    continue;
 		}
 
-		int swiz;
-
-		/* with TCL we always seem to route 4 components */
-		if (hw_tcl_on)
-		  count = 4;
-		else
-		  count = VB->AttribPtr[_TNL_ATTRIB_TEX(i)]->size;
-
-		switch(count) {
-		case 4: swiz = R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3); break;
-		case 3: swiz = R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(R300_RS_SEL_K1); break;
-		default:
-		case 1:
-		case 2: swiz = R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(R300_RS_SEL_K0) | R300_RS_SEL_Q(R300_RS_SEL_K1); break;
-		};
-
-		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= swiz | R300_RS_TEX_PTR(rs_tex_count);
+		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count);
 		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
 		InputsRead &= ~(FRAG_BIT_TEX0 << i);
-		rs_tex_count += count;
+		rs_tex_count += 4;
+		++tex_ip;
+		++fp_reg;
+	}
+
+	if (InputsRead & FRAG_BIT_WPOS) {
+		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count);
+		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
+		InputsRead &= ~FRAG_BIT_WPOS;
+		rs_tex_count += 4;
 		++tex_ip;
 		++fp_reg;
 	}
 
 	if (InputsRead & FRAG_BIT_FOGC) {
 		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_FOGC, _TNL_ATTRIB_FOG)) {
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |=  R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) |  R300_RS_TEX_PTR(rs_tex_count);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(R300_RS_SEL_K0) | R300_RS_SEL_R(R300_RS_SEL_K0);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_Q(R300_RS_SEL_K1) | R300_RS_TEX_PTR(rs_tex_count);
 			r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
 			InputsRead &= ~FRAG_BIT_FOGC;
 			rs_tex_count += 4;
@@ -1591,27 +1544,19 @@ static void r300SetupRSUnit(GLcontext * ctx)
 		}
 	}
 
-	if (InputsRead & FRAG_BIT_WPOS) {
-		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |=  R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) |  R300_RS_TEX_PTR(rs_tex_count);
-		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
-		InputsRead &= ~FRAG_BIT_WPOS;
-		rs_tex_count += 4;
-		++tex_ip;
-		++fp_reg;
-	}
-	InputsRead &= ~FRAG_BIT_WPOS;
-
 	/* Setup default color if no color or tex was set */
 	if (rs_tex_count == 0 && col_ip == 0) {
-		r300->hw.rr.cmd[R300_RR_INST_0] = R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(0) | R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
+		r300->hw.rr.cmd[R300_RR_INST_0] = R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_ADDR(0);
+		r300->hw.ri.cmd[R300_RI_INTERP_0] = R300_RS_COL_PTR(0) | R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
 		++col_ip;
 	}
 
 	high_rr = (col_ip > tex_ip) ? col_ip : tex_ip;
-	r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT)  | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN;
+	r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT) | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN;
 	r300->hw.rc.cmd[2] |= high_rr - 1;
 
-	r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R300_RS_INST_0, high_rr);
+	r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_INST_0, high_rr);
+	r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_IP_0, high_rr);
 
 	if (InputsRead)
 		WARN_ONCE("Don't know how to satisfy InputsRead=0x%08x\n", InputsRead);
@@ -1620,26 +1565,21 @@ static void r300SetupRSUnit(GLcontext * ctx)
 static void r500SetupRSUnit(GLcontext * ctx)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
-        TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *VB = &tnl->vb;
 	union r300_outputs_written OutputsWritten;
 	GLuint InputsRead;
 	int fp_reg, high_rr;
 	int col_ip, tex_ip;
 	int rs_tex_count = 0;
-	int i, count, col_fmt;
+	int i, col_fmt, hw_tcl_on;
+
+	hw_tcl_on = r300->options.hw_tcl_enabled;
 
 	if (hw_tcl_on)
-		OutputsWritten.vp_outputs = CURRENT_VERTEX_SHADER(ctx)->key.OutputsWritten;
+		OutputsWritten.vp_outputs = r300->selected_vp->key.OutputsWritten;
 	else
-		RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->state.render_inputs_bitset);
+		RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset);
 
-	if (ctx->FragmentProgram._Current)
-		InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
-	else {
-		fprintf(stderr, "No ctx->FragmentProgram._Current!!\n");
-		return;		/* This should only ever happen once.. */
-	}
+	InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
 
 	R300_STATECHANGE(r300, ri);
 	R300_STATECHANGE(r300, rc);
@@ -1658,15 +1598,7 @@ static void r500SetupRSUnit(GLcontext * ctx)
 
 	if (InputsRead & FRAG_BIT_COL0) {
 		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL0, _TNL_ATTRIB_COLOR0)) {
-			count = VB->AttribPtr[_TNL_ATTRIB_COLOR0]->size;
-			if (count == 4)
-			    col_fmt = R300_RS_COL_FMT_RGBA;
-			else if (count == 3)
-			    col_fmt = R300_RS_COL_FMT_RGB1;
-			else
-			    col_fmt = R300_RS_COL_FMT_0001;
-
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(col_fmt);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 			r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R500_RS_INST_COL_ID(col_ip) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_reg);
 			InputsRead &= ~FRAG_BIT_COL0;
 			++col_ip;
@@ -1678,15 +1610,7 @@ static void r500SetupRSUnit(GLcontext * ctx)
 
 	if (InputsRead & FRAG_BIT_COL1) {
 		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL1, _TNL_ATTRIB_COLOR1)) {
-			count = VB->AttribPtr[_TNL_ATTRIB_COLOR1]->size;
-			if (count == 4)
-			    col_fmt = R300_RS_COL_FMT_RGBA;
-			else if (count == 3)
-			    col_fmt = R300_RS_COL_FMT_RGB1;
-			else
-			    col_fmt = R300_RS_COL_FMT_0001;
-
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(col_fmt);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 			r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R500_RS_INST_COL_ID(col_ip) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_reg);
 			InputsRead &= ~FRAG_BIT_COL1;
 			++col_ip;
@@ -1696,7 +1620,7 @@ static void r500SetupRSUnit(GLcontext * ctx)
 		}
 	}
 
-
+	/* We always route 4 texcoord components */
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
 		if (! ( InputsRead & FRAG_BIT_TEX(i) ) )
 		    continue;
@@ -1706,55 +1630,37 @@ static void r500SetupRSUnit(GLcontext * ctx)
 		    continue;
 		}
 
-		int swiz = 0;
-
-		/* with TCL we always seem to route 4 components */
-		if (hw_tcl_on)
-		  count = 4;
-		else
-		  count = VB->AttribPtr[_TNL_ATTRIB_TEX(i)]->size;
-
-		if (count == 4) {
-			swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT;
-			swiz |= (rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT;
-			swiz |= (rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT;
-			swiz |= (rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT;
-		} else if (count == 3) {
-			swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT;
-			swiz |= (rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT;
-			swiz |= (rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT;
-			swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT;
-		} else if (count == 2) {
-			swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT;
-			swiz |= (rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT;
-			swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT;
-			swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT;
-		} else if (count == 1) {
-			swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT;
-			swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT;
-			swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT;
-			swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT;
-		} else {
-			swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT;
-			swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT;
-			swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT;
-			swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT;
-		}
+		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) |
+			((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) |
+			((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) |
+			((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT);
 
-		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= swiz;
 		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
 		InputsRead &= ~(FRAG_BIT_TEX0 << i);
-		rs_tex_count += count;
+		rs_tex_count += 4;
+		++tex_ip;
+		++fp_reg;
+	}
+
+	if (InputsRead & FRAG_BIT_WPOS) {
+		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) |
+			((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) |
+			((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) |
+			((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT);
+
+		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
+		InputsRead &= ~FRAG_BIT_WPOS;
+		rs_tex_count += 4;
 		++tex_ip;
 		++fp_reg;
 	}
 
 	if (InputsRead & FRAG_BIT_FOGC) {
 		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_FOGC, _TNL_ATTRIB_FOG)) {
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) |
-				((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) |
-				((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) |
-				((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT);
+			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= (rs_tex_count << R500_RS_IP_TEX_PTR_S_SHIFT) |
+				(R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) |
+				(R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) |
+				(R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT);
 
 			r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
 			InputsRead &= ~FRAG_BIT_FOGC;
@@ -1766,87 +1672,27 @@ static void r500SetupRSUnit(GLcontext * ctx)
 		}
 	}
 
-	if (InputsRead & FRAG_BIT_WPOS) {
-		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) |
-				((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) |
-				((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) |
-				((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT);
-
-		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
-		InputsRead &= ~FRAG_BIT_WPOS;
-		rs_tex_count += 4;
-		++tex_ip;
-		++fp_reg;
-	}
-
 	/* Setup default color if no color or tex was set */
 	if (rs_tex_count == 0 && col_ip == 0) {
-		r300->hw.rr.cmd[R300_RR_INST_0] |= R500_RS_INST_COL_ID(0) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(0) | R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
+		r300->hw.rr.cmd[R300_RR_INST_0] = R500_RS_INST_COL_ID(0) | R500_RS_INST_COL_ADDR(0);
+		r300->hw.ri.cmd[R300_RI_INTERP_0] = R500_RS_COL_PTR(0) | R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
 		++col_ip;
 	}
 
 	high_rr = (col_ip > tex_ip) ? col_ip : tex_ip;
-	r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT)  | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN;
-	r300->hw.rc.cmd[2] |= 0xC0 | (high_rr - 1);
+	r300->hw.rc.cmd[1] = (rs_tex_count << R300_IT_COUNT_SHIFT) | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN;
+	r300->hw.rc.cmd[2] = 0xC0 | (high_rr - 1);
 
-	r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R500_RS_INST_0, high_rr);
+	r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_INST_0, high_rr);
+	r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_IP_0, high_rr);
 
 	if (InputsRead)
 		WARN_ONCE("Don't know how to satisfy InputsRead=0x%08x\n", InputsRead);
 }
 
-
-
-
-#define bump_vpu_count(ptr, new_count)   do{\
-	drm_r300_cmd_header_t* _p=((drm_r300_cmd_header_t*)(ptr));\
-	int _nc=(new_count)/4; \
-	assert(_nc < 256); \
-	if(_nc>_p->vpu.count)_p->vpu.count=_nc;\
-	}while(0)
-
-static INLINE void r300SetupVertexProgramFragment(r300ContextPtr r300, int dest, struct r300_vertex_shader_fragment *vsf)
-{
-	int i;
-
-	if (vsf->length == 0)
-		return;
-
-	if (vsf->length & 0x3) {
-		fprintf(stderr, "VERTEX_SHADER_FRAGMENT must have length divisible by 4\n");
-		_mesa_exit(-1);
-	}
-
-	switch ((dest >> 8) & 0xf) {
-	case 0:
-		R300_STATECHANGE(r300, vpi);
-		for (i = 0; i < vsf->length; i++)
-			r300->hw.vpi.cmd[R300_VPI_INSTR_0 + i + 4 * (dest & 0xff)] = (vsf->body.d[i]);
-		bump_vpu_count(r300->hw.vpi.cmd, vsf->length + 4 * (dest & 0xff));
-		break;
-
-	case 2:
-		R300_STATECHANGE(r300, vpp);
-		for (i = 0; i < vsf->length; i++)
-			r300->hw.vpp.cmd[R300_VPP_PARAM_0 + i + 4 * (dest & 0xff)] = (vsf->body.d[i]);
-		bump_vpu_count(r300->hw.vpp.cmd, vsf->length + 4 * (dest & 0xff));
-		break;
-	case 4:
-		R300_STATECHANGE(r300, vps);
-		for (i = 0; i < vsf->length; i++)
-			r300->hw.vps.cmd[1 + i + 4 * (dest & 0xff)] = (vsf->body.d[i]);
-		bump_vpu_count(r300->hw.vps.cmd, vsf->length + 4 * (dest & 0xff));
-		break;
-	default:
-		fprintf(stderr, "%s:%s don't know how to handle dest %04x\n", __FILE__, __FUNCTION__, dest);
-		_mesa_exit(-1);
-	}
-}
-
 #define MIN3(a, b, c)	((a) < (b) ? MIN2(a, c) : MIN2(b, c))
 
-
-static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
+void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
 			GLuint output_count, GLuint temp_count)
 {
     int vtx_mem_size;
@@ -1870,7 +1716,7 @@ static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
     pvs_num_cntrls = MIN2(6, vtx_mem_size/temp_count);
 
     R300_STATECHANGE(rmesa, vap_cntl);
-    if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
+    if (rmesa->options.hw_tcl_enabled) {
 	rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] =
 	    (pvs_num_slots << R300_PVS_NUM_SLOTS_SHIFT) |
 	    (pvs_num_cntrls << R300_PVS_NUM_CNTLRS_SHIFT) |
@@ -1900,114 +1746,6 @@ static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
 
 }
 
-static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa)
-{
-	struct r300_vertex_shader_state *prog = &(rmesa->state.vertex_shader);
-	GLuint o_reg = 0;
-	GLuint i_reg = 0;
-	int i;
-	int inst_count = 0;
-	int param_count = 0;
-	int program_end = 0;
-
-	for (i = VERT_ATTRIB_POS; i < VERT_ATTRIB_MAX; i++) {
-		if (rmesa->state.sw_tcl_inputs[i] != -1) {
-			prog->program.body.i[program_end + 0] = PVS_OP_DST_OPERAND(VE_MULTIPLY, GL_FALSE, GL_FALSE, o_reg++, VSF_FLAG_ALL, PVS_DST_REG_OUT);
-			prog->program.body.i[program_end + 1] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
-			prog->program.body.i[program_end + 2] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
-			prog->program.body.i[program_end + 3] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_REG_INPUT, VSF_FLAG_NONE);
-			program_end += 4;
-			i_reg++;
-		}
-	}
-
-	prog->program.length = program_end;
-
-	r300SetupVertexProgramFragment(rmesa, R300_PVS_CODE_START,
-				       &(prog->program));
-	inst_count = (prog->program.length / 4) - 1;
-
-	r300VapCntl(rmesa, i_reg, o_reg, 0);
-
-	R300_STATECHANGE(rmesa, pvs);
-	rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] =
-	    (0 << R300_PVS_FIRST_INST_SHIFT) |
-	    (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) |
-	    (inst_count << R300_PVS_LAST_INST_SHIFT);
-	rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] =
-	    (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) |
-	    (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT);
-	rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] =
-	    (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT);
-}
-
-static int bit_count (int x)
-{
-    x = ((x & 0xaaaaaaaaU) >> 1) + (x & 0x55555555U);
-    x = ((x & 0xccccccccU) >> 2) + (x & 0x33333333U);
-    x = (x >> 16) + (x & 0xffff);
-    x = ((x & 0xf0f0) >> 4) + (x & 0x0f0f);
-    return (x >> 8) + (x & 0x00ff);
-}
-
-static void r300SetupRealVertexProgram(r300ContextPtr rmesa)
-{
-	GLcontext *ctx = rmesa->radeon.glCtx;
-	struct r300_vertex_program *prog = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
-	int inst_count = 0;
-	int param_count = 0;
-
-	/* FIXME: r300SetupVertexProgramFragment */
-	R300_STATECHANGE(rmesa, vpp);
-	param_count =
-	    r300VertexProgUpdateParams(ctx,
-				       (struct r300_vertex_program_cont *)
-				       ctx->VertexProgram._Current,
-				       (float *)&rmesa->hw.vpp.
-				       cmd[R300_VPP_PARAM_0]);
-	bump_vpu_count(rmesa->hw.vpp.cmd, param_count);
-	param_count /= 4;
-
-	r300SetupVertexProgramFragment(rmesa, R300_PVS_CODE_START, &(prog->program));
-	inst_count = (prog->program.length / 4) - 1;
-
-	r300VapCntl(rmesa, bit_count(prog->key.InputsRead),
-		    bit_count(prog->key.OutputsWritten), prog->num_temporaries);
-
-	R300_STATECHANGE(rmesa, pvs);
-	rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] =
-	  (0 << R300_PVS_FIRST_INST_SHIFT) |
-	  (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) |
-	  (inst_count << R300_PVS_LAST_INST_SHIFT);
-	rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] =
-	  (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) |
-	  (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT);
-	rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] =
-	  (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT);
-}
-
-static void r300SetupVertexProgram(r300ContextPtr rmesa)
-{
-	GLcontext *ctx = rmesa->radeon.glCtx;
-
-	/* Reset state, in case we don't use something */
-	((drm_r300_cmd_header_t *) rmesa->hw.vpp.cmd)->vpu.count = 0;
-	((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0;
-	((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0;
-
-	/* Not sure why this doesnt work...
-	   0x400 area might have something to do with pixel shaders as it appears right after pfs programming.
-	   0x406 is set to { 0.0, 0.0, 1.0, 0.0 } most of the time but should change with smooth points and in other rare cases. */
-	//setup_vertex_shader_fragment(rmesa, 0x406, &unk4);
-	if (hw_tcl_on && ((struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx))->translated) {
-		r300SetupRealVertexProgram(rmesa);
-	} else {
-		/* FIXME: This needs to be replaced by vertex shader generation code. */
-		r300SetupDefaultVertexProgram(rmesa);
-	}
-
-}
-
 /**
  * Enable/Disable states.
  *
@@ -2015,20 +1753,13 @@ static void r300SetupVertexProgram(r300ContextPtr rmesa)
  */
 static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state)
 {
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	if (RADEON_DEBUG & DEBUG_STATE)
 		fprintf(stderr, "%s( %s = %s )\n", __FUNCTION__,
 			_mesa_lookup_enum_by_nr(cap),
 			state ? "GL_TRUE" : "GL_FALSE");
 
 	switch (cap) {
-	case GL_TEXTURE_1D:
-	case GL_TEXTURE_2D:
-	case GL_TEXTURE_3D:
-		/* empty */
-		break;
-	case GL_FOG:
-		/* empty */
-		break;
 	case GL_ALPHA_TEST:
 		r300SetAlphaState(ctx);
 		break;
@@ -2046,22 +1777,46 @@ static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state)
 	case GL_CLIP_PLANE5:
 		r300SetClipPlaneState(ctx, cap, state);
 		break;
+	case GL_CULL_FACE:
+		r300UpdateCulling(ctx);
+		break;
 	case GL_DEPTH_TEST:
 		r300SetDepthState(ctx);
 		break;
-	case GL_STENCIL_TEST:
-		r300SetStencilState(ctx, state);
+	case GL_LINE_SMOOTH:
+		if (rmesa->options.conformance_mode)
+			r300SwitchFallback(ctx, R300_FALLBACK_LINE_SMOOTH, ctx->Line.SmoothFlag);
 		break;
-	case GL_CULL_FACE:
-		r300UpdateCulling(ctx);
+	case GL_LINE_STIPPLE:
+		if (rmesa->options.conformance_mode)
+			r300SwitchFallback(ctx, R300_FALLBACK_LINE_STIPPLE, ctx->Line.StippleFlag);
+		break;
+	case GL_POINT_SMOOTH:
+		if (rmesa->options.conformance_mode)
+			r300SwitchFallback(ctx, R300_FALLBACK_POINT_SMOOTH, ctx->Point.SmoothFlag);
+		break;
+	case GL_POLYGON_SMOOTH:
+		if (rmesa->options.conformance_mode)
+			r300SwitchFallback(ctx, R300_FALLBACK_POLYGON_SMOOTH, ctx->Polygon.SmoothFlag);
+		break;
+	case GL_POLYGON_STIPPLE:
+		if (rmesa->options.conformance_mode)
+			r300SwitchFallback(ctx, R300_FALLBACK_POLYGON_STIPPLE, ctx->Polygon.StippleFlag);
 		break;
 	case GL_POLYGON_OFFSET_POINT:
 	case GL_POLYGON_OFFSET_LINE:
 	case GL_POLYGON_OFFSET_FILL:
 		r300SetPolygonOffsetState(ctx, state);
 		break;
+	case GL_SCISSOR_TEST:
+		radeon_firevertices(&rmesa->radeon);
+		rmesa->radeon.state.scissor.enabled = state;
+		radeonUpdateScissor( ctx );
+		break;
+	case GL_STENCIL_TEST:
+		r300SetStencilState(ctx, state);
+		break;
 	default:
-		radeonEnable(ctx, cap, state);
 		break;
 	}
 }
@@ -2072,15 +1827,14 @@ static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state)
 static void r300ResetHwState(r300ContextPtr r300)
 {
 	GLcontext *ctx = r300->radeon.glCtx;
-	int has_tcl = 1;
+	int has_tcl;
 
-	if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
-		has_tcl = 0;
+	has_tcl = r300->options.hw_tcl_enabled;
 
 	if (RADEON_DEBUG & DEBUG_STATE)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
-	r300UpdateWindow(ctx);
+	radeon_firevertices(&r300->radeon);
 
 	r300ColorMask(ctx,
 		      ctx->Color.ColorMask[RCOMP],
@@ -2102,8 +1856,6 @@ static void r300ResetHwState(r300ContextPtr r300)
 
 	r300UpdateCulling(ctx);
 
-	r300UpdateTextureState(ctx);
-
 	r300SetBlendState(ctx);
 	r300SetLogicOpState(ctx);
 
@@ -2182,8 +1934,8 @@ static void r300ResetHwState(r300ContextPtr r300)
 	}
 
 	/* XXX: Enable anti-aliasing? */
-	r300->hw.gb_misc.cmd[R300_GB_MISC_AA_CONFIG] = GB_AA_CONFIG_AA_DISABLE;
-	r300->hw.gb_misc.cmd[R300_GB_MISC_SELECT] = 0;
+	r300->hw.gb_misc2.cmd[R300_GB_MISC2_AA_CONFIG] = GB_AA_CONFIG_AA_DISABLE;
+	r300->hw.gb_misc2.cmd[R300_GB_MISC2_SELECT] = 0;
 
 	r300->hw.ga_point_s0.cmd[1] = r300PackFloat32(0.0);
 	r300->hw.ga_point_s0.cmd[2] = r300PackFloat32(0.0);
@@ -2242,20 +1994,6 @@ static void r300ResetHwState(r300ContextPtr r300)
 
 	r300BlendColor(ctx, ctx->Color.BlendColor);
 
-	/* Again, r300ClearBuffer uses this */
-	r300->hw.cb.cmd[R300_CB_OFFSET] =
-	    r300->radeon.state.color.drawOffset +
-	    r300->radeon.radeonScreen->fbLocation;
-	r300->hw.cb.cmd[R300_CB_PITCH] = r300->radeon.state.color.drawPitch;
-
-	if (r300->radeon.radeonScreen->cpp == 4)
-		r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_ARGB8888;
-	else
-		r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_RGB565;
-
-	if (r300->radeon.sarea->tiling_enabled)
-		r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_TILE_ENABLE;
-
 	r300->hw.rb3d_dither_ctl.cmd[1] = 0;
 	r300->hw.rb3d_dither_ctl.cmd[2] = 0;
 	r300->hw.rb3d_dither_ctl.cmd[3] = 0;
@@ -2268,44 +2006,18 @@ static void r300ResetHwState(r300ContextPtr r300)
 
 	r300->hw.rb3d_aaresolve_ctl.cmd[1] = 0;
 
-	r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[1] = 0x00000000;
-	r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[2] = 0xffffffff;
-
-	r300->hw.zb.cmd[R300_ZB_OFFSET] =
-	    r300->radeon.radeonScreen->depthOffset +
-	    r300->radeon.radeonScreen->fbLocation;
-	r300->hw.zb.cmd[R300_ZB_PITCH] = r300->radeon.radeonScreen->depthPitch;
-
-	if (r300->radeon.sarea->tiling_enabled) {
-		/* XXX: Turn off when clearing buffers ? */
-		r300->hw.zb.cmd[R300_ZB_PITCH] |= R300_DEPTHMACROTILE_ENABLE;
-
-		if (ctx->Visual.depthBits == 24)
-			r300->hw.zb.cmd[R300_ZB_PITCH] |=
-			    R300_DEPTHMICROTILE_TILED;
-	}
+    r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[1] = 0x00000000;
+    r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[2] = 0xffffffff;
 
 	r300->hw.zb_depthclearvalue.cmd[1] = 0;
 
-	switch (ctx->Visual.depthBits) {
-	case 16:
-		r300->hw.zstencil_format.cmd[1] = R300_DEPTHFORMAT_16BIT_INT_Z;
-		break;
-	case 24:
-		r300->hw.zstencil_format.cmd[1] = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
-		break;
-	default:
-		fprintf(stderr, "Error: Unsupported depth %d... exiting\n", ctx->Visual.depthBits);
-		_mesa_exit(-1);
-	}
-
 	r300->hw.zstencil_format.cmd[2] = R300_ZTOP_DISABLE;
 	r300->hw.zstencil_format.cmd[3] = 0x00000003;
 	r300->hw.zstencil_format.cmd[4] = 0x00000000;
 	r300SetEarlyZState(ctx);
 
-	r300->hw.unk4F30.cmd[1] = 0;
-	r300->hw.unk4F30.cmd[2] = 0;
+	r300->hw.zb_zmask.cmd[1] = 0;
+	r300->hw.zb_zmask.cmd[2] = 0;
 
 	r300->hw.zb_hiz_offset.cmd[1] = 0;
 
@@ -2319,20 +2031,26 @@ static void r300ResetHwState(r300ContextPtr r300)
 		r300->hw.vps.cmd[R300_VPS_ZERO_3] = 0;
 	}
 
-	r300->hw.all_dirty = GL_TRUE;
+	r300->radeon.hw.all_dirty = GL_TRUE;
 }
 
 void r300UpdateShaders(r300ContextPtr rmesa)
 {
 	GLcontext *ctx;
-	struct r300_vertex_program *vp;
+	struct r300_fragment_program *fp;
 	int i;
 
 	ctx = rmesa->radeon.glCtx;
+	fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
 
-	if (rmesa->NewGLState && hw_tcl_on) {
-		rmesa->NewGLState = 0;
+	/* should only happenen once, just after context is created */
+	/* TODO: shouldn't we fallback to sw here? */
+	if (!fp) {
+		_mesa_fprintf(stderr, "No ctx->FragmentProgram._Current!!\n");
+		return;
+	}
 
+	if (rmesa->radeon.NewGLState && rmesa->options.hw_tcl_enabled) {
 		for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
 			rmesa->temp_attrib[i] =
 			    TNL_CONTEXT(ctx)->vb.AttribPtr[i];
@@ -2348,21 +2066,16 @@ void r300UpdateShaders(r300ContextPtr rmesa)
 		}
 
 		r300SelectVertexShader(rmesa);
-		vp = (struct r300_vertex_program *)
-		    CURRENT_VERTEX_SHADER(ctx);
-		/*if (vp->translated == GL_FALSE)
-		   r300TranslateVertexShader(vp); */
-		if (vp->translated == GL_FALSE) {
-			fprintf(stderr, "Failing back to sw-tcl\n");
-			hw_tcl_on = future_hw_tcl_on = 0;
-			r300ResetHwState(rmesa);
-
-			r300UpdateStateParameters(ctx, _NEW_PROGRAM |
-                                                  _NEW_PROGRAM_CONSTANTS);
-			return;
-		}
+		r300SwitchFallback(ctx, R300_FALLBACK_VERTEX_PROGRAM, rmesa->selected_vp->error);
 	}
+
+	if (!fp->translated || rmesa->radeon.NewGLState)
+		r300TranslateFragmentShader(ctx, ctx->FragmentProgram._Current);
+
+	r300SwitchFallback(ctx, R300_FALLBACK_FRAGMENT_PROGRAM, fp->error);
+
 	r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+	rmesa->radeon.NewGLState = 0;
 }
 
 static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx,
@@ -2386,35 +2099,23 @@ static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx,
 }
 
 
-static void r300SetupPixelShader(r300ContextPtr rmesa)
+static void r300SetupPixelShader(GLcontext *ctx)
 {
-	GLcontext *ctx = rmesa->radeon.glCtx;
-	struct r300_fragment_program *fp = (struct r300_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
 	struct r300_fragment_program_code *code;
 	int i, k;
 
-	if (!fp)		/* should only happenen once, just after context is created */
-		return;
-
-	r300TranslateFragmentShader(rmesa, fp);
-	if (!fp->translated) {
-		fprintf(stderr, "%s: No valid fragment shader, exiting\n",
-			__FUNCTION__);
-		return;
-	}
-	code = &fp->code;
-
-	r300SetupTextures(ctx);
+	code = &fp->code.r300;
 
 	R300_STATECHANGE(rmesa, fpi[0]);
 	R300_STATECHANGE(rmesa, fpi[1]);
 	R300_STATECHANGE(rmesa, fpi[2]);
 	R300_STATECHANGE(rmesa, fpi[3]);
-	rmesa->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_INST_0, code->alu.length);
-	rmesa->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_ADDR_0, code->alu.length);
-	rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_INST_0, code->alu.length);
-	rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_ADDR_0, code->alu.length);
+	rmesa->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_RGB_INST_0, code->alu.length);
+	rmesa->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_RGB_ADDR_0, code->alu.length);
+	rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_ALPHA_INST_0, code->alu.length);
+	rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_ALPHA_ADDR_0, code->alu.length);
 	for (i = 0; i < code->alu.length; i++) {
 		rmesa->hw.fpi[0].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst0;
 		rmesa->hw.fpi[1].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst1;
@@ -2445,10 +2146,10 @@ static void r300SetupPixelShader(r300ContextPtr rmesa)
 	}
 
 	R300_STATECHANGE(rmesa, fpp);
-	rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(R300_PFS_PARAM_0_X, code->const_nr * 4);
+	rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_PFS_PARAM_0_X, code->const_nr * 4);
 	for (i = 0; i < code->const_nr; i++) {
 		const GLfloat *constant = get_fragmentprogram_constant(ctx,
-			&fp->mesa_program.Base, code->constant[i]);
+			&fp->Base.Base, code->constant[i]);
 		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat24(constant[0]);
 		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat24(constant[1]);
 		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat24(constant[2]);
@@ -2470,29 +2171,17 @@ static void r300SetupPixelShader(r300ContextPtr rmesa)
 	if(_nc>_p->r500fp.count)_p->r500fp.count=_nc;\
 } while(0)
 
-static void r500SetupPixelShader(r300ContextPtr rmesa)
+static void r500SetupPixelShader(GLcontext *ctx)
 {
-	GLcontext *ctx = rmesa->radeon.glCtx;
-	struct r500_fragment_program *fp = (struct r500_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
 	int i;
 	struct r500_fragment_program_code *code;
 
-	if (!fp)		/* should only happenen once, just after context is created */
-		return;
-
 	((drm_r300_cmd_header_t *) rmesa->hw.r500fp.cmd)->r500fp.count = 0;
 	((drm_r300_cmd_header_t *) rmesa->hw.r500fp_const.cmd)->r500fp.count = 0;
 
-	r500TranslateFragmentShader(rmesa, fp);
-	if (!fp->translated) {
-		fprintf(stderr, "%s: No valid fragment shader, exiting\n",
-			__FUNCTION__);
-		return;
-	}
-	code = &fp->code;
-
-	r300SetupTextures(ctx);
+	code = &fp->code.r500;
 
 	R300_STATECHANGE(rmesa, fp);
 	rmesa->hw.fp.cmd[R500_FP_PIXSIZE] = code->max_temp_idx;
@@ -2522,58 +2211,96 @@ static void r500SetupPixelShader(r300ContextPtr rmesa)
 	R300_STATECHANGE(rmesa, r500fp_const);
 	for (i = 0; i < code->const_nr; i++) {
 		const GLfloat *constant = get_fragmentprogram_constant(ctx,
-			&fp->mesa_program.Base, code->constant[i]);
+			&fp->Base.Base, code->constant[i]);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat32(constant[0]);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat32(constant[1]);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat32(constant[2]);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 3] = r300PackFloat32(constant[3]);
 	}
 	bump_r500fp_const_count(rmesa->hw.r500fp_const.cmd, code->const_nr * 4);
+}
+
+void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten)
+{
+	r300ContextPtr rmesa = R300_CONTEXT( ctx );
+	struct vertex_attribute *attrs = rmesa->vbuf.attribs;
+	int i, j, reg_count;
+	uint32_t *vir0 = &rmesa->hw.vir[0].cmd[1];
+	uint32_t *vir1 = &rmesa->hw.vir[1].cmd[1];
+
+	for (i = 0; i < R300_VIR_CMDSIZE-1; ++i)
+		vir0[i] = vir1[i] = 0;
+
+	for (i = 0, j = 0; i < rmesa->vbuf.num_attribs; ++i) {
+		int tmp;
+
+		tmp = attrs[i].data_type | (attrs[i].dst_loc << R300_DST_VEC_LOC_SHIFT);
+		if (attrs[i]._signed)
+			tmp |= R300_SIGNED;
+		if (attrs[i].normalize)
+			tmp |= R300_NORMALIZE;
+
+		if (i % 2 == 0) {
+			vir0[j] = tmp << R300_DATA_TYPE_0_SHIFT;
+			vir1[j] = attrs[i].swizzle | (attrs[i].write_mask << R300_WRITE_ENA_SHIFT);
+		} else {
+			vir0[j] |= tmp << R300_DATA_TYPE_1_SHIFT;
+			vir1[j] |= (attrs[i].swizzle | (attrs[i].write_mask << R300_WRITE_ENA_SHIFT)) << R300_SWIZZLE1_SHIFT;
+			++j;
+		}
+	}
+
+	reg_count = (rmesa->vbuf.num_attribs + 1) >> 1;
+	if (rmesa->vbuf.num_attribs % 2 != 0) {
+		vir0[reg_count-1] |= R300_LAST_VEC << R300_DATA_TYPE_0_SHIFT;
+	} else {
+		vir0[reg_count-1] |= R300_LAST_VEC << R300_DATA_TYPE_1_SHIFT;
+	}
+
+	R300_STATECHANGE(rmesa, vir[0]);
+	R300_STATECHANGE(rmesa, vir[1]);
+	R300_STATECHANGE(rmesa, vof);
+	R300_STATECHANGE(rmesa, vic);
+
+	if (rmesa->radeon.radeonScreen->kernel_mm) {
+		rmesa->hw.vir[0].cmd[0] &= 0xC000FFFF;
+		rmesa->hw.vir[1].cmd[0] &= 0xC000FFFF;
+		rmesa->hw.vir[0].cmd[0] |= (reg_count & 0x3FFF) << 16;
+		rmesa->hw.vir[1].cmd[0] |= (reg_count & 0x3FFF) << 16;
+	} else {
+		((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count = reg_count;
+		((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count = reg_count;
+	}
 
+	rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
+	rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
+	rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead);
+	rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = r300VAPOutputCntl1(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead);
 }
 
 void r300UpdateShaderStates(r300ContextPtr rmesa)
 {
 	GLcontext *ctx;
 	ctx = rmesa->radeon.glCtx;
+	struct r300_fragment_program *r300_fp;
 
-	r300UpdateTextureState(ctx);
-	r300SetEarlyZState(ctx);
+	r300_fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
 
-	/* w_fmt value is set to get best performance
-	 * see p.130 R5xx 3D acceleration guide v1.3 */
-	GLuint w_fmt, fgdepthsrc;
-	if (current_fragment_program_writes_depth(ctx)) {
-		fgdepthsrc = R300_FG_DEPTH_SRC_SHADER;
-		w_fmt = R300_W_FMT_W24 | R300_W_SRC_US;
-	} else {
-		fgdepthsrc = R300_FG_DEPTH_SRC_SCAN;
-		w_fmt = R300_W_FMT_W0 | R300_W_SRC_US;
-	}
+	/* should only happenen once, just after context is created */
+	if (!r300_fp)
+		return;
 
-	if (w_fmt != rmesa->hw.us_out_fmt.cmd[5]) {
-		R300_STATECHANGE(rmesa, us_out_fmt);
-		rmesa->hw.us_out_fmt.cmd[5] = w_fmt;
-	}
+	r300SetEarlyZState(ctx);
 
-	if (fgdepthsrc != rmesa->hw.fg_depth_src.cmd[1]) {
-		R300_STATECHANGE(rmesa, fg_depth_src);
-		rmesa->hw.fg_depth_src.cmd[1] = fgdepthsrc;
-	}
+	r300SetupTextures(ctx);
 
-	if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
-		r500SetupPixelShader(rmesa);
-	else
-		r300SetupPixelShader(rmesa);
+	rmesa->vtbl.SetupPixelShader(ctx);
 
-	if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
-		r500SetupRSUnit(ctx);
-	else
-		r300SetupRSUnit(ctx);
+	rmesa->vtbl.SetupRSUnit(ctx);
 
-	if ((rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL))
+	if (rmesa->options.hw_tcl_enabled) {
 		r300SetupVertexProgram(rmesa);
-
+	}
 }
 
 /**
@@ -2587,15 +2314,18 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state)
 	_swsetup_InvalidateState(ctx, new_state);
 	_vbo_InvalidateState(ctx, new_state);
 	_tnl_InvalidateState(ctx, new_state);
-	_ae_invalidate_state(ctx, new_state);
 
 	if (new_state & _NEW_BUFFERS) {
-		r300UpdateDrawBuffer(ctx);
+		_mesa_update_framebuffer(ctx);
+		/* this updates the DrawBuffer's Width/Height if it's a FBO */
+		_mesa_update_draw_buffer_bounds(ctx);
+
+		R300_STATECHANGE(r300, cb);
 	}
 
 	r300UpdateStateParameters(ctx, new_state);
 
-	r300->NewGLState |= new_state;
+	r300->radeon.NewGLState |= new_state;
 }
 
 /**
@@ -2605,58 +2335,12 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state)
  */
 void r300InitState(r300ContextPtr r300)
 {
-	GLcontext *ctx = r300->radeon.glCtx;
-	GLuint depth_fmt;
-
-	radeonInitState(&r300->radeon);
-
-	switch (ctx->Visual.depthBits) {
-	case 16:
-		r300->state.depth.scale = 1.0 / (GLfloat) 0xffff;
-		depth_fmt = R300_DEPTHFORMAT_16BIT_INT_Z;
-		break;
-	case 24:
-		r300->state.depth.scale = 1.0 / (GLfloat) 0xffffff;
-		depth_fmt = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
-		break;
-	default:
-		fprintf(stderr, "Error: Unsupported depth %d... exiting\n",
-			ctx->Visual.depthBits);
-		_mesa_exit(-1);
-	}
-
-	/* Only have hw stencil when depth buffer is 24 bits deep */
-	r300->state.stencil.hw_stencil = (ctx->Visual.stencilBits > 0 &&
-					  ctx->Visual.depthBits == 24);
-
-	memset(&(r300->state.texture), 0, sizeof(r300->state.texture));
-
 	r300ResetHwState(r300);
 }
 
 static void r300RenderMode(GLcontext * ctx, GLenum mode)
 {
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	(void)rmesa;
-	(void)mode;
-}
-
-void r300UpdateClipPlanes( GLcontext *ctx )
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	GLuint p;
-
-	for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
-		if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
-			GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p];
-
-			R300_STATECHANGE( rmesa, vpucp[p] );
-			rmesa->hw.vpucp[p].cmd[R300_VPUCP_X] = ip[0];
-			rmesa->hw.vpucp[p].cmd[R300_VPUCP_Y] = ip[1];
-			rmesa->hw.vpucp[p].cmd[R300_VPUCP_Z] = ip[2];
-			rmesa->hw.vpucp[p].cmd[R300_VPUCP_W] = ip[3];
-		}
-	}
+	r300SwitchFallback(ctx, R300_FALLBACK_RENDER_MODE, ctx->RenderMode != GL_RENDER);
 }
 
 /**
@@ -2664,7 +2348,6 @@ void r300UpdateClipPlanes( GLcontext *ctx )
  */
 void r300InitStateFuncs(struct dd_function_table *functions)
 {
-	radeonInitStateFuncs(functions);
 
 	functions->UpdateState = r300InvalidateState;
 	functions->AlphaFunc = r300AlphaFunc;
@@ -2700,4 +2383,25 @@ void r300InitStateFuncs(struct dd_function_table *functions)
 	functions->RenderMode = r300RenderMode;
 
 	functions->ClipPlane = r300ClipPlane;
+	functions->Scissor = radeonScissor;
+
+	functions->DrawBuffer		= radeonDrawBuffer;
+	functions->ReadBuffer		= radeonReadBuffer;
+}
+
+void r300InitShaderFunctions(r300ContextPtr r300)
+{
+	if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+		r300->vtbl.SetupRSUnit = r500SetupRSUnit;
+		r300->vtbl.SetupPixelShader = r500SetupPixelShader;
+		r300->vtbl.SetupFragmentShaderTextures = r500SetupFragmentShaderTextures;
+		r300->vtbl.BuildFragmentProgramHwCode = r500BuildFragmentProgramHwCode;
+		r300->vtbl.FragmentProgramDump = r500FragmentProgramDump;
+	} else {
+		r300->vtbl.SetupRSUnit = r300SetupRSUnit;
+		r300->vtbl.SetupPixelShader = r300SetupPixelShader;
+		r300->vtbl.SetupFragmentShaderTextures = r300SetupFragmentShaderTextures;
+		r300->vtbl.BuildFragmentProgramHwCode = r300BuildFragmentProgramHwCode;
+		r300->vtbl.FragmentProgramDump = r300FragmentProgramDump;
+	}
 }
diff --git a/src/mesa/drivers/dri/r300/r300_state.h b/src/mesa/drivers/dri/r300/r300_state.h
index 0589ab7cad9..23282894202 100644
--- a/src/mesa/drivers/dri/r300/r300_state.h
+++ b/src/mesa/drivers/dri/r300/r300_state.h
@@ -39,42 +39,25 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #define R300_NEWPRIM( rmesa )			\
   do {						\
-    if ( rmesa->dma.flush )			\
-      rmesa->dma.flush( rmesa );		\
+  if ( rmesa->radeon.dma.flush )			\
+    rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	\
   } while (0)
 
 #define R300_STATECHANGE(r300, atom) \
 	do {						\
 	  R300_NEWPRIM(r300);				\
 		r300->hw.atom.dirty = GL_TRUE;		\
-		r300->hw.is_dirty = GL_TRUE;		\
+		r300->radeon.hw.is_dirty = GL_TRUE;		\
 	} while(0)
 
-#define R300_PRINT_STATE(r300, atom) \
-		r300PrintStateAtom(r300, &r300->hw.atom)
-
-/* Fire the buffered vertices no matter what.
-   TODO: This has not been implemented yet
- */
-#define R300_FIREVERTICES( r300 )			\
-do {							\
-    \
-   if ( (r300)->cmdbuf.count_used || (r300)->dma.flush ) {	\
-      r300Flush( (r300)->radeon.glCtx );		\
-   }							\
-    \
-} while (0)
-
-// r300_state.c
-extern int future_hw_tcl_on;
-void _tnl_UpdateFixedFunctionProgram (GLcontext * ctx);
 void r300UpdateViewportOffset (GLcontext * ctx);
 void r300UpdateDrawBuffer (GLcontext * ctx);
 void r300UpdateStateParameters (GLcontext * ctx, GLuint new_state);
 void r300UpdateShaders (r300ContextPtr rmesa);
 void r300UpdateShaderStates (r300ContextPtr rmesa);
 void r300InitState (r300ContextPtr r300);
-void r300UpdateClipPlanes (GLcontext * ctx);
 void r300InitStateFuncs (struct dd_function_table *functions);
+void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, GLuint output_count, GLuint temp_count);
+void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten);
 
 #endif				/* __R300_STATE_H__ */
diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c
index ba3621b16b2..ce4179208eb 100644
--- a/src/mesa/drivers/dri/r300/r300_swtcl.c
+++ b/src/mesa/drivers/dri/r300/r300_swtcl.c
@@ -28,362 +28,237 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /*
  * Authors:
  *   Dave Airlie <[email protected]>
+ *   Maciej Cencora <[email protected]>
  */
 
-/* derived from r200 swtcl path */
-
-
-
-#include "main/glheader.h"
-#include "main/mtypes.h"
-#include "main/colormac.h"
-#include "main/enums.h"
-#include "main/image.h"
-#include "main/imports.h"
-#include "main/light.h"
-#include "main/macros.h"
-
-#include "swrast/s_context.h"
-#include "swrast/s_fog.h"
-#include "swrast_setup/swrast_setup.h"
-#include "math/m_translate.h"
 #include "tnl/tnl.h"
-#include "tnl/t_context.h"
 #include "tnl/t_pipeline.h"
 
-#include "r300_context.h"
-#include "r300_swtcl.h"
 #include "r300_state.h"
-#include "r300_ioctl.h"
+#include "r300_swtcl.h"
 #include "r300_emit.h"
-#include "r300_mem.h"
-
-static void flush_last_swtcl_prim( r300ContextPtr rmesa  );
+#include "r300_tex.h"
+#include "r300_render.h"
 
-
-void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, GLuint offset);
-void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr);
 #define EMIT_ATTR( ATTR, STYLE )					\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = (ATTR);	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = (STYLE);	\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR);	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE);	\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
 } while (0)
 
 #define EMIT_PAD( N )							\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = 0;		\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = EMIT_PAD;	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].offset = (N);		\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0;		\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD;	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N);		\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
+} while (0)
+
+#define ADD_ATTR(_attr, _format, _dst_loc, _swizzle, _write_mask, _normalize) \
+do { \
+	attrs[num_attrs].element = (_attr); \
+	attrs[num_attrs].data_type = (_format); \
+	attrs[num_attrs].dst_loc = (_dst_loc); \
+	attrs[num_attrs].swizzle = (_swizzle); \
+	attrs[num_attrs].write_mask = (_write_mask); \
+	attrs[num_attrs]._signed = 0; \
+	attrs[num_attrs].normalize = (_normalize); \
+	++num_attrs; \
 } while (0)
 
-static void r300SetVertexFormat( GLcontext *ctx )
+void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead,  GLuint *_OutputsWritten)
 {
 	r300ContextPtr rmesa = R300_CONTEXT( ctx );
 	TNLcontext *tnl = TNL_CONTEXT(ctx);
 	struct vertex_buffer *VB = &tnl->vb;
-	DECLARE_RENDERINPUTS(index_bitset);
-	GLuint InputsRead = 0, OutputsWritten = 0;
-	int vap_fmt_1 = 0;
-	int offset = 0;
-	int vte = 0;
-	int fog_id;
-	GLint inputs[VERT_ATTRIB_MAX];
-	GLint tab[VERT_ATTRIB_MAX];
-	int swizzle[VERT_ATTRIB_MAX][4];
-	GLuint i, nr;
-	GLuint sz;
-
-	DECLARE_RENDERINPUTS(render_inputs_bitset);
-	RENDERINPUTS_COPY(render_inputs_bitset, tnl->render_inputs_bitset);
-	RENDERINPUTS_COPY( index_bitset, tnl->render_inputs_bitset );
-	RENDERINPUTS_COPY(rmesa->state.render_inputs_bitset, render_inputs_bitset);
-
-	vte = rmesa->hw.vte.cmd[1];
-	vte &= ~(R300_VTX_XY_FMT | R300_VTX_Z_FMT | R300_VTX_W0_FMT);
-	/* Important:
-	 */
-	if ( VB->NdcPtr != NULL ) {
-		VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
-		vte |= R300_VTX_XY_FMT | R300_VTX_Z_FMT;
-	}
-	else {
-		VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
-		vte |= R300_VTX_W0_FMT;
-	}
-
-	assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
-	rmesa->swtcl.vertex_attr_count = 0;
-
-	/* EMIT_ATTR's must be in order as they tell t_vertex.c how to
-	 * build up a hardware vertex.
-	 */
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POS)) {
-		sz = VB->AttribPtr[VERT_ATTRIB_POS]->size;
-		InputsRead |= 1 << VERT_ATTRIB_POS;
-		OutputsWritten |= 1 << VERT_RESULT_HPOS;
-		EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_1F + sz - 1 );
-		offset = sz;
-	} else {
-		offset = 4;
-		EMIT_PAD(4 * sizeof(float));
-	}
-/*
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POINTSIZE )) {
-		EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F );
-		offset += 1;
-	}
-*/
-	if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_COLOR0)) {
-		sz = VB->AttribPtr[VERT_ATTRIB_COLOR0]->size;
-	        rmesa->swtcl.coloroffset = offset;
+	int first_free_tex = 0;
+	GLuint InputsRead = 0;
+	GLuint OutputsWritten = 0;
+	int num_attrs = 0;
+	GLuint fp_reads = ctx->FragmentProgram._Current->Base.InputsRead;
+	struct vertex_attribute *attrs = rmesa->vbuf.attribs;
+
+	rmesa->swtcl.coloroffset = rmesa->swtcl.specoffset = 0;
+	rmesa->radeon.swtcl.vertex_attr_count = 0;
+
+	/* We always want non Ndc coords format */
+	VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
+
+	/* Always write position vector */
+	InputsRead |= 1 << VERT_ATTRIB_POS;
+	OutputsWritten |= 1 << VERT_RESULT_HPOS;
+	EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F );
+	ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_POS, SWIZZLE_XYZW, MASK_XYZW, 0);
+	rmesa->swtcl.coloroffset = 4;
+
+	if (fp_reads & FRAG_BIT_COL0) {
 		InputsRead |= 1 << VERT_ATTRIB_COLOR0;
 		OutputsWritten |= 1 << VERT_RESULT_COL0;
-		EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_1F + sz - 1 );
-		offset += sz;
+#if MESA_LITTLE_ENDIAN
+		EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA );
+		ADD_ATTR(VERT_ATTRIB_COLOR0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR0, SWIZZLE_XYZW, MASK_XYZW, 1);
+#else
+		EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR );
+		ADD_ATTR(VERT_ATTRIB_COLOR0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR0, SWIZZLE_XYZW, MASK_XYZW, 1);
+#endif
 	}
 
-	rmesa->swtcl.specoffset = 0;
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 )) {
-		sz = VB->AttribPtr[VERT_ATTRIB_COLOR1]->size;
-		rmesa->swtcl.specoffset = offset;
-		EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_1F + sz - 1 );
+	if (fp_reads & FRAG_BIT_COL1) {
+		GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
 		InputsRead |= 1 << VERT_ATTRIB_COLOR1;
 		OutputsWritten |= 1 << VERT_RESULT_COL1;
+#if MESA_LITTLE_ENDIAN
+		EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4UB_4F_RGBA );
+		ADD_ATTR(VERT_ATTRIB_COLOR1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR1, swiz, MASK_XYZW, 1);
+#else
+		EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4UB_4F_ABGR );
+		ADD_ATTR(VERT_ATTRIB_COLOR1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR1, swiz, MASK_XYZW, 1);
+#endif
+		rmesa->swtcl.specoffset = rmesa->swtcl.coloroffset + 1;
 	}
 
-	fog_id = -1;
-	if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_FOG)) {
-		/* find first free tex coord slot */
-		if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-			int i;
-			for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-				if (!RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-					fog_id = i;
-					break;
-				}
-			}
-		} else {
-			fog_id = 0;
-		}
-
-		if (fog_id == -1) {
-			fprintf(stderr, "\tout of free texcoords to do fog\n");
-			_mesa_exit(-1);
+	if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
+		VB->AttribPtr[VERT_ATTRIB_GENERIC0] = VB->ColorPtr[1];
+		OutputsWritten |= 1 << VERT_RESULT_BFC0;
+#if MESA_LITTLE_ENDIAN
+		EMIT_ATTR( _TNL_ATTRIB_GENERIC0, EMIT_4UB_4F_RGBA );
+		ADD_ATTR(VERT_ATTRIB_GENERIC0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR2, SWIZZLE_XYZW, MASK_XYZW, 1);
+#else
+		EMIT_ATTR( _TNL_ATTRIB_GENERIC0, EMIT_4UB_4F_ABGR );
+		ADD_ATTR(VERT_ATTRIB_GENERIC0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR2, SWIZZLE_XYZW, MASK_XYZW, 1);
+#endif
+		if (fp_reads & FRAG_BIT_COL1) {
+			VB->AttribPtr[VERT_ATTRIB_GENERIC1] = VB->SecondaryColorPtr[1];
+			GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
+			OutputsWritten |= 1 << VERT_RESULT_BFC1;
+#if MESA_LITTLE_ENDIAN
+			EMIT_ATTR( _TNL_ATTRIB_GENERIC1, EMIT_4UB_4F_RGBA );
+			ADD_ATTR(VERT_ATTRIB_GENERIC1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR3, swiz, MASK_XYZW, 1);
+#else
+			EMIT_ATTR( _TNL_ATTRIB_GENERIC1, EMIT_4UB_4F_ABGR );
+			ADD_ATTR(VERT_ATTRIB_GENERIC1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR3, swiz, MASK_XYZW, 1);
+#endif
 		}
+	}
 
-		sz = VB->AttribPtr[VERT_ATTRIB_FOG]->size;
-		EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F + sz - 1);
-		InputsRead |= 1 << VERT_ATTRIB_FOG;
-		OutputsWritten |= 1 << VERT_RESULT_FOGC;
-		vap_fmt_1 |= sz << (3 * fog_id);
+	if (RENDERINPUTS_TEST(tnl->render_inputs_bitset, _TNL_ATTRIB_POINTSIZE )) {
+		GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO);
+		InputsRead |= 1 << VERT_ATTRIB_POINT_SIZE;
+		OutputsWritten |= 1 << VERT_RESULT_PSIZ;
+		EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F );
+		ADD_ATTR(VERT_ATTRIB_POINT_SIZE, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_POINT_SIZE, swiz, MASK_X, 0);
 	}
 
-	if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
+	/**
+	 *  Sending only one texcoord component may lead to lock up,
+	 *  so for all textures always output 4 texcoord components to RS.
+	 */
+	{
 		int i;
-
+		GLuint swiz, format, hw_format;
 		for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-			if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-				sz = VB->TexCoordPtr[i]->size;
+			if (fp_reads & FRAG_BIT_TEX(i)) {
+				switch (VB->TexCoordPtr[i]->size) {
+					case 1:
+						format = EMIT_1F;
+						hw_format = R300_DATA_TYPE_FLOAT_1;
+						swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE);
+						break;
+					case 2:
+						format = EMIT_2F;
+						hw_format = R300_DATA_TYPE_FLOAT_2;
+						swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ONE);
+						break;
+					case 3:
+						format = EMIT_3F;
+						hw_format = R300_DATA_TYPE_FLOAT_3;
+						swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
+						break;
+					case 4:
+						format = EMIT_4F;
+						hw_format = R300_DATA_TYPE_FLOAT_4;
+						swiz = SWIZZLE_XYZW;
+						break;
+					default:
+						continue;
+				}
 				InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i);
 				OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i);
-				EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_1F + sz - 1 );
-				vap_fmt_1 |= sz << (3 * i);
+				EMIT_ATTR(_TNL_ATTRIB_TEX(i), format);
+				ADD_ATTR(VERT_ATTRIB_TEX0 + i, hw_format, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0);
+				++first_free_tex;
 			}
 		}
 	}
 
 	/* RS can't put fragment position on the pixel stack, so stuff it in texcoord if needed */
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POS) && (ctx->FragmentProgram._Current->Base.InputsRead & FRAG_BIT_WPOS)) {
-		int first_free_tex = -1;
-		if (fog_id >= 0) {
-			first_free_tex = fog_id+1;
-		} else {
-			if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-				int i;
-				for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-					if (!RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-						first_free_tex = i;
-						break;
-					}
-				}
-			} else {
-				first_free_tex = 0;
-			}
-		}
-
-		if (first_free_tex == -1) {
+	if (fp_reads & FRAG_BIT_WPOS) {
+		if (first_free_tex >= ctx->Const.MaxTextureUnits) {
 			fprintf(stderr, "\tout of free texcoords to write w pos\n");
 			_mesa_exit(-1);
 		}
 
-		sz = VB->AttribPtr[VERT_ATTRIB_POS]->size;
 		InputsRead |= 1 << (VERT_ATTRIB_TEX0 + first_free_tex);
 		OutputsWritten |= 1 << (VERT_RESULT_TEX0 + first_free_tex);
-		EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_1F + sz - 1 );
-		vap_fmt_1 |= sz << (3 * first_free_tex);
-	}
-
-	for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) {
-		if (InputsRead & (1 << i)) {
-			inputs[i] = nr++;
-		} else {
-			inputs[i] = -1;
-		}
+		EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F );
+		ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_TEX(first_free_tex), SWIZZLE_XYZW, MASK_XYZW, 0);
+		++first_free_tex;
 	}
 
-	/* Fixed, apply to vir0 only */
-	if (InputsRead & (1 << VERT_ATTRIB_POS))
-		inputs[VERT_ATTRIB_POS] = 0;
-	if (InputsRead & (1 << VERT_ATTRIB_COLOR0))
-		inputs[VERT_ATTRIB_COLOR0] = 2;
-	if (InputsRead & (1 << VERT_ATTRIB_COLOR1))
-		inputs[VERT_ATTRIB_COLOR1] = 3;
-	if (InputsRead & (1 << VERT_ATTRIB_FOG))
-		inputs[VERT_ATTRIB_FOG] = 6 + fog_id;
-	for (i = VERT_ATTRIB_TEX0; i <= VERT_ATTRIB_TEX7; i++)
-		if (InputsRead & (1 << i))
-			inputs[i] = 6 + (i - VERT_ATTRIB_TEX0);
-
-	for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) {
-		if (InputsRead & (1 << i)) {
-			tab[nr++] = i;
+	if (fp_reads & FRAG_BIT_FOGC) {
+		if (first_free_tex >= ctx->Const.MaxTextureUnits) {
+			fprintf(stderr, "\tout of free texcoords to write fog coordinate\n");
+			_mesa_exit(-1);
 		}
-	}
-
-	for (i = 0; i < nr; i++) {
-		int ci;
 
-		swizzle[i][0] = SWIZZLE_ZERO;
-		swizzle[i][1] = SWIZZLE_ZERO;
-		swizzle[i][2] = SWIZZLE_ZERO;
-		swizzle[i][3] = SWIZZLE_ONE;
-
-		for (ci = 0; ci < VB->AttribPtr[tab[i]]->size; ci++) {
-			swizzle[i][ci] = ci;
-		}
+		InputsRead |= 1 << VERT_ATTRIB_FOG;
+		OutputsWritten |= 1 << VERT_RESULT_FOGC;
+		GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO);
+		EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F );
+		ADD_ATTR(VERT_ATTRIB_FOG, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0);
 	}
 
 	R300_NEWPRIM(rmesa);
-	R300_STATECHANGE(rmesa, vir[0]);
-	((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count =
-		r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0],
-				   VB->AttribPtr, inputs, tab, nr);
-	R300_STATECHANGE(rmesa, vir[1]);
-	((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count =
-		r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle,
-				   nr);
-
-	R300_STATECHANGE(rmesa, vic);
-	rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
-	rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
-
-	R300_STATECHANGE(rmesa, vof);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = vap_fmt_1;
-
-	rmesa->swtcl.vertex_size =
-		_tnl_install_attrs( ctx,
-				    rmesa->swtcl.vertex_attrs,
-				    rmesa->swtcl.vertex_attr_count,
-				    NULL, 0 );
-
-	rmesa->swtcl.vertex_size /= 4;
+	rmesa->vbuf.num_attribs = num_attrs;
+	*_InputsRead = InputsRead;
+	*_OutputsWritten = OutputsWritten;
 
-	RENDERINPUTS_COPY( rmesa->tnl_index_bitset, index_bitset );
-
-
-	R300_STATECHANGE(rmesa, vte);
-	rmesa->hw.vte.cmd[1] = vte;
-	rmesa->hw.vte.cmd[2] = rmesa->swtcl.vertex_size;
+	RENDERINPUTS_COPY(rmesa->render_inputs_bitset, tnl->render_inputs_bitset);
 }
 
-
-/* Flush vertices in the current dma region.
- */
-static void flush_last_swtcl_prim( r300ContextPtr rmesa  )
+static void r300PrepareVertices(GLcontext *ctx)
 {
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	rmesa->dma.flush = NULL;
-
-	if (rmesa->dma.current.buf) {
-		struct r300_dma_region *current = &rmesa->dma.current;
-		GLuint current_offset = GET_START(current);
-
-		assert (current->start +
-			rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-			current->ptr);
-
-		if (rmesa->dma.current.start != rmesa->dma.current.ptr) {
-
-			r300EnsureCmdBufSpace( rmesa, rmesa->hw.max_state_size + (12*sizeof(int)), __FUNCTION__);
-
-			r300EmitState(rmesa);
-
-			r300EmitVertexAOS( rmesa,
-					   rmesa->swtcl.vertex_size,
-					   current_offset);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	GLuint InputsRead, OutputsWritten;
 
-			r300EmitVbufPrim( rmesa,
-					  rmesa->swtcl.hw_primitive,
-					  rmesa->swtcl.numverts);
+	r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten);
+	r300SetupVAP(ctx, InputsRead, OutputsWritten);
 
-			r300EmitCacheFlush(rmesa);
-		}
+	rmesa->radeon.swtcl.vertex_size =
+		_tnl_install_attrs( ctx,
+				    rmesa->radeon.swtcl.vertex_attrs,
+				    rmesa->radeon.swtcl.vertex_attr_count,
+				    NULL, 0 );
 
-		rmesa->swtcl.numverts = 0;
-		current->start = current->ptr;
-	}
+	rmesa->radeon.swtcl.vertex_size /= 4;
 }
 
-/* Alloc space in the current dma region.
- */
-static void *
-r300AllocDmaLowVerts( r300ContextPtr rmesa, int nverts, int vsize )
-{
-	GLuint bytes = vsize * nverts;
-
-	if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end )
-		r300RefillCurrentDmaRegion( rmesa, bytes);
-
-	if (!rmesa->dma.flush) {
-		rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
-		rmesa->dma.flush = flush_last_swtcl_prim;
-	}
-
-	ASSERT( vsize == rmesa->swtcl.vertex_size * 4 );
-	ASSERT( rmesa->dma.flush == flush_last_swtcl_prim );
-	ASSERT( rmesa->dma.current.start +
-		rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-		rmesa->dma.current.ptr );
-
-	{
-		GLubyte *head = (GLubyte *) (rmesa->dma.current.address + rmesa->dma.current.ptr);
-		rmesa->dma.current.ptr += bytes;
-		rmesa->swtcl.numverts += nverts;
-		return head;
-	}
-}
 
 static GLuint reduced_prim[] = {
-  GL_POINTS,
-  GL_LINES,
-  GL_LINES,
-  GL_LINES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
+	GL_POINTS,
+	GL_LINES,
+	GL_LINES,
+	GL_LINES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
 };
 
 static void r300RasterPrimitive( GLcontext *ctx, GLuint prim );
-static void r300RenderPrimitive( GLcontext *ctx, GLenum prim );
-//static void r300ResetLineStipple( GLcontext *ctx );
 
 /***********************************************************************
  *                    Emit primitives as inline vertices               *
@@ -405,15 +280,13 @@ static void r300RenderPrimitive( GLcontext *ctx, GLenum prim );
 #undef LOCAL_VARS
 #undef ALLOC_VERTS
 #define CTX_ARG r300ContextPtr rmesa
-#define GET_VERTEX_DWORDS() rmesa->swtcl.vertex_size
-#define ALLOC_VERTS( n, size ) r300AllocDmaLowVerts( rmesa, n, size * 4 )
+#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
+#define ALLOC_VERTS( n, size ) rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 )
 #define LOCAL_VARS						\
    r300ContextPtr rmesa = R300_CONTEXT(ctx);		\
-   const char *r300verts = (char *)rmesa->swtcl.verts;
+   const char *r300verts = (char *)rmesa->radeon.swtcl.verts;
 #define VERT(x) (r300Vertex *)(r300verts + ((x) * vertsize * sizeof(int)))
 #define VERTEX r300Vertex
-#define DO_DEBUG_VERTS (1 && (RADEON_DEBUG & DEBUG_VERTS))
-#define PRINT_VERTEX(x)
 #undef TAG
 #define TAG(x) r300_##x
 #include "tnl_dd/t_dd_triemit.h"
@@ -433,9 +306,8 @@ static void r300RenderPrimitive( GLcontext *ctx, GLenum prim );
  *              Build render functions from dd templates               *
  ***********************************************************************/
 
-#define R300_TWOSIDE_BIT	0x01
-#define R300_UNFILLED_BIT	0x02
-#define R300_MAX_TRIFUNC	0x04
+#define R300_UNFILLED_BIT	0x01
+#define R300_MAX_TRIFUNC	0x02
 
 static struct {
    tnl_points_func	        points;
@@ -446,9 +318,9 @@ static struct {
 
 #define DO_FALLBACK  0
 #define DO_UNFILLED (IND & R300_UNFILLED_BIT)
-#define DO_TWOSIDE  (IND & R300_TWOSIDE_BIT)
+#define DO_TWOSIDE   0
 #define DO_FLAT      0
-#define DO_OFFSET     0
+#define DO_OFFSET    0
 #define DO_TRI       1
 #define DO_QUAD      1
 #define DO_LINE      1
@@ -468,33 +340,39 @@ static struct {
 #define VERT_Y(_v) _v->v.y
 #define VERT_Z(_v) _v->v.z
 #define AREA_IS_CCW( a ) (a < 0)
-#define GET_VERTEX(e) (rmesa->swtcl.verts + (e*rmesa->swtcl.vertex_size*sizeof(int)))
-
-/* Only used to pull back colors into vertices (ie, we know color is
- * floating point).
- */
-#define R300_COLOR( dst, src )				\
-do {							\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[0], (src)[2]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[1], (src)[1]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[2], (src)[0]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[3], (src)[3]);	\
+#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + (e*rmesa->radeon.swtcl.vertex_size*sizeof(int)))
+
+#define VERT_SET_RGBA( v, c ) \
+do { \
+   r300_color_t *color = (r300_color_t *)&((v)->ui[coloroffset]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->alpha, (c)[3]); \
 } while (0)
 
-#define VERT_SET_RGBA( v, c )    if (coloroffset) R300_COLOR( v->ub4[coloroffset], c )
-#define VERT_COPY_RGBA( v0, v1 ) if (coloroffset) v0->ui[coloroffset] = v1->ui[coloroffset]
-#define VERT_SAVE_RGBA( idx )    if (coloroffset) color[idx] = v[idx]->ui[coloroffset]
-#define VERT_RESTORE_RGBA( idx ) if (coloroffset) v[idx]->ui[coloroffset] = color[idx]
+#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset]
+
+#define VERT_SET_SPEC( v0, c ) \
+do { \
+   if (specoffset) { \
+   UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.red, (c)[0]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.green, (c)[1]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.blue, (c)[2]); \
+   } \
+} while (0)
 
-#define R300_SPEC( dst, src )				\
-do {							\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[0], (src)[2]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[1], (src)[1]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[2], (src)[0]);	\
+#define VERT_COPY_SPEC( v0, v1 ) \
+do { \
+   if (specoffset) { \
+       v0->v.specular.red = v1->v.specular.red; \
+       v0->v.specular.green = v1->v.specular.green; \
+       v0->v.specular.blue = v1->v.specular.blue; \
+   } \
 } while (0)
 
-#define VERT_SET_SPEC( v, c )    if (specoffset) R300_SPEC( v->ub4[specoffset], c )
-#define VERT_COPY_SPEC( v0, v1 ) if (specoffset) COPY_3V(v0->ub4[specoffset], v1->ub4[specoffset])
+#define VERT_SAVE_RGBA( idx )    color[idx] = v[idx]->ui[coloroffset]
+#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx]
 #define VERT_SAVE_SPEC( idx )    if (specoffset) spec[idx] = v[idx]->ui[specoffset]
 #define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx]
 
@@ -514,7 +392,7 @@ do {							\
  ***********************************************************************/
 
 #define RASTERIZE(x) r300RasterPrimitive( ctx, reduced_prim[x] )
-#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive
+#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive
 #undef TAG
 #define TAG(x) x
 #include "tnl_dd/t_dd_unfilled.h"
@@ -530,26 +408,15 @@ do {							\
 #define TAG(x) x
 #include "tnl_dd/t_dd_tritmp.h"
 
-#define IND (R300_TWOSIDE_BIT)
-#define TAG(x) x##_twoside
-#include "tnl_dd/t_dd_tritmp.h"
-
 #define IND (R300_UNFILLED_BIT)
 #define TAG(x) x##_unfilled
 #include "tnl_dd/t_dd_tritmp.h"
 
-#define IND (R300_TWOSIDE_BIT|R300_UNFILLED_BIT)
-#define TAG(x) x##_twoside_unfilled
-#include "tnl_dd/t_dd_tritmp.h"
-
-
 
 static void init_rast_tab( void )
 {
    init();
-   init_twoside();
    init_unfilled();
-   init_twoside_unfilled();
 }
 
 /**********************************************************************/
@@ -571,8 +438,8 @@ static void init_rast_tab( void )
 #undef LOCAL_VARS
 #define LOCAL_VARS						\
    r300ContextPtr rmesa = R300_CONTEXT(ctx);		\
-   const GLuint vertsize = rmesa->swtcl.vertex_size;		\
-   const char *r300verts = (char *)rmesa->swtcl.verts;		\
+   const GLuint vertsize = rmesa->radeon.swtcl.vertex_size;		\
+   const char *r300verts = (char *)rmesa->radeon.swtcl.verts;		\
    const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
    const GLboolean stipple = ctx->Line.StippleFlag;		\
    (void) elt; (void) stipple;
@@ -601,10 +468,9 @@ static void r300ChooseRenderState( GLcontext *ctx )
 	GLuint index = 0;
 	GLuint flags = ctx->_TriangleCaps;
 
-	if (flags & DD_TRI_LIGHT_TWOSIDE) index |= R300_TWOSIDE_BIT;
 	if (flags & DD_TRI_UNFILLED)      index |= R300_UNFILLED_BIT;
 
-	if (index != rmesa->swtcl.RenderIndex) {
+	if (index != rmesa->radeon.swtcl.RenderIndex) {
 		tnl->Driver.Render.Points = rast_tab[index].points;
 		tnl->Driver.Render.Line = rast_tab[index].line;
 		tnl->Driver.Render.ClippedLine = rast_tab[index].line;
@@ -621,30 +487,32 @@ static void r300ChooseRenderState( GLcontext *ctx )
 			tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
 		}
 
-		rmesa->swtcl.RenderIndex = index;
+		rmesa->radeon.swtcl.RenderIndex = index;
 	}
 }
 
 
-static void r300RenderStart(GLcontext *ctx)
+void r300RenderStart(GLcontext *ctx)
 {
-        r300ContextPtr rmesa = R300_CONTEXT( ctx );
+	r300ContextPtr rmesa = R300_CONTEXT( ctx );
 
 	r300ChooseRenderState(ctx);
-	r300SetVertexFormat(ctx);
+	r300PrepareVertices(ctx);
+
+	r300ValidateBuffers(ctx);
 
 	r300UpdateShaders(rmesa);
 	r300UpdateShaderStates(rmesa);
 
 	r300EmitCacheFlush(rmesa);
 
-	if (rmesa->dma.flush != 0 &&
-	    rmesa->dma.flush != flush_last_swtcl_prim)
-		rmesa->dma.flush( rmesa );
-
+	/* investigate if we can put back flush optimisation if needed */
+	if (rmesa->radeon.dma.flush != NULL) {
+		rmesa->radeon.dma.flush(ctx);
+	}
 }
 
-static void r300RenderFinish(GLcontext *ctx)
+void r300RenderFinish(GLcontext *ctx)
 {
 }
 
@@ -652,28 +520,26 @@ static void r300RasterPrimitive( GLcontext *ctx, GLuint hwprim )
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 
-	if (rmesa->swtcl.hw_primitive != hwprim) {
-	        R300_NEWPRIM( rmesa );
-		rmesa->swtcl.hw_primitive = hwprim;
+	if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
+		R300_NEWPRIM( rmesa );
+		rmesa->radeon.swtcl.hw_primitive = hwprim;
 	}
 }
 
-static void r300RenderPrimitive(GLcontext *ctx, GLenum prim)
+void r300RenderPrimitive(GLcontext *ctx, GLenum prim)
 {
 
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	rmesa->swtcl.render_primitive = prim;
+	rmesa->radeon.swtcl.render_primitive = prim;
 
 	if ((prim == GL_TRIANGLES) && (ctx->_TriangleCaps & DD_TRI_UNFILLED))
-	  return;
+		return;
 
 	r300RasterPrimitive( ctx, reduced_prim[prim] );
 }
 
-static void r300ResetLineStipple(GLcontext *ctx)
+void r300ResetLineStipple(GLcontext *ctx)
 {
-
-
 }
 
 void r300InitSwtcl(GLcontext *ctx)
@@ -699,50 +565,68 @@ void r300InitSwtcl(GLcontext *ctx)
 	_tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12,
 			    48 * sizeof(GLfloat) );
 
-	rmesa->swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
-	rmesa->swtcl.RenderIndex = ~0;
-	rmesa->swtcl.render_primitive = GL_TRIANGLES;
-	rmesa->swtcl.hw_primitive = 0;
+	rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
+	rmesa->radeon.swtcl.RenderIndex = ~0;
+	rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES;
+	rmesa->radeon.swtcl.hw_primitive = 0;
 
 	_tnl_invalidate_vertex_state( ctx, ~0 );
 	_tnl_invalidate_vertices( ctx, ~0 );
-	RENDERINPUTS_ZERO( rmesa->tnl_index_bitset );
 
 	_tnl_need_projected_coords( ctx, GL_FALSE );
-	r300ChooseRenderState(ctx);
 }
 
 void r300DestroySwtcl(GLcontext *ctx)
 {
 }
 
-void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, GLuint offset)
+static void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, struct radeon_bo *bo, GLuint offset)
 {
-	int cmd_reserved = 0;
-	int cmd_written = 0;
+	BATCH_LOCALS(&rmesa->radeon);
 
-	drm_radeon_cmd_header_t *cmd = NULL;
 	if (RADEON_DEBUG & DEBUG_VERTS)
-	  fprintf(stderr, "%s:  vertex_size %d, offset 0x%x \n",
-		  __FUNCTION__, vertex_size, offset);
-
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2), 2);
-	e32(1);
-	e32(vertex_size | (vertex_size << 8));
-	e32(offset);
+		fprintf(stderr, "%s:  vertex_size %d, offset 0x%x \n",
+			__FUNCTION__, vertex_size, offset);
+
+	BEGIN_BATCH(7);
+	OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2);
+	OUT_BATCH(1);
+	OUT_BATCH(vertex_size | (vertex_size << 8));
+	OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+	END_BATCH();
 }
 
-void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr)
+static void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr)
 {
-
-	int cmd_reserved = 0;
-	int cmd_written = 0;
+	BATCH_LOCALS(&rmesa->radeon);
 	int type, num_verts;
-	drm_radeon_cmd_header_t *cmd = NULL;
 
 	type = r300PrimitiveType(rmesa, primitive);
 	num_verts = r300NumVerts(rmesa, vertex_nr, primitive);
 
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0);
-	e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type);
+	BEGIN_BATCH(3);
+	OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
+	OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type);
+	END_BATCH();
+}
+
+void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+
+	rcommonEnsureCmdBufSpace(&rmesa->radeon,
+			   rmesa->radeon.hw.max_state_size + (12*sizeof(int)),
+			   __FUNCTION__);
+	radeonEmitState(&rmesa->radeon);
+    r300_emit_scissor(ctx);
+	r300EmitVertexAOS(rmesa,
+			rmesa->radeon.swtcl.vertex_size,
+			rmesa->radeon.dma.current,
+			current_offset);
+
+	r300EmitVbufPrim(rmesa,
+		   rmesa->radeon.swtcl.hw_primitive,
+		   rmesa->radeon.swtcl.numverts);
+	r300EmitCacheFlush(rmesa);
+	COMMIT_BATCH();
 }
diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.h b/src/mesa/drivers/dri/r300/r300_swtcl.h
index 55df53c1adf..c271d265468 100644
--- a/src/mesa/drivers/dri/r300/r300_swtcl.h
+++ b/src/mesa/drivers/dri/r300/r300_swtcl.h
@@ -39,7 +39,27 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "swrast/swrast.h"
 #include "r300_context.h"
 
+/*
+ * Here are definitions of OVM locations of vertex attributes for non TCL hw
+ */
+#define SWTCL_OVM_POS 0
+#define SWTCL_OVM_COLOR0 2
+#define SWTCL_OVM_COLOR1 3
+#define SWTCL_OVM_COLOR2 4
+#define SWTCL_OVM_COLOR3 5
+#define SWTCL_OVM_TEX(n) ((n) + 6)
+#define SWTCL_OVM_POINT_SIZE 15
+
+extern void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *InputsRead,  GLuint *OutputsWritten);
+
 extern void r300InitSwtcl( GLcontext *ctx );
 extern void r300DestroySwtcl( GLcontext *ctx );
 
+extern void r300RenderStart(GLcontext *ctx);
+extern void r300RenderFinish(GLcontext *ctx);
+extern void r300RenderPrimitive(GLcontext *ctx, GLenum prim);
+extern void r300ResetLineStipple(GLcontext *ctx);
+
+extern void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset);
+
 #endif
diff --git a/src/mesa/drivers/dri/r300/r300_tex.c b/src/mesa/drivers/dri/r300/r300_tex.c
index 7c699ec572c..0af5bb4f469 100644
--- a/src/mesa/drivers/dri/r300/r300_tex.c
+++ b/src/mesa/drivers/dri/r300/r300_tex.c
@@ -38,6 +38,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/context.h"
 #include "main/enums.h"
 #include "main/image.h"
+#include "main/mipmap.h"
 #include "main/simple_list.h"
 #include "main/texformat.h"
 #include "main/texstore.h"
@@ -49,6 +50,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_context.h"
 #include "r300_state.h"
 #include "r300_ioctl.h"
+#include "radeon_mipmap_tree.h"
 #include "r300_tex.h"
 
 #include "xmlpool.h"
@@ -77,20 +79,20 @@ static unsigned int translate_wrap_mode(GLenum wrapmode)
  *
  * \param t Texture object whose wrap modes are to be set
  */
-static void r300UpdateTexWrap(r300TexObjPtr t)
+static void r300UpdateTexWrap(radeonTexObjPtr t)
 {
-	struct gl_texture_object *tObj = t->base.tObj;
+	struct gl_texture_object *tObj = &t->base;
 
-	t->filter &=
+	t->pp_txfilter &=
 	    ~(R300_TX_WRAP_S_MASK | R300_TX_WRAP_T_MASK | R300_TX_WRAP_R_MASK);
 
-	t->filter |= translate_wrap_mode(tObj->WrapS) << R300_TX_WRAP_S_SHIFT;
+	t->pp_txfilter |= translate_wrap_mode(tObj->WrapS) << R300_TX_WRAP_S_SHIFT;
 
 	if (tObj->Target != GL_TEXTURE_1D) {
-		t->filter |= translate_wrap_mode(tObj->WrapT) << R300_TX_WRAP_T_SHIFT;
+		t->pp_txfilter |= translate_wrap_mode(tObj->WrapT) << R300_TX_WRAP_T_SHIFT;
 
 		if (tObj->Target == GL_TEXTURE_3D)
-			t->filter |= translate_wrap_mode(tObj->WrapR) << R300_TX_WRAP_R_SHIFT;
+			t->pp_txfilter |= translate_wrap_mode(tObj->WrapR) << R300_TX_WRAP_R_SHIFT;
 	}
 }
 
@@ -117,10 +119,13 @@ static GLuint aniso_filter(GLfloat anisotropy)
  * \param magf Texture magnification mode
  * \param anisotropy Maximum anisotropy level
  */
-static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat anisotropy)
+static void r300SetTexFilter(radeonTexObjPtr t, GLenum minf, GLenum magf, GLfloat anisotropy)
 {
-	t->filter &= ~(R300_TX_MIN_FILTER_MASK | R300_TX_MIN_FILTER_MIP_MASK | R300_TX_MAG_FILTER_MASK | R300_TX_MAX_ANISO_MASK);
-	t->filter_1 &= ~R300_EDGE_ANISO_EDGE_ONLY;
+	/* Force revalidation to account for switches from/to mipmapping. */
+	t->validated = GL_FALSE;
+
+	t->pp_txfilter &= ~(R300_TX_MIN_FILTER_MASK | R300_TX_MIN_FILTER_MIP_MASK | R300_TX_MAG_FILTER_MASK | R300_TX_MAX_ANISO_MASK);
+	t->pp_txfilter_1 &= ~R300_EDGE_ANISO_EDGE_ONLY;
 
 	/* Note that EXT_texture_filter_anisotropic is extremely vague about
 	 * how anisotropic filtering interacts with the "normal" filter modes.
@@ -128,7 +133,7 @@ static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat
 	 * filter settings completely. This includes driconf's settings.
 	 */
 	if (anisotropy >= 2.0 && (minf != GL_NEAREST) && (magf != GL_NEAREST)) {
-		t->filter |= R300_TX_MAG_FILTER_ANISO
+		t->pp_txfilter |= R300_TX_MAG_FILTER_ANISO
 			| R300_TX_MIN_FILTER_ANISO
 			| R300_TX_MIN_FILTER_MIP_LINEAR
 			| aniso_filter(anisotropy);
@@ -139,22 +144,22 @@ static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat
 
 	switch (minf) {
 	case GL_NEAREST:
-		t->filter |= R300_TX_MIN_FILTER_NEAREST;
+		t->pp_txfilter |= R300_TX_MIN_FILTER_NEAREST;
 		break;
 	case GL_LINEAR:
-		t->filter |= R300_TX_MIN_FILTER_LINEAR;
+		t->pp_txfilter |= R300_TX_MIN_FILTER_LINEAR;
 		break;
 	case GL_NEAREST_MIPMAP_NEAREST:
-		t->filter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_NEAREST;
+		t->pp_txfilter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_NEAREST;
 		break;
 	case GL_NEAREST_MIPMAP_LINEAR:
-		t->filter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_LINEAR;
+		t->pp_txfilter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_LINEAR;
 		break;
 	case GL_LINEAR_MIPMAP_NEAREST:
-		t->filter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_NEAREST;
+		t->pp_txfilter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_NEAREST;
 		break;
 	case GL_LINEAR_MIPMAP_LINEAR:
-		t->filter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_LINEAR;
+		t->pp_txfilter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_LINEAR;
 		break;
 	}
 
@@ -163,15 +168,15 @@ static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat
 	 */
 	switch (magf) {
 	case GL_NEAREST:
-		t->filter |= R300_TX_MAG_FILTER_NEAREST;
+		t->pp_txfilter |= R300_TX_MAG_FILTER_NEAREST;
 		break;
 	case GL_LINEAR:
-		t->filter |= R300_TX_MAG_FILTER_LINEAR;
+		t->pp_txfilter |= R300_TX_MAG_FILTER_LINEAR;
 		break;
 	}
 }
 
-static void r300SetTexBorderColor(r300TexObjPtr t, const GLfloat color[4])
+static void r300SetTexBorderColor(radeonTexObjPtr t, const GLfloat color[4])
 {
 	GLubyte c[4];
 	CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]);
@@ -182,729 +187,6 @@ static void r300SetTexBorderColor(r300TexObjPtr t, const GLfloat color[4])
 }
 
 /**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-static r300TexObjPtr r300AllocTexObj(struct gl_texture_object *texObj)
-{
-	r300TexObjPtr t;
-
-	t = CALLOC_STRUCT(r300_tex_obj);
-	texObj->DriverData = t;
-	if (t != NULL) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE) {
-			fprintf(stderr, "%s( %p, %p )\n", __FUNCTION__,
-				(void *)texObj, (void *)t);
-		}
-
-		/* Initialize non-image-dependent parts of the state:
-		 */
-		t->base.tObj = texObj;
-		t->border_fallback = GL_FALSE;
-
-		make_empty_list(&t->base);
-
-		r300UpdateTexWrap(t);
-		r300SetTexFilter(t, texObj->MinFilter, texObj->MagFilter, texObj->MaxAnisotropy);
-		r300SetTexBorderColor(t, texObj->BorderColor);
-	}
-
-	return t;
-}
-
-/* try to find a format which will only need a memcopy */
-static const struct gl_texture_format *r300Choose8888TexFormat(GLenum srcFormat,
-							       GLenum srcType)
-{
-	const GLuint ui = 1;
-	const GLubyte littleEndian = *((const GLubyte *)&ui);
-
-	if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
-	    (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
-	    (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
-	    (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && littleEndian)) {
-		return &_mesa_texformat_rgba8888;
-	} else if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
-		   (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && littleEndian) ||
-		   (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
-		   (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && !littleEndian)) {
-		return &_mesa_texformat_rgba8888_rev;
-	} else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
-					    srcType == GL_UNSIGNED_INT_8_8_8_8)) {
-		return &_mesa_texformat_argb8888_rev;
-	} else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && littleEndian) ||
-					    srcType == GL_UNSIGNED_INT_8_8_8_8_REV)) {
-		return &_mesa_texformat_argb8888;
-	} else
-		return _dri_texformat_argb8888;
-}
-
-static const struct gl_texture_format *r300ChooseTextureFormat(GLcontext * ctx,
-							       GLint
-							       internalFormat,
-							       GLenum format,
-							       GLenum type)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	const GLboolean do32bpt =
-	    (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32);
-	const GLboolean force16bpt =
-	    (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16);
-	(void)format;
-
-#if 0
-	fprintf(stderr, "InternalFormat=%s(%d) type=%s format=%s\n",
-		_mesa_lookup_enum_by_nr(internalFormat), internalFormat,
-		_mesa_lookup_enum_by_nr(type), _mesa_lookup_enum_by_nr(format));
-	fprintf(stderr, "do32bpt=%d force16bpt=%d\n", do32bpt, force16bpt);
-#endif
-
-	switch (internalFormat) {
-	case 4:
-	case GL_RGBA:
-	case GL_COMPRESSED_RGBA:
-		switch (type) {
-		case GL_UNSIGNED_INT_10_10_10_2:
-		case GL_UNSIGNED_INT_2_10_10_10_REV:
-			return do32bpt ? _dri_texformat_argb8888 :
-			    _dri_texformat_argb1555;
-		case GL_UNSIGNED_SHORT_4_4_4_4:
-		case GL_UNSIGNED_SHORT_4_4_4_4_REV:
-			return _dri_texformat_argb4444;
-		case GL_UNSIGNED_SHORT_5_5_5_1:
-		case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-			return _dri_texformat_argb1555;
-		default:
-			return do32bpt ? r300Choose8888TexFormat(format, type) :
-			    _dri_texformat_argb4444;
-		}
-
-	case 3:
-	case GL_RGB:
-	case GL_COMPRESSED_RGB:
-		switch (type) {
-		case GL_UNSIGNED_SHORT_4_4_4_4:
-		case GL_UNSIGNED_SHORT_4_4_4_4_REV:
-			return _dri_texformat_argb4444;
-		case GL_UNSIGNED_SHORT_5_5_5_1:
-		case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-			return _dri_texformat_argb1555;
-		case GL_UNSIGNED_SHORT_5_6_5:
-		case GL_UNSIGNED_SHORT_5_6_5_REV:
-			return _dri_texformat_rgb565;
-		default:
-			return do32bpt ? _dri_texformat_argb8888 :
-			    _dri_texformat_rgb565;
-		}
-
-	case GL_RGBA8:
-	case GL_RGB10_A2:
-	case GL_RGBA12:
-	case GL_RGBA16:
-		return !force16bpt ?
-		    r300Choose8888TexFormat(format,
-					    type) : _dri_texformat_argb4444;
-
-	case GL_RGBA4:
-	case GL_RGBA2:
-		return _dri_texformat_argb4444;
-
-	case GL_RGB5_A1:
-		return _dri_texformat_argb1555;
-
-	case GL_RGB8:
-	case GL_RGB10:
-	case GL_RGB12:
-	case GL_RGB16:
-		return !force16bpt ? _dri_texformat_argb8888 :
-		    _dri_texformat_rgb565;
-
-	case GL_RGB5:
-	case GL_RGB4:
-	case GL_R3_G3_B2:
-		return _dri_texformat_rgb565;
-
-	case GL_ALPHA:
-	case GL_ALPHA4:
-	case GL_ALPHA8:
-	case GL_ALPHA12:
-	case GL_ALPHA16:
-	case GL_COMPRESSED_ALPHA:
-		return _dri_texformat_a8;
-
-	case 1:
-	case GL_LUMINANCE:
-	case GL_LUMINANCE4:
-	case GL_LUMINANCE8:
-	case GL_LUMINANCE12:
-	case GL_LUMINANCE16:
-	case GL_COMPRESSED_LUMINANCE:
-		return _dri_texformat_l8;
-
-	case 2:
-	case GL_LUMINANCE_ALPHA:
-	case GL_LUMINANCE4_ALPHA4:
-	case GL_LUMINANCE6_ALPHA2:
-	case GL_LUMINANCE8_ALPHA8:
-	case GL_LUMINANCE12_ALPHA4:
-	case GL_LUMINANCE12_ALPHA12:
-	case GL_LUMINANCE16_ALPHA16:
-	case GL_COMPRESSED_LUMINANCE_ALPHA:
-		return _dri_texformat_al88;
-
-	case GL_INTENSITY:
-	case GL_INTENSITY4:
-	case GL_INTENSITY8:
-	case GL_INTENSITY12:
-	case GL_INTENSITY16:
-	case GL_COMPRESSED_INTENSITY:
-		return _dri_texformat_i8;
-
-	case GL_YCBCR_MESA:
-		if (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
-		    type == GL_UNSIGNED_BYTE)
-			return &_mesa_texformat_ycbcr;
-		else
-			return &_mesa_texformat_ycbcr_rev;
-
-	case GL_RGB_S3TC:
-	case GL_RGB4_S3TC:
-	case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
-		return &_mesa_texformat_rgb_dxt1;
-
-	case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
-		return &_mesa_texformat_rgba_dxt1;
-
-	case GL_RGBA_S3TC:
-	case GL_RGBA4_S3TC:
-	case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
-		return &_mesa_texformat_rgba_dxt3;
-
-	case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
-		return &_mesa_texformat_rgba_dxt5;
-
-	case GL_ALPHA16F_ARB:
-		return &_mesa_texformat_alpha_float16;
-	case GL_ALPHA32F_ARB:
-		return &_mesa_texformat_alpha_float32;
-	case GL_LUMINANCE16F_ARB:
-		return &_mesa_texformat_luminance_float16;
-	case GL_LUMINANCE32F_ARB:
-		return &_mesa_texformat_luminance_float32;
-	case GL_LUMINANCE_ALPHA16F_ARB:
-		return &_mesa_texformat_luminance_alpha_float16;
-	case GL_LUMINANCE_ALPHA32F_ARB:
-		return &_mesa_texformat_luminance_alpha_float32;
-	case GL_INTENSITY16F_ARB:
-		return &_mesa_texformat_intensity_float16;
-	case GL_INTENSITY32F_ARB:
-		return &_mesa_texformat_intensity_float32;
-	case GL_RGB16F_ARB:
-		return &_mesa_texformat_rgba_float16;
-	case GL_RGB32F_ARB:
-		return &_mesa_texformat_rgba_float32;
-	case GL_RGBA16F_ARB:
-		return &_mesa_texformat_rgba_float16;
-	case GL_RGBA32F_ARB:
-		return &_mesa_texformat_rgba_float32;
-
-	case GL_DEPTH_COMPONENT:
-	case GL_DEPTH_COMPONENT16:
-	case GL_DEPTH_COMPONENT24:
-	case GL_DEPTH_COMPONENT32:
-#if 0
-		switch (type) {
-		case GL_UNSIGNED_BYTE:
-		case GL_UNSIGNED_SHORT:
-			return &_mesa_texformat_z16;
-		case GL_UNSIGNED_INT:
-			return &_mesa_texformat_z32;
-		case GL_UNSIGNED_INT_24_8_EXT:
-		default:
-			return &_mesa_texformat_z24_s8;
-		}
-#else
-		return &_mesa_texformat_z16;
-#endif
-
-	default:
-		_mesa_problem(ctx,
-			      "unexpected internalFormat 0x%x in r300ChooseTextureFormat",
-			      (int)internalFormat);
-		return NULL;
-	}
-
-	return NULL;		/* never get here */
-}
-
-static GLboolean
-r300ValidateClientStorage(GLcontext * ctx, GLenum target,
-			  GLint internalFormat,
-			  GLint srcWidth, GLint srcHeight,
-			  GLenum format, GLenum type, const void *pixels,
-			  const struct gl_pixelstore_attrib *packing,
-			  struct gl_texture_object *texObj,
-			  struct gl_texture_image *texImage)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-
-	if (RADEON_DEBUG & DEBUG_TEXTURE)
-		fprintf(stderr, "intformat %s format %s type %s\n",
-			_mesa_lookup_enum_by_nr(internalFormat),
-			_mesa_lookup_enum_by_nr(format),
-			_mesa_lookup_enum_by_nr(type));
-
-	if (!ctx->Unpack.ClientStorage)
-		return 0;
-
-	if (ctx->_ImageTransferState ||
-	    texImage->IsCompressed || texObj->GenerateMipmap)
-		return 0;
-
-	/* This list is incomplete, may be different on ppc???
-	 */
-	switch (internalFormat) {
-	case GL_RGBA:
-		if (format == GL_BGRA && type == GL_UNSIGNED_INT_8_8_8_8_REV) {
-			texImage->TexFormat = _dri_texformat_argb8888;
-		} else
-			return 0;
-		break;
-
-	case GL_RGB:
-		if (format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5) {
-			texImage->TexFormat = _dri_texformat_rgb565;
-		} else
-			return 0;
-		break;
-
-	case GL_YCBCR_MESA:
-		if (format == GL_YCBCR_MESA &&
-		    type == GL_UNSIGNED_SHORT_8_8_REV_APPLE) {
-			texImage->TexFormat = &_mesa_texformat_ycbcr_rev;
-		} else if (format == GL_YCBCR_MESA &&
-			   (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
-			    type == GL_UNSIGNED_BYTE)) {
-			texImage->TexFormat = &_mesa_texformat_ycbcr;
-		} else
-			return 0;
-		break;
-
-	default:
-		return 0;
-	}
-
-	/* Could deal with these packing issues, but currently don't:
-	 */
-	if (packing->SkipPixels ||
-	    packing->SkipRows || packing->SwapBytes || packing->LsbFirst) {
-		return 0;
-	}
-
-	GLint srcRowStride = _mesa_image_row_stride(packing, srcWidth,
-						    format, type);
-
-	if (RADEON_DEBUG & DEBUG_TEXTURE)
-		fprintf(stderr, "%s: srcRowStride %d/%x\n",
-			__FUNCTION__, srcRowStride, srcRowStride);
-
-	/* Could check this later in upload, pitch restrictions could be
-	 * relaxed, but would need to store the image pitch somewhere,
-	 * as packing details might change before image is uploaded:
-	 */
-	if (!r300IsGartMemory(rmesa, pixels, srcHeight * srcRowStride)
-	    || (srcRowStride & 63))
-		return 0;
-
-	/* Have validated that _mesa_transfer_teximage would be a straight
-	 * memcpy at this point.  NOTE: future calls to TexSubImage will
-	 * overwrite the client data.  This is explicitly mentioned in the
-	 * extension spec.
-	 */
-	texImage->Data = (void *)pixels;
-	texImage->IsClientData = GL_TRUE;
-	texImage->RowStride = srcRowStride / texImage->TexFormat->TexelBytes;
-
-	return 1;
-}
-
-static void r300TexImage1D(GLcontext * ctx, GLenum target, GLint level,
-			   GLint internalFormat,
-			   GLint width, GLint border,
-			   GLenum format, GLenum type, const GLvoid * pixels,
-			   const struct gl_pixelstore_attrib *packing,
-			   struct gl_texture_object *texObj,
-			   struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-
-	if (t) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
-			return;
-		}
-	}
-
-	/* Note, this will call ChooseTextureFormat */
-	_mesa_store_teximage1d(ctx, target, level, internalFormat,
-			       width, border, format, type, pixels,
-			       &ctx->Unpack, texObj, texImage);
-
-	t->dirty_images[0] |= (1 << level);
-}
-
-static void r300TexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
-			      GLint xoffset,
-			      GLsizei width,
-			      GLenum format, GLenum type,
-			      const GLvoid * pixels,
-			      const struct gl_pixelstore_attrib *packing,
-			      struct gl_texture_object *texObj,
-			      struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-
-	assert(t);		/* this _should_ be true */
-	if (t) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D");
-			return;
-		}
-	}
-
-	_mesa_store_texsubimage1d(ctx, target, level, xoffset, width,
-				  format, type, pixels, packing, texObj,
-				  texImage);
-
-	t->dirty_images[0] |= (1 << level);
-}
-
-static void r300TexImage2D(GLcontext * ctx, GLenum target, GLint level,
-			   GLint internalFormat,
-			   GLint width, GLint height, GLint border,
-			   GLenum format, GLenum type, const GLvoid * pixels,
-			   const struct gl_pixelstore_attrib *packing,
-			   struct gl_texture_object *texObj,
-			   struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-	GLuint face;
-
-	/* which cube face or ordinary 2D image */
-	switch (target) {
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-		face =
-		    (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-		ASSERT(face < 6);
-		break;
-	default:
-		face = 0;
-	}
-
-	if (t != NULL) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
-			return;
-		}
-	}
-
-	texImage->IsClientData = GL_FALSE;
-
-	if (r300ValidateClientStorage(ctx, target,
-				      internalFormat,
-				      width, height,
-				      format, type, pixels,
-				      packing, texObj, texImage)) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: Using client storage\n",
-				__FUNCTION__);
-	} else {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: Using normal storage\n",
-				__FUNCTION__);
-
-		/* Normal path: copy (to cached memory) and eventually upload
-		 * via another copy to GART memory and then a blit...  Could
-		 * eliminate one copy by going straight to (permanent) GART.
-		 *
-		 * Note, this will call r300ChooseTextureFormat.
-		 */
-		_mesa_store_teximage2d(ctx, target, level, internalFormat,
-				       width, height, border, format, type,
-				       pixels, &ctx->Unpack, texObj, texImage);
-
-		t->dirty_images[face] |= (1 << level);
-	}
-}
-
-static void r300TexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
-			      GLint xoffset, GLint yoffset,
-			      GLsizei width, GLsizei height,
-			      GLenum format, GLenum type,
-			      const GLvoid * pixels,
-			      const struct gl_pixelstore_attrib *packing,
-			      struct gl_texture_object *texObj,
-			      struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-	GLuint face;
-
-	/* which cube face or ordinary 2D image */
-	switch (target) {
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-		face =
-		    (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-		ASSERT(face < 6);
-		break;
-	default:
-		face = 0;
-	}
-
-	assert(t);		/* this _should_ be true */
-	if (t) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D");
-			return;
-		}
-	}
-
-	_mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
-				  height, format, type, pixels, packing, texObj,
-				  texImage);
-
-	t->dirty_images[face] |= (1 << level);
-}
-
-static void r300CompressedTexImage2D(GLcontext * ctx, GLenum target,
-				     GLint level, GLint internalFormat,
-				     GLint width, GLint height, GLint border,
-				     GLsizei imageSize, const GLvoid * data,
-				     struct gl_texture_object *texObj,
-				     struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-	GLuint face;
-
-	/* which cube face or ordinary 2D image */
-	switch (target) {
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-		face =
-		    (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-		ASSERT(face < 6);
-		break;
-	default:
-		face = 0;
-	}
-
-	if (t != NULL) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY,
-				    "glCompressedTexImage2D");
-			return;
-		}
-	}
-
-	texImage->IsClientData = GL_FALSE;
-
-	/* can't call this, different parameters. Would never evaluate to true anyway currently */
-#if 0
-	if (r300ValidateClientStorage(ctx, target,
-				      internalFormat,
-				      width, height,
-				      format, type, pixels,
-				      packing, texObj, texImage)) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: Using client storage\n",
-				__FUNCTION__);
-	} else
-#endif
-	{
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: Using normal storage\n",
-				__FUNCTION__);
-
-		/* Normal path: copy (to cached memory) and eventually upload
-		 * via another copy to GART memory and then a blit...  Could
-		 * eliminate one copy by going straight to (permanent) GART.
-		 *
-		 * Note, this will call r300ChooseTextureFormat.
-		 */
-		_mesa_store_compressed_teximage2d(ctx, target, level,
-						  internalFormat, width, height,
-						  border, imageSize, data,
-						  texObj, texImage);
-
-		t->dirty_images[face] |= (1 << level);
-	}
-}
-
-static void r300CompressedTexSubImage2D(GLcontext * ctx, GLenum target,
-					GLint level, GLint xoffset,
-					GLint yoffset, GLsizei width,
-					GLsizei height, GLenum format,
-					GLsizei imageSize, const GLvoid * data,
-					struct gl_texture_object *texObj,
-					struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-	GLuint face;
-
-	/* which cube face or ordinary 2D image */
-	switch (target) {
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-	case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-		face =
-		    (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-		ASSERT(face < 6);
-		break;
-	default:
-		face = 0;
-	}
-
-	assert(t);		/* this _should_ be true */
-	if (t) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY,
-				    "glCompressedTexSubImage3D");
-			return;
-		}
-	}
-
-	_mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset,
-					     yoffset, width, height, format,
-					     imageSize, data, texObj, texImage);
-
-	t->dirty_images[face] |= (1 << level);
-}
-
-static void r300TexImage3D(GLcontext * ctx, GLenum target, GLint level,
-			   GLint internalFormat,
-			   GLint width, GLint height, GLint depth,
-			   GLint border,
-			   GLenum format, GLenum type, const GLvoid * pixels,
-			   const struct gl_pixelstore_attrib *packing,
-			   struct gl_texture_object *texObj,
-			   struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-
-	if (t) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage3D");
-			return;
-		}
-	}
-
-	texImage->IsClientData = GL_FALSE;
-
-#if 0
-	if (r300ValidateClientStorage(ctx, target,
-				      internalFormat,
-				      width, height,
-				      format, type, pixels,
-				      packing, texObj, texImage)) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: Using client storage\n",
-				__FUNCTION__);
-	} else
-#endif
-	{
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: Using normal storage\n",
-				__FUNCTION__);
-
-		/* Normal path: copy (to cached memory) and eventually upload
-		 * via another copy to GART memory and then a blit...  Could
-		 * eliminate one copy by going straight to (permanent) GART.
-		 *
-		 * Note, this will call r300ChooseTextureFormat.
-		 */
-		_mesa_store_teximage3d(ctx, target, level, internalFormat,
-				       width, height, depth, border,
-				       format, type, pixels,
-				       &ctx->Unpack, texObj, texImage);
-
-		t->dirty_images[0] |= (1 << level);
-	}
-}
-
-static void
-r300TexSubImage3D(GLcontext * ctx, GLenum target, GLint level,
-		  GLint xoffset, GLint yoffset, GLint zoffset,
-		  GLsizei width, GLsizei height, GLsizei depth,
-		  GLenum format, GLenum type,
-		  const GLvoid * pixels,
-		  const struct gl_pixelstore_attrib *packing,
-		  struct gl_texture_object *texObj,
-		  struct gl_texture_image *texImage)
-{
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
-
-/*     fprintf(stderr, "%s\n", __FUNCTION__); */
-
-	assert(t);		/* this _should_ be true */
-	if (t) {
-		driSwapOutTextureObject(t);
-	} else {
-		t = (driTextureObject *) r300AllocTexObj(texObj);
-		if (!t) {
-			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage3D");
-			return;
-		}
-		texObj->DriverData = t;
-	}
-
-	_mesa_store_texsubimage3d(ctx, target, level, xoffset, yoffset, zoffset,
-				  width, height, depth,
-				  format, type, pixels, packing, texObj,
-				  texImage);
-
-	t->dirty_images[0] |= (1 << level);
-}
-
-/**
  * Changes variables and flags for a state update, which will happen at the
  * next UpdateTextureState
  */
@@ -913,7 +195,7 @@ static void r300TexParameter(GLcontext * ctx, GLenum target,
 			     struct gl_texture_object *texObj,
 			     GLenum pname, const GLfloat * params)
 {
-	r300TexObjPtr t = (r300TexObjPtr) texObj->DriverData;
+	radeonTexObj* t = radeon_tex_obj(texObj);
 
 	if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) {
 		fprintf(stderr, "%s( %s )\n", __FUNCTION__,
@@ -946,7 +228,11 @@ static void r300TexParameter(GLcontext * ctx, GLenum target,
 		 * we just have to rely on loading the right subset of mipmap levels
 		 * to simulate a clamped LOD.
 		 */
-		driSwapOutTextureObject((driTextureObject *) t);
+		if (t->mt) {
+			radeon_miptree_unreference(t->mt);
+			t->mt = 0;
+			t->validated = GL_FALSE;
+		}
 		break;
 
 	case GL_DEPTH_TEXTURE_MODE:
@@ -969,27 +255,10 @@ static void r300TexParameter(GLcontext * ctx, GLenum target,
 	}
 }
 
-static void r300BindTexture(GLcontext * ctx, GLenum target,
-			    struct gl_texture_object *texObj)
-{
-	if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) {
-		fprintf(stderr, "%s( %p ) unit=%d\n", __FUNCTION__,
-			(void *)texObj, ctx->Texture.CurrentUnit);
-	}
-
-	if ((target == GL_TEXTURE_1D)
-	    || (target == GL_TEXTURE_2D)
-	    || (target == GL_TEXTURE_3D)
-	    || (target == GL_TEXTURE_CUBE_MAP)
-	    || (target == GL_TEXTURE_RECTANGLE_NV)) {
-		assert(texObj->DriverData != NULL);
-	}
-}
-
 static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	driTextureObject *t = (driTextureObject *) texObj->DriverData;
+	radeonTexObj* t = radeon_tex_obj(texObj);
 
 	if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) {
 		fprintf(stderr, "%s( %p (target = %s) )\n", __FUNCTION__,
@@ -997,14 +266,24 @@ static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 			_mesa_lookup_enum_by_nr(texObj->Target));
 	}
 
-	if (t != NULL) {
-		if (rmesa) {
-			R300_FIREVERTICES(rmesa);
-		}
+	if (rmesa) {
+		int i;
+		radeon_firevertices(&rmesa->radeon);
+
+		for(i = 0; i < R300_MAX_TEXTURE_UNITS; ++i)
+			if (rmesa->hw.textures[i] == t)
+				rmesa->hw.textures[i] = 0;
+	}
 
-		driDestroyTextureObject(t);
+	if (t->bo) {
+		radeon_bo_unref(t->bo);
+		t->bo = NULL;
+	}
+
+	if (t->mt) {
+		radeon_miptree_unreference(t->mt);
+		t->mt = 0;
 	}
-	/* Free mipmap images and the texture object itself */
 	_mesa_delete_texture_object(ctx, texObj);
 }
 
@@ -1013,8 +292,6 @@ static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
  * Called via ctx->Driver.NewTextureObject.
  * Note: this function will be called during context creation to
  * allocate the default texture objects.
- * Note: we could use containment here to 'derive' the driver-specific
- * texture object from the core mesa gl_texture_object.  Not done at this time.
  * Fixup MaxAnisotropy according to user preference.
  */
 static struct gl_texture_object *r300NewTextureObject(GLcontext * ctx,
@@ -1022,14 +299,23 @@ static struct gl_texture_object *r300NewTextureObject(GLcontext * ctx,
 						      GLenum target)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct gl_texture_object *obj;
-	obj = _mesa_new_texture_object(ctx, name, target);
-	if (!obj)
-		return NULL;
-	obj->MaxAnisotropy = rmesa->initialMaxAnisotropy;
+	radeonTexObj* t = CALLOC_STRUCT(radeon_tex_obj);
+
 
-	r300AllocTexObj(obj);
-	return obj;
+	if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) {
+		fprintf(stderr, "%s( %p (target = %s) )\n", __FUNCTION__,
+			t, _mesa_lookup_enum_by_nr(target));
+	}
+
+	_mesa_initialize_texture_object(&t->base, name, target);
+	t->base.MaxAnisotropy = rmesa->radeon.initialMaxAnisotropy;
+
+	/* Initialize hardware state */
+	r300UpdateTexWrap(t);
+	r300SetTexFilter(t, t->base.MinFilter, t->base.MagFilter, t->base.MaxAnisotropy);
+	r300SetTexBorderColor(t, t->base.BorderColor);
+
+	return &t->base;
 }
 
 void r300InitTextureFuncs(struct dd_function_table *functions)
@@ -1037,22 +323,30 @@ void r300InitTextureFuncs(struct dd_function_table *functions)
 	/* Note: we only plug in the functions we implement in the driver
 	 * since _mesa_init_driver_functions() was already called.
 	 */
-	functions->ChooseTextureFormat = r300ChooseTextureFormat;
-	functions->TexImage1D = r300TexImage1D;
-	functions->TexImage2D = r300TexImage2D;
-	functions->TexImage3D = r300TexImage3D;
-	functions->TexSubImage1D = r300TexSubImage1D;
-	functions->TexSubImage2D = r300TexSubImage2D;
-	functions->TexSubImage3D = r300TexSubImage3D;
+	functions->NewTextureImage = radeonNewTextureImage;
+	functions->FreeTexImageData = radeonFreeTexImageData;
+	functions->MapTexture = radeonMapTexture;
+	functions->UnmapTexture = radeonUnmapTexture;
+
+	functions->ChooseTextureFormat = radeonChooseTextureFormat_mesa;
+	functions->TexImage1D = radeonTexImage1D;
+	functions->TexImage2D = radeonTexImage2D;
+	functions->TexImage3D = radeonTexImage3D;
+	functions->TexSubImage1D = radeonTexSubImage1D;
+	functions->TexSubImage2D = radeonTexSubImage2D;
+	functions->TexSubImage3D = radeonTexSubImage3D;
+	functions->GetTexImage = radeonGetTexImage;
+	functions->GetCompressedTexImage = radeonGetCompressedTexImage;
 	functions->NewTextureObject = r300NewTextureObject;
-	functions->BindTexture = r300BindTexture;
 	functions->DeleteTexture = r300DeleteTexture;
 	functions->IsTextureResident = driIsTextureResident;
 
 	functions->TexParameter = r300TexParameter;
 
-	functions->CompressedTexImage2D = r300CompressedTexImage2D;
-	functions->CompressedTexSubImage2D = r300CompressedTexSubImage2D;
+	functions->CompressedTexImage2D = radeonCompressedTexImage2D;
+	functions->CompressedTexSubImage2D = radeonCompressedTexSubImage2D;
+
+	functions->GenerateMipmap = radeonGenerateMipmap;
 
 	driInitTextureFormats();
 }
diff --git a/src/mesa/drivers/dri/r300/r300_tex.h b/src/mesa/drivers/dri/r300/r300_tex.h
index b86d45bfe05..8a653ea2d11 100644
--- a/src/mesa/drivers/dri/r300/r300_tex.h
+++ b/src/mesa/drivers/dri/r300/r300_tex.h
@@ -37,16 +37,17 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 extern void r300SetDepthTexMode(struct gl_texture_object *tObj);
 
+extern void r300SetTexBuffer(__DRIcontext *pDRICtx, GLint target,
+			     __DRIdrawable *dPriv);
+
+extern void r300SetTexBuffer2(__DRIcontext *pDRICtx, GLint target,
+			      GLint format, __DRIdrawable *dPriv);
+
 extern void r300SetTexOffset(__DRIcontext *pDRICtx, GLint texname,
 			     unsigned long long offset, GLint depth,
 			     GLuint pitch);
 
-extern void r300UpdateTextureState(GLcontext * ctx);
-
-extern int r300UploadTexImages(r300ContextPtr rmesa, r300TexObjPtr t,
-			       GLuint face);
-
-extern void r300DestroyTexObj(r300ContextPtr rmesa, r300TexObjPtr t);
+extern GLboolean r300ValidateBuffers(GLcontext * ctx);
 
 extern void r300InitTextureFuncs(struct dd_function_table *functions);
 
diff --git a/src/mesa/drivers/dri/r300/r300_texmem.c b/src/mesa/drivers/dri/r300/r300_texmem.c
deleted file mode 100644
index a89ab83d948..00000000000
--- a/src/mesa/drivers/dri/r300/r300_texmem.c
+++ /dev/null
@@ -1,568 +0,0 @@
-/**************************************************************************
-
-Copyright (C) Tungsten Graphics 2002.  All Rights Reserved.
-The Weather Channel, Inc. funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86
-license. This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation on the rights to use, copy, modify, merge, publish,
-distribute, sub license, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-NON-INFRINGEMENT. IN NO EVENT SHALL ATI, VA LINUX SYSTEMS AND/OR THEIR
-SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR
-IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-
-**************************************************************************/
-
-/**
- * \file
- *
- * \author Gareth Hughes <[email protected]>
- *
- * \author Kevin E. Martin <[email protected]>
- */
-
-#include <errno.h>
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/context.h"
-#include "main/colormac.h"
-#include "main/macros.h"
-#include "main/simple_list.h"
-#include "main/texobj.h"
-#include "radeon_reg.h"		/* gets definition for usleep */
-#include "r300_context.h"
-#include "r300_state.h"
-#include "r300_cmdbuf.h"
-#include "radeon_ioctl.h"
-#include "r300_tex.h"
-#include "r300_ioctl.h"
-#include <unistd.h>		/* for usleep() */
-
-#ifdef USER_BUFFERS
-#include "r300_mem.h"
-#endif
-
-/**
- * Destroy any device-dependent state associated with the texture.  This may
- * include NULLing out hardware state that points to the texture.
- */
-void r300DestroyTexObj(r300ContextPtr rmesa, r300TexObjPtr t)
-{
-	int i;
-
-	if (RADEON_DEBUG & DEBUG_TEXTURE) {
-		fprintf(stderr, "%s( %p, %p )\n", __FUNCTION__,
-			(void *)t, (void *)t->base.tObj);
-	}
-
-	for (i = 0; i < rmesa->radeon.glCtx->Const.MaxTextureUnits; i++) {
-		if (rmesa->state.texture.unit[i].texobj == t->base.tObj) {
-			_mesa_reference_texobj(&rmesa->state.texture.unit[i].texobj, NULL);
-		}
-	}
-}
-
-/* ------------------------------------------------------------
- * Texture image conversions
- */
-
-static void r300UploadGARTClientSubImage(r300ContextPtr rmesa,
-					 r300TexObjPtr t,
-					 struct gl_texture_image *texImage,
-					 GLint hwlevel,
-					 GLint x, GLint y,
-					 GLint width, GLint height)
-{
-	const struct gl_texture_format *texFormat = texImage->TexFormat;
-	GLuint srcPitch, dstPitch;
-	int blit_format;
-	int srcOffset;
-
-	/*
-	 * XXX it appears that we always upload the full image, not a subimage.
-	 * I.e. x==0, y==0, width=texWidth, height=texWidth.  If this is ever
-	 * changed, the src pitch will have to change.
-	 */
-	switch (texFormat->TexelBytes) {
-	case 1:
-		blit_format = R300_CP_COLOR_FORMAT_CI8;
-		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-		break;
-	case 2:
-		blit_format = R300_CP_COLOR_FORMAT_RGB565;
-		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-		break;
-	case 4:
-		blit_format = R300_CP_COLOR_FORMAT_ARGB8888;
-		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-		break;
-	case 8:
-	case 16:
-		blit_format = R300_CP_COLOR_FORMAT_CI8;
-		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
-		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
-		break;
-	default:
-		return;
-	}
-
-	t->image[0][hwlevel].data = texImage->Data;
-	srcOffset = r300GartOffsetFromVirtual(rmesa, texImage->Data);
-
-	assert(srcOffset != ~0);
-
-	/* Don't currently need to cope with small pitches?
-	 */
-	width = texImage->Width;
-	height = texImage->Height;
-
-	if (texFormat->TexelBytes > 4) {
-		width *= texFormat->TexelBytes;
-	}
-
-	r300EmitWait(rmesa, R300_WAIT_3D);
-
-	r300EmitBlit(rmesa, blit_format,
-		     srcPitch,
-		     srcOffset,
-		     dstPitch,
-		     t->bufAddr,
-		     x,
-		     y,
-		     t->image[0][hwlevel].x + x,
-		     t->image[0][hwlevel].y + y, width, height);
-
-	r300EmitWait(rmesa, R300_WAIT_2D);
-}
-
-static void r300UploadRectSubImage(r300ContextPtr rmesa,
-				   r300TexObjPtr t,
-				   struct gl_texture_image *texImage,
-				   GLint x, GLint y, GLint width, GLint height)
-{
-	const struct gl_texture_format *texFormat = texImage->TexFormat;
-	int blit_format, dstPitch, done;
-
-	switch (texFormat->TexelBytes) {
-	case 1:
-		blit_format = R300_CP_COLOR_FORMAT_CI8;
-		break;
-	case 2:
-		blit_format = R300_CP_COLOR_FORMAT_RGB565;
-		break;
-	case 4:
-		blit_format = R300_CP_COLOR_FORMAT_ARGB8888;
-		break;
-	case 8:
-	case 16:
-		blit_format = R300_CP_COLOR_FORMAT_CI8;
-		break;
-	default:
-		return;
-	}
-
-	t->image[0][0].data = texImage->Data;
-
-	/* Currently don't need to cope with small pitches.
-	 */
-	width = texImage->Width;
-	height = texImage->Height;
-	dstPitch = t->pitch;
-
-	if (texFormat->TexelBytes > 4) {
-		width *= texFormat->TexelBytes;
-	}
-
-	if (rmesa->prefer_gart_client_texturing && texImage->IsClientData) {
-		/* In this case, could also use GART texturing.  This is
-		 * currently disabled, but has been tested & works.
-		 */
-		t->offset = r300GartOffsetFromVirtual(rmesa, texImage->Data);
-		t->pitch = texImage->RowStride * texFormat->TexelBytes - 32;
-
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr,
-				"Using GART texturing for rectangular client texture\n");
-
-		/* Release FB memory allocated for this image:
-		 */
-		/* FIXME This may not be correct as driSwapOutTextureObject sets
-		 * FIXME dirty_images.  It may be fine, though.
-		 */
-		if (t->base.memBlock) {
-			driSwapOutTextureObject((driTextureObject *) t);
-		}
-	} else if (texImage->IsClientData) {
-		/* Data already in GART memory, with usable pitch.
-		 */
-		GLuint srcPitch;
-		srcPitch = texImage->RowStride * texFormat->TexelBytes;
-		r300EmitBlit(rmesa,
-			     blit_format,
-			     srcPitch,
-			     r300GartOffsetFromVirtual(rmesa, texImage->Data),
-			     dstPitch, t->bufAddr, 0, 0, 0, 0, width, height);
-	} else {
-		/* Data not in GART memory, or bad pitch.
-		 */
-		for (done = 0; done < height;) {
-			struct r300_dma_region region;
-			int lines =
-			    MIN2(height - done, RADEON_BUFFER_SIZE / dstPitch);
-			int src_pitch;
-			char *tex;
-
-			src_pitch = texImage->RowStride * texFormat->TexelBytes;
-
-			tex = (char *)texImage->Data + done * src_pitch;
-
-			memset(&region, 0, sizeof(region));
-			r300AllocDmaRegion(rmesa, &region, lines * dstPitch,
-					   1024);
-
-			/* Copy texdata to dma:
-			 */
-			if (RADEON_DEBUG & DEBUG_TEXTURE)
-				fprintf(stderr,
-					"%s: src_pitch %d dst_pitch %d\n",
-					__FUNCTION__, src_pitch, dstPitch);
-
-			if (src_pitch == dstPitch) {
-				memcpy(region.address + region.start, tex,
-				       lines * src_pitch);
-			} else {
-				char *buf = region.address + region.start;
-				int i;
-				for (i = 0; i < lines; i++) {
-					memcpy(buf, tex, src_pitch);
-					buf += dstPitch;
-					tex += src_pitch;
-				}
-			}
-
-			r300EmitWait(rmesa, R300_WAIT_3D);
-
-			/* Blit to framebuffer
-			 */
-			r300EmitBlit(rmesa,
-				     blit_format,
-				     dstPitch, GET_START(&region),
-				     dstPitch | (t->tile_bits >> 16),
-				     t->bufAddr, 0, 0, 0, done, width, lines);
-
-			r300EmitWait(rmesa, R300_WAIT_2D);
-#ifdef USER_BUFFERS
-			r300_mem_use(rmesa, region.buf->id);
-#endif
-
-			r300ReleaseDmaRegion(rmesa, &region, __FUNCTION__);
-			done += lines;
-		}
-	}
-}
-
-/**
- * Upload the texture image associated with texture \a t at the specified
- * level at the address relative to \a start.
- */
-static void r300UploadSubImage(r300ContextPtr rmesa, r300TexObjPtr t,
-			       GLint hwlevel,
-			       GLint x, GLint y, GLint width, GLint height,
-			       GLuint face)
-{
-	struct gl_texture_image *texImage = NULL;
-	GLuint offset;
-	GLint imageWidth, imageHeight;
-	GLint ret;
-	drm_radeon_texture_t tex;
-	drm_radeon_tex_image_t tmp;
-	const int level = hwlevel + t->base.firstLevel;
-
-	if (RADEON_DEBUG & DEBUG_TEXTURE) {
-		fprintf(stderr,
-			"%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n",
-			__FUNCTION__, (void *)t, (void *)t->base.tObj, level,
-			width, height, face);
-	}
-
-	ASSERT(face < 6);
-
-	/* Ensure we have a valid texture to upload */
-	if ((hwlevel < 0) || (hwlevel >= R300_MAX_TEXTURE_LEVELS)) {
-		_mesa_problem(NULL, "bad texture level in %s", __FUNCTION__);
-		return;
-	}
-
-	texImage = t->base.tObj->Image[face][level];
-
-	if (!texImage) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: texImage %d is NULL!\n",
-				__FUNCTION__, level);
-		return;
-	}
-	if (!texImage->Data) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: image data is NULL!\n",
-				__FUNCTION__);
-		return;
-	}
-
-	if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-		assert(level == 0);
-		assert(hwlevel == 0);
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "%s: image data is rectangular\n",
-				__FUNCTION__);
-		r300UploadRectSubImage(rmesa, t, texImage, x, y, width, height);
-		return;
-	} else if (texImage->IsClientData) {
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr,
-				"%s: image data is in GART client storage\n",
-				__FUNCTION__);
-		r300UploadGARTClientSubImage(rmesa, t, texImage, hwlevel, x, y,
-					     width, height);
-		return;
-	} else if (RADEON_DEBUG & DEBUG_TEXTURE)
-		fprintf(stderr, "%s: image data is in normal memory\n",
-			__FUNCTION__);
-
-	imageWidth = texImage->Width;
-	imageHeight = texImage->Height;
-
-	offset = t->bufAddr;
-
-	if (RADEON_DEBUG & (DEBUG_TEXTURE | DEBUG_IOCTL)) {
-		GLint imageX = 0;
-		GLint imageY = 0;
-		GLint blitX = t->image[face][hwlevel].x;
-		GLint blitY = t->image[face][hwlevel].y;
-		GLint blitWidth = t->image[face][hwlevel].width;
-		GLint blitHeight = t->image[face][hwlevel].height;
-		fprintf(stderr, "   upload image: %d,%d at %d,%d\n",
-			imageWidth, imageHeight, imageX, imageY);
-		fprintf(stderr, "   upload  blit: %d,%d at %d,%d\n",
-			blitWidth, blitHeight, blitX, blitY);
-		fprintf(stderr, "       blit ofs: 0x%07x level: %d/%d\n",
-			(GLuint) offset, hwlevel, level);
-	}
-
-	t->image[face][hwlevel].data = texImage->Data;
-
-	/* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct.
-	 * NOTE: we're always use a 1KB-wide blit and I8 texture format.
-	 * We used to use 1, 2 and 4-byte texels and used to use the texture
-	 * width to dictate the blit width - but that won't work for compressed
-	 * textures. (Brian)
-	 * NOTE: can't do that with texture tiling. (sroland)
-	 */
-	tex.offset = offset;
-	tex.image = &tmp;
-	/* copy (x,y,width,height,data) */
-	memcpy(&tmp, &t->image[face][hwlevel], sizeof(tmp));
-
-	if (texImage->TexFormat->TexelBytes > 4) {
-		const int log2TexelBytes =
-		    (3 + (texImage->TexFormat->TexelBytes >> 4));
-		tex.format = RADEON_TXFORMAT_I8;	/* any 1-byte texel format */
-		tex.pitch =
-		    MAX2((texImage->Width * texImage->TexFormat->TexelBytes) /
-			 64, 1);
-		tex.height = imageHeight;
-		tex.width = imageWidth << log2TexelBytes;
-		tex.offset += (tmp.x << log2TexelBytes) & ~1023;
-		tmp.x = tmp.x % (1024 >> log2TexelBytes);
-		tmp.width = tmp.width << log2TexelBytes;
-	} else if (texImage->TexFormat->TexelBytes) {
-		/* use multi-byte upload scheme */
-		tex.height = imageHeight;
-		tex.width = imageWidth;
-		switch (texImage->TexFormat->TexelBytes) {
-		case 1:
-			tex.format = RADEON_TXFORMAT_I8;
-			break;
-		case 2:
-			tex.format = RADEON_TXFORMAT_AI88;
-			break;
-		case 4:
-			tex.format = RADEON_TXFORMAT_ARGB8888;
-			break;
-		}
-		tex.pitch =
-		    MAX2((texImage->Width * texImage->TexFormat->TexelBytes) /
-			 64, 1);
-		tex.offset += tmp.x & ~1023;
-		tmp.x = tmp.x % 1024;
-
-		if (t->tile_bits & R300_TXO_MICRO_TILE) {
-			/* need something like "tiled coordinates" ? */
-			tmp.y = tmp.x / (tex.pitch * 128) * 2;
-			tmp.x =
-			    tmp.x % (tex.pitch * 128) / 2 /
-			    texImage->TexFormat->TexelBytes;
-			tex.pitch |= RADEON_DST_TILE_MICRO >> 22;
-		} else {
-			tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1);
-		}
-#if 1
-		if ((t->tile_bits & R300_TXO_MACRO_TILE) &&
-		    (texImage->Width * texImage->TexFormat->TexelBytes >= 256)
-		    && ((!(t->tile_bits & R300_TXO_MICRO_TILE)
-			 && (texImage->Height >= 8))
-			|| (texImage->Height >= 16))) {
-			/* weird: R200 disables macro tiling if mip width is smaller than 256 bytes,
-			   OR if height is smaller than 8 automatically, but if micro tiling is active
-			   the limit is height 16 instead ? */
-			tex.pitch |= RADEON_DST_TILE_MACRO >> 22;
-		}
-#endif
-	} else {
-		/* In case of for instance 8x8 texture (2x2 dxt blocks),
-		   padding after the first two blocks is needed (only
-		   with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */
-		/* set tex.height to 1/4 since 1 "macropixel" (dxt-block)
-		   has 4 real pixels. Needed so the kernel module reads
-		   the right amount of data. */
-		tex.format = RADEON_TXFORMAT_I8;	/* any 1-byte texel format */
-		tex.pitch = (R300_BLIT_WIDTH_BYTES / 64);
-		tex.height = (imageHeight + 3) / 4;
-		tex.width = (imageWidth + 3) / 4;
-		if ((t->format & R300_TX_FORMAT_DXT1) == R300_TX_FORMAT_DXT1) {
-			tex.width *= 8;
-		} else {
-			tex.width *= 16;
-		}
-	}
-
-	LOCK_HARDWARE(&rmesa->radeon);
-	do {
-		ret =
-		    drmCommandWriteRead(rmesa->radeon.dri.fd,
-					DRM_RADEON_TEXTURE, &tex,
-					sizeof(drm_radeon_texture_t));
-		if (ret) {
-			if (RADEON_DEBUG & DEBUG_IOCTL)
-				fprintf(stderr,
-					"DRM_RADEON_TEXTURE:  again!\n");
-			usleep(1);
-		}
-	} while (ret == -EAGAIN);
-
-	UNLOCK_HARDWARE(&rmesa->radeon);
-
-	if (ret) {
-		fprintf(stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret);
-		fprintf(stderr, "   offset=0x%08x\n", offset);
-		fprintf(stderr, "   image width=%d height=%d\n",
-			imageWidth, imageHeight);
-		fprintf(stderr, "    blit width=%d height=%d data=%p\n",
-			t->image[face][hwlevel].width,
-			t->image[face][hwlevel].height,
-			t->image[face][hwlevel].data);
-		_mesa_exit(-1);
-	}
-}
-
-/**
- * Upload the texture images associated with texture \a t.  This might
- * require the allocation of texture memory.
- *
- * \param rmesa Context pointer
- * \param t Texture to be uploaded
- * \param face Cube map face to be uploaded.  Zero for non-cube maps.
- */
-
-int r300UploadTexImages(r300ContextPtr rmesa, r300TexObjPtr t, GLuint face)
-{
-	const int numLevels = t->base.lastLevel - t->base.firstLevel + 1;
-
-	if (t->image_override)
-		return 0;
-
-	if (RADEON_DEBUG & (DEBUG_TEXTURE | DEBUG_IOCTL)) {
-		fprintf(stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__,
-			(void *)rmesa->radeon.glCtx, (void *)t->base.tObj,
-			t->base.totalSize, t->base.firstLevel,
-			t->base.lastLevel);
-	}
-
-	if (t->base.totalSize == 0)
-		return 0;
-
-	if (RADEON_DEBUG & DEBUG_SYNC) {
-		fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
-		radeonFinish(rmesa->radeon.glCtx);
-	}
-
-	LOCK_HARDWARE(&rmesa->radeon);
-
-	if (t->base.memBlock == NULL) {
-		int heap;
-
-		heap = driAllocateTexture(rmesa->texture_heaps, rmesa->nr_heaps,
-					  (driTextureObject *) t);
-		if (heap == -1) {
-			UNLOCK_HARDWARE(&rmesa->radeon);
-			return -1;
-		}
-
-		/* Set the base offset of the texture image */
-		t->bufAddr = rmesa->radeon.radeonScreen->texOffset[heap]
-		    + t->base.memBlock->ofs;
-		t->offset = t->bufAddr;
-
-		if (!(t->base.tObj->Image[0][0]->IsClientData)) {
-			/* hope it's safe to add that here... */
-			t->offset |= t->tile_bits;
-		}
-	}
-
-	/* Let the world know we've used this memory recently.
-	 */
-	driUpdateTextureLRU((driTextureObject *) t);
-	UNLOCK_HARDWARE(&rmesa->radeon);
-
-	/* Upload any images that are new */
-	if (t->base.dirty_images[face]) {
-		int i;
-		for (i = 0; i < numLevels; i++) {
-			if ((t->base.
-			     dirty_images[face] & (1 <<
-						   (i + t->base.firstLevel))) !=
-			    0) {
-				r300UploadSubImage(rmesa, t, i, 0, 0,
-						   t->image[face][i].width,
-						   t->image[face][i].height,
-						   face);
-			}
-		}
-		t->base.dirty_images[face] = 0;
-	}
-
-	if (RADEON_DEBUG & DEBUG_SYNC) {
-		fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
-		radeonFinish(rmesa->radeon.glCtx);
-	}
-
-	return 0;
-}
diff --git a/src/mesa/drivers/dri/r300/r300_texstate.c b/src/mesa/drivers/dri/r300/r300_texstate.c
index f6ae4b675b8..6d6a90aa88a 100644
--- a/src/mesa/drivers/dri/r300/r300_texstate.c
+++ b/src/mesa/drivers/dri/r300/r300_texstate.c
@@ -47,7 +47,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_context.h"
 #include "r300_state.h"
 #include "r300_ioctl.h"
-#include "radeon_ioctl.h"
+#include "radeon_mipmap_tree.h"
 #include "r300_tex.h"
 #include "r300_reg.h"
 
@@ -117,6 +117,7 @@ static const struct tx_table {
 	_ASSIGN(INTENSITY_FLOAT16, R300_EASY_TX_FORMAT(X, X, X, X, FL_I16)),
 	_ASSIGN(Z16, R300_EASY_TX_FORMAT(X, X, X, X, X16)),
 	_ASSIGN(Z24_S8, R300_EASY_TX_FORMAT(X, X, X, X, X24_Y8)),
+	_ASSIGN(S8_Z24, R300_EASY_TX_FORMAT(Y, Y, Y, Y, X24_Y8)),
 	_ASSIGN(Z32, R300_EASY_TX_FORMAT(X, X, X, X, X32)),
 	/* *INDENT-ON* */
 };
@@ -143,13 +144,12 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj)
 		},
 	};
 	const GLuint *format;
-	r300TexObjPtr t;
+	radeonTexObjPtr t;
 
 	if (!tObj)
 		return;
 
-	t = (r300TexObjPtr) tObj->DriverData;
-
+	t = radeon_tex_obj(tObj);
 
 	switch (tObj->Image[0][tObj->BaseLevel]->TexFormat->MesaFormat) {
 	case MESA_FORMAT_Z16:
@@ -171,13 +171,13 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj)
 
 	switch (tObj->DepthMode) {
 	case GL_LUMINANCE:
-		t->format = format[0];
+		t->pp_txformat = format[0];
 		break;
 	case GL_INTENSITY:
-		t->format = format[1];
+		t->pp_txformat = format[1];
 		break;
 	case GL_ALPHA:
-		t->format = format[2];
+		t->pp_txformat = format[2];
 		break;
 	default:
 		/* Error...which should have already been caught by higher
@@ -190,406 +190,134 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj)
 
 
 /**
- * Compute sizes and fill in offset and blit information for the given
- * image (determined by \p face and \p level).
- *
- * \param curOffset points to the offset at which the image is to be stored
- * and is updated by this function according to the size of the image.
- */
-static void compute_tex_image_offset(
-	struct gl_texture_object *tObj,
-	GLuint face,
-	GLint level,
-	GLint* curOffset)
-{
-	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
-	const struct gl_texture_image* texImage;
-	GLuint blitWidth = R300_BLIT_WIDTH_BYTES;
-	GLuint texelBytes;
-	GLuint size;
-
-	texImage = tObj->Image[0][level + t->base.firstLevel];
-	if (!texImage)
-		return;
-
-	texelBytes = texImage->TexFormat->TexelBytes;
-
-	/* find image size in bytes */
-	if (texImage->IsCompressed) {
-		if ((t->format & R300_TX_FORMAT_DXT1) ==
-			R300_TX_FORMAT_DXT1) {
-			// fprintf(stderr,"DXT 1 %d %08X\n", texImage->Width, t->format);
-			if ((texImage->Width + 3) < 8)	/* width one block */
-				size = texImage->CompressedSize * 4;
-			else if ((texImage->Width + 3) < 16)
-				size = texImage->CompressedSize * 2;
-			else
-				size = texImage->CompressedSize;
-		} else {
-			/* DXT3/5, 16 bytes per block */
-			WARN_ONCE
-				("DXT 3/5 suffers from multitexturing problems!\n");
-			// fprintf(stderr,"DXT 3/5 %d\n", texImage->Width);
-			if ((texImage->Width + 3) < 8)
-				size = texImage->CompressedSize * 2;
-			else
-				size = texImage->CompressedSize;
-		}
-	} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-		size =
-			((texImage->Width * texelBytes +
-			63) & ~63) * texImage->Height;
-		blitWidth = 64 / texelBytes;
-	} else if (t->tile_bits & R300_TXO_MICRO_TILE) {
-		/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
-			though the actual offset may be different (if texture is less than
-			32 bytes width) to the untiled case */
-		int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
-		size =
-			(w * ((texImage->Height + 1) / 2)) *
-			texImage->Depth;
-		blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-	} else {
-		int w = (texImage->Width * texelBytes + 31) & ~31;
-		size = w * texImage->Height * texImage->Depth;
-		blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-	}
-	assert(size > 0);
-
-	if (RADEON_DEBUG & DEBUG_TEXTURE)
-		fprintf(stderr, "w=%d h=%d d=%d tb=%d intFormat=%d\n",
-			texImage->Width, texImage->Height,
-			texImage->Depth,
-			texImage->TexFormat->TexelBytes,
-			texImage->InternalFormat);
-
-	/* All images are aligned to a 32-byte offset */
-	*curOffset = (*curOffset + 0x1f) & ~0x1f;
-
-	if (texelBytes) {
-		/* fix x and y coords up later together with offset */
-		t->image[face][level].x = *curOffset;
-		t->image[face][level].y = 0;
-		t->image[face][level].width =
-			MIN2(size / texelBytes, blitWidth);
-		t->image[face][level].height =
-			(size / texelBytes) / t->image[face][level].width;
-	} else {
-		t->image[face][level].x = *curOffset % R300_BLIT_WIDTH_BYTES;
-		t->image[face][level].y = *curOffset / R300_BLIT_WIDTH_BYTES;
-		t->image[face][level].width =
-			MIN2(size, R300_BLIT_WIDTH_BYTES);
-		t->image[face][level].height = size / t->image[face][level].width;
-	}
-
-	if (RADEON_DEBUG & DEBUG_TEXTURE)
-		fprintf(stderr,
-			"level %d, face %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
-			level, face, texImage->Width, texImage->Height,
-			t->image[face][level].x, t->image[face][level].y,
-			t->image[face][level].width, t->image[face][level].height,
-			size, *curOffset);
-
-	*curOffset += size;
-}
-
-
-
-/**
- * This function computes the number of bytes of storage needed for
- * the given texture object (all mipmap levels, all cube faces).
- * The \c image[face][level].x/y/width/height parameters for upload/blitting
- * are computed here.  \c filter, \c format, etc. will be set here
- * too.
+ * Compute the cached hardware register values for the given texture object.
  *
  * \param rmesa Context pointer
- * \param tObj GL texture object whose images are to be posted to
- *                 hardware state.
+ * \param t the r300 texture object
  */
-static void r300SetTexImages(r300ContextPtr rmesa,
-			     struct gl_texture_object *tObj)
+static void setup_hardware_state(r300ContextPtr rmesa, radeonTexObj *t)
 {
-	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
-	const struct gl_texture_image *baseImage =
-	    tObj->Image[0][tObj->BaseLevel];
-	GLint curOffset;
-	GLint i, texelBytes;
-	GLint numLevels;
-	GLint log2Width, log2Height, log2Depth;
-
-	/* Set the hardware texture format
-	 */
+	const struct gl_texture_image *firstImage;
+	int firstlevel = t->mt ? t->mt->firstLevel : 0;
+	    
+	firstImage = t->base.Image[0][firstlevel];
+
 	if (!t->image_override
-	    && VALID_FORMAT(baseImage->TexFormat->MesaFormat)) {
-		if (baseImage->TexFormat->BaseFormat == GL_DEPTH_COMPONENT) {
-			r300SetDepthTexMode(tObj);
+	    && VALID_FORMAT(firstImage->TexFormat->MesaFormat)) {
+		if (firstImage->TexFormat->BaseFormat == GL_DEPTH_COMPONENT) {
+			r300SetDepthTexMode(&t->base);
 		} else {
-			t->format = tx_table[baseImage->TexFormat->MesaFormat].format;
+			t->pp_txformat = tx_table[firstImage->TexFormat->MesaFormat].format;
 		}
 
-		t->filter |= tx_table[baseImage->TexFormat->MesaFormat].filter;
+		t->pp_txfilter |= tx_table[firstImage->TexFormat->MesaFormat].filter;
 	} else if (!t->image_override) {
 		_mesa_problem(NULL, "unexpected texture format in %s",
 			      __FUNCTION__);
 		return;
 	}
 
-	texelBytes = baseImage->TexFormat->TexelBytes;
-
-	/* Compute which mipmap levels we really want to send to the hardware.
-	 */
-	driCalculateTextureFirstLastLevel((driTextureObject *) t);
-	log2Width = tObj->Image[0][t->base.firstLevel]->WidthLog2;
-	log2Height = tObj->Image[0][t->base.firstLevel]->HeightLog2;
-	log2Depth = tObj->Image[0][t->base.firstLevel]->DepthLog2;
-
-	numLevels = t->base.lastLevel - t->base.firstLevel + 1;
+	if (t->image_override && t->bo)
+		return;
 
-	assert(numLevels <= R300_MAX_TEXTURE_LEVELS);
+	t->pp_txsize = (((firstImage->Width - 1) << R300_TX_WIDTHMASK_SHIFT)
+			| ((firstImage->Height - 1) << R300_TX_HEIGHTMASK_SHIFT)
+			| ((firstImage->DepthLog2) << R300_TX_DEPTHMASK_SHIFT)
+			| ((t->mt->lastLevel - t->mt->firstLevel) << R300_TX_MAX_MIP_LEVEL_SHIFT));
 
-	/* Calculate mipmap offsets and dimensions for blitting (uploading)
-	 * The idea is that we lay out the mipmap levels within a block of
-	 * memory organized as a rectangle of width BLIT_WIDTH_BYTES.
-	 */
 	t->tile_bits = 0;
 
-	/* figure out if this texture is suitable for tiling. */
-#if 0				/* Disabled for now */
-	if (texelBytes) {
-		if ((tObj->Target != GL_TEXTURE_RECTANGLE_NV) &&
-		    /* texrect might be able to use micro tiling too in theory? */
-		    (baseImage->Height > 1)) {
-
-			/* allow 32 (bytes) x 1 mip (which will use two times the space
-			   the non-tiled version would use) max if base texture is large enough */
-			if ((numLevels == 1) ||
-			    (((baseImage->Width * texelBytes /
-			       baseImage->Height) <= 32)
-			     && (baseImage->Width * texelBytes > 64))
-			    ||
-			    ((baseImage->Width * texelBytes /
-			      baseImage->Height) <= 16)) {
-				t->tile_bits |= R300_TXO_MICRO_TILE;
-			}
-		}
-
-		if (tObj->Target != GL_TEXTURE_RECTANGLE_NV) {
-			/* we can set macro tiling even for small textures, they will be untiled anyway */
-			t->tile_bits |= R300_TXO_MACRO_TILE;
-		}
-	}
-#endif
-
-	curOffset = 0;
+	if (t->base.Target == GL_TEXTURE_CUBE_MAP)
+		t->pp_txformat |= R300_TX_FORMAT_CUBIC_MAP;
+	if (t->base.Target == GL_TEXTURE_3D)
+		t->pp_txformat |= R300_TX_FORMAT_3D;
 
-	if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-		ASSERT(log2Width == log2Height);
-		t->format |= R300_TX_FORMAT_CUBIC_MAP;
 
-		for(i = 0; i < numLevels; i++) {
-			GLuint face;
-			for(face = 0; face < 6; face++)
-				compute_tex_image_offset(tObj, face, i, &curOffset);
-		}
-	} else {
-		if (tObj->Target == GL_TEXTURE_3D)
-                	t->format |= R300_TX_FORMAT_3D;
-
-		for (i = 0; i < numLevels; i++)
-			compute_tex_image_offset(tObj, 0, i, &curOffset);
-	}
-
-	/* Align the total size of texture memory block.
-	 */
-	t->base.totalSize =
-	    (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
-
-	t->size =
-	    (((tObj->Image[0][t->base.firstLevel]->Width -
-	       1) << R300_TX_WIDTHMASK_SHIFT)
-	     | ((tObj->Image[0][t->base.firstLevel]->Height - 1) <<
-		R300_TX_HEIGHTMASK_SHIFT)
-	     | ((tObj->Image[0][t->base.firstLevel]->DepthLog2) <<
-		R300_TX_DEPTHMASK_SHIFT))
-	    | ((numLevels - 1) << R300_TX_MAX_MIP_LEVEL_SHIFT);
-
-	t->pitch = 0;
-
-	/* Only need to round to nearest 32 for textures, but the blitter
-	 * requires 64-byte aligned pitches, and we may/may not need the
-	 * blitter.   NPOT only!
-	 */
-	if (baseImage->IsCompressed) {
-		t->pitch |=
-		    (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63);
-	} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-		unsigned int align = (64 / texelBytes) - 1;
-		t->pitch |= ((tObj->Image[0][t->base.firstLevel]->Width *
-			     texelBytes) + 63) & ~(63);
-		t->size |= R300_TX_SIZE_TXPITCH_EN;
+	if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) {
+		unsigned int align = (64 / t->mt->bpp) - 1;
+		t->pp_txsize |= R300_TX_SIZE_TXPITCH_EN;
 		if (!t->image_override)
-			t->pitch_reg =
-			    (((tObj->Image[0][t->base.firstLevel]->Width) +
-			      align) & ~align) - 1;
-	} else {
-		t->pitch |=
-		    ((tObj->Image[0][t->base.firstLevel]->Width *
-		      texelBytes) + 63) & ~(63);
+			t->pp_txpitch = ((firstImage->Width + align) & ~align) - 1;
 	}
 
 	if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
-	    if (tObj->Image[0][t->base.firstLevel]->Width > 2048)
-		t->pitch_reg |= R500_TXWIDTH_BIT11;
-	    if (tObj->Image[0][t->base.firstLevel]->Height > 2048)
-		t->pitch_reg |= R500_TXHEIGHT_BIT11;
+	    if (firstImage->Width > 2048)
+		t->pp_txpitch |= R500_TXWIDTH_BIT11;
+	    if (firstImage->Height > 2048)
+		t->pp_txpitch |= R500_TXHEIGHT_BIT11;
 	}
 }
 
-/* ================================================================
- * Texture unit state management
+/**
+ * Ensure the given texture is ready for rendering.
+ *
+ * Mostly this means populating the texture object's mipmap tree.
  */
-
-static GLboolean r300EnableTexture2D(GLcontext * ctx, int unit)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-	struct gl_texture_object *tObj = texUnit->_Current;
-	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
-
-	ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D);
-
-	if (t->base.dirty_images[0]) {
-		R300_FIREVERTICES(rmesa);
-
-		r300SetTexImages(rmesa, tObj);
-		r300UploadTexImages(rmesa, (r300TexObjPtr) tObj->DriverData, 0);
-		if (!t->base.memBlock && !t->image_override)
-			return GL_FALSE;
-	}
-
-	return GL_TRUE;
-}
-
-static GLboolean r300EnableTexture3D(GLcontext * ctx, int unit)
+static GLboolean r300_validate_texture(GLcontext * ctx, struct gl_texture_object *texObj)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-	struct gl_texture_object *tObj = texUnit->_Current;
-	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
+	radeonTexObj *t = radeon_tex_obj(texObj);
 
-	ASSERT(tObj->Target == GL_TEXTURE_3D);
-
-	/* r300 does not support mipmaps for 3D textures. */
-	if ((tObj->MinFilter != GL_NEAREST) && (tObj->MinFilter != GL_LINEAR)) {
+	if (!radeon_validate_texture_miptree(ctx, texObj))
 		return GL_FALSE;
-	}
 
-	if (t->base.dirty_images[0]) {
-		R300_FIREVERTICES(rmesa);
-		r300SetTexImages(rmesa, tObj);
-		r300UploadTexImages(rmesa, (r300TexObjPtr) tObj->DriverData, 0);
-		if (!t->base.memBlock)
-			return GL_FALSE;
-	}
+	/* Configure the hardware registers (more precisely, the cached version
+	 * of the hardware registers). */
+	setup_hardware_state(rmesa, t);
 
+	t->validated = GL_TRUE;
 	return GL_TRUE;
 }
 
-static GLboolean r300EnableTextureCube(GLcontext * ctx, int unit)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-	struct gl_texture_object *tObj = texUnit->_Current;
-	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
-	GLuint face;
-
-	ASSERT(tObj->Target == GL_TEXTURE_CUBE_MAP);
-
-	if (t->base.dirty_images[0] || t->base.dirty_images[1] ||
-	    t->base.dirty_images[2] || t->base.dirty_images[3] ||
-	    t->base.dirty_images[4] || t->base.dirty_images[5]) {
-		/* flush */
-		R300_FIREVERTICES(rmesa);
-		/* layout memory space, once for all faces */
-		r300SetTexImages(rmesa, tObj);
-	}
-
-	/* upload (per face) */
-	for (face = 0; face < 6; face++) {
-		if (t->base.dirty_images[face]) {
-			r300UploadTexImages(rmesa,
-					    (r300TexObjPtr) tObj->DriverData,
-					    face);
-		}
-	}
-
-	if (!t->base.memBlock) {
-		/* texmem alloc failed, use s/w fallback */
-		return GL_FALSE;
-	}
-
-	return GL_TRUE;
-}
-
-static GLboolean r300EnableTextureRect(GLcontext * ctx, int unit)
+/**
+ * Ensure all enabled and complete textures are uploaded along with any buffers being used.
+ */
+GLboolean r300ValidateBuffers(GLcontext * ctx)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-	struct gl_texture_object *tObj = texUnit->_Current;
-	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
-
-	ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV);
+	struct radeon_renderbuffer *rrb;
+	int i;
 
-	if (t->base.dirty_images[0]) {
-		R300_FIREVERTICES(rmesa);
+	radeon_validate_reset_bos(&rmesa->radeon);
 
-		r300SetTexImages(rmesa, tObj);
-		r300UploadTexImages(rmesa, (r300TexObjPtr) tObj->DriverData, 0);
-		if (!t->base.memBlock && !t->image_override &&
-		    !rmesa->prefer_gart_client_texturing)
-			return GL_FALSE;
+	rrb = radeon_get_colorbuffer(&rmesa->radeon);
+	/* color buffer */
+	if (rrb && rrb->bo) {
+		radeon_validate_bo(&rmesa->radeon, rrb->bo,
+				   0, RADEON_GEM_DOMAIN_VRAM);
 	}
 
-	return GL_TRUE;
-}
-
-static GLboolean r300UpdateTexture(GLcontext * ctx, int unit)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-	struct gl_texture_object *tObj = texUnit->_ReallyEnabled ?
-		texUnit->_Current : NULL;
-	r300TexObjPtr t = tObj ? (r300TexObjPtr) tObj->DriverData : NULL;
-
-	/* Fallback if there's a texture border */
-	if (tObj && tObj->Image[0][tObj->BaseLevel]->Border > 0) {
-		tObj = NULL;
-		t = NULL;
+	/* depth buffer */
+	rrb = radeon_get_depthbuffer(&rmesa->radeon);
+	if (rrb && rrb->bo) {
+		radeon_validate_bo(&rmesa->radeon, rrb->bo,
+				   0, RADEON_GEM_DOMAIN_VRAM);
 	}
+	
+	for (i = 0; i < ctx->Const.MaxTextureImageUnits; ++i) {
+		radeonTexObj *t;
 
-	/* Update state if this is a different texture object to last
-	 * time.
-	 */
-	if (rmesa->state.texture.unit[unit].texobj != tObj) {
-		if (rmesa->state.texture.unit[unit].texobj != NULL) {
-			r300TexObjPtr t_old = (r300TexObjPtr) rmesa->state.texture.unit[unit].texobj->DriverData;
-
-			/* The old texture is no longer bound to this texture unit.
-			 * Mark it as such.
-			 */
-
-			t_old->base.bound &= ~(1 << unit);
-		}
-
-		_mesa_reference_texobj(&rmesa->state.texture.unit[unit].texobj, tObj);
+		if (!ctx->Texture.Unit[i]._ReallyEnabled)
+			continue;
 
-		if (t) {
-			t->base.bound |= (1 << unit);
-			driUpdateTextureLRU(&t->base);	/* XXX: should be locked! */
+		if (!r300_validate_texture(ctx, ctx->Texture.Unit[i]._Current)) {
+			_mesa_warning(ctx,
+				      "failed to validate texture for unit %d.\n",
+				      i);
 		}
+		t = radeon_tex_obj(ctx->Texture.Unit[i]._Current);
+		if (t->image_override && t->bo)
+			radeon_validate_bo(&rmesa->radeon, t->bo,
+					   RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
+
+		else if (t->mt->bo)
+			radeon_validate_bo(&rmesa->radeon, t->mt->bo,
+					   RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
 	}
+	if (rmesa->radeon.dma.current)
+		radeon_validate_bo(&rmesa->radeon, rmesa->radeon.dma.current, RADEON_GEM_DOMAIN_GTT, 0);
 
-	return !t || !t->border_fallback;
+	return radeon_revalidate_bos(ctx);
 }
 
 void r300SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
@@ -598,78 +326,166 @@ void r300SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 	r300ContextPtr rmesa = pDRICtx->driverPrivate;
 	struct gl_texture_object *tObj =
 	    _mesa_lookup_texture(rmesa->radeon.glCtx, texname);
-	r300TexObjPtr t;
+	radeonTexObjPtr t = radeon_tex_obj(tObj);
 	uint32_t pitch_val;
 
 	if (!tObj)
 		return;
 
-	t = (r300TexObjPtr) tObj->DriverData;
-
 	t->image_override = GL_TRUE;
 
 	if (!offset)
 		return;
 
-	t->offset = offset;
-	t->pitch_reg &= (1 << 13) -1;
+	t->bo = NULL;
+	t->override_offset = offset;
+	t->pp_txpitch &= (1 << 13) -1;
 	pitch_val = pitch;
 
 	switch (depth) {
 	case 32:
-		t->format = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8);
-		t->filter |= tx_table[2].filter;
+		t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8);
+		t->pp_txfilter |= tx_table[2].filter;
 		pitch_val /= 4;
 		break;
 	case 24:
 	default:
-		t->format = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8);
-		t->filter |= tx_table[4].filter;
+		t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8);
+		t->pp_txfilter |= tx_table[4].filter;
 		pitch_val /= 4;
 		break;
 	case 16:
-		t->format = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5);
-		t->filter |= tx_table[5].filter;
+		t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5);
+		t->pp_txfilter |= tx_table[5].filter;
 		pitch_val /= 2;
 		break;
 	}
 	pitch_val--;
 
-	t->pitch_reg |= pitch_val;
+	t->pp_txpitch |= pitch_val;
 }
 
-static GLboolean r300UpdateTextureUnit(GLcontext * ctx, int unit)
+void r300SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format, __DRIdrawable *dPriv)
 {
-	struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-
-	if (texUnit->_ReallyEnabled & (TEXTURE_RECT_BIT)) {
-		return (r300EnableTextureRect(ctx, unit) &&
-			r300UpdateTexture(ctx, unit));
-	} else if (texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT)) {
-		return (r300EnableTexture2D(ctx, unit) &&
-			r300UpdateTexture(ctx, unit));
-	} else if (texUnit->_ReallyEnabled & (TEXTURE_3D_BIT)) {
-		return (r300EnableTexture3D(ctx, unit) &&
-			r300UpdateTexture(ctx, unit));
-	} else if (texUnit->_ReallyEnabled & (TEXTURE_CUBE_BIT)) {
-		return (r300EnableTextureCube(ctx, unit) &&
-			r300UpdateTexture(ctx, unit));
-	} else if (texUnit->_ReallyEnabled) {
-		return GL_FALSE;
-	} else {
-		return r300UpdateTexture(ctx, unit);
+	struct gl_texture_unit *texUnit;
+	struct gl_texture_object *texObj;
+	struct gl_texture_image *texImage;
+	struct radeon_renderbuffer *rb;
+	radeon_texture_image *rImage;
+	radeonContextPtr radeon;
+	r300ContextPtr rmesa;
+	struct radeon_framebuffer *rfb;
+	radeonTexObjPtr t;
+	uint32_t pitch_val;
+	uint32_t internalFormat, type, format;
+
+	type = GL_BGRA;
+	format = GL_UNSIGNED_BYTE;
+	internalFormat = (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT ? 3 : 4);
+
+	radeon = pDRICtx->driverPrivate;
+	rmesa = pDRICtx->driverPrivate;
+
+	rfb = dPriv->driverPrivate;
+        texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit];
+	texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target);
+        texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0);
+
+	rImage = get_radeon_texture_image(texImage);
+	t = radeon_tex_obj(texObj);
+        if (t == NULL) {
+    	    return;
+    	}
+
+	radeon_update_renderbuffers(pDRICtx, dPriv);
+	/* back & depth buffer are useless free them right away */
+	rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer;
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+        rb->bo = NULL;
+	}
+	rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+		rb->bo = NULL;
 	}
+	rb = rfb->color_rb[0];
+	if (rb->bo == NULL) {
+		/* Failed to BO for the buffer */
+		return;
+	}
+	
+	_mesa_lock_texture(radeon->glCtx, texObj);
+	if (t->bo) {
+		radeon_bo_unref(t->bo);
+		t->bo = NULL;
+	}
+	if (rImage->bo) {
+		radeon_bo_unref(rImage->bo);
+		rImage->bo = NULL;
+	}
+	if (t->mt) {
+		radeon_miptree_unreference(t->mt);
+		t->mt = NULL;
+	}
+	if (rImage->mt) {
+		radeon_miptree_unreference(rImage->mt);
+		rImage->mt = NULL;
+	}
+	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
+				   rb->width, rb->height, 1, 0, rb->cpp);
+	texImage->RowStride = rb->pitch / rb->cpp;
+	texImage->TexFormat = radeonChooseTextureFormat(radeon->glCtx,
+							internalFormat,
+							type, format, 0);
+	rImage->bo = rb->bo;
+	radeon_bo_ref(rImage->bo);
+	t->bo = rb->bo;
+	radeon_bo_ref(t->bo);
+	t->tile_bits = 0;
+	t->image_override = GL_TRUE;
+	t->override_offset = 0;
+	t->pp_txpitch &= (1 << 13) -1;
+	pitch_val = rb->pitch;
+	switch (rb->cpp) {
+	case 4:
+		if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT)
+			t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8);
+		else
+			t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8);
+		t->pp_txfilter |= tx_table[2].filter;
+		pitch_val /= 4;
+		break;
+	case 3:
+	default:
+		t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8);
+		t->pp_txfilter |= tx_table[4].filter;
+		pitch_val /= 4;
+		break;
+	case 2:
+		t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5);
+		t->pp_txfilter |= tx_table[5].filter;
+		pitch_val /= 2;
+		break;
+	}
+	pitch_val--;
+	t->pp_txsize = ((rb->width - 1) << R300_TX_WIDTHMASK_SHIFT) |
+              ((rb->height - 1) << R300_TX_HEIGHTMASK_SHIFT);
+	t->pp_txsize |= R300_TX_SIZE_TXPITCH_EN;
+	t->pp_txpitch |= pitch_val;
+
+	if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+	    if (rb->width > 2048)
+		t->pp_txpitch |= R500_TXWIDTH_BIT11;
+	    if (rb->height > 2048)
+		t->pp_txpitch |= R500_TXHEIGHT_BIT11;
+	}
+	t->validated = GL_TRUE;
+	_mesa_unlock_texture(radeon->glCtx, texObj);
+	return;
 }
 
-void r300UpdateTextureState(GLcontext * ctx)
+void r300SetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
 {
-	int i;
-
-	for (i = 0; i < 8; i++) {
-		if (!r300UpdateTextureUnit(ctx, i)) {
-			_mesa_warning(ctx,
-				      "failed to update texture state for unit %d.\n",
-				      i);
-		}
-	}
+        r300SetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv);
 }
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index 146daa367cd..c41a8fdd621 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -34,10 +34,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "shader/program.h"
 #include "shader/prog_instruction.h"
 #include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
 #include "shader/prog_statevars.h"
 #include "tnl/tnl.h"
 
 #include "r300_context.h"
+#include "r300_state.h"
 
 /* TODO: Get rid of t_src_class call */
 #define CMP_SRCS(a, b) ((a.RelAddr != b.RelAddr) || (a.Index != b.Index && \
@@ -64,7 +66,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 		int u_temp_used = (VSF_MAX_FRAGMENT_TEMPS - 1) - u_temp_i; \
 		if((vp->num_temporaries + u_temp_used) > VSF_MAX_FRAGMENT_TEMPS) { \
 			WARN_ONCE("Ran out of temps, num temps %d, us %d\n", vp->num_temporaries, u_temp_used); \
-			vp->native = GL_FALSE; \
+			vp->error = GL_TRUE; \
 		} \
 		u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1; \
 	} while (0)
@@ -214,21 +216,8 @@ static void vp_dump_inputs(struct r300_vertex_program *vp, char *caller)
 static unsigned long t_src_index(struct r300_vertex_program *vp,
 				 struct prog_src_register *src)
 {
-	int i;
-	int max_reg = -1;
-
 	if (src->File == PROGRAM_INPUT) {
-		if (vp->inputs[src->Index] != -1)
-			return vp->inputs[src->Index];
-
-		for (i = 0; i < VERT_ATTRIB_MAX; i++)
-			if (vp->inputs[i] > max_reg)
-				max_reg = vp->inputs[i];
-
-		vp->inputs[src->Index] = max_reg + 1;
-
-		//vp_dump_inputs(vp, __FUNCTION__);
-
+		assert(vp->inputs[src->Index] != -1);
 		return vp->inputs[src->Index];
 	} else {
 		if (src->Index < 0) {
@@ -943,11 +932,17 @@ static GLuint *r300TranslateOpcodeXPD(struct r300_vertex_program *vp,
 static void t_inputs_outputs(struct r300_vertex_program *vp)
 {
 	int i;
-	int cur_reg = 0;
+	int cur_reg;
 
-	for (i = 0; i < VERT_ATTRIB_MAX; i++)
-		vp->inputs[i] = -1;
+	cur_reg = -1;
+	for (i = 0; i < VERT_ATTRIB_MAX; i++) {
+		if (vp->key.InputsRead & (1 << i))
+			vp->inputs[i] = ++cur_reg;
+		else
+			vp->inputs[i] = -1;
+	}
 
+	cur_reg = 0;
 	for (i = 0; i < VERT_RESULT_MAX; i++)
 		vp->outputs[i] = -1;
 
@@ -961,26 +956,36 @@ static void t_inputs_outputs(struct r300_vertex_program *vp)
 		vp->outputs[VERT_RESULT_PSIZ] = cur_reg++;
 	}
 
+	/* If we're writing back facing colors we need to send
+	 * four colors to make front/back face colors selection work.
+	 * If the vertex program doesn't write all 4 colors, lets
+	 * pretend it does by skipping output index reg so the colors
+	 * get written into appropriate output vectors.
+	 */
 	if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL0)) {
 		vp->outputs[VERT_RESULT_COL0] = cur_reg++;
+	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+		vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+		cur_reg++;
 	}
 
 	if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL1)) {
-		vp->outputs[VERT_RESULT_COL1] =
-		    vp->outputs[VERT_RESULT_COL0] + 1;
-		cur_reg = vp->outputs[VERT_RESULT_COL1] + 1;
+		vp->outputs[VERT_RESULT_COL1] = cur_reg++;
+	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+		vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+		cur_reg++;
 	}
 
 	if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) {
-		vp->outputs[VERT_RESULT_BFC0] =
-		    vp->outputs[VERT_RESULT_COL0] + 2;
-		cur_reg = vp->outputs[VERT_RESULT_BFC0] + 2;
+		vp->outputs[VERT_RESULT_BFC0] = cur_reg++;
+	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+		cur_reg++;
 	}
 
 	if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
-		vp->outputs[VERT_RESULT_BFC1] =
-		    vp->outputs[VERT_RESULT_COL0] + 3;
-		cur_reg = vp->outputs[VERT_RESULT_BFC1] + 1;
+		vp->outputs[VERT_RESULT_BFC1] = cur_reg++;
+	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) {
+		cur_reg++;
 	}
 
 	for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) {
@@ -1007,14 +1012,13 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp,
 	struct prog_src_register src[3];
 
 	vp->pos_end = 0;	/* Not supported yet */
-	vp->program.length = 0;
-	/*vp->num_temporaries=mesa_vp->Base.NumTemporaries; */
+	vp->hw_code.length = 0;
 	vp->translated = GL_TRUE;
-	vp->native = GL_TRUE;
+	vp->error = GL_FALSE;
 
 	t_inputs_outputs(vp);
 
-	for (inst = vp->program.body.i; vpi->Opcode != OPCODE_END;
+	for (inst = vp->hw_code.body.d; vpi->Opcode != OPCODE_END;
 	     vpi++, inst += 4) {
 
 		FREE_TEMPS();
@@ -1176,38 +1180,15 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp,
 						      &u_temp_i);
 			break;
 		default:
-			assert(0);
+			vp->error = GL_TRUE;
 			break;
 		}
 	}
 
-	/* Some outputs may be artificially added, to match the inputs
-	   of the fragment program. Blank the outputs here. */
-	for (i = 0; i < VERT_RESULT_MAX; i++) {
-		if (vp->key.OutputsAdded & (1 << i)) {
-			inst[0] = PVS_OP_DST_OPERAND(VE_ADD,
-						     GL_FALSE,
-						     GL_FALSE,
-						     vp->outputs[i],
-						     VSF_FLAG_ALL,
-						     PVS_DST_REG_OUT);
-			inst[1] = __CONST(0, SWIZZLE_ZERO);
-			inst[2] = __CONST(0, SWIZZLE_ZERO);
-			inst[3] = __CONST(0, SWIZZLE_ZERO);
-			inst += 4;
-		}
+	vp->hw_code.length = (inst - vp->hw_code.body.d);
+	if (vp->hw_code.length >= VSF_MAX_FRAGMENT_LENGTH) {
+		vp->error = GL_TRUE;
 	}
-
-	vp->program.length = (inst - vp->program.body.i);
-	if (vp->program.length >= VSF_MAX_FRAGMENT_LENGTH) {
-		vp->program.length = 0;
-		vp->native = GL_FALSE;
-	}
-#if 0
-	fprintf(stderr, "hw program:\n");
-	for (i = 0; i < vp->program.length; i++)
-		fprintf(stderr, "%08x\n", vp->program.body.d[i]);
-#endif
 }
 
 /* DP4 version seems to trigger some hw peculiarity */
@@ -1386,6 +1367,49 @@ static struct r300_vertex_program *build_program(struct r300_vertex_program_key
 		pos_as_texcoord(vp, &mesa_vp->Base);
 	}
 
+	if (RADEON_DEBUG & DEBUG_VERTS) {
+		fprintf(stderr, "Vertex program after native rewrite:\n");
+		_mesa_print_program(&mesa_vp->Base);
+		fflush(stdout);
+	}
+
+	/* Some outputs may be artificially added, to match the inputs of the fragment program.
+	 * Issue 16 of vertex program spec says that all vertex attributes that are unwritten by
+	 * vertex program are undefined, so just use MOV [vertex_result], CONST[0]
+	 */
+	{
+		int i, count = 0;
+		for (i = 0; i < VERT_RESULT_MAX; ++i) {
+			if (vp->key.OutputsAdded & (1 << i)) {
+				++count;
+			}
+		}
+
+		if (count > 0) {
+			struct prog_instruction *inst;
+
+			_mesa_insert_instructions(&mesa_vp->Base, mesa_vp->Base.NumInstructions - 1, count);
+			inst = &mesa_vp->Base.Instructions[mesa_vp->Base.NumInstructions - 1 - count];
+
+			for (i = 0; i < VERT_RESULT_MAX; ++i) {
+				if (vp->key.OutputsAdded & (1 << i)) {
+					inst->Opcode = OPCODE_MOV;
+
+					inst->DstReg.File = PROGRAM_OUTPUT;
+					inst->DstReg.Index = i;
+					inst->DstReg.WriteMask = WRITEMASK_XYZW;
+					inst->DstReg.CondMask = COND_TR;
+
+					inst->SrcReg[0].File = PROGRAM_CONSTANT;
+					inst->SrcReg[0].Index = 0;
+					inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+					++inst;
+				}
+			}
+		}
+	}
+
 	assert(mesa_vp->Base.NumInstructions);
 	vp->num_temporaries = mesa_vp->Base.NumTemporaries;
 	r300TranslateVertexShader(vp, mesa_vp->Base.Instructions);
@@ -1432,7 +1456,12 @@ void r300SelectVertexShader(r300ContextPtr r300)
 		wpos_idx = i;
 	}
 
-	add_outputs(&wanted_key, VERT_RESULT_HPOS);
+	if (vpc->mesa_program.IsPositionInvariant) {
+		wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS);
+		wanted_key.OutputsWritten |= (1 << VERT_RESULT_HPOS);
+	} else {
+		add_outputs(&wanted_key, VERT_RESULT_HPOS);
+	}
 
 	if (InputsRead & FRAG_BIT_COL0) {
 		add_outputs(&wanted_key, VERT_RESULT_COL0);
@@ -1442,27 +1471,103 @@ void r300SelectVertexShader(r300ContextPtr r300)
 		add_outputs(&wanted_key, VERT_RESULT_COL1);
 	}
 
+	if (InputsRead & FRAG_BIT_FOGC) {
+		add_outputs(&wanted_key, VERT_RESULT_FOGC);
+	}
+
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
 		if (InputsRead & (FRAG_BIT_TEX0 << i)) {
 			add_outputs(&wanted_key, VERT_RESULT_TEX0 + i);
 		}
 	}
 
-	if (vpc->mesa_program.IsPositionInvariant) {
-		/* we wan't position don't we ? */
-		wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS);
-	}
-
 	for (vp = vpc->progs; vp; vp = vp->next)
 		if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key))
 		    == 0) {
 			r300->selected_vp = vp;
 			return;
 		}
-	//_mesa_print_program(&vpc->mesa_program.Base);
+
+	if (RADEON_DEBUG & DEBUG_VERTS) {
+		fprintf(stderr, "Initial vertex program:\n");
+		_mesa_print_program(&vpc->mesa_program.Base);
+		fflush(stdout);
+	}
 
 	vp = build_program(&wanted_key, &vpc->mesa_program, wpos_idx);
 	vp->next = vpc->progs;
 	vpc->progs = vp;
 	r300->selected_vp = vp;
 }
+
+#define bump_vpu_count(ptr, new_count)   do { \
+		drm_r300_cmd_header_t* _p=((drm_r300_cmd_header_t*)(ptr)); \
+		int _nc=(new_count)/4; \
+		assert(_nc < 256); \
+		if(_nc>_p->vpu.count)_p->vpu.count=_nc; \
+	} while(0)
+
+static void r300EmitVertexProgram(r300ContextPtr r300, int dest, struct r300_vertex_shader_hw_code *code)
+{
+	int i;
+
+	assert((code->length > 0) && (code->length % 4 == 0));
+
+	switch ((dest >> 8) & 0xf) {
+		case 0:
+			R300_STATECHANGE(r300, vpi);
+			for (i = 0; i < code->length; i++)
+				r300->hw.vpi.cmd[R300_VPI_INSTR_0 + i + 4 * (dest & 0xff)] = (code->body.d[i]);
+			bump_vpu_count(r300->hw.vpi.cmd, code->length + 4 * (dest & 0xff));
+			break;
+		case 2:
+			R300_STATECHANGE(r300, vpp);
+			for (i = 0; i < code->length; i++)
+				r300->hw.vpp.cmd[R300_VPP_PARAM_0 + i + 4 * (dest & 0xff)] = (code->body.d[i]);
+			bump_vpu_count(r300->hw.vpp.cmd, code->length + 4 * (dest & 0xff));
+			break;
+		case 4:
+			R300_STATECHANGE(r300, vps);
+			for (i = 0; i < code->length; i++)
+				r300->hw.vps.cmd[1 + i + 4 * (dest & 0xff)] = (code->body.d[i]);
+			bump_vpu_count(r300->hw.vps.cmd, code->length + 4 * (dest & 0xff));
+			break;
+		default:
+			fprintf(stderr, "%s:%s don't know how to handle dest %04x\n", __FILE__, __FUNCTION__, dest);
+			_mesa_exit(-1);
+	}
+}
+
+void r300SetupVertexProgram(r300ContextPtr rmesa)
+{
+	GLcontext *ctx = rmesa->radeon.glCtx;
+	struct r300_vertex_program *prog = rmesa->selected_vp;
+	int inst_count = 0;
+	int param_count = 0;
+	
+	/* Reset state, in case we don't use something */
+	((drm_r300_cmd_header_t *) rmesa->hw.vpp.cmd)->vpu.count = 0;
+	((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0;
+	((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0;
+	
+	R300_STATECHANGE(rmesa, vpp);
+	param_count = r300VertexProgUpdateParams(ctx,
+								(struct r300_vertex_program_cont *)
+								ctx->VertexProgram._Current,
+								(float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
+	bump_vpu_count(rmesa->hw.vpp.cmd, param_count);
+	param_count /= 4;
+
+	r300EmitVertexProgram(rmesa, R300_PVS_CODE_START, &(prog->hw_code));
+	inst_count = (prog->hw_code.length / 4) - 1;
+
+	r300VapCntl(rmesa, _mesa_bitcount(prog->key.InputsRead),
+				 _mesa_bitcount(prog->key.OutputsWritten), prog->num_temporaries);
+
+	R300_STATECHANGE(rmesa, pvs);
+	rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = (0 << R300_PVS_FIRST_INST_SHIFT) | (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) |
+				(inst_count << R300_PVS_LAST_INST_SHIFT);
+
+	rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] = (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT);
+	rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] = (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT);
+}
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.h b/src/mesa/drivers/dri/r300/r300_vertprog.h
index 2f35f02bc84..b552e3fb1bd 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.h
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.h
@@ -32,4 +32,6 @@
 
 #endif
 
+void r300SetupVertexProgram(r300ContextPtr rmesa);
+
 #endif
diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.c b/src/mesa/drivers/dri/r300/r500_fragprog.c
index 300559d0b4d..4d58cf21622 100644
--- a/src/mesa/drivers/dri/r300/r500_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r500_fragprog.c
@@ -27,10 +27,6 @@
 
 #include "r500_fragprog.h"
 
-#include "radeon_nqssadce.h"
-#include "radeon_program_alu.h"
-
-
 static void reset_srcreg(struct prog_src_register* reg)
 {
 	_mesa_bzero(reg, sizeof(*reg));
@@ -58,12 +54,12 @@ static struct prog_src_register shadow_ambient(struct gl_program *program, int t
  *  - introduce a temporary register when write masks are needed
  *
  */
-static GLboolean transform_TEX(
+GLboolean r500_transform_TEX(
 	struct radeon_transform_context *t,
 	struct prog_instruction* orig_inst, void* data)
 {
-	struct r500_fragment_program_compiler *compiler =
-		(struct r500_fragment_program_compiler*)data;
+	struct r300_fragment_program_compiler *compiler =
+		(struct r300_fragment_program_compiler*)data;
 	struct prog_instruction inst = *orig_inst;
 	struct prog_instruction* tgt;
 	GLboolean destredirect = GL_FALSE;
@@ -188,121 +184,7 @@ static GLboolean transform_TEX(
 	return GL_TRUE;
 }
 
-
-static void update_params(r300ContextPtr r300, struct r500_fragment_program *fp)
-{
-	struct gl_fragment_program *mp = &fp->mesa_program;
-
-	/* Ask Mesa nicely to fill in ParameterValues for us */
-	if (mp->Base.Parameters)
-		_mesa_load_state_parameters(r300->radeon.glCtx, mp->Base.Parameters);
-}
-
-
-/**
- * Transform the program to support fragment.position.
- *
- * Introduce a small fragment at the start of the program that will be
- * the only code that directly reads the FRAG_ATTRIB_WPOS input.
- * All other code pieces that reference that input will be rewritten
- * to read from a newly allocated temporary.
- *
- * \todo if/when r5xx supports the radeon_program architecture, this is a
- * likely candidate for code sharing.
- */
-static void insert_WPOS_trailer(struct r500_fragment_program_compiler *compiler)
-{
-	GLuint InputsRead = compiler->fp->mesa_program.Base.InputsRead;
-
-	if (!(InputsRead & FRAG_BIT_WPOS))
-		return;
-
-	static gl_state_index tokens[STATE_LENGTH] = {
-		STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
-	};
-	struct prog_instruction *fpi;
-	GLuint window_index;
-	int i = 0;
-	GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY);
-
-	_mesa_insert_instructions(compiler->program, 0, 3);
-	fpi = compiler->program->Instructions;
-
-	/* perspective divide */
-	fpi[i].Opcode = OPCODE_RCP;
-
-	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
-	fpi[i].DstReg.Index = tempregi;
-	fpi[i].DstReg.WriteMask = WRITEMASK_W;
-	fpi[i].DstReg.CondMask = COND_TR;
-
-	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
-	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
-	fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
-	i++;
-
-	fpi[i].Opcode = OPCODE_MUL;
-
-	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
-	fpi[i].DstReg.Index = tempregi;
-	fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
-	fpi[i].DstReg.CondMask = COND_TR;
-
-	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
-	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
-	fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-
-	fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
-	fpi[i].SrcReg[1].Index = tempregi;
-	fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
-	i++;
-
-	/* viewport transformation */
-	window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens);
-
-	fpi[i].Opcode = OPCODE_MAD;
-
-	fpi[i].DstReg.File = PROGRAM_TEMPORARY;
-	fpi[i].DstReg.Index = tempregi;
-	fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
-	fpi[i].DstReg.CondMask = COND_TR;
-
-	fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
-	fpi[i].SrcReg[0].Index = tempregi;
-	fpi[i].SrcReg[0].Swizzle =
-	    MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-
-	fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
-	fpi[i].SrcReg[1].Index = window_index;
-	fpi[i].SrcReg[1].Swizzle =
-	    MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-
-	fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
-	fpi[i].SrcReg[2].Index = window_index;
-	fpi[i].SrcReg[2].Swizzle =
-	    MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-	i++;
-
-	for (; i < compiler->program->NumInstructions; ++i) {
-		int reg;
-		for (reg = 0; reg < 3; reg++) {
-			if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT &&
-			    fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) {
-				fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY;
-				fpi[i].SrcReg[reg].Index = tempregi;
-			}
-		}
-	}
-}
-
-
-static void nqssadce_init(struct nqssadce_state* s)
-{
-	s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW;
-	s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W;
-}
-
-static GLboolean is_native_swizzle(GLuint opcode, struct prog_src_register reg)
+GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg)
 {
 	GLuint relevant;
 	int i;
@@ -314,22 +196,20 @@ static GLboolean is_native_swizzle(GLuint opcode, struct prog_src_register reg)
 		if (reg.Abs)
 			return GL_FALSE;
 
+		if (opcode == OPCODE_KIL && (reg.Swizzle != SWIZZLE_NOOP || reg.Negate != NEGATE_NONE))
+			return GL_FALSE;
+
 		if (reg.Negate)
 			reg.Negate ^= NEGATE_XYZW;
 
-		if (opcode == OPCODE_KIL) {
-			if (reg.Swizzle != SWIZZLE_NOOP)
-				return GL_FALSE;
-		} else {
-			for(i = 0; i < 4; ++i) {
-				GLuint swz = GET_SWZ(reg.Swizzle, i);
-				if (swz == SWIZZLE_NIL) {
-					reg.Negate &= ~(1 << i);
-					continue;
-				}
-				if (swz >= 4)
-					return GL_FALSE;
+		for(i = 0; i < 4; ++i) {
+			GLuint swz = GET_SWZ(reg.Swizzle, i);
+			if (swz == SWIZZLE_NIL) {
+				reg.Negate &= ~(1 << i);
+				continue;
 			}
+			if (swz >= 4)
+				return GL_FALSE;
 		}
 
 		if (reg.Negate)
@@ -367,8 +247,7 @@ static GLboolean is_native_swizzle(GLuint opcode, struct prog_src_register reg)
  * The only thing we *cannot* do in an ALU instruction is per-component
  * negation. Therefore, we split the MOV into two instructions when necessary.
  */
-static void nqssadce_build_swizzle(struct nqssadce_state *s,
-	struct prog_dst_register dst, struct prog_src_register src)
+void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src)
 {
 	struct prog_instruction *inst;
 	GLuint negatebase[2] = { 0, 0 };
@@ -392,130 +271,12 @@ static void nqssadce_build_swizzle(struct nqssadce_state *s,
 		inst->DstReg = dst;
 		inst->DstReg.WriteMask = negatebase[i];
 		inst->SrcReg[0] = src;
+		inst->SrcReg[0].Negate = (i == 0) ? NEGATE_NONE : NEGATE_XYZW;
 		inst++;
 		s->IP++;
 	}
 }
 
-static GLuint build_dtm(GLuint depthmode)
-{
-	switch(depthmode) {
-	default:
-	case GL_LUMINANCE: return 0;
-	case GL_INTENSITY: return 1;
-	case GL_ALPHA: return 2;
-	}
-}
-
-static GLuint build_func(GLuint comparefunc)
-{
-	return comparefunc - GL_NEVER;
-}
-
-
-/**
- * Collect all external state that is relevant for compiling the given
- * fragment program.
- */
-static void build_state(
-	r300ContextPtr r300,
-	struct r500_fragment_program *fp,
-	struct r500_fragment_program_external_state *state)
-{
-	int unit;
-
-	_mesa_bzero(state, sizeof(*state));
-
-	for(unit = 0; unit < 16; ++unit) {
-		if (fp->mesa_program.Base.ShadowSamplers & (1 << unit)) {
-			struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
-
-			state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode);
-			state->unit[unit].texture_compare_func = build_func(tex->CompareFunc);
-		}
-	}
-}
-
-static void dump_program(struct r500_fragment_program_code *code);
-
-void r500TranslateFragmentShader(r300ContextPtr r300,
-				 struct r500_fragment_program *fp)
-{
-	struct r500_fragment_program_external_state state;
-
-	build_state(r300, fp, &state);
-	if (_mesa_memcmp(&fp->state, &state, sizeof(state))) {
-		/* TODO: cache compiled programs */
-		fp->translated = GL_FALSE;
-		_mesa_memcpy(&fp->state, &state, sizeof(state));
-	}
-
-	if (!fp->translated) {
-		struct r500_fragment_program_compiler compiler;
-
-		compiler.r300 = r300;
-		compiler.fp = fp;
-		compiler.code = &fp->code;
-		compiler.program = _mesa_clone_program(r300->radeon.glCtx, &fp->mesa_program.Base);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Compiler: Initial program:\n");
-			_mesa_print_program(compiler.program);
-		}
-
-		insert_WPOS_trailer(&compiler);
-
-		struct radeon_program_transformation transformations[] = {
-			{ &transform_TEX, &compiler },
-			{ &radeonTransformALU, 0 },
-			{ &radeonTransformDeriv, 0 },
-			{ &radeonTransformTrigScale, 0 }
-		};
-		radeonLocalTransform(r300->radeon.glCtx, compiler.program,
-			4, transformations);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Compiler: after native rewrite:\n");
-			_mesa_print_program(compiler.program);
-		}
-
-		struct radeon_nqssadce_descr nqssadce = {
-			.Init = &nqssadce_init,
-			.IsNativeSwizzle = &is_native_swizzle,
-			.BuildSwizzle = &nqssadce_build_swizzle,
-			.RewriteDepthOut = GL_TRUE
-		};
-		radeonNqssaDce(r300->radeon.glCtx, compiler.program, &nqssadce);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Compiler: after NqSSA-DCE:\n");
-			_mesa_print_program(compiler.program);
-		}
-
-		fp->translated = r500FragmentProgramEmit(&compiler);
-
-		/* Subtle: Rescue any parameters that have been added during transformations */
-		_mesa_free_parameter_list(fp->mesa_program.Base.Parameters);
-		fp->mesa_program.Base.Parameters = compiler.program->Parameters;
-		compiler.program->Parameters = 0;
-
-		_mesa_reference_program(r300->radeon.glCtx, &compiler.program, 0);
-
-		r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM |
-                                          _NEW_PROGRAM_CONSTANTS);
-
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			if (fp->translated) {
-				_mesa_printf("Machine-readable code:\n");
-				dump_program(&fp->code);
-			}
-		}
-
-	}
-
-	update_params(r300, fp);
-
-}
 
 static char *toswiz(int swiz_val) {
   switch(swiz_val) {
@@ -614,9 +375,9 @@ static char *to_texop(int val)
   return NULL;
 }
 
-static void dump_program(struct r500_fragment_program_code *code)
+void r500FragmentProgramDump(union rX00_fragment_program_code *c)
 {
-
+  struct r500_fragment_program_code *code = &c->r500;
   fprintf(stderr, "R500 Fragment Program:\n--------\n");
 
   int n;
diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.h b/src/mesa/drivers/dri/r300/r500_fragprog.h
index 1e45538f807..1179bf66073 100644
--- a/src/mesa/drivers/dri/r300/r500_fragprog.h
+++ b/src/mesa/drivers/dri/r300/r500_fragprog.h
@@ -33,30 +33,20 @@
 #ifndef __R500_FRAGPROG_H_
 #define __R500_FRAGPROG_H_
 
-#include "main/glheader.h"
-#include "main/macros.h"
-#include "main/enums.h"
 #include "shader/prog_parameter.h"
-#include "shader/prog_print.h"
-#include "shader/program.h"
 #include "shader/prog_instruction.h"
 
 #include "r300_context.h"
-#include "r300_state.h"
-#include "radeon_program.h"
+#include "radeon_nqssadce.h"
 
-struct r500_fragment_program;
+extern GLboolean r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler);
 
-extern void r500TranslateFragmentShader(r300ContextPtr r300,
-					struct r500_fragment_program *fp);
+extern void r500FragmentProgramDump(union rX00_fragment_program_code *c);
 
-struct r500_fragment_program_compiler {
-	r300ContextPtr r300;
-	struct r500_fragment_program *fp;
-	struct r500_fragment_program_code *code;
-	struct gl_program *program;
-};
+extern GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg);
 
-extern GLboolean r500FragmentProgramEmit(struct r500_fragment_program_compiler *compiler);
+extern void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src);
+
+extern GLboolean r500_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/r500_fragprog_emit.c
index 4631235f0d3..30f4514897e 100644
--- a/src/mesa/drivers/dri/r300/r500_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/r500_fragprog_emit.c
@@ -49,8 +49,8 @@
 
 
 #define PROG_CODE \
-	struct r500_fragment_program_compiler *c = (struct r500_fragment_program_compiler*)data; \
-	struct r500_fragment_program_code *code = c->code
+	struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \
+	struct r500_fragment_program_code *code = &c->code->r500
 
 #define error(fmt, args...) do {			\
 		fprintf(stderr, "%s::%s(): " fmt "\n",	\
@@ -72,7 +72,7 @@ static GLboolean emit_const(void *data, GLuint file, GLuint idx, GLuint *hwindex
 	}
 
 	if (*hwindex >= code->const_nr) {
-		if (*hwindex >= PFS_NUM_CONST_REGS) {
+		if (*hwindex >= R500_PFS_NUM_CONST_REGS) {
 			error("Out of hw constants!\n");
 			return GL_FALSE;
 		}
@@ -299,9 +299,9 @@ static const struct radeon_pair_handler pair_handler = {
 	.MaxHwTemps = 128
 };
 
-GLboolean r500FragmentProgramEmit(struct r500_fragment_program_compiler *compiler)
+GLboolean r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler)
 {
-	struct r500_fragment_program_code *code = compiler->code;
+	struct r500_fragment_program_code *code = &compiler->code->r500;
 
 	_mesa_bzero(code, sizeof(*code));
 	code->max_temp_idx = 1;
diff --git a/src/mesa/drivers/dri/r300/radeon_context.c b/src/mesa/drivers/dri/r300/radeon_context.c
deleted file mode 100644
index 5267fe9a774..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_context.c
+++ /dev/null
@@ -1,330 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/**
- * \file radeon_context.c
- * Common context initialization.
- *
- * \author Keith Whitwell <[email protected]>
- */
-
-#include <dlfcn.h>
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/context.h"
-#include "main/state.h"
-#include "main/matrix.h"
-#include "main/framebuffer.h"
-
-#include "drivers/common/driverfuncs.h"
-#include "swrast/swrast.h"
-
-#include "radeon_screen.h"
-#include "radeon_ioctl.h"
-#include "radeon_macros.h"
-#include "radeon_reg.h"
-
-#include "radeon_state.h"
-#include "r300_state.h"
-
-#include "utils.h"
-#include "vblank.h"
-#include "xmlpool.h"		/* for symbolic values of enum-type options */
-
-#define DRIVER_DATE "20060815"
-
-
-/* Return various strings for glGetString().
- */
-static const GLubyte *radeonGetString(GLcontext * ctx, GLenum name)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-	static char buffer[128];
-
-	switch (name) {
-	case GL_VENDOR:
-		if (IS_R300_CLASS(radeon->radeonScreen))
-			return (GLubyte *) "DRI R300 Project";
-		else
-			return (GLubyte *) "Tungsten Graphics, Inc.";
-
-	case GL_RENDERER:
-	{
-		unsigned offset;
-		GLuint agp_mode = (radeon->radeonScreen->card_type==RADEON_CARD_PCI) ? 0 :
-			radeon->radeonScreen->AGPMode;
-		const char* chipname;
-
-		if (IS_R300_CLASS(radeon->radeonScreen))
-			chipname = "R300";
-		else
-			chipname = "R200";
-
-		offset = driGetRendererString(buffer, chipname, DRIVER_DATE,
-					      agp_mode);
-
-		if (IS_R300_CLASS(radeon->radeonScreen)) {
-		sprintf(&buffer[offset], " %sTCL",
-			(radeon->radeonScreen->chip_flags & RADEON_CHIPSET_TCL)
-			? "" : "NO-");
-		} else {
-			sprintf(&buffer[offset], " %sTCL",
-			!(radeon->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE)
-			? "" : "NO-");
-		}
-
-		return (GLubyte *) buffer;
-	}
-
-	default:
-		return NULL;
-	}
-}
-
-/* Initialize the driver's misc functions.
- */
-static void radeonInitDriverFuncs(struct dd_function_table *functions)
-{
-	functions->GetString = radeonGetString;
-}
-
-
-/**
- * Create and initialize all common fields of the context,
- * including the Mesa context itself.
- */
-GLboolean radeonInitContext(radeonContextPtr radeon,
-			    struct dd_function_table* functions,
-			    const __GLcontextModes * glVisual,
-			    __DRIcontextPrivate * driContextPriv,
-			    void *sharedContextPrivate)
-{
-	__DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
-	radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private);
-	GLcontext* ctx;
-	GLcontext* shareCtx;
-	int fthrottle_mode;
-
-	/* Fill in additional standard functions. */
-	radeonInitDriverFuncs(functions);
-
-	radeon->radeonScreen = screen;
-	/* Allocate and initialize the Mesa context */
-	if (sharedContextPrivate)
-		shareCtx = ((radeonContextPtr)sharedContextPrivate)->glCtx;
-	else
-		shareCtx = NULL;
-	radeon->glCtx = _mesa_create_context(glVisual, shareCtx,
-					    functions, (void *)radeon);
-	if (!radeon->glCtx)
-		return GL_FALSE;
-
-	ctx = radeon->glCtx;
-	driContextPriv->driverPrivate = radeon;
-
-	/* DRI fields */
-	radeon->dri.context = driContextPriv;
-	radeon->dri.screen = sPriv;
-	radeon->dri.drawable = NULL;
-	radeon->dri.readable = NULL;
-	radeon->dri.hwContext = driContextPriv->hHWContext;
-	radeon->dri.hwLock = &sPriv->pSAREA->lock;
-	radeon->dri.fd = sPriv->fd;
-	radeon->dri.drmMinor = sPriv->drm_version.minor;
-
-	radeon->sarea = (drm_radeon_sarea_t *) ((GLubyte *) sPriv->pSAREA +
-					       screen->sarea_priv_offset);
-
-	/* Setup IRQs */
-	fthrottle_mode = driQueryOptioni(&radeon->optionCache, "fthrottle_mode");
-	radeon->iw.irq_seq = -1;
-	radeon->irqsEmitted = 0;
-	radeon->do_irqs = (fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS &&
-			  radeon->radeonScreen->irq);
-
-	radeon->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS);
-
-	if (!radeon->do_irqs)
-		fprintf(stderr,
-			"IRQ's not enabled, falling back to %s: %d %d\n",
-			radeon->do_usleeps ? "usleeps" : "busy waits",
-			fthrottle_mode, radeon->radeonScreen->irq);
-
-	(*sPriv->systemTime->getUST) (&radeon->swap_ust);
-
-	return GL_TRUE;
-}
-
-
-/**
- * Cleanup common context fields.
- * Called by r200DestroyContext/r300DestroyContext
- */
-void radeonCleanupContext(radeonContextPtr radeon)
-{
-	/* _mesa_destroy_context() might result in calls to functions that
-	 * depend on the DriverCtx, so don't set it to NULL before.
-	 *
-	 * radeon->glCtx->DriverCtx = NULL;
-	 */
-
-	/* free the Mesa context */
-	_mesa_destroy_context(radeon->glCtx);
-
-	if (radeon->state.scissor.pClipRects) {
-		FREE(radeon->state.scissor.pClipRects);
-		radeon->state.scissor.pClipRects = 0;
-	}
-}
-
-
-/**
- * Swap front and back buffer.
- */
-void radeonSwapBuffers(__DRIdrawablePrivate * dPriv)
-{
-	if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
-		radeonContextPtr radeon;
-		GLcontext *ctx;
-
-		radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-		ctx = radeon->glCtx;
-
-		if (ctx->Visual.doubleBufferMode) {
-			_mesa_notifySwapBuffers(ctx);	/* flush pending rendering comands */
-			if (radeon->doPageFlip) {
-				radeonPageFlip(dPriv);
-			} else {
-			    radeonCopyBuffer(dPriv, NULL);
-			}
-		}
-	} else {
-		/* XXX this shouldn't be an error but we can't handle it for now */
-		_mesa_problem(NULL, "%s: drawable has no context!",
-			      __FUNCTION__);
-	}
-}
-
-void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
-			 int x, int y, int w, int h )
-{
-    if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
-	radeonContextPtr radeon;
-	GLcontext *ctx;
-
-	radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-	ctx = radeon->glCtx;
-
-	if (ctx->Visual.doubleBufferMode) {
-	    drm_clip_rect_t rect;
-	    rect.x1 = x + dPriv->x;
-	    rect.y1 = (dPriv->h - y - h) + dPriv->y;
-	    rect.x2 = rect.x1 + w;
-	    rect.y2 = rect.y1 + h;
-	    _mesa_notifySwapBuffers(ctx);	/* flush pending rendering comands */
-	    radeonCopyBuffer(dPriv, &rect);
-	}
-    } else {
-	/* XXX this shouldn't be an error but we can't handle it for now */
-	_mesa_problem(NULL, "%s: drawable has no context!",
-		      __FUNCTION__);
-    }
-}
-
-/* Force the context `c' to be the current context and associate with it
- * buffer `b'.
- */
-GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv,
-			    __DRIdrawablePrivate * driDrawPriv,
-			    __DRIdrawablePrivate * driReadPriv)
-{
-	if (driContextPriv) {
-		radeonContextPtr radeon =
-			(radeonContextPtr) driContextPriv->driverPrivate;
-
-		if (RADEON_DEBUG & DEBUG_DRI)
-			fprintf(stderr, "%s ctx %p\n", __FUNCTION__,
-				radeon->glCtx);
-
-		if (radeon->dri.drawable != driDrawPriv) {
-			if (driDrawPriv->swap_interval == (unsigned)-1) {
-				driDrawPriv->vblFlags =
-					(radeon->radeonScreen->irq != 0)
-					? driGetDefaultVBlankFlags(&radeon->
-								   optionCache)
-					: VBLANK_FLAG_NO_IRQ;
-
-				driDrawableInitVBlank(driDrawPriv);
-			}
-		}
-
-		radeon->dri.readable = driReadPriv;
-
-		if (radeon->dri.drawable != driDrawPriv ||
-		    radeon->lastStamp != driDrawPriv->lastStamp) {
-			radeon->dri.drawable = driDrawPriv;
-
-			radeonSetCliprects(radeon);
-			r300UpdateViewportOffset(radeon->glCtx);
-		}
-
-		_mesa_make_current(radeon->glCtx,
-				    (GLframebuffer *) driDrawPriv->
-				    driverPrivate,
-				    (GLframebuffer *) driReadPriv->
-				    driverPrivate);
-
-		_mesa_update_state(radeon->glCtx);		
-
-		radeonUpdatePageFlipping(radeon);
-	} else {
-		if (RADEON_DEBUG & DEBUG_DRI)
-			fprintf(stderr, "%s ctx is null\n", __FUNCTION__);
-		_mesa_make_current(0, 0, 0);
-	}
-
-	if (RADEON_DEBUG & DEBUG_DRI)
-		fprintf(stderr, "End %s\n", __FUNCTION__);
-	return GL_TRUE;
-}
-
-/* Force the context `c' to be unbound from its buffer.
- */
-GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv)
-{
-	radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate;
-
-	if (RADEON_DEBUG & DEBUG_DRI)
-		fprintf(stderr, "%s ctx %p\n", __FUNCTION__,
-			radeon->glCtx);
-
-	return GL_TRUE;
-}
-
diff --git a/src/mesa/drivers/dri/r300/radeon_context.h b/src/mesa/drivers/dri/r300/radeon_context.h
index 47cbc22a725..250570f6b89 100644
--- a/src/mesa/drivers/dri/r300/radeon_context.h
+++ b/src/mesa/drivers/dri/r300/radeon_context.h
@@ -49,20 +49,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "drm.h"
 #include "dri_util.h"
 
-struct radeon_context;
-typedef struct radeon_context radeonContextRec;
-typedef struct radeon_context *radeonContextPtr;
-
-/* Rasterizing fallbacks */
-/* See correponding strings in r200_swtcl.c */
-#define RADEON_FALLBACK_TEXTURE		0x0001
-#define RADEON_FALLBACK_DRAW_BUFFER	0x0002
-#define RADEON_FALLBACK_STENCIL		0x0004
-#define RADEON_FALLBACK_RENDER_MODE	0x0008
-#define RADEON_FALLBACK_BLEND_EQ	0x0010
-#define RADEON_FALLBACK_BLEND_FUNC	0x0020
-#define RADEON_FALLBACK_DISABLE		0x0040
-#define RADEON_FALLBACK_BORDER_MODE	0x0080
+#include "radeon_screen.h"
 
 #if R200_MERGED
 extern void radeonFallback(GLcontext * ctx, GLuint bit, GLboolean mode);
@@ -79,155 +66,11 @@ extern void radeonFallback(GLcontext * ctx, GLuint bit, GLboolean mode);
 /* TCL fallbacks */
 extern void radeonTclFallback(GLcontext * ctx, GLuint bit, GLboolean mode);
 
-#define RADEON_TCL_FALLBACK_RASTER		0x0001	/* rasterization */
-#define RADEON_TCL_FALLBACK_UNFILLED		0x0002	/* unfilled tris */
-#define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE	0x0004	/* twoside tris */
-#define RADEON_TCL_FALLBACK_MATERIAL		0x0008	/* material in vb */
-#define RADEON_TCL_FALLBACK_TEXGEN_0		0x0010	/* texgen, unit 0 */
-#define RADEON_TCL_FALLBACK_TEXGEN_1		0x0020	/* texgen, unit 1 */
-#define RADEON_TCL_FALLBACK_TEXGEN_2		0x0040	/* texgen, unit 2 */
-#define RADEON_TCL_FALLBACK_TEXGEN_3		0x0080	/* texgen, unit 3 */
-#define RADEON_TCL_FALLBACK_TEXGEN_4		0x0100	/* texgen, unit 4 */
-#define RADEON_TCL_FALLBACK_TEXGEN_5		0x0200	/* texgen, unit 5 */
-#define RADEON_TCL_FALLBACK_TCL_DISABLE		0x0400	/* user disable */
-#define RADEON_TCL_FALLBACK_BITMAP		0x0800	/* draw bitmap with points */
-#define RADEON_TCL_FALLBACK_VERTEX_PROGRAM	0x1000	/* vertex program active */
-
 #if R200_MERGED
 #define TCL_FALLBACK( ctx, bit, mode )	radeonTclFallback( ctx, bit, mode )
 #else
 #define TCL_FALLBACK( ctx, bit, mode )	;
 #endif
 
-struct radeon_dri_mirror {
-	__DRIcontextPrivate *context;	/* DRI context */
-	__DRIscreenPrivate *screen;	/* DRI screen */
-	/**
-	 * DRI drawable bound to this context for drawing.
-	 */
-	__DRIdrawablePrivate *drawable;
-
-	/**
-	 * DRI drawable bound to this context for reading.
-	 */
-	__DRIdrawablePrivate *readable;
-
-	drm_context_t hwContext;
-	drm_hw_lock_t *hwLock;
-	int fd;
-	int drmMinor;
-};
-
-/**
- * Derived state for internal purposes.
- */
-struct radeon_scissor_state {
-	drm_clip_rect_t rect;
-	GLboolean enabled;
-
-	GLuint numClipRects;	/* Cliprects active */
-	GLuint numAllocedClipRects;	/* Cliprects available */
-	drm_clip_rect_t *pClipRects;
-};
-
-struct radeon_colorbuffer_state {
-	GLuint clear;
-	GLint drawOffset, drawPitch;
-};
-
-struct radeon_state {
-	struct radeon_colorbuffer_state color;
-	struct radeon_scissor_state scissor;
-};
-
-/**
- * Common per-context variables shared by R200 and R300.
- * R200- and R300-specific code "derive" their own context from this
- * structure.
- */
-struct radeon_context {
-	GLcontext *glCtx;	/* Mesa context */
-	radeonScreenPtr radeonScreen;	/* Screen private DRI data */
-
-	/* Fallback state */
-	GLuint Fallback;
-	GLuint TclFallback;
-
-	/* Page flipping */
-	GLuint doPageFlip;
-
-	/* Drawable, cliprect and scissor information */
-	GLuint numClipRects;	/* Cliprects for the draw buffer */
-	drm_clip_rect_t *pClipRects;
-	unsigned int lastStamp;
-	GLboolean lost_context;
-	drm_radeon_sarea_t *sarea;	/* Private SAREA data */
-
-	/* Mirrors of some DRI state */
-	struct radeon_dri_mirror dri;
-
-	/* Busy waiting */
-	GLuint do_usleeps;
-	GLuint do_irqs;
-	GLuint irqsEmitted;
-	drm_radeon_irq_wait_t iw;
-
-	/* buffer swap */
-	int64_t swap_ust;
-	int64_t swap_missed_ust;
-
-	GLuint swap_count;
-	GLuint swap_missed_count;
-
-	/* Derived state */
-	struct radeon_state state;
-
-	/* Configuration cache
-	 */
-	driOptionCache optionCache;
-};
-
-#define RADEON_CONTEXT(glctx) ((radeonContextPtr)(ctx->DriverCtx))
-
-extern void radeonSwapBuffers(__DRIdrawablePrivate * dPriv);
-extern void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
-				int x, int y, int w, int h);
-extern GLboolean radeonInitContext(radeonContextPtr radeon,
-				   struct dd_function_table *functions,
-				   const __GLcontextModes * glVisual,
-				   __DRIcontextPrivate * driContextPriv,
-				   void *sharedContextPrivate);
-extern void radeonCleanupContext(radeonContextPtr radeon);
-extern GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv,
-				   __DRIdrawablePrivate * driDrawPriv,
-				   __DRIdrawablePrivate * driReadPriv);
-extern GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv);
-
-/* ================================================================
- * Debugging:
- */
-#define DO_DEBUG		1
-
-#if DO_DEBUG
-extern int RADEON_DEBUG;
-#else
-#define RADEON_DEBUG		0
-#endif
-
-#define DEBUG_TEXTURE	0x0001
-#define DEBUG_STATE	0x0002
-#define DEBUG_IOCTL	0x0004
-#define DEBUG_PRIMS	0x0008
-#define DEBUG_VERTS	0x0010
-#define DEBUG_FALLBACKS	0x0020
-#define DEBUG_VFMT	0x0040
-#define DEBUG_CODEGEN	0x0080
-#define DEBUG_VERBOSE	0x0100
-#define DEBUG_DRI       0x0200
-#define DEBUG_DMA       0x0400
-#define DEBUG_SANITY    0x0800
-#define DEBUG_SYNC      0x1000
-#define DEBUG_PIXEL     0x2000
-#define DEBUG_MEMORY    0x4000
 
 #endif				/* __RADEON_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/r300/radeon_ioctl.c b/src/mesa/drivers/dri/r300/radeon_ioctl.c
deleted file mode 100644
index f042a7b943a..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_ioctl.c
+++ /dev/null
@@ -1,396 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
-
-#include <sched.h>
-#include <errno.h>
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/macros.h"
-#include "main/context.h"
-#include "swrast/swrast.h"
-#include "r300_context.h"
-#include "radeon_ioctl.h"
-#include "r300_ioctl.h"
-#include "r300_state.h"
-#include "radeon_reg.h"
-
-#include "drirenderbuffer.h"
-#include "vblank.h"
-
-static void radeonWaitForIdle(radeonContextPtr radeon);
-
-/* ================================================================
- * SwapBuffers with client-side throttling
- */
-
-static uint32_t radeonGetLastFrame(radeonContextPtr radeon)
-{
-	drm_radeon_getparam_t gp;
-	int ret;
-	uint32_t frame = 0;
-
-	gp.param = RADEON_PARAM_LAST_FRAME;
-	gp.value = (int *)&frame;
-	ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM,
-				  &gp, sizeof(gp));
-	if (ret) {
-		fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__,
-			ret);
-		exit(1);
-	}
-
-	return frame;
-}
-
-uint32_t radeonGetAge(radeonContextPtr radeon)
-{
-	drm_radeon_getparam_t gp;
-	int ret;
-	uint32_t age = 0;
-
-	gp.param = RADEON_PARAM_LAST_CLEAR;
-	gp.value = (int *)&age;
-	ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM,
-				  &gp, sizeof(gp));
-	if (ret) {
-		fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__,
-			ret);
-		exit(1);
-	}
-
-	return age;
-}
-
-static void radeonEmitIrqLocked(radeonContextPtr radeon)
-{
-	drm_radeon_irq_emit_t ie;
-	int ret;
-
-	ie.irq_seq = &radeon->iw.irq_seq;
-	ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_IRQ_EMIT,
-				  &ie, sizeof(ie));
-	if (ret) {
-		fprintf(stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__,
-			ret);
-		exit(1);
-	}
-}
-
-static void radeonWaitIrq(radeonContextPtr radeon)
-{
-	int ret;
-
-	do {
-		ret = drmCommandWrite(radeon->dri.fd, DRM_RADEON_IRQ_WAIT,
-				      &radeon->iw, sizeof(radeon->iw));
-	} while (ret && (errno == EINTR || errno == EBUSY));
-
-	if (ret) {
-		fprintf(stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__,
-			ret);
-		exit(1);
-	}
-}
-
-static void radeonWaitForFrameCompletion(radeonContextPtr radeon)
-{
-	drm_radeon_sarea_t *sarea = radeon->sarea;
-
-	if (radeon->do_irqs) {
-		if (radeonGetLastFrame(radeon) < sarea->last_frame) {
-			if (!radeon->irqsEmitted) {
-				while (radeonGetLastFrame(radeon) <
-				       sarea->last_frame) ;
-			} else {
-				UNLOCK_HARDWARE(radeon);
-				radeonWaitIrq(radeon);
-				LOCK_HARDWARE(radeon);
-			}
-			radeon->irqsEmitted = 10;
-		}
-
-		if (radeon->irqsEmitted) {
-			radeonEmitIrqLocked(radeon);
-			radeon->irqsEmitted--;
-		}
-	} else {
-		while (radeonGetLastFrame(radeon) < sarea->last_frame) {
-			UNLOCK_HARDWARE(radeon);
-			if (radeon->do_usleeps)
-				DO_USLEEP(1);
-			LOCK_HARDWARE(radeon);
-		}
-	}
-}
-
-/* Copy the back color buffer to the front color buffer.
- */
-void radeonCopyBuffer(__DRIdrawablePrivate * dPriv,
-		      const drm_clip_rect_t	 * rect)
-{
-	radeonContextPtr radeon;
-	GLint nbox, i, ret;
-	GLboolean missed_target;
-	int64_t ust;
-	__DRIscreenPrivate *psp = dPriv->driScreenPriv;
-
-	assert(dPriv);
-	assert(dPriv->driContextPriv);
-	assert(dPriv->driContextPriv->driverPrivate);
-
-	radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-
-	if (RADEON_DEBUG & DEBUG_IOCTL) {
-		fprintf(stderr, "\n%s( %p )\n\n", __FUNCTION__,
-			(void *)radeon->glCtx);
-	}
-
-	r300Flush(radeon->glCtx);
-
-	LOCK_HARDWARE(radeon);
-
-	/* Throttle the frame rate -- only allow one pending swap buffers
-	 * request at a time.
-	 */
-	radeonWaitForFrameCompletion(radeon);
-	if (!rect)
-	{
-	    UNLOCK_HARDWARE(radeon);
-	    driWaitForVBlank(dPriv, &missed_target);
-	    LOCK_HARDWARE(radeon);
-	}
-
-	nbox = dPriv->numClipRects;	/* must be in locked region */
-
-	for (i = 0; i < nbox;) {
-		GLint nr = MIN2(i + RADEON_NR_SAREA_CLIPRECTS, nbox);
-		drm_clip_rect_t *box = dPriv->pClipRects;
-		drm_clip_rect_t *b = radeon->sarea->boxes;
-		GLint n = 0;
-
-		for ( ; i < nr ; i++ ) {
-
-		    *b = box[i];
-
-		    if (rect)
-		    {
-			if (rect->x1 > b->x1)
-			    b->x1 = rect->x1;
-			if (rect->y1 > b->y1)
-			    b->y1 = rect->y1;
-			if (rect->x2 < b->x2)
-			    b->x2 = rect->x2;
-			if (rect->y2 < b->y2)
-			    b->y2 = rect->y2;
-
-			if (b->x1 >= b->x2 || b->y1 >= b->y2)
-			    continue;
-		    }
-
-		    b++;
-		    n++;
-		}
-		radeon->sarea->nbox = n;
-
-		if (!n)
-		   continue;
-
-		ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_SWAP);
-
-		if (ret) {
-			fprintf(stderr, "DRM_RADEON_SWAP: return = %d\n",
-				ret);
-			UNLOCK_HARDWARE(radeon);
-			exit(1);
-		}
-	}
-
-	UNLOCK_HARDWARE(radeon);
-	if (!rect)
-	{
-	    ((r300ContextPtr)radeon)->hw.all_dirty = GL_TRUE;
-
-	    radeon->swap_count++;
-	    (*psp->systemTime->getUST) (&ust);
-	    if (missed_target) {
-		radeon->swap_missed_count++;
-		radeon->swap_missed_ust = ust - radeon->swap_ust;
-	    }
-
-	    radeon->swap_ust = ust;
-
-	    sched_yield();
-	}
-}
-
-void radeonPageFlip(__DRIdrawablePrivate * dPriv)
-{
-	radeonContextPtr radeon;
-	GLint ret;
-	GLboolean missed_target;
-	__DRIscreenPrivate *psp = dPriv->driScreenPriv;
-
-	assert(dPriv);
-	assert(dPriv->driContextPriv);
-	assert(dPriv->driContextPriv->driverPrivate);
-
-	radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-
-	if (RADEON_DEBUG & DEBUG_IOCTL) {
-		fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__,
-			radeon->sarea->pfCurrentPage);
-	}
-
-	r300Flush(radeon->glCtx);
-	LOCK_HARDWARE(radeon);
-
-	if (!dPriv->numClipRects) {
-		UNLOCK_HARDWARE(radeon);
-		usleep(10000);	/* throttle invisible client 10ms */
-		return;
-	}
-
-	/* Need to do this for the perf box placement:
-	 */
-	{
-		drm_clip_rect_t *box = dPriv->pClipRects;
-		drm_clip_rect_t *b = radeon->sarea->boxes;
-		b[0] = box[0];
-		radeon->sarea->nbox = 1;
-	}
-
-	/* Throttle the frame rate -- only allow a few pending swap buffers
-	 * request at a time.
-	 */
-	radeonWaitForFrameCompletion(radeon);
-	UNLOCK_HARDWARE(radeon);
-	driWaitForVBlank(dPriv, &missed_target);
-	if (missed_target) {
-		radeon->swap_missed_count++;
-		(void)(*psp->systemTime->getUST) (&radeon->swap_missed_ust);
-	}
-	LOCK_HARDWARE(radeon);
-
-	ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_FLIP);
-
-	UNLOCK_HARDWARE(radeon);
-
-	if (ret) {
-		fprintf(stderr, "DRM_RADEON_FLIP: return = %d\n", ret);
-		exit(1);
-	}
-
-	radeon->swap_count++;
-	(void)(*psp->systemTime->getUST) (&radeon->swap_ust);
-
-        driFlipRenderbuffers(radeon->glCtx->WinSysDrawBuffer, 
-                             radeon->sarea->pfCurrentPage);
-
-	if (radeon->sarea->pfCurrentPage == 1) {
-		radeon->state.color.drawOffset = radeon->radeonScreen->frontOffset;
-		radeon->state.color.drawPitch = radeon->radeonScreen->frontPitch;
-	} else {
-		radeon->state.color.drawOffset = radeon->radeonScreen->backOffset;
-		radeon->state.color.drawPitch = radeon->radeonScreen->backPitch;
-	}
-
-	if (IS_R300_CLASS(radeon->radeonScreen)) {
-		r300ContextPtr r300 = (r300ContextPtr)radeon;
-		R300_STATECHANGE(r300, cb);
-		r300->hw.cb.cmd[R300_CB_OFFSET] = r300->radeon.state.color.drawOffset + 
-						r300->radeon.radeonScreen->fbLocation;
-		r300->hw.cb.cmd[R300_CB_PITCH] = r300->radeon.state.color.drawPitch;
-		
-		if (r300->radeon.radeonScreen->cpp == 4)
-			r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_ARGB8888;
-		else
-			r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_RGB565;
-	
-		if (r300->radeon.sarea->tiling_enabled)
-			r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_TILE_ENABLE;
-	}
-}
-
-void radeonWaitForIdleLocked(radeonContextPtr radeon)
-{
-	int ret;
-	int i = 0;
-
-	do {
-		ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_CP_IDLE);
-		if (ret)
-			DO_USLEEP(1);
-	} while (ret && ++i < 100);
-
-	if (ret < 0) {
-		UNLOCK_HARDWARE(radeon);
-		fprintf(stderr, "Error: R300 timed out... exiting\n");
-		exit(-1);
-	}
-}
-
-static void radeonWaitForIdle(radeonContextPtr radeon)
-{
-	LOCK_HARDWARE(radeon);
-	radeonWaitForIdleLocked(radeon);
-	UNLOCK_HARDWARE(radeon);
-}
-
-void radeonFlush(GLcontext * ctx)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-
-	if (IS_R300_CLASS(radeon->radeonScreen))
-		r300Flush(ctx);
-}
-
-
-/* Make sure all commands have been sent to the hardware and have
- * completed processing.
- */
-void radeonFinish(GLcontext * ctx)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-
-	radeonFlush(ctx);
-
-	if (radeon->do_irqs) {
-		LOCK_HARDWARE(radeon);
-		radeonEmitIrqLocked(radeon);
-		UNLOCK_HARDWARE(radeon);
-		radeonWaitIrq(radeon);
-	} else
-		radeonWaitForIdle(radeon);
-}
diff --git a/src/mesa/drivers/dri/r300/radeon_ioctl.h b/src/mesa/drivers/dri/r300/radeon_ioctl.h
deleted file mode 100644
index 3add775b822..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_ioctl.h
+++ /dev/null
@@ -1,57 +0,0 @@
-/*
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
-
-#ifndef __RADEON_IOCTL_H__
-#define __RADEON_IOCTL_H__
-
-#include "main/simple_list.h"
-#include "radeon_dri.h"
-#include "radeon_lock.h"
-
-#include "xf86drm.h"
-#include "drm.h"
-#if 0
-#include "r200context.h"
-#endif
-#include "radeon_drm.h"
-
-extern void radeonCopyBuffer(__DRIdrawablePrivate * drawable,
-			     const drm_clip_rect_t	* rect);
-extern void radeonPageFlip(__DRIdrawablePrivate * drawable);
-extern void radeonFlush(GLcontext * ctx);
-extern void radeonFinish(GLcontext * ctx);
-extern void radeonWaitForIdleLocked(radeonContextPtr radeon);
-extern uint32_t radeonGetAge(radeonContextPtr radeon);
-
-#endif				/* __RADEON_IOCTL_H__ */
diff --git a/src/mesa/drivers/dri/r300/radeon_lock.c b/src/mesa/drivers/dri/r300/radeon_lock.c
deleted file mode 100644
index 4f47afd5dc6..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_lock.c
+++ /dev/null
@@ -1,137 +0,0 @@
-/**************************************************************************
-
-Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
-                     VA Linux Systems Inc., Fremont, California.
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Gareth Hughes <[email protected]>
- *   Keith Whitwell <[email protected]>
- *   Kevin E. Martin <[email protected]>
- */
-
-#include "radeon_lock.h"
-#include "radeon_ioctl.h"
-#include "radeon_state.h"
-#include "r300_context.h"
-#include "r300_state.h"
-
-#include "main/framebuffer.h"
-
-#include "drirenderbuffer.h"
-
-#if DEBUG_LOCKING
-char *prevLockFile = NULL;
-int prevLockLine = 0;
-#endif
-
-/* Turn on/off page flipping according to the flags in the sarea:
- */
-void radeonUpdatePageFlipping(radeonContextPtr rmesa)
-{
-	int use_back;
-
-	rmesa->doPageFlip = rmesa->sarea->pfState;
-	if (rmesa->glCtx->WinSysDrawBuffer) {
-		driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer,
-				     rmesa->sarea->pfCurrentPage);
-		r300UpdateDrawBuffer(rmesa->glCtx);
-	}
-
-	use_back = rmesa->glCtx->DrawBuffer ?
-	    (rmesa->glCtx->DrawBuffer->_ColorDrawBufferIndexes[0] ==
-	     BUFFER_BACK_LEFT) : 1;
-	use_back ^= (rmesa->sarea->pfCurrentPage == 1);
-
-	if (use_back) {
-		rmesa->state.color.drawOffset =
-		    rmesa->radeonScreen->backOffset;
-		rmesa->state.color.drawPitch = rmesa->radeonScreen->backPitch;
-	} else {
-		rmesa->state.color.drawOffset =
-		    rmesa->radeonScreen->frontOffset;
-		rmesa->state.color.drawPitch =
-		    rmesa->radeonScreen->frontPitch;
-	}
-}
-
-/* Update the hardware state.  This is called if another context has
- * grabbed the hardware lock, which includes the X server.  This
- * function also updates the driver's window state after the X server
- * moves, resizes or restacks a window -- the change will be reflected
- * in the drawable position and clip rects.  Since the X server grabs
- * the hardware lock when it changes the window state, this routine will
- * automatically be called after such a change.
- */
-void radeonGetLock(radeonContextPtr rmesa, GLuint flags)
-{
-	__DRIdrawablePrivate *const drawable = rmesa->dri.drawable;
-	__DRIdrawablePrivate *const readable = rmesa->dri.readable;
-	__DRIscreenPrivate *sPriv = rmesa->dri.screen;
-	drm_radeon_sarea_t *sarea = rmesa->sarea;
-	r300ContextPtr r300 = (r300ContextPtr) rmesa;
-
-	assert(drawable != NULL);
-
-	drmGetLock(rmesa->dri.fd, rmesa->dri.hwContext, flags);
-
-	/* The window might have moved, so we might need to get new clip
-	 * rects.
-	 *
-	 * NOTE: This releases and regrabs the hw lock to allow the X server
-	 * to respond to the DRI protocol request for new drawable info.
-	 * Since the hardware state depends on having the latest drawable
-	 * clip rects, all state checking must be done _after_ this call.
-	 */
-	DRI_VALIDATE_DRAWABLE_INFO(sPriv, drawable);
-	if (drawable != readable) {
-		DRI_VALIDATE_DRAWABLE_INFO(sPriv, readable);
-	}
-
-	if (rmesa->lastStamp != drawable->lastStamp) {
-		radeonUpdatePageFlipping(rmesa);
-		radeonSetCliprects(rmesa);
-		r300UpdateViewportOffset(rmesa->glCtx);
-		driUpdateFramebufferSize(rmesa->glCtx, drawable);
-	}
-
-	if (sarea->ctx_owner != rmesa->dri.hwContext) {
-		int i;
-
-		sarea->ctx_owner = rmesa->dri.hwContext;
-		for (i = 0; i < r300->nr_heaps; i++) {
-			DRI_AGE_TEXTURES(r300->texture_heaps[i]);
-		}
-	}
-
-	rmesa->lost_context = GL_TRUE;
-}
diff --git a/src/mesa/drivers/dri/r300/radeon_lock.h b/src/mesa/drivers/dri/r300/radeon_lock.h
deleted file mode 100644
index a344837f475..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_lock.h
+++ /dev/null
@@ -1,115 +0,0 @@
-/**************************************************************************
-
-Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
-                     VA Linux Systems Inc., Fremont, California.
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Gareth Hughes <[email protected]>
- *   Keith Whitwell <[email protected]>
- *   Kevin E. Martin <[email protected]>
- */
-
-#ifndef __RADEON_LOCK_H__
-#define __RADEON_LOCK_H__
-
-#include "radeon_context.h"
-
-extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags);
-extern void radeonUpdatePageFlipping(radeonContextPtr rmesa);
-
-/* Turn DEBUG_LOCKING on to find locking conflicts.
- */
-#define DEBUG_LOCKING	0
-
-#if DEBUG_LOCKING
-extern char *prevLockFile;
-extern int prevLockLine;
-
-#define DEBUG_LOCK()							\
-   do {									\
-      prevLockFile = (__FILE__);					\
-      prevLockLine = (__LINE__);					\
-   } while (0)
-
-#define DEBUG_RESET()							\
-   do {									\
-      prevLockFile = 0;							\
-      prevLockLine = 0;							\
-   } while (0)
-
-#define DEBUG_CHECK_LOCK()						\
-   do {									\
-      if (prevLockFile) {						\
-	 fprintf(stderr,						\
-		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
-		  prevLockFile, prevLockLine, __FILE__, __LINE__);	\
-	 exit(1);							\
-      }									\
-   } while (0)
-
-#else
-
-#define DEBUG_LOCK()
-#define DEBUG_RESET()
-#define DEBUG_CHECK_LOCK()
-
-#endif
-
-/*
- * !!! We may want to separate locks from locks with validation.  This
- * could be used to improve performance for those things commands that
- * do not do any drawing !!!
- */
-
-/* Lock the hardware and validate our state.
- */
-#define LOCK_HARDWARE( rmesa )						\
-	do {								\
-		char __ret = 0;						\
-		DEBUG_CHECK_LOCK();					\
-		DRM_CAS((rmesa)->dri.hwLock, (rmesa)->dri.hwContext,	\
-			(DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret); \
-		if (__ret)						\
-			radeonGetLock((rmesa), 0);			\
-		DEBUG_LOCK();						\
-	} while (0)
-
-#define UNLOCK_HARDWARE( rmesa )					\
-	do {								\
-		DRM_UNLOCK((rmesa)->dri.fd,				\
-			(rmesa)->dri.hwLock,				\
-			(rmesa)->dri.hwContext);			\
-		DEBUG_RESET();						\
-	} while (0)
-
-#endif				/* __RADEON_LOCK_H__ */
diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.c b/src/mesa/drivers/dri/r300/radeon_program_pair.c
index 2e21f7bf666..906d36e5226 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_pair.c
+++ b/src/mesa/drivers/dri/r300/radeon_program_pair.c
@@ -35,7 +35,7 @@
 
 #include "radeon_program_pair.h"
 
-#include "radeon_context.h"
+#include "radeon_common.h"
 
 #include "shader/prog_print.h"
 
@@ -47,7 +47,6 @@
 
 struct pair_state_instruction {
 	GLuint IsTex:1; /**< Is a texture instruction */
-	GLuint IsOutput:1; /**< Is output instruction */
 	GLuint NeedRGB:1; /**< Needs the RGB ALU */
 	GLuint NeedAlpha:1; /**< Needs the Alpha ALU */
 	GLuint IsTranscendent:1; /**< Is a special transcendent instruction */
@@ -124,7 +123,6 @@ struct pair_state {
 	GLboolean Debug;
 	GLboolean Verbose;
 	void *UserData;
-	GLubyte NumKillInsts;
 
 	/**
 	 * Translate Mesa registers to hardware registers
@@ -151,11 +149,6 @@ struct pair_state {
 	struct pair_state_instruction *ReadyTEX;
 
 	/**
-	 * Linked list of deferred instructions
-	 */
-	struct pair_state_instruction *DeferredInsts;
-
-	/**
 	 * Pool of @ref reg_value structures for fast allocation.
 	 */
 	struct reg_value *ValuePool;
@@ -238,9 +231,7 @@ static void instruction_ready(struct pair_state *s, int ip)
 	if (s->Verbose)
 		_mesa_printf("instruction_ready(%i)\n", ip);
 
-	if (s->NumKillInsts > 0 && pairinst->IsOutput)
-		add_pairinst_to_list(&s->DeferredInsts, pairinst);
-	else if (pairinst->IsTex)
+	if (pairinst->IsTex)
 		add_pairinst_to_list(&s->ReadyTEX, pairinst);
 	else if (!pairinst->NeedAlpha)
 		add_pairinst_to_list(&s->ReadyRGB, pairinst);
@@ -348,8 +339,6 @@ static void classify_instruction(struct pair_state *s,
 		error("Unknown opcode %d\n", inst->Opcode);
 		break;
 	}
-
-	pairinst->IsOutput = (inst->DstReg.File == PROGRAM_OUTPUT);
 }
 
 
@@ -613,16 +602,14 @@ static void emit_all_tex(struct pair_state *s)
 		struct prog_instruction *inst = s->Program->Instructions + ip;
 		commit_instruction(s, ip);
 
-		if (inst->Opcode == OPCODE_KIL)
-			--s->NumKillInsts;
-		else
+		if (inst->Opcode != OPCODE_KIL)
 			inst->DstReg.Index = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
-
 		inst->SrcReg[0].Index = get_hw_reg(s, inst->SrcReg[0].File, inst->SrcReg[0].Index);
 
 		if (s->Debug) {
 			_mesa_printf("   ");
 			_mesa_print_instruction(inst);
+			fflush(stdout);
 		}
 		s->Error = s->Error || !s->Handler->EmitTex(s->UserData, inst);
 	}
@@ -875,17 +862,6 @@ static void emit_alu(struct pair_state *s)
 	s->Error = s->Error || !s->Handler->EmitPaired(s->UserData, &pair);
 }
 
-static GLubyte countKillInsts(struct gl_program *prog)
-{
-	GLubyte i, count = 0;
-
-	for (i = 0; i < prog->NumInstructions; ++i) {
-		if (prog->Instructions[i].Opcode == OPCODE_KIL)
-			++count;
-	}
-
-	return count;
-}
 
 GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program,
 	const struct radeon_pair_handler* handler, void *userdata)
@@ -899,7 +875,6 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program,
 	s.UserData = userdata;
 	s.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE;
 	s.Verbose = GL_FALSE && s.Debug;
-	s.NumKillInsts = countKillInsts(program);
 
 	s.Instructions = (struct pair_state_instruction*)_mesa_calloc(
 		sizeof(struct pair_state_instruction)*s.Program->NumInstructions);
@@ -918,21 +893,6 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program,
 		if (s.ReadyTEX)
 			emit_all_tex(&s);
 
-		if (!s.NumKillInsts) {
-			struct pair_state_instruction *pairinst = s.DeferredInsts;
-			while (pairinst) {
-				if (!pairinst->NeedAlpha)
-					add_pairinst_to_list(&s.ReadyRGB, pairinst);
-				else if (!pairinst->NeedRGB)
-					add_pairinst_to_list(&s.ReadyAlpha, pairinst);
-				else
-					add_pairinst_to_list(&s.ReadyFullALU, pairinst);
-
-				pairinst = pairinst->NextReady;
-			}
-			s.DeferredInsts = NULL;
-		}
-
 		while(s.ReadyFullALU || s.ReadyRGB || s.ReadyAlpha)
 			emit_alu(&s);
 	}
diff --git a/src/mesa/drivers/dri/r300/radeon_span.c b/src/mesa/drivers/dri/r300/radeon_span.c
deleted file mode 100644
index 16f9fb99e67..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_span.c
+++ /dev/null
@@ -1,349 +0,0 @@
-/**************************************************************************
-
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
-                     VA Linux Systems Inc., Fremont, California.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Kevin E. Martin <[email protected]>
- *   Gareth Hughes <[email protected]>
- *   Keith Whitwell <[email protected]>
- *
- */
-
-#include "main/glheader.h"
-#include "swrast/swrast.h"
-
-#include "r300_state.h"
-#include "radeon_ioctl.h"
-#include "r300_ioctl.h"
-#include "radeon_span.h"
-
-#include "drirenderbuffer.h"
-
-#define DBG 0
-
-/*
- * Note that all information needed to access pixels in a renderbuffer
- * should be obtained through the gl_renderbuffer parameter, not per-context
- * information.
- */
-#define LOCAL_VARS						\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   const __DRIdrawablePrivate *dPriv = drb->dPriv;		\
-   const GLuint bottom = dPriv->h - 1;				\
-   GLubyte *buf = (GLubyte *) drb->flippedData			\
-      + (dPriv->y * drb->flippedPitch + dPriv->x) * drb->cpp;	\
-   GLuint p;							\
-   (void) p;
-
-#define LOCAL_DEPTH_VARS				\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;	\
-   const __DRIdrawablePrivate *dPriv = drb->dPriv;	\
-   const GLuint bottom = dPriv->h - 1;			\
-   GLuint xo = dPriv->x;				\
-   GLuint yo = dPriv->y;				\
-   GLubyte *buf = (GLubyte *) drb->Base.Data;
-
-#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
-
-#define Y_FLIP(Y) (bottom - (Y))
-
-#define HW_LOCK()
-
-#define HW_UNLOCK()
-
-/* ================================================================
- * Color buffer
- */
-
-/* 16 bit, RGB565 color spanline and pixel functions
- */
-#define SPANTMP_PIXEL_FMT GL_RGB
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
-
-#define TAG(x)    radeon##x##_RGB565
-#define TAG2(x,y) radeon##x##_RGB565##y
-#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 2)
-#include "spantmp2.h"
-
-/* 32 bit, ARGB8888 color spanline and pixel functions
- */
-#define SPANTMP_PIXEL_FMT GL_BGRA
-#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
-
-#define TAG(x)    radeon##x##_ARGB8888
-#define TAG2(x,y) radeon##x##_ARGB8888##y
-#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 4)
-#include "spantmp2.h"
-
-/* ================================================================
- * Depth buffer
- */
-
-/* The Radeon family has depth tiling on all the time, so we have to convert
- * the x,y coordinates into the memory bus address (mba) in the same
- * manner as the engine.  In each case, the linear block address (ba)
- * is calculated, and then wired with x and y to produce the final
- * memory address.
- * The chip will do address translation on its own if the surface registers
- * are set up correctly. It is not quite enough to get it working with hyperz
- * too...
- */
-
-static GLuint radeon_mba_z32(const driRenderbuffer * drb, GLint x, GLint y)
-{
-	GLuint pitch = drb->pitch;
-	if (drb->depthHasSurface) {
-		return 4 * (x + y * pitch);
-	} else {
-		GLuint ba, address = 0;	/* a[0..1] = 0           */
-
-#ifdef COMPILE_R300
-		ba = (y / 8) * (pitch / 8) + (x / 8);
-#else
-		ba = (y / 16) * (pitch / 16) + (x / 16);
-#endif
-
-		address |= (x & 0x7) << 2;	/* a[2..4] = x[0..2]     */
-		address |= (y & 0x3) << 5;	/* a[5..6] = y[0..1]     */
-		address |= (((x & 0x10) >> 2) ^ (y & 0x4)) << 5;	/* a[7]    = x[4] ^ y[2] */
-		address |= (ba & 0x3) << 8;	/* a[8..9] = ba[0..1]    */
-
-		address |= (y & 0x8) << 7;	/* a[10]   = y[3]        */
-		address |= (((x & 0x8) << 1) ^ (y & 0x10)) << 7;	/* a[11]   = x[3] ^ y[4] */
-		address |= (ba & ~0x3) << 10;	/* a[12..] = ba[2..]     */
-
-		return address;
-	}
-}
-
-static INLINE GLuint
-radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y)
-{
-	GLuint pitch = drb->pitch;
-	if (drb->depthHasSurface) {
-		return 2 * (x + y * pitch);
-	} else {
-		GLuint ba, address = 0;	/* a[0]    = 0           */
-
-		ba = (y / 16) * (pitch / 32) + (x / 32);
-
-		address |= (x & 0x7) << 1;	/* a[1..3] = x[0..2]     */
-		address |= (y & 0x7) << 4;	/* a[4..6] = y[0..2]     */
-		address |= (x & 0x8) << 4;	/* a[7]    = x[3]        */
-		address |= (ba & 0x3) << 8;	/* a[8..9] = ba[0..1]    */
-		address |= (y & 0x8) << 7;	/* a[10]   = y[3]        */
-		address |= ((x & 0x10) ^ (y & 0x10)) << 7;	/* a[11]   = x[4] ^ y[4] */
-		address |= (ba & ~0x3) << 10;	/* a[12..] = ba[2..]     */
-
-		return address;
-	}
-}
-
-/* 16-bit depth buffer functions
- */
-#define VALUE_TYPE GLushort
-
-#define WRITE_DEPTH( _x, _y, d )					\
-   *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )) = d;
-
-#define READ_DEPTH( d, _x, _y )						\
-   d = *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo ));
-
-#define TAG(x) radeon##x##_z16
-#include "depthtmp.h"
-
-/* 24 bit depth, 8 bit stencil depthbuffer functions
- *
- * Careful: It looks like the R300 uses ZZZS byte order while the R200
- * uses SZZZ for 24 bit depth, 8 bit stencil mode.
- */
-#define VALUE_TYPE GLuint
-
-#ifdef COMPILE_R300
-#define WRITE_DEPTH( _x, _y, d )					\
-do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0x000000ff;							\
-   tmp |= ((d << 8) & 0xffffff00);					\
-   *(GLuint *)(buf + offset) = tmp;					\
-} while (0)
-#else
-#define WRITE_DEPTH( _x, _y, d )					\
-do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0xff000000;							\
-   tmp |= ((d) & 0x00ffffff);						\
-   *(GLuint *)(buf + offset) = tmp;					\
-} while (0)
-#endif
-
-#ifdef COMPILE_R300
-#define READ_DEPTH( d, _x, _y )						\
-  do { \
-    d = (*(GLuint *)(buf + radeon_mba_z32( drb, _x + xo,		\
-					 _y + yo )) & 0xffffff00) >> 8; \
-  }while(0)
-#else
-#define READ_DEPTH( d, _x, _y )						\
-   d = *(GLuint *)(buf + radeon_mba_z32( drb, _x + xo,			\
-					 _y + yo )) & 0x00ffffff;
-#endif
-
-#define TAG(x) radeon##x##_z24_s8
-#include "depthtmp.h"
-
-/* ================================================================
- * Stencil buffer
- */
-
-/* 24 bit depth, 8 bit stencil depthbuffer functions
- */
-#ifdef COMPILE_R300
-#define WRITE_STENCIL( _x, _y, d )					\
-do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0xffffff00;							\
-   tmp |= (d) & 0xff;							\
-   *(GLuint *)(buf + offset) = tmp;					\
-} while (0)
-#else
-#define WRITE_STENCIL( _x, _y, d )					\
-do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   tmp &= 0x00ffffff;							\
-   tmp |= (((d) & 0xff) << 24);						\
-   *(GLuint *)(buf + offset) = tmp;					\
-} while (0)
-#endif
-
-#ifdef COMPILE_R300
-#define READ_STENCIL( d, _x, _y )					\
-do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   d = tmp & 0x000000ff;						\
-} while (0)
-#else
-#define READ_STENCIL( d, _x, _y )					\
-do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
-   d = (tmp & 0xff000000) >> 24;					\
-} while (0)
-#endif
-
-#define TAG(x) radeon##x##_z24_s8
-#include "stenciltmp.h"
-
-/* Move locking out to get reasonable span performance (10x better
- * than doing this in HW_LOCK above).  WaitForIdle() is the main
- * culprit.
- */
-
-static void radeonSpanRenderStart(GLcontext * ctx)
-{
-	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-#ifdef COMPILE_R300
-	r300ContextPtr r300 = (r300ContextPtr) rmesa;
-	R300_FIREVERTICES(r300);
-#else
-	RADEON_FIREVERTICES(rmesa);
-#endif
-	LOCK_HARDWARE(rmesa);
-	radeonWaitForIdleLocked(rmesa);
-
-	/* Read the first pixel in the frame buffer.  This should
-	 * be a noop, right?  In fact without this conform fails as reading
-	 * from the framebuffer sometimes produces old results -- the
-	 * on-card read cache gets mixed up and doesn't notice that the
-	 * framebuffer has been updated.
-	 *
-	 * Note that we should probably be reading some otherwise unused
-	 * region of VRAM, otherwise we might get incorrect results when
-	 * reading pixels from the top left of the screen.
-	 *
-	 * I found this problem on an R420 with glean's texCube test.
-	 * Note that the R200 span code also *writes* the first pixel in the
-	 * framebuffer, but I've found this to be unnecessary.
-	 *  -- Nicolai Hähnle, June 2008
-	 */
-	{
-		int p;
-		driRenderbuffer *drb =
-			(driRenderbuffer *) ctx->WinSysDrawBuffer->_ColorDrawBuffers[0];
-		volatile int *buf =
-			(volatile int *)(rmesa->dri.screen->pFB + drb->offset);
-		p = *buf;
-	}
-}
-
-static void radeonSpanRenderFinish(GLcontext * ctx)
-{
-	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-	_swrast_flush(ctx);
-	UNLOCK_HARDWARE(rmesa);
-}
-
-void radeonInitSpanFuncs(GLcontext * ctx)
-{
-	struct swrast_device_driver *swdd =
-	    _swrast_GetDeviceDriverReference(ctx);
-	swdd->SpanRenderStart = radeonSpanRenderStart;
-	swdd->SpanRenderFinish = radeonSpanRenderFinish;
-}
-
-/**
- * Plug in the Get/Put routines for the given driRenderbuffer.
- */
-void radeonSetSpanFunctions(driRenderbuffer * drb, const GLvisual * vis)
-{
-	if (drb->Base.InternalFormat == GL_RGBA) {
-		if (vis->redBits == 5 && vis->greenBits == 6
-		    && vis->blueBits == 5) {
-			radeonInitPointers_RGB565(&drb->Base);
-		} else {
-			radeonInitPointers_ARGB8888(&drb->Base);
-		}
-	} else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) {
-		radeonInitDepthPointers_z16(&drb->Base);
-	} else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) {
-		radeonInitDepthPointers_z24_s8(&drb->Base);
-	} else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) {
-		radeonInitStencilPointers_z24_s8(&drb->Base);
-	}
-}
diff --git a/src/mesa/drivers/dri/r300/radeon_state.c b/src/mesa/drivers/dri/r300/radeon_state.c
deleted file mode 100644
index c401da6c544..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_state.c
+++ /dev/null
@@ -1,244 +0,0 @@
-/**************************************************************************
-
-Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
-
-The Weather Channel (TM) funded Tungsten Graphics to develop the
-initial release of the Radeon 8500 driver under the XFree86 license.
-This notice must be preserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- */
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/api_arrayelt.h"
-#include "main/enums.h"
-#include "main/framebuffer.h"
-#include "main/colormac.h"
-#include "main/light.h"
-
-#include "swrast/swrast.h"
-#include "vbo/vbo.h"
-#include "tnl/tnl.h"
-#include "tnl/t_pipeline.h"
-#include "swrast_setup/swrast_setup.h"
-
-#include "radeon_ioctl.h"
-#include "radeon_state.h"
-#include "r300_ioctl.h"
-
-
-/* =============================================================
- * Scissoring
- */
-
-static GLboolean intersect_rect(drm_clip_rect_t * out,
-				drm_clip_rect_t * a, drm_clip_rect_t * b)
-{
-	*out = *a;
-	if (b->x1 > out->x1)
-		out->x1 = b->x1;
-	if (b->y1 > out->y1)
-		out->y1 = b->y1;
-	if (b->x2 < out->x2)
-		out->x2 = b->x2;
-	if (b->y2 < out->y2)
-		out->y2 = b->y2;
-	if (out->x1 >= out->x2)
-		return GL_FALSE;
-	if (out->y1 >= out->y2)
-		return GL_FALSE;
-	return GL_TRUE;
-}
-
-void radeonRecalcScissorRects(radeonContextPtr radeon)
-{
-	drm_clip_rect_t *out;
-	int i;
-
-	/* Grow cliprect store?
-	 */
-	if (radeon->state.scissor.numAllocedClipRects < radeon->numClipRects) {
-		while (radeon->state.scissor.numAllocedClipRects <
-		       radeon->numClipRects) {
-			radeon->state.scissor.numAllocedClipRects += 1;	/* zero case */
-			radeon->state.scissor.numAllocedClipRects *= 2;
-		}
-
-		if (radeon->state.scissor.pClipRects)
-			FREE(radeon->state.scissor.pClipRects);
-
-		radeon->state.scissor.pClipRects =
-		    MALLOC(radeon->state.scissor.numAllocedClipRects *
-			   sizeof(drm_clip_rect_t));
-
-		if (radeon->state.scissor.pClipRects == NULL) {
-			radeon->state.scissor.numAllocedClipRects = 0;
-			return;
-		}
-	}
-
-	out = radeon->state.scissor.pClipRects;
-	radeon->state.scissor.numClipRects = 0;
-
-	for (i = 0; i < radeon->numClipRects; i++) {
-		if (intersect_rect(out,
-				   &radeon->pClipRects[i],
-				   &radeon->state.scissor.rect)) {
-			radeon->state.scissor.numClipRects++;
-			out++;
-		}
-	}
-}
-
-void radeonUpdateScissor(GLcontext* ctx)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-
-	if (radeon->dri.drawable) {
-		__DRIdrawablePrivate *dPriv = radeon->dri.drawable;
-		int x1 = dPriv->x + ctx->Scissor.X;
-		int y1 = dPriv->y + dPriv->h - (ctx->Scissor.Y + ctx->Scissor.Height);
-
-		radeon->state.scissor.rect.x1 = x1;
-		radeon->state.scissor.rect.y1 = y1;
-		radeon->state.scissor.rect.x2 = x1 + ctx->Scissor.Width;
-		radeon->state.scissor.rect.y2 = y1 + ctx->Scissor.Height;
-
-		radeonRecalcScissorRects(radeon);
-	}
-}
-
-static void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h)
-{
-	if (ctx->Scissor.Enabled) {
-		/* We don't pipeline cliprect changes */
-		r300Flush(ctx);
-		radeonUpdateScissor(ctx);
-	}
-}
-
-
-/**
- * Update cliprects and scissors.
- */
-void radeonSetCliprects(radeonContextPtr radeon)
-{
-	__DRIdrawablePrivate *const drawable = radeon->dri.drawable;
-	__DRIdrawablePrivate *const readable = radeon->dri.readable;
-	GLframebuffer *const draw_fb = (GLframebuffer*)drawable->driverPrivate;
-	GLframebuffer *const read_fb = (GLframebuffer*)readable->driverPrivate;
-
-	if (draw_fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) {
-		/* Can't ignore 2d windows if we are page flipping. */
-		if (drawable->numBackClipRects == 0 || radeon->doPageFlip ||
-		    radeon->sarea->pfCurrentPage == 1) {
-			radeon->numClipRects = drawable->numClipRects;
-			radeon->pClipRects = drawable->pClipRects;
-		} else {
-			radeon->numClipRects = drawable->numBackClipRects;
-			radeon->pClipRects = drawable->pBackClipRects;
-		}
-	} else {
-		/* front buffer (or none, or multiple buffers */
-		radeon->numClipRects = drawable->numClipRects;
-		radeon->pClipRects = drawable->pClipRects;
-	}
-
-	if ((draw_fb->Width != drawable->w) ||
-	    (draw_fb->Height != drawable->h)) {
-		_mesa_resize_framebuffer(radeon->glCtx, draw_fb,
-					 drawable->w, drawable->h);
-		draw_fb->Initialized = GL_TRUE;
-	}
-
-	if (drawable != readable) {
-		if ((read_fb->Width != readable->w) ||
-		    (read_fb->Height != readable->h)) {
-			_mesa_resize_framebuffer(radeon->glCtx, read_fb,
-						 readable->w, readable->h);
-			read_fb->Initialized = GL_TRUE;
-		}
-	}
-
-	if (radeon->state.scissor.enabled)
-		radeonRecalcScissorRects(radeon);
-
-	radeon->lastStamp = drawable->lastStamp;
-}
-
-
-/**
- * Handle common enable bits.
- * Called as a fallback by r200Enable/r300Enable.
- */
-void radeonEnable(GLcontext* ctx, GLenum cap, GLboolean state)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-
-	switch(cap) {
-	case GL_SCISSOR_TEST:
-		/* We don't pipeline cliprect & scissor changes */
-		r300Flush(ctx);
-
-		radeon->state.scissor.enabled = state;
-		radeonUpdateScissor(ctx);
-		break;
-
-	default:
-		return;
-	}
-}
-
-
-/**
- * Initialize default state.
- * This function is called once at context init time from
- * r200InitState/r300InitState
- */
-void radeonInitState(radeonContextPtr radeon)
-{
-	radeon->Fallback = 0;
-
-	if (radeon->glCtx->Visual.doubleBufferMode && radeon->sarea->pfCurrentPage == 0) {
-		radeon->state.color.drawOffset = radeon->radeonScreen->backOffset;
-		radeon->state.color.drawPitch = radeon->radeonScreen->backPitch;
-	} else {
-		radeon->state.color.drawOffset = radeon->radeonScreen->frontOffset;
-		radeon->state.color.drawPitch = radeon->radeonScreen->frontPitch;
-	}
-}
-
-
-/**
- * Initialize common state functions.
- * Called by r200InitStateFuncs/r300InitStateFuncs
- */
-void radeonInitStateFuncs(struct dd_function_table *functions)
-{
-	functions->Scissor = radeonScissor;
-}
diff --git a/src/mesa/drivers/dri/r300/radeon_state.h b/src/mesa/drivers/dri/r300/radeon_state.h
deleted file mode 100644
index 821cb40c7eb..00000000000
--- a/src/mesa/drivers/dri/r300/radeon_state.h
+++ /dev/null
@@ -1,43 +0,0 @@
-/*
-Copyright (C) 2004 Nicolai Haehnle.  All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Nicolai Haehnle <[email protected]>
- */
-
-#ifndef __RADEON_STATE_H__
-#define __RADEON_STATE_H__
-
-extern void radeonRecalcScissorRects(radeonContextPtr radeon);
-extern void radeonSetCliprects(radeonContextPtr radeon);
-extern void radeonUpdateScissor(GLcontext* ctx);
-
-extern void radeonEnable(GLcontext* ctx, GLenum cap, GLboolean state);
-
-extern void radeonInitState(radeonContextPtr radeon);
-extern void radeonInitStateFuncs(struct dd_function_table* functions);
-
-#endif
diff --git a/src/mesa/drivers/dri/radeon/Makefile b/src/mesa/drivers/dri/radeon/Makefile
index f223b2d9228..ba409ba8130 100644
--- a/src/mesa/drivers/dri/radeon/Makefile
+++ b/src/mesa/drivers/dri/radeon/Makefile
@@ -4,25 +4,37 @@
 TOP = ../../../../..
 include $(TOP)/configs/current
 
+CFLAGS += $(RADEON_CFLAGS)
+
 LIBNAME = radeon_dri.so
 
 MINIGLX_SOURCES = server/radeon_dri.c 
 
+RADEON_COMMON_SOURCES = \
+	radeon_texture.c \
+	radeon_common_context.c \
+	radeon_common.c \
+	radeon_dma.c \
+	radeon_lock.c \
+	radeon_bo_legacy.c \
+	radeon_cs_legacy.c \
+	radeon_mipmap_tree.c \
+	radeon_span.c \
+	radeon_fbo.c
+
 DRIVER_SOURCES = \
 	radeon_context.c \
 	radeon_ioctl.c \
-	radeon_lock.c \
 	radeon_screen.c \
 	radeon_state.c \
 	radeon_state_init.c \
 	radeon_tex.c \
-	radeon_texmem.c \
 	radeon_texstate.c \
 	radeon_tcl.c \
 	radeon_swtcl.c \
-	radeon_span.c \
 	radeon_maos.c \
-	radeon_sanity.c 
+	radeon_sanity.c \
+	$(RADEON_COMMON_SOURCES)
 
 C_SOURCES = \
 	$(COMMON_SOURCES) \
@@ -30,6 +42,8 @@ C_SOURCES = \
 
 DRIVER_DEFINES = -DRADEON_COMMON=0
 
+DRI_LIB_DEPS += $(RADEON_LDFLAGS)
+
 X86_SOURCES = 
 
 include ../Makefile.template
diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_drm.h b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h
new file mode 100644
index 00000000000..1ed13f17955
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h
@@ -0,0 +1,182 @@
+/* 
+ * Copyright © 2008 Jérôme Glisse
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
+ * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ */
+/*
+ * Authors:
+ *      Jérôme Glisse <[email protected]>
+ */
+#ifndef RADEON_BO_H
+#define RADEON_BO_H
+
+#include <stdio.h>
+#include <stdint.h>
+//#include "radeon_track.h"
+
+/* bo object */
+#define RADEON_BO_FLAGS_MACRO_TILE  1
+#define RADEON_BO_FLAGS_MICRO_TILE  2
+
+struct radeon_bo_manager;
+
+struct radeon_bo {
+    uint32_t                    alignment;
+    uint32_t                    handle;
+    uint32_t                    size;
+    uint32_t                    domains;
+    uint32_t                    flags;
+    unsigned                    cref;
+#ifdef RADEON_BO_TRACK
+    struct radeon_track         *track;
+#endif
+    void                        *ptr;
+    struct radeon_bo_manager    *bom;
+    uint32_t                    space_accounted;
+};
+
+/* bo functions */
+struct radeon_bo_funcs {
+    struct radeon_bo *(*bo_open)(struct radeon_bo_manager *bom,
+                                 uint32_t handle,
+                                 uint32_t size,
+                                 uint32_t alignment,
+                                 uint32_t domains,
+                                 uint32_t flags);
+    void (*bo_ref)(struct radeon_bo *bo);
+    struct radeon_bo *(*bo_unref)(struct radeon_bo *bo);
+    int (*bo_map)(struct radeon_bo *bo, int write);
+    int (*bo_unmap)(struct radeon_bo *bo);
+    int (*bo_wait)(struct radeon_bo *bo);
+};
+
+struct radeon_bo_manager {
+    struct radeon_bo_funcs  *funcs;
+    int                     fd;
+
+#ifdef RADEON_BO_TRACK
+    struct radeon_tracker   tracker;
+#endif
+};
+    
+static inline void _radeon_bo_debug(struct radeon_bo *bo,
+                                    const char *op,
+                                    const char *file,
+                                    const char *func,
+                                    int line)
+{
+    fprintf(stderr, "%s %p 0x%08X 0x%08X 0x%08X [%s %s %d]\n",
+            op, bo, bo->handle, bo->size, bo->cref, file, func, line);
+}
+
+static inline struct radeon_bo *_radeon_bo_open(struct radeon_bo_manager *bom,
+                                                uint32_t handle,
+                                                uint32_t size,
+                                                uint32_t alignment,
+                                                uint32_t domains,
+                                                uint32_t flags,
+                                                const char *file,
+                                                const char *func,
+                                                int line)
+{
+    struct radeon_bo *bo;
+
+    bo = bom->funcs->bo_open(bom, handle, size, alignment, domains, flags);
+#ifdef RADEON_BO_TRACK
+    if (bo) {
+        bo->track = radeon_tracker_add_track(&bom->tracker, bo->handle);
+        radeon_track_add_event(bo->track, file, func, "open", line);
+    }
+#endif
+    return bo;
+}
+
+static inline void _radeon_bo_ref(struct radeon_bo *bo,
+                                  const char *file,
+                                  const char *func,
+                                  int line)
+{
+    bo->cref++;
+#ifdef RADEON_BO_TRACK
+    radeon_track_add_event(bo->track, file, func, "ref", line); 
+#endif
+    bo->bom->funcs->bo_ref(bo);
+}
+
+static inline struct radeon_bo *_radeon_bo_unref(struct radeon_bo *bo,
+                                                 const char *file,
+                                                 const char *func,
+                                                 int line)
+{
+    bo->cref--;
+#ifdef RADEON_BO_TRACK
+    radeon_track_add_event(bo->track, file, func, "unref", line);
+    if (bo->cref <= 0) {
+        radeon_tracker_remove_track(&bo->bom->tracker, bo->track);
+        bo->track = NULL;
+    }
+#endif
+    return bo->bom->funcs->bo_unref(bo);
+}
+
+static inline int _radeon_bo_map(struct radeon_bo *bo,
+                                 int write,
+                                 const char *file,
+                                 const char *func,
+                                 int line)
+{
+    return bo->bom->funcs->bo_map(bo, write);
+}
+
+static inline int _radeon_bo_unmap(struct radeon_bo *bo,
+                                   const char *file,
+                                   const char *func,
+                                   int line)
+{
+    return bo->bom->funcs->bo_unmap(bo);
+}
+
+static inline int _radeon_bo_wait(struct radeon_bo *bo,
+                                  const char *file,
+                                  const char *func,
+                                  int line)
+{
+    return bo->bom->funcs->bo_wait(bo);
+}
+
+#define radeon_bo_open(bom, h, s, a, d, f)\
+    _radeon_bo_open(bom, h, s, a, d, f, __FILE__, __FUNCTION__, __LINE__)
+#define radeon_bo_ref(bo)\
+    _radeon_bo_ref(bo, __FILE__, __FUNCTION__, __LINE__)
+#define radeon_bo_unref(bo)\
+    _radeon_bo_unref(bo, __FILE__, __FUNCTION__, __LINE__)
+#define radeon_bo_map(bo, w)\
+    _radeon_bo_map(bo, w, __FILE__, __FUNCTION__, __LINE__)
+#define radeon_bo_unmap(bo)\
+    _radeon_bo_unmap(bo, __FILE__, __FUNCTION__, __LINE__)
+#define radeon_bo_debug(bo, opcode)\
+    _radeon_bo_debug(bo, opcode, __FILE__, __FUNCTION__, __LINE__)
+#define radeon_bo_wait(bo) \
+    _radeon_bo_wait(bo, __FILE__, __func__, __LINE__)
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c
new file mode 100644
index 00000000000..6a8da402b1b
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c
@@ -0,0 +1,830 @@
+/* 
+ * Copyright © 2008 Nicolai Haehnle
+ * Copyright © 2008 Dave Airlie
+ * Copyright © 2008 Jérôme Glisse
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ */
+/*
+ * Authors:
+ *      Aapo Tahkola <[email protected]>
+ *      Nicolai Haehnle <[email protected]>
+ *      Dave Airlie
+ *      Jérôme Glisse <[email protected]>
+ */
+#include <stdio.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+#include <unistd.h>
+#include <sys/mman.h>
+#include <sys/ioctl.h>
+#include "xf86drm.h"
+#include "texmem.h"
+#include "main/simple_list.h"
+
+#include "drm.h"
+#include "radeon_drm.h"
+#include "radeon_common.h"
+#include "radeon_bocs_wrapper.h"
+#include "radeon_macros.h"
+
+/* no seriously texmem.c is this screwed up */
+struct bo_legacy_texture_object {
+    driTextureObject    base;
+    struct bo_legacy *parent;
+};
+
+struct bo_legacy {
+    struct radeon_bo    base;
+    int                 map_count;
+    uint32_t            pending;
+    int                 is_pending;
+    int                 static_bo;
+    uint32_t            offset;
+    struct bo_legacy_texture_object *tobj;
+    int                 validated;
+    int                 dirty;
+    void                *ptr;
+    struct bo_legacy    *next, *prev;
+    struct bo_legacy    *pnext, *pprev;
+};
+
+struct bo_manager_legacy {
+    struct radeon_bo_manager    base;
+    unsigned                    nhandle;
+    unsigned                    nfree_handles;
+    unsigned                    cfree_handles;
+    uint32_t                    current_age;
+    struct bo_legacy            bos;
+    struct bo_legacy            pending_bos;
+    uint32_t                    fb_location;
+    uint32_t                    texture_offset;
+    unsigned                    dma_alloc_size;
+    uint32_t                    dma_buf_count;
+    unsigned                    cpendings;
+    driTextureObject            texture_swapped;
+    driTexHeap                  *texture_heap;
+    struct radeon_screen        *screen;
+    unsigned                    *free_handles;
+};
+
+static void bo_legacy_tobj_destroy(void *data, driTextureObject *t)
+{
+    struct bo_legacy_texture_object *tobj = (struct bo_legacy_texture_object *)t;
+    
+    if (tobj->parent) {
+        tobj->parent->tobj = NULL;
+        tobj->parent->validated = 0;
+    }
+}
+
+static void inline clean_handles(struct bo_manager_legacy *bom)
+{
+  while (bom->cfree_handles > 0 &&
+	 !bom->free_handles[bom->cfree_handles - 1])
+    bom->cfree_handles--;
+
+}
+static int legacy_new_handle(struct bo_manager_legacy *bom, uint32_t *handle)
+{
+    uint32_t tmp;
+
+    *handle = 0;
+    if (bom->nhandle == 0xFFFFFFFF) {
+        return -EINVAL;
+    }
+    if (bom->cfree_handles > 0) {
+        tmp = bom->free_handles[--bom->cfree_handles];
+	clean_handles(bom);
+    } else {
+        bom->cfree_handles = 0;
+        tmp = bom->nhandle++;
+    }
+    assert(tmp);
+    *handle = tmp;
+    return 0;
+}
+
+static int legacy_free_handle(struct bo_manager_legacy *bom, uint32_t handle)
+{
+    uint32_t *handles;
+
+    if (!handle) {
+        return 0;
+    }
+    if (handle == (bom->nhandle - 1)) {
+        int i;
+
+        bom->nhandle--;
+        for (i = bom->cfree_handles - 1; i >= 0; i--) {
+            if (bom->free_handles[i] == (bom->nhandle - 1)) {
+                bom->nhandle--;
+                bom->free_handles[i] = 0;
+            }
+        }
+        clean_handles(bom);
+        return 0;
+    }
+    if (bom->cfree_handles < bom->nfree_handles) {
+        bom->free_handles[bom->cfree_handles++] = handle;
+        return 0;
+    }
+    bom->nfree_handles += 0x100;
+    handles = (uint32_t*)realloc(bom->free_handles, bom->nfree_handles * 4);
+    if (handles == NULL) {
+        bom->nfree_handles -= 0x100;
+        return -ENOMEM;
+    }
+    bom->free_handles = handles;
+    bom->free_handles[bom->cfree_handles++] = handle;
+    return 0;
+}
+
+static void legacy_get_current_age(struct bo_manager_legacy *boml)
+{
+    drm_radeon_getparam_t gp;
+    unsigned char *RADEONMMIO = NULL;
+    int r;
+
+    if (IS_R300_CLASS(boml->screen)) {
+    	gp.param = RADEON_PARAM_LAST_CLEAR;
+    	gp.value = (int *)&boml->current_age;
+    	r = drmCommandWriteRead(boml->base.fd, DRM_RADEON_GETPARAM,
+       	                     &gp, sizeof(gp));
+    	if (r) {
+       	 fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, r);
+         exit(1);
+       }
+    } else {
+        RADEONMMIO = boml->screen->mmio.map;
+        boml->current_age = boml->screen->scratch[3];
+        boml->current_age = INREG(RADEON_GUI_SCRATCH_REG3);
+    }
+}
+
+static int legacy_is_pending(struct radeon_bo *bo)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+
+    if (bo_legacy->is_pending <= 0) {
+        bo_legacy->is_pending = 0;
+        return 0;
+    }
+    if (boml->current_age >= bo_legacy->pending) {
+        if (boml->pending_bos.pprev == bo_legacy) {
+            boml->pending_bos.pprev = bo_legacy->pprev;
+        }
+        bo_legacy->pprev->pnext = bo_legacy->pnext;
+        if (bo_legacy->pnext) {
+            bo_legacy->pnext->pprev = bo_legacy->pprev;
+        }
+	assert(bo_legacy->is_pending <= bo->cref);
+        while (bo_legacy->is_pending--) {
+	    bo = radeon_bo_unref(bo);
+	    if (!bo)
+	      break;
+        }
+	if (bo)
+	  bo_legacy->is_pending = 0;
+        boml->cpendings--;
+        return 0;
+    }
+    return 1;
+}
+
+static int legacy_wait_pending(struct radeon_bo *bo)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+
+    if (!bo_legacy->is_pending) {
+        return 0;
+    }
+    /* FIXME: lockup and userspace busy looping that's all the folks */
+    legacy_get_current_age(boml);
+    while (legacy_is_pending(bo)) {
+        usleep(10);
+        legacy_get_current_age(boml);
+    }
+    return 0;
+}
+
+static void legacy_track_pending(struct bo_manager_legacy *boml, int debug)
+{
+    struct bo_legacy *bo_legacy;
+    struct bo_legacy *next;
+
+    legacy_get_current_age(boml);
+    bo_legacy = boml->pending_bos.pnext;
+    while (bo_legacy) {
+        if (debug)
+	  fprintf(stderr,"pending %p %d %d %d\n", bo_legacy, bo_legacy->base.size,
+		  boml->current_age, bo_legacy->pending);
+        next = bo_legacy->pnext;
+        if (legacy_is_pending(&(bo_legacy->base))) {
+        }
+        bo_legacy = next;
+    } 
+}
+
+static int legacy_wait_any_pending(struct bo_manager_legacy *boml)
+{
+    struct bo_legacy *bo_legacy;
+
+    legacy_get_current_age(boml);
+    bo_legacy = boml->pending_bos.pnext;
+    if (!bo_legacy)
+      return -1;
+    legacy_wait_pending(&bo_legacy->base);
+    return 0;
+}
+
+static void legacy_kick_all_buffers(struct bo_manager_legacy *boml)
+{
+    struct bo_legacy *legacy;
+
+    legacy = boml->bos.next;
+    while (legacy != &boml->bos) {
+	if (legacy->tobj) {
+	    if (legacy->validated) {
+		driDestroyTextureObject(&legacy->tobj->base);
+		legacy->tobj = 0;
+		legacy->validated = 0;
+	    }
+	}
+	legacy = legacy->next;
+    }
+}
+
+static struct bo_legacy *bo_allocate(struct bo_manager_legacy *boml,
+                                     uint32_t size,
+                                     uint32_t alignment,
+                                     uint32_t domains,
+                                     uint32_t flags)
+{
+    struct bo_legacy *bo_legacy;
+    static int pgsize;
+
+    if (pgsize == 0)
+        pgsize = getpagesize() - 1;
+
+    size = (size + pgsize) & ~pgsize;
+
+    bo_legacy = (struct bo_legacy*)calloc(1, sizeof(struct bo_legacy));
+    if (bo_legacy == NULL) {
+        return NULL;
+    }
+    bo_legacy->base.bom = (struct radeon_bo_manager*)boml;
+    bo_legacy->base.handle = 0;
+    bo_legacy->base.size = size;
+    bo_legacy->base.alignment = alignment;
+    bo_legacy->base.domains = domains;
+    bo_legacy->base.flags = flags;
+    bo_legacy->base.ptr = NULL;
+    bo_legacy->map_count = 0;
+    bo_legacy->next = NULL;
+    bo_legacy->prev = NULL;
+    bo_legacy->pnext = NULL;
+    bo_legacy->pprev = NULL;
+    bo_legacy->next = boml->bos.next;
+    bo_legacy->prev = &boml->bos;
+    boml->bos.next = bo_legacy;
+    if (bo_legacy->next) {
+        bo_legacy->next->prev = bo_legacy;
+    }
+    return bo_legacy;
+}
+
+static int bo_dma_alloc(struct radeon_bo *bo)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    drm_radeon_mem_alloc_t alloc;
+    unsigned size;
+    int base_offset;
+    int r;
+
+    /* align size on 4Kb */
+    size = (((4 * 1024) - 1) + bo->size) & ~((4 * 1024) - 1);
+    alloc.region = RADEON_MEM_REGION_GART;
+    alloc.alignment = bo_legacy->base.alignment;
+    alloc.size = size;
+    alloc.region_offset = &base_offset;
+    r = drmCommandWriteRead(bo->bom->fd,
+                            DRM_RADEON_ALLOC,
+                            &alloc,
+                            sizeof(alloc));
+    if (r) {
+        /* ptr is set to NULL if dma allocation failed */
+        bo_legacy->ptr = NULL;
+        return r;
+    }
+    bo_legacy->ptr = boml->screen->gartTextures.map + base_offset;
+    bo_legacy->offset = boml->screen->gart_texture_offset + base_offset;
+    bo->size = size;
+    boml->dma_alloc_size += size;
+    boml->dma_buf_count++;
+    return 0;
+}
+
+static int bo_dma_free(struct radeon_bo *bo)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    drm_radeon_mem_free_t memfree;
+    int r;
+
+    if (bo_legacy->ptr == NULL) {
+        /* ptr is set to NULL if dma allocation failed */
+        return 0;
+    }
+    legacy_get_current_age(boml);
+    memfree.region = RADEON_MEM_REGION_GART;
+    memfree.region_offset  = bo_legacy->offset;
+    memfree.region_offset -= boml->screen->gart_texture_offset;
+    r = drmCommandWrite(boml->base.fd,
+                        DRM_RADEON_FREE,
+                        &memfree,
+                        sizeof(memfree));
+    if (r) {
+        fprintf(stderr, "Failed to free bo[%p] at %08x\n",
+                &bo_legacy->base, memfree.region_offset);
+        fprintf(stderr, "ret = %s\n", strerror(-r));
+        return r;
+    }
+    boml->dma_alloc_size -= bo_legacy->base.size;
+    boml->dma_buf_count--;
+    return 0;
+}
+
+static void bo_free(struct bo_legacy *bo_legacy)
+{
+    struct bo_manager_legacy *boml;
+
+    if (bo_legacy == NULL) {
+        return;
+    }
+    boml = (struct bo_manager_legacy *)bo_legacy->base.bom;
+    bo_legacy->prev->next = bo_legacy->next;
+    if (bo_legacy->next) {
+        bo_legacy->next->prev = bo_legacy->prev;
+    }
+    if (!bo_legacy->static_bo) {
+        legacy_free_handle(boml, bo_legacy->base.handle);
+        if (bo_legacy->base.domains & RADEON_GEM_DOMAIN_GTT) {
+            /* dma buffers */
+            bo_dma_free(&bo_legacy->base);
+        } else {
+  	    driDestroyTextureObject(&bo_legacy->tobj->base);
+	    bo_legacy->tobj = NULL;
+            /* free backing store */
+            free(bo_legacy->ptr);
+        }
+    }
+    memset(bo_legacy, 0 , sizeof(struct bo_legacy));
+    free(bo_legacy);
+}
+
+static struct radeon_bo *bo_open(struct radeon_bo_manager *bom,
+                                 uint32_t handle,
+                                 uint32_t size,
+                                 uint32_t alignment,
+                                 uint32_t domains,
+                                 uint32_t flags)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom;
+    struct bo_legacy *bo_legacy;
+    int r;
+
+    if (handle) {
+        bo_legacy = boml->bos.next;
+        while (bo_legacy) {
+            if (bo_legacy->base.handle == handle) {
+                radeon_bo_ref(&(bo_legacy->base));
+                return (struct radeon_bo*)bo_legacy;
+            }
+            bo_legacy = bo_legacy->next;
+        }
+        return NULL;
+    }
+
+    bo_legacy = bo_allocate(boml, size, alignment, domains, flags);
+    bo_legacy->static_bo = 0;
+    r = legacy_new_handle(boml, &bo_legacy->base.handle);
+    if (r) {
+        bo_free(bo_legacy);
+        return NULL;
+    }
+    if (bo_legacy->base.domains & RADEON_GEM_DOMAIN_GTT) {
+    retry:
+        legacy_track_pending(boml, 0);
+        /* dma buffers */
+
+        r = bo_dma_alloc(&(bo_legacy->base));
+        if (r) {
+	  if (legacy_wait_any_pending(boml) == -1) {
+            bo_free(bo_legacy);
+	    return NULL;
+	  }
+	  goto retry;
+	  return NULL;
+        }
+    } else {
+        bo_legacy->ptr = malloc(bo_legacy->base.size);
+        if (bo_legacy->ptr == NULL) {
+            bo_free(bo_legacy);
+            return NULL;
+        }
+    }
+    radeon_bo_ref(&(bo_legacy->base));
+    return (struct radeon_bo*)bo_legacy;
+}
+
+static void bo_ref(struct radeon_bo *bo)
+{
+}
+
+static struct radeon_bo *bo_unref(struct radeon_bo *bo)
+{
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+
+    if (bo->cref <= 0) {
+        bo_legacy->prev->next = bo_legacy->next;
+        if (bo_legacy->next) {
+            bo_legacy->next->prev = bo_legacy->prev;
+        }
+        if (!bo_legacy->is_pending) {
+            bo_free(bo_legacy);
+        }
+        return NULL;
+    }
+    return bo;
+}
+
+static int bo_map(struct radeon_bo *bo, int write)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    
+    legacy_wait_pending(bo);
+    bo_legacy->validated = 0;
+    bo_legacy->dirty = 1;
+    bo_legacy->map_count++;
+    bo->ptr = bo_legacy->ptr;
+    /* Read the first pixel in the frame buffer.  This should
+     * be a noop, right?  In fact without this conform fails as reading
+     * from the framebuffer sometimes produces old results -- the
+     * on-card read cache gets mixed up and doesn't notice that the
+     * framebuffer has been updated.
+     *
+     * Note that we should probably be reading some otherwise unused
+     * region of VRAM, otherwise we might get incorrect results when
+     * reading pixels from the top left of the screen.
+     *
+     * I found this problem on an R420 with glean's texCube test.
+     * Note that the R200 span code also *writes* the first pixel in the
+     * framebuffer, but I've found this to be unnecessary.
+     *  -- Nicolai Hähnle, June 2008
+     */
+    if (!(bo->domains & RADEON_GEM_DOMAIN_GTT)) {
+        int p;
+        volatile int *buf = (int*)boml->screen->driScreen->pFB;
+        p = *buf;
+    }
+    return 0;
+}
+
+static int bo_unmap(struct radeon_bo *bo)
+{
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+
+    if (--bo_legacy->map_count > 0) {
+        return 0;
+    }
+    bo->ptr = NULL;
+    return 0;
+}
+
+static struct radeon_bo_funcs bo_legacy_funcs = {
+    bo_open,
+    bo_ref,
+    bo_unref,
+    bo_map,
+    bo_unmap
+};
+
+static int bo_vram_validate(struct radeon_bo *bo,
+                            uint32_t *soffset,
+                            uint32_t *eoffset)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    int r;
+    int retry_count = 0, pending_retry = 0;
+    
+    if (!bo_legacy->tobj) {
+	bo_legacy->tobj = CALLOC(sizeof(struct bo_legacy_texture_object));
+	bo_legacy->tobj->parent = bo_legacy;
+	make_empty_list(&bo_legacy->tobj->base);
+	bo_legacy->tobj->base.totalSize = bo->size;
+    retry:
+        r = driAllocateTexture(&boml->texture_heap, 1,
+                               &bo_legacy->tobj->base);
+        if (r) {
+		pending_retry = 0;
+		while(boml->cpendings && pending_retry++ < 10000) {
+			legacy_track_pending(boml, 0);
+			retry_count++;
+			if (retry_count > 2) {
+				free(bo_legacy->tobj);
+				bo_legacy->tobj = NULL;
+				fprintf(stderr, "Ouch! vram_validate failed %d\n", r);
+				return -1;
+			}
+			goto retry;
+		}
+	}
+        bo_legacy->offset = boml->texture_offset +
+                            bo_legacy->tobj->base.memBlock->ofs;
+        bo_legacy->dirty = 1;
+    }
+
+    assert(bo_legacy->tobj->base.memBlock);
+
+    if (bo_legacy->tobj)
+	driUpdateTextureLRU(&bo_legacy->tobj->base);
+
+    if (bo_legacy->dirty || bo_legacy->tobj->base.dirty_images[0]) {
+        /* Copy to VRAM using a blit.
+         * All memory is 4K aligned. We're using 1024 pixels wide blits.
+         */
+        drm_radeon_texture_t tex;
+        drm_radeon_tex_image_t tmp;
+        int ret;
+
+        tex.offset = bo_legacy->offset;
+        tex.image = &tmp;
+        assert(!(tex.offset & 1023));
+
+        tmp.x = 0;
+        tmp.y = 0;
+        if (bo->size < 4096) {
+            tmp.width = (bo->size + 3) / 4;
+            tmp.height = 1;
+        } else {
+            tmp.width = 1024;
+            tmp.height = (bo->size + 4095) / 4096;
+        }
+        tmp.data = bo_legacy->ptr;
+        tex.format = RADEON_TXFORMAT_ARGB8888;
+        tex.width = tmp.width;
+        tex.height = tmp.height;
+        tex.pitch = MAX2(tmp.width / 16, 1);
+        do {
+            ret = drmCommandWriteRead(bo->bom->fd,
+                                      DRM_RADEON_TEXTURE,
+                                      &tex,
+                                      sizeof(drm_radeon_texture_t));
+            if (ret) {
+                if (RADEON_DEBUG & DEBUG_IOCTL)
+                    fprintf(stderr, "DRM_RADEON_TEXTURE:  again!\n");
+                usleep(1);
+            }
+        } while (ret == -EAGAIN);
+        bo_legacy->dirty = 0;
+	bo_legacy->tobj->base.dirty_images[0] = 0;
+    }
+    return 0;
+}
+
+/* 
+ *  radeon_bo_legacy_validate -
+ *  returns:
+ *  0 - all good
+ *  -EINVAL - mapped buffer can't be validated
+ *  -EAGAIN - restart validation we've kicked all the buffers out
+ */
+int radeon_bo_legacy_validate(struct radeon_bo *bo,
+                              uint32_t *soffset,
+                              uint32_t *eoffset)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    int r;
+    int retries = 0;
+
+    if (bo_legacy->map_count) {
+        fprintf(stderr, "bo(%p, %d) is mapped (%d) can't valide it.\n",
+                bo, bo->size, bo_legacy->map_count);
+        return -EINVAL;
+    }
+    if (bo_legacy->static_bo || bo_legacy->validated) {
+        *soffset = bo_legacy->offset;
+        *eoffset = bo_legacy->offset + bo->size;
+        return 0;
+    }
+    if (!(bo->domains & RADEON_GEM_DOMAIN_GTT)) {
+
+        r = bo_vram_validate(bo, soffset, eoffset);
+        if (r) {
+	    legacy_track_pending(boml, 0);
+	    legacy_kick_all_buffers(boml);
+	    retries++;
+	    if (retries == 2) {
+		fprintf(stderr,"legacy bo: failed to get relocations into aperture\n");
+		assert(0);
+		exit(-1);
+	    }
+	    return -EAGAIN;
+        }
+    }
+    *soffset = bo_legacy->offset;
+    *eoffset = bo_legacy->offset + bo->size;
+    bo_legacy->validated = 1;
+    return 0;
+}
+
+void radeon_bo_legacy_pending(struct radeon_bo *bo, uint32_t pending)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bo->bom;
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+
+    bo_legacy->pending = pending;
+    bo_legacy->is_pending++;
+    /* add to pending list */
+    radeon_bo_ref(bo);
+    if (bo_legacy->is_pending > 1) {
+        return;    
+    }
+    bo_legacy->pprev = boml->pending_bos.pprev;
+    bo_legacy->pnext = NULL;
+    bo_legacy->pprev->pnext = bo_legacy;
+    boml->pending_bos.pprev = bo_legacy;
+    boml->cpendings++;
+}
+
+void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom;
+    struct bo_legacy *bo_legacy;
+
+    if (bom == NULL) {
+        return;
+    }
+    bo_legacy = boml->bos.next;
+    while (bo_legacy) {
+        struct bo_legacy *next;
+
+        next = bo_legacy->next;
+        bo_free(bo_legacy);
+        bo_legacy = next;
+    }
+    driDestroyTextureHeap(boml->texture_heap);
+    free(boml->free_handles);
+    free(boml);
+}
+
+static struct bo_legacy *radeon_legacy_bo_alloc_static(struct bo_manager_legacy *bom,
+						       int size, uint32_t offset)
+{
+    struct bo_legacy *bo;
+
+    bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+    if (bo == NULL)
+	return NULL;
+    bo->static_bo = 1;
+    bo->offset = offset + bom->fb_location;
+    bo->base.handle = bo->offset;
+    bo->ptr = bom->screen->driScreen->pFB + offset;
+    if (bo->base.handle > bom->nhandle) {
+        bom->nhandle = bo->base.handle + 1;
+    }
+    radeon_bo_ref(&(bo->base));
+    return bo;
+}
+
+struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *scrn)
+{
+    struct bo_manager_legacy *bom;
+    struct bo_legacy *bo;
+    unsigned size;
+
+    bom = (struct bo_manager_legacy*)
+          calloc(1, sizeof(struct bo_manager_legacy));
+    if (bom == NULL) {
+        return NULL;
+    }
+
+    make_empty_list(&bom->texture_swapped);
+
+    bom->texture_heap = driCreateTextureHeap(0,
+                                             bom,
+                                             scrn->texSize[0],
+                                             12,
+                                             RADEON_NR_TEX_REGIONS,
+                                             (drmTextureRegionPtr)scrn->sarea->tex_list[0],
+                                             &scrn->sarea->tex_age[0],
+                                             &bom->texture_swapped,
+                                             sizeof(struct bo_legacy_texture_object),
+                                             &bo_legacy_tobj_destroy);
+    bom->texture_offset = scrn->texOffset[0];
+
+    bom->base.funcs = &bo_legacy_funcs;
+    bom->base.fd = scrn->driScreen->fd;
+    bom->bos.next = NULL;
+    bom->bos.prev = NULL;
+    bom->pending_bos.pprev = &bom->pending_bos;
+    bom->pending_bos.pnext = NULL;
+    bom->screen = scrn;
+    bom->fb_location = scrn->fbLocation;
+    bom->nhandle = 1;
+    bom->cfree_handles = 0;
+    bom->nfree_handles = 0x400;
+    bom->free_handles = (uint32_t*)malloc(bom->nfree_handles * 4);
+    if (bom->free_handles == NULL) {
+        radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
+        return NULL;
+    }
+
+    /* biggest framebuffer size */
+    size = 4096*4096*4; 
+
+    /* allocate front */
+    bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->frontOffset);
+    if (!bo) {
+        radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
+        return NULL;
+    }
+    if (scrn->sarea->tiling_enabled) {
+        bo->base.flags = RADEON_BO_FLAGS_MACRO_TILE;
+    }
+
+    /* allocate back */
+    bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->backOffset);
+    if (!bo) {
+        radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
+        return NULL;
+    }
+    if (scrn->sarea->tiling_enabled) {
+        bo->base.flags = RADEON_BO_FLAGS_MACRO_TILE;
+    }
+
+    /* allocate depth */
+    bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->depthOffset);
+    if (!bo) {
+        radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
+        return NULL;
+    }
+    bo->base.flags = 0;
+    if (scrn->sarea->tiling_enabled) {
+        bo->base.flags |= RADEON_BO_FLAGS_MACRO_TILE;
+        bo->base.flags |= RADEON_BO_FLAGS_MICRO_TILE;
+    }
+    return (struct radeon_bo_manager*)bom;
+}
+
+void radeon_bo_legacy_texture_age(struct radeon_bo_manager *bom)
+{
+    struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom;
+    DRI_AGE_TEXTURES(boml->texture_heap);
+}
+
+unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo)
+{
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+
+    if (bo_legacy->static_bo || (bo->domains & RADEON_GEM_DOMAIN_GTT)) {
+        return 0;
+    }
+    return bo->size;
+}
+
+int radeon_legacy_bo_is_static(struct radeon_bo *bo)
+{
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    return bo_legacy->static_bo;
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h
new file mode 100644
index 00000000000..9187cd72013
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h
@@ -0,0 +1,47 @@
+/* 
+ * Copyright © 2008 Nicolai Haehnle
+ * Copyright © 2008 Jérôme Glisse
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ */
+/*
+ * Authors:
+ *      Aapo Tahkola <[email protected]>
+ *      Nicolai Haehnle <[email protected]>
+ *      Jérôme Glisse <[email protected]>
+ */
+#ifndef RADEON_BO_LEGACY_H
+#define RADEON_BO_LEGACY_H
+
+#include "radeon_screen.h"
+
+void radeon_bo_legacy_pending(struct radeon_bo *bo, uint32_t pending);
+int radeon_bo_legacy_validate(struct radeon_bo *bo,
+                              uint32_t *soffset,
+                              uint32_t *eoffset);
+struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *scrn);
+void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom);
+void radeon_bo_legacy_texture_age(struct radeon_bo_manager *bom);
+unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo);
+
+int radeon_legacy_bo_is_static(struct radeon_bo *bo);
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h b/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h
new file mode 100644
index 00000000000..6f1a0b45358
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h
@@ -0,0 +1,85 @@
+#ifndef RADEON_CS_WRAPPER_H
+#define RADEON_CS_WRAPPER_H
+
+#ifndef RADEON_PARAM_DEVICE_ID
+#define RADEON_PARAM_DEVICE_ID 16
+#endif
+
+#ifndef RADEON_INFO_DEVICE_ID
+#define RADEON_INFO_DEVICE_ID 0
+#endif
+#ifndef RADEON_INFO_NUM_GB_PIPES
+#define RADEON_INFO_NUM_GB_PIPES 0
+#endif
+
+#ifndef DRM_RADEON_INFO
+#define DRM_RADEON_INFO 0x1
+#endif
+
+#ifdef HAVE_LIBDRM_RADEON
+
+#include "radeon_bo.h"
+#include "radeon_bo_gem.h"
+#include "radeon_cs.h"
+#include "radeon_cs_gem.h"
+
+#else
+#include <stdint.h>
+
+#define RADEON_GEM_DOMAIN_CPU 0x1   // Cached CPU domain
+#define RADEON_GEM_DOMAIN_GTT 0x2   // GTT or cache flushed
+#define RADEON_GEM_DOMAIN_VRAM 0x4  // VRAM domain
+
+/* to be used to build locally in mesa with no libdrm bits */
+#include "../radeon/radeon_bo_drm.h"
+#include "../radeon/radeon_cs_drm.h"
+
+#ifndef DRM_RADEON_GEM_INFO
+#define DRM_RADEON_GEM_INFO 0x1c
+
+struct drm_radeon_gem_info {
+        uint64_t gart_size;
+        uint64_t vram_size;
+        uint64_t vram_visible;
+};
+
+struct drm_radeon_info {
+	uint32_t request;
+	uint32_t pad;
+	uint32_t value;
+};
+#endif
+
+
+static inline uint32_t radeon_gem_name_bo(struct radeon_bo *dummy)
+{
+  return 0;
+}
+
+static inline void *radeon_bo_manager_gem_ctor(int fd)
+{
+  return NULL;
+}
+
+static inline void radeon_bo_manager_gem_dtor(void *dummy)
+{
+}
+
+static inline void *radeon_cs_manager_gem_ctor(int fd)
+{
+  return NULL;
+}
+
+static inline void radeon_cs_manager_gem_dtor(void *dummy)
+{
+}
+
+static inline void radeon_tracker_print(void *ptr, int io)
+{
+}
+#endif
+
+#include "radeon_bo_legacy.h"
+#include "radeon_cs_legacy.h"
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h b/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h
new file mode 100644
index 00000000000..4b5116c4741
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_cmdbuf.h
@@ -0,0 +1,143 @@
+#ifndef COMMON_CMDBUF_H
+#define COMMON_CMDBUF_H
+
+#include "radeon_bocs_wrapper.h"
+
+void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller);
+int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller);
+int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller);
+void rcommonInitCmdBuf(radeonContextPtr rmesa);
+void rcommonDestroyCmdBuf(radeonContextPtr rmesa);
+
+void rcommonBeginBatch(radeonContextPtr rmesa,
+		       int n,
+		       int dostate,
+		       const char *file,
+		       const char *function,
+		       int line);
+
+#define RADEON_CP_PACKET3_NOP                       0xC0001000
+#define RADEON_CP_PACKET3_NEXT_CHAR                 0xC0001900
+#define RADEON_CP_PACKET3_PLY_NEXTSCAN              0xC0001D00
+#define RADEON_CP_PACKET3_SET_SCISSORS              0xC0001E00
+#define RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM     0xC0002300
+#define RADEON_CP_PACKET3_LOAD_MICROCODE            0xC0002400
+#define RADEON_CP_PACKET3_WAIT_FOR_IDLE             0xC0002600
+#define RADEON_CP_PACKET3_3D_DRAW_VBUF              0xC0002800
+#define RADEON_CP_PACKET3_3D_DRAW_IMMD              0xC0002900
+#define RADEON_CP_PACKET3_3D_DRAW_INDX              0xC0002A00
+#define RADEON_CP_PACKET3_LOAD_PALETTE              0xC0002C00
+#define RADEON_CP_PACKET3_3D_LOAD_VBPNTR            0xC0002F00
+#define RADEON_CP_PACKET3_CNTL_PAINT                0xC0009100
+#define RADEON_CP_PACKET3_CNTL_BITBLT               0xC0009200
+#define RADEON_CP_PACKET3_CNTL_SMALLTEXT            0xC0009300
+#define RADEON_CP_PACKET3_CNTL_HOSTDATA_BLT         0xC0009400
+#define RADEON_CP_PACKET3_CNTL_POLYLINE             0xC0009500
+#define RADEON_CP_PACKET3_CNTL_POLYSCANLINES        0xC0009800
+#define RADEON_CP_PACKET3_CNTL_PAINT_MULTI          0xC0009A00
+#define RADEON_CP_PACKET3_CNTL_BITBLT_MULTI         0xC0009B00
+#define RADEON_CP_PACKET3_CNTL_TRANS_BITBLT         0xC0009C00
+
+#define CP_PACKET2  (2 << 30)
+#define CP_PACKET0(reg, n)	(RADEON_CP_PACKET0 | ((n)<<16) | ((reg)>>2))
+#define CP_PACKET0_ONE(reg, n)	(RADEON_CP_PACKET0 | RADEON_CP_PACKET0_ONE_REG_WR | ((n)<<16) | ((reg)>>2))
+#define CP_PACKET3( pkt, n )						\
+	(RADEON_CP_PACKET3 | (pkt) | ((n) << 16))
+
+/**
+ * Every function writing to the command buffer needs to declare this
+ * to get the necessary local variables.
+ */
+#define BATCH_LOCALS(rmesa) \
+	const radeonContextPtr b_l_rmesa = rmesa
+
+/**
+ * Prepare writing n dwords to the command buffer,
+ * including producing any necessary state emits on buffer wraparound.
+ */
+#define BEGIN_BATCH(n) rcommonBeginBatch(b_l_rmesa, n, 1, __FILE__, __FUNCTION__, __LINE__)
+
+/**
+ * Same as BEGIN_BATCH, but do not cause automatic state emits.
+ */
+#define BEGIN_BATCH_NO_AUTOSTATE(n) rcommonBeginBatch(b_l_rmesa, n, 0, __FILE__, __FUNCTION__, __LINE__)
+
+/**
+ * Write one dword to the command buffer.
+ */
+#define OUT_BATCH(data) \
+	do { \
+        radeon_cs_write_dword(b_l_rmesa->cmdbuf.cs, data);\
+	} while(0)
+
+/**
+ * Write a relocated dword to the command buffer.
+ */
+#define OUT_BATCH_RELOC(data, bo, offset, rd, wd, flags) 	\
+	do { 							\
+        if (0 && offset) {					\
+            fprintf(stderr, "(%s:%s:%d) offset : %d\n",		\
+            __FILE__, __FUNCTION__, __LINE__, offset);		\
+        }							\
+        radeon_cs_write_dword(b_l_rmesa->cmdbuf.cs, offset);	\
+        radeon_cs_write_reloc(b_l_rmesa->cmdbuf.cs, 		\
+                              bo, rd, wd, flags);		\
+	if (!b_l_rmesa->radeonScreen->kernel_mm) 		\
+		b_l_rmesa->cmdbuf.cs->section_cdw += 2;		\
+	} while(0)
+
+
+/**
+ * Write n dwords from ptr to the command buffer.
+ */
+#define OUT_BATCH_TABLE(ptr,n) \
+	do { \
+		int _i; \
+        for (_i=0; _i < n; _i++) {\
+            radeon_cs_write_dword(b_l_rmesa->cmdbuf.cs, ptr[_i]);\
+        }\
+	} while(0)
+
+/**
+ * Finish writing dwords to the command buffer.
+ * The number of (direct or indirect) OUT_BATCH calls between the previous
+ * BEGIN_BATCH and END_BATCH must match the number specified at BEGIN_BATCH time.
+ */
+#define END_BATCH() \
+	do { \
+        radeon_cs_end(b_l_rmesa->cmdbuf.cs, __FILE__, __FUNCTION__, __LINE__);\
+	} while(0)
+
+/**
+ * After the last END_BATCH() of rendering, this indicates that flushing
+ * the command buffer now is okay.
+ */
+#define COMMIT_BATCH() \
+	do { \
+	} while(0)
+
+
+/** Single register write to command buffer; requires 2 dwords. */
+#define OUT_BATCH_REGVAL(reg, val) \
+	OUT_BATCH(cmdpacket0(b_l_rmesa->radeonScreen, (reg), 1)); \
+	OUT_BATCH((val))
+
+/** Continuous register range write to command buffer; requires 1 dword,
+ * expects count dwords afterwards for register contents. */
+#define OUT_BATCH_REGSEQ(reg, count) \
+	OUT_BATCH(cmdpacket0(b_l_rmesa->radeonScreen, (reg), (count)));
+
+/** Write a 32 bit float to the ring; requires 1 dword. */
+#define OUT_BATCH_FLOAT32(f) \
+	OUT_BATCH(radeonPackFloat32((f)));
+
+
+/* Fire the buffered vertices no matter what.
+ */
+static INLINE void radeon_firevertices(radeonContextPtr radeon)
+{
+   if (radeon->cmdbuf.cs->cdw || radeon->dma.flush )
+      radeonFlush(radeon->glCtx);
+}
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_common.c b/src/mesa/drivers/dri/radeon/radeon_common.c
new file mode 100644
index 00000000000..466eda784ef
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_common.c
@@ -0,0 +1,1495 @@
+/**************************************************************************
+
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ */
+
+/*
+   - Scissor implementation
+   - buffer swap/copy ioctls
+   - finish/flush
+   - state emission
+   - cmdbuffer management
+*/
+
+#include <errno.h>
+#include "main/glheader.h"
+#include "main/imports.h"
+#include "main/context.h"
+#include "main/api_arrayelt.h"
+#include "main/enums.h"
+#include "main/colormac.h"
+#include "main/light.h"
+#include "main/framebuffer.h"
+#include "main/simple_list.h"
+#include "main/renderbuffer.h"
+#include "swrast/swrast.h"
+#include "vbo/vbo.h"
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+#include "swrast_setup/swrast_setup.h"
+
+#include "main/blend.h"
+#include "main/bufferobj.h"
+#include "main/buffers.h"
+#include "main/depth.h"
+#include "main/polygon.h"
+#include "main/shaders.h"
+#include "main/texstate.h"
+#include "main/varray.h"
+#include "glapi/dispatch.h"
+#include "swrast/swrast.h"
+#include "main/stencil.h"
+#include "main/matrix.h"
+#include "main/attrib.h"
+#include "main/enable.h"
+#include "main/viewport.h"
+
+#include "dri_util.h"
+#include "vblank.h"
+
+#include "radeon_common.h"
+#include "radeon_bocs_wrapper.h"
+#include "radeon_lock.h"
+#include "radeon_drm.h"
+#include "radeon_mipmap_tree.h"
+
+#define DEBUG_CMDBUF         0
+
+/* =============================================================
+ * Scissoring
+ */
+
+static GLboolean intersect_rect(drm_clip_rect_t * out,
+				drm_clip_rect_t * a, drm_clip_rect_t * b)
+{
+	*out = *a;
+	if (b->x1 > out->x1)
+		out->x1 = b->x1;
+	if (b->y1 > out->y1)
+		out->y1 = b->y1;
+	if (b->x2 < out->x2)
+		out->x2 = b->x2;
+	if (b->y2 < out->y2)
+		out->y2 = b->y2;
+	if (out->x1 >= out->x2)
+		return GL_FALSE;
+	if (out->y1 >= out->y2)
+		return GL_FALSE;
+	return GL_TRUE;
+}
+
+void radeonRecalcScissorRects(radeonContextPtr radeon)
+{
+	drm_clip_rect_t *out;
+	int i;
+
+	/* Grow cliprect store?
+	 */
+	if (radeon->state.scissor.numAllocedClipRects < radeon->numClipRects) {
+		while (radeon->state.scissor.numAllocedClipRects <
+		       radeon->numClipRects) {
+			radeon->state.scissor.numAllocedClipRects += 1;	/* zero case */
+			radeon->state.scissor.numAllocedClipRects *= 2;
+		}
+
+		if (radeon->state.scissor.pClipRects)
+			FREE(radeon->state.scissor.pClipRects);
+
+		radeon->state.scissor.pClipRects =
+			MALLOC(radeon->state.scissor.numAllocedClipRects *
+			       sizeof(drm_clip_rect_t));
+
+		if (radeon->state.scissor.pClipRects == NULL) {
+			radeon->state.scissor.numAllocedClipRects = 0;
+			return;
+		}
+	}
+
+	out = radeon->state.scissor.pClipRects;
+	radeon->state.scissor.numClipRects = 0;
+
+	for (i = 0; i < radeon->numClipRects; i++) {
+		if (intersect_rect(out,
+				   &radeon->pClipRects[i],
+				   &radeon->state.scissor.rect)) {
+			radeon->state.scissor.numClipRects++;
+			out++;
+		}
+	}
+}
+
+void radeon_get_cliprects(radeonContextPtr radeon,
+			  struct drm_clip_rect **cliprects,
+			  unsigned int *num_cliprects,
+			  int *x_off, int *y_off)
+{
+	__DRIdrawablePrivate *dPriv = radeon_get_drawable(radeon);
+	struct radeon_framebuffer *rfb = dPriv->driverPrivate;
+
+	if (radeon->constant_cliprect) {
+		radeon->fboRect.x1 = 0;
+		radeon->fboRect.y1 = 0;
+		radeon->fboRect.x2 = radeon->glCtx->DrawBuffer->Width;
+		radeon->fboRect.y2 = radeon->glCtx->DrawBuffer->Height;
+
+		*cliprects = &radeon->fboRect;
+		*num_cliprects = 1;
+		*x_off = 0;
+		*y_off = 0;
+	} else if (radeon->front_cliprects ||
+		   rfb->pf_active || dPriv->numBackClipRects == 0) {
+		*cliprects = dPriv->pClipRects;
+		*num_cliprects = dPriv->numClipRects;
+		*x_off = dPriv->x;
+		*y_off = dPriv->y;
+	} else {
+		*num_cliprects = dPriv->numBackClipRects;
+		*cliprects = dPriv->pBackClipRects;
+		*x_off = dPriv->backX;
+		*y_off = dPriv->backY;
+	}
+}
+
+/**
+ * Update cliprects and scissors.
+ */
+void radeonSetCliprects(radeonContextPtr radeon)
+{
+	__DRIdrawablePrivate *const drawable = radeon_get_drawable(radeon);
+	__DRIdrawablePrivate *const readable = radeon_get_readable(radeon);
+	struct radeon_framebuffer *const draw_rfb = drawable->driverPrivate;
+	struct radeon_framebuffer *const read_rfb = readable->driverPrivate;
+	int x_off, y_off;
+
+	radeon_get_cliprects(radeon, &radeon->pClipRects,
+			     &radeon->numClipRects, &x_off, &y_off);
+
+	if ((draw_rfb->base.Width != drawable->w) ||
+	    (draw_rfb->base.Height != drawable->h)) {
+		_mesa_resize_framebuffer(radeon->glCtx, &draw_rfb->base,
+					 drawable->w, drawable->h);
+		draw_rfb->base.Initialized = GL_TRUE;
+	}
+
+	if (drawable != readable) {
+		if ((read_rfb->base.Width != readable->w) ||
+		    (read_rfb->base.Height != readable->h)) {
+			_mesa_resize_framebuffer(radeon->glCtx, &read_rfb->base,
+						 readable->w, readable->h);
+			read_rfb->base.Initialized = GL_TRUE;
+		}
+	}
+
+	if (radeon->state.scissor.enabled)
+		radeonRecalcScissorRects(radeon);
+
+}
+
+
+
+void radeonUpdateScissor( GLcontext *ctx )
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+	if ( radeon_get_drawable(rmesa) ) {
+		__DRIdrawablePrivate *dPriv = radeon_get_drawable(rmesa);
+
+		int x = ctx->Scissor.X;
+		int y = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height;
+		int w = ctx->Scissor.X + ctx->Scissor.Width - 1;
+		int h = dPriv->h - ctx->Scissor.Y - 1;
+
+		rmesa->state.scissor.rect.x1 = x + dPriv->x;
+		rmesa->state.scissor.rect.y1 = y + dPriv->y;
+		rmesa->state.scissor.rect.x2 = w + dPriv->x + 1;
+		rmesa->state.scissor.rect.y2 = h + dPriv->y + 1;
+
+		radeonRecalcScissorRects( rmesa );
+	}
+}
+
+/* =============================================================
+ * Scissoring
+ */
+
+void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	if (ctx->Scissor.Enabled) {
+		/* We don't pipeline cliprect changes */
+		radeon_firevertices(radeon);
+		radeonUpdateScissor(ctx);
+	}
+}
+
+
+/* ================================================================
+ * SwapBuffers with client-side throttling
+ */
+
+static uint32_t radeonGetLastFrame(radeonContextPtr radeon)
+{
+	drm_radeon_getparam_t gp;
+	int ret;
+	uint32_t frame = 0;
+
+	gp.param = RADEON_PARAM_LAST_FRAME;
+	gp.value = (int *)&frame;
+	ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM,
+				  &gp, sizeof(gp));
+	if (ret) {
+		fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__,
+			ret);
+		exit(1);
+	}
+
+	return frame;
+}
+
+uint32_t radeonGetAge(radeonContextPtr radeon)
+{
+	drm_radeon_getparam_t gp;
+	int ret;
+	uint32_t age;
+
+	gp.param = RADEON_PARAM_LAST_CLEAR;
+	gp.value = (int *)&age;
+	ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM,
+				  &gp, sizeof(gp));
+	if (ret) {
+		fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__,
+			ret);
+		exit(1);
+	}
+
+	return age;
+}
+
+static void radeonEmitIrqLocked(radeonContextPtr radeon)
+{
+	drm_radeon_irq_emit_t ie;
+	int ret;
+
+	ie.irq_seq = &radeon->iw.irq_seq;
+	ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_IRQ_EMIT,
+				  &ie, sizeof(ie));
+	if (ret) {
+		fprintf(stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__,
+			ret);
+		exit(1);
+	}
+}
+
+static void radeonWaitIrq(radeonContextPtr radeon)
+{
+	int ret;
+
+	do {
+		ret = drmCommandWrite(radeon->dri.fd, DRM_RADEON_IRQ_WAIT,
+				      &radeon->iw, sizeof(radeon->iw));
+	} while (ret && (errno == EINTR || errno == EBUSY));
+
+	if (ret) {
+		fprintf(stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__,
+			ret);
+		exit(1);
+	}
+}
+
+static void radeonWaitForFrameCompletion(radeonContextPtr radeon)
+{
+	drm_radeon_sarea_t *sarea = radeon->sarea;
+
+	if (radeon->do_irqs) {
+		if (radeonGetLastFrame(radeon) < sarea->last_frame) {
+			if (!radeon->irqsEmitted) {
+				while (radeonGetLastFrame(radeon) <
+				       sarea->last_frame) ;
+			} else {
+				UNLOCK_HARDWARE(radeon);
+				radeonWaitIrq(radeon);
+				LOCK_HARDWARE(radeon);
+			}
+			radeon->irqsEmitted = 10;
+		}
+
+		if (radeon->irqsEmitted) {
+			radeonEmitIrqLocked(radeon);
+			radeon->irqsEmitted--;
+		}
+	} else {
+		while (radeonGetLastFrame(radeon) < sarea->last_frame) {
+			UNLOCK_HARDWARE(radeon);
+			if (radeon->do_usleeps)
+				DO_USLEEP(1);
+			LOCK_HARDWARE(radeon);
+		}
+	}
+}
+
+/* wait for idle */
+void radeonWaitForIdleLocked(radeonContextPtr radeon)
+{
+	int ret;
+	int i = 0;
+
+	do {
+		ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_CP_IDLE);
+		if (ret)
+			DO_USLEEP(1);
+	} while (ret && ++i < 100);
+
+	if (ret < 0) {
+		UNLOCK_HARDWARE(radeon);
+		fprintf(stderr, "Error: R300 timed out... exiting\n");
+		exit(-1);
+	}
+}
+
+static void radeonWaitForIdle(radeonContextPtr radeon)
+{
+	if (!radeon->radeonScreen->driScreen->dri2.enabled) {
+        LOCK_HARDWARE(radeon);
+	    radeonWaitForIdleLocked(radeon);
+	    UNLOCK_HARDWARE(radeon);
+    }
+}
+
+static void radeon_flip_renderbuffers(struct radeon_framebuffer *rfb)
+{
+	int current_page = rfb->pf_current_page;
+	int next_page = (current_page + 1) % rfb->pf_num_pages;
+	struct gl_renderbuffer *tmp_rb;
+
+	/* Exchange renderbuffers if necessary but make sure their
+	 * reference counts are preserved.
+	 */
+	if (rfb->color_rb[current_page] &&
+	    rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer !=
+	    &rfb->color_rb[current_page]->base) {
+		tmp_rb = NULL;
+		_mesa_reference_renderbuffer(&tmp_rb,
+					     rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer);
+		tmp_rb = &rfb->color_rb[current_page]->base;
+		_mesa_reference_renderbuffer(&rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer, tmp_rb);
+		_mesa_reference_renderbuffer(&tmp_rb, NULL);
+	}
+
+	if (rfb->color_rb[next_page] &&
+	    rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer !=
+	    &rfb->color_rb[next_page]->base) {
+		tmp_rb = NULL;
+		_mesa_reference_renderbuffer(&tmp_rb,
+					     rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer);
+		tmp_rb = &rfb->color_rb[next_page]->base;
+		_mesa_reference_renderbuffer(&rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer, tmp_rb);
+		_mesa_reference_renderbuffer(&tmp_rb, NULL);
+	}
+}
+
+/* Copy the back color buffer to the front color buffer.
+ */
+void radeonCopyBuffer( __DRIdrawablePrivate *dPriv,
+		       const drm_clip_rect_t	  *rect)
+{
+	radeonContextPtr rmesa;
+	struct radeon_framebuffer *rfb;
+	GLint nbox, i, ret;
+
+	assert(dPriv);
+	assert(dPriv->driContextPriv);
+	assert(dPriv->driContextPriv->driverPrivate);
+
+	rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+
+	LOCK_HARDWARE(rmesa);
+
+	rfb = dPriv->driverPrivate;
+
+	if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+		fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *) rmesa->glCtx );
+	}
+
+	nbox = dPriv->numClipRects; /* must be in locked region */
+
+	for ( i = 0 ; i < nbox ; ) {
+		GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox );
+		drm_clip_rect_t *box = dPriv->pClipRects;
+		drm_clip_rect_t *b = rmesa->sarea->boxes;
+		GLint n = 0;
+
+		for ( ; i < nr ; i++ ) {
+
+			*b = box[i];
+
+			if (rect)
+			{
+				if (rect->x1 > b->x1)
+					b->x1 = rect->x1;
+				if (rect->y1 > b->y1)
+					b->y1 = rect->y1;
+				if (rect->x2 < b->x2)
+					b->x2 = rect->x2;
+				if (rect->y2 < b->y2)
+					b->y2 = rect->y2;
+
+				if (b->x1 >= b->x2 || b->y1 >= b->y2)
+					continue;
+			}
+
+			b++;
+			n++;
+		}
+		rmesa->sarea->nbox = n;
+
+		if (!n)
+			continue;
+
+		ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP );
+
+		if ( ret ) {
+			fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret );
+			UNLOCK_HARDWARE( rmesa );
+			exit( 1 );
+		}
+	}
+
+	UNLOCK_HARDWARE( rmesa );
+}
+
+static int radeonScheduleSwap(__DRIdrawablePrivate *dPriv, GLboolean *missed_target)
+{
+	radeonContextPtr rmesa;
+
+	rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+	radeon_firevertices(rmesa);
+
+	LOCK_HARDWARE( rmesa );
+
+	if (!dPriv->numClipRects) {
+		UNLOCK_HARDWARE(rmesa);
+		usleep(10000);	/* throttle invisible client 10ms */
+		return 0;
+	}
+
+	radeonWaitForFrameCompletion(rmesa);
+
+	UNLOCK_HARDWARE(rmesa);
+	driWaitForVBlank(dPriv, missed_target);
+
+	return 0;
+}
+
+static GLboolean radeonPageFlip( __DRIdrawablePrivate *dPriv )
+{
+	radeonContextPtr radeon;
+	GLint ret;
+	__DRIscreenPrivate *psp;
+	struct radeon_renderbuffer *rrb;
+	struct radeon_framebuffer *rfb;
+
+	assert(dPriv);
+	assert(dPriv->driContextPriv);
+	assert(dPriv->driContextPriv->driverPrivate);
+
+	radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+	rfb = dPriv->driverPrivate;
+	rrb = (void *)rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
+
+	psp = dPriv->driScreenPriv;
+
+	LOCK_HARDWARE(radeon);
+
+	if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+		fprintf(stderr, "%s: pfCurrentPage: %d %d\n", __FUNCTION__,
+			radeon->sarea->pfCurrentPage, radeon->sarea->pfState);
+	}
+	drm_clip_rect_t *box = dPriv->pClipRects;
+	drm_clip_rect_t *b = radeon->sarea->boxes;
+	b[0] = box[0];
+	radeon->sarea->nbox = 1;
+
+	ret = drmCommandNone( radeon->dri.fd, DRM_RADEON_FLIP );
+
+	UNLOCK_HARDWARE(radeon);
+
+	if ( ret ) {
+		fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret );
+		return GL_FALSE;
+	}
+
+	if (!rfb->pf_active)
+		return GL_FALSE;
+
+	rfb->pf_current_page = radeon->sarea->pfCurrentPage;
+	radeon_flip_renderbuffers(rfb);
+	radeon_draw_buffer(radeon->glCtx, &rfb->base);
+
+	return GL_TRUE;
+}
+
+
+/**
+ * Swap front and back buffer.
+ */
+void radeonSwapBuffers(__DRIdrawablePrivate * dPriv)
+{
+	int64_t ust;
+	__DRIscreenPrivate *psp;
+
+	if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+		radeonContextPtr radeon;
+		GLcontext *ctx;
+
+		radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+		ctx = radeon->glCtx;
+
+		if (ctx->Visual.doubleBufferMode) {
+			GLboolean missed_target;
+			struct radeon_framebuffer *rfb = dPriv->driverPrivate;
+			_mesa_notifySwapBuffers(ctx);/* flush pending rendering comands */
+
+			radeonScheduleSwap(dPriv, &missed_target);
+
+			if (rfb->pf_active) {
+				radeonPageFlip(dPriv);
+			} else {
+				radeonCopyBuffer(dPriv, NULL);
+			}
+
+			psp = dPriv->driScreenPriv;
+
+			rfb->swap_count++;
+			(*psp->systemTime->getUST)( & ust );
+			if ( missed_target ) {
+				rfb->swap_missed_count++;
+				rfb->swap_missed_ust = ust - rfb->swap_ust;
+			}
+
+			rfb->swap_ust = ust;
+			radeon->hw.all_dirty = GL_TRUE;
+		}
+	} else {
+		/* XXX this shouldn't be an error but we can't handle it for now */
+		_mesa_problem(NULL, "%s: drawable has no context!",
+			      __FUNCTION__);
+	}
+}
+
+void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
+			 int x, int y, int w, int h )
+{
+	if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+		radeonContextPtr radeon;
+		GLcontext *ctx;
+
+		radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+		ctx = radeon->glCtx;
+
+		if (ctx->Visual.doubleBufferMode) {
+			drm_clip_rect_t rect;
+			rect.x1 = x + dPriv->x;
+			rect.y1 = (dPriv->h - y - h) + dPriv->y;
+			rect.x2 = rect.x1 + w;
+			rect.y2 = rect.y1 + h;
+			_mesa_notifySwapBuffers(ctx);	/* flush pending rendering comands */
+			radeonCopyBuffer(dPriv, &rect);
+		}
+	} else {
+		/* XXX this shouldn't be an error but we can't handle it for now */
+		_mesa_problem(NULL, "%s: drawable has no context!",
+			      __FUNCTION__);
+	}
+}
+
+void radeon_draw_buffer(GLcontext *ctx, struct gl_framebuffer *fb)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	struct radeon_renderbuffer *rrbDepth = NULL, *rrbStencil = NULL,
+		*rrbColor = NULL;
+	uint32_t offset = 0;
+
+
+	if (!fb) {
+		/* this can happen during the initial context initialization */
+		return;
+	}
+
+	/* radeons only handle 1 color draw so far */
+	if (fb->_NumColorDrawBuffers != 1) {
+		radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE);
+		return;
+	}
+
+	/* Do this here, note core Mesa, since this function is called from
+	 * many places within the driver.
+	 */
+	if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) {
+		/* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */
+		_mesa_update_framebuffer(ctx);
+		/* this updates the DrawBuffer's Width/Height if it's a FBO */
+		_mesa_update_draw_buffer_bounds(ctx);
+	}
+
+	if (fb->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) {
+		/* this may occur when we're called by glBindFrameBuffer() during
+		 * the process of someone setting up renderbuffers, etc.
+		 */
+		/*_mesa_debug(ctx, "DrawBuffer: incomplete user FBO\n");*/
+		return;
+	}
+
+	if (fb->Name)
+		;/* do something depthy/stencily TODO */
+
+
+		/* none */
+	if (fb->Name == 0) {
+		if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) {
+			rrbColor = radeon_renderbuffer(fb->Attachment[BUFFER_FRONT_LEFT].Renderbuffer);
+			radeon->front_cliprects = GL_TRUE;
+			radeon->front_buffer_dirty = GL_TRUE;
+		} else {
+			rrbColor = radeon_renderbuffer(fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer);
+			radeon->front_cliprects = GL_FALSE;
+		}
+	} else {
+		/* user FBO in theory */
+		struct radeon_renderbuffer *rrb;
+		rrb = radeon_renderbuffer(fb->_ColorDrawBuffers[0]);
+		if (rrb) {
+			offset = rrb->draw_offset;
+			rrbColor = rrb;
+		}
+		radeon->constant_cliprect = GL_TRUE;
+	}
+
+	if (rrbColor == NULL)
+		radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE);
+	else
+		radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE);
+
+
+	if (fb->_DepthBuffer && fb->_DepthBuffer->Wrapped) {
+		rrbDepth = radeon_renderbuffer(fb->_DepthBuffer->Wrapped);
+		if (rrbDepth && rrbDepth->bo) {
+			radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_FALSE);
+		} else {
+			radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_TRUE);
+		}
+	} else {
+		radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_FALSE);
+		rrbDepth = NULL;
+	}
+
+	if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) {
+		rrbStencil = radeon_renderbuffer(fb->_DepthBuffer->Wrapped);
+		if (rrbStencil && rrbStencil->bo) {
+			radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_FALSE);
+			/* need to re-compute stencil hw state */
+			if (!rrbDepth)
+				rrbDepth = rrbStencil;
+		} else {
+			radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_TRUE);
+		}
+	} else {
+		radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_FALSE);
+		if (ctx->Driver.Enable != NULL)
+			ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
+		else
+			ctx->NewState |= _NEW_STENCIL;
+	}
+
+	/* Update culling direction which changes depending on the
+	 * orientation of the buffer:
+	 */
+	if (ctx->Driver.FrontFace)
+		ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
+	else
+		ctx->NewState |= _NEW_POLYGON;
+
+	/*
+	 * Update depth test state
+	 */
+	if (ctx->Driver.Enable) {
+		ctx->Driver.Enable(ctx, GL_DEPTH_TEST,
+				   (ctx->Depth.Test && fb->Visual.depthBits > 0));
+		/* Need to update the derived ctx->Stencil._Enabled first */
+		_mesa_update_stencil(ctx);
+		ctx->Driver.Enable(ctx, GL_STENCIL_TEST,
+				   (ctx->Stencil._Enabled && fb->Visual.stencilBits > 0));
+	} else {
+		ctx->NewState |= (_NEW_DEPTH | _NEW_STENCIL);
+	}
+
+	_mesa_reference_renderbuffer(&radeon->state.depth.rb, &rrbDepth->base);
+	_mesa_reference_renderbuffer(&radeon->state.color.rb, &rrbColor->base);
+	radeon->state.color.draw_offset = offset;
+
+#if 0
+	/* update viewport since it depends on window size */
+	if (ctx->Driver.Viewport) {
+		ctx->Driver.Viewport(ctx, ctx->Viewport.X, ctx->Viewport.Y,
+				     ctx->Viewport.Width, ctx->Viewport.Height);
+	} else {
+
+	}
+#endif
+	ctx->NewState |= _NEW_VIEWPORT;
+
+	/* Set state we know depends on drawable parameters:
+	 */
+	radeonUpdateScissor(ctx);
+	radeon->NewGLState |= _NEW_SCISSOR;
+
+	if (ctx->Driver.DepthRange)
+		ctx->Driver.DepthRange(ctx,
+				       ctx->Viewport.Near,
+				       ctx->Viewport.Far);
+
+	/* Update culling direction which changes depending on the
+	 * orientation of the buffer:
+	 */
+	if (ctx->Driver.FrontFace)
+		ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
+	else
+		ctx->NewState |= _NEW_POLYGON;
+}
+
+/**
+ * Called via glDrawBuffer.
+ */
+void radeonDrawBuffer( GLcontext *ctx, GLenum mode )
+{
+	if (RADEON_DEBUG & DEBUG_DRI)
+		fprintf(stderr, "%s %s\n", __FUNCTION__,
+			_mesa_lookup_enum_by_nr( mode ));
+
+	if (ctx->DrawBuffer->Name == 0) {
+		radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+
+		const GLboolean was_front_buffer_rendering =
+			radeon->is_front_buffer_rendering;
+
+		radeon->is_front_buffer_rendering = (mode == GL_FRONT_LEFT) ||
+                                            (mode == GL_FRONT);
+
+      /* If we weren't front-buffer rendering before but we are now, make sure
+       * that the front-buffer has actually been allocated.
+       */
+		if (!was_front_buffer_rendering && radeon->is_front_buffer_rendering) {
+			radeon_update_renderbuffers(radeon->dri.context,
+				radeon->dri.context->driDrawablePriv);
+      }
+	}
+
+	radeon_draw_buffer(ctx, ctx->DrawBuffer);
+}
+
+void radeonReadBuffer( GLcontext *ctx, GLenum mode )
+{
+	/* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
+	if (ctx->ReadBuffer == ctx->DrawBuffer) {
+		/* This will update FBO completeness status.
+		 * A framebuffer will be incomplete if the GL_READ_BUFFER setting
+		 * refers to a missing renderbuffer.  Calling glReadBuffer can set
+		 * that straight and can make the drawing buffer complete.
+		 */
+		radeon_draw_buffer(ctx, ctx->DrawBuffer);
+	}
+}
+
+
+/* Turn on/off page flipping according to the flags in the sarea:
+ */
+void radeonUpdatePageFlipping(radeonContextPtr radeon)
+{
+	struct radeon_framebuffer *rfb = radeon_get_drawable(radeon)->driverPrivate;
+
+	rfb->pf_active = radeon->sarea->pfState;
+	rfb->pf_current_page = radeon->sarea->pfCurrentPage;
+	rfb->pf_num_pages = 2;
+	radeon_flip_renderbuffers(rfb);
+	radeon_draw_buffer(radeon->glCtx, radeon->glCtx->DrawBuffer);
+}
+
+void radeon_window_moved(radeonContextPtr radeon)
+{
+	if (!radeon->radeonScreen->driScreen->dri2.enabled) {
+		radeonUpdatePageFlipping(radeon);
+	}
+	radeonSetCliprects(radeon);
+}
+
+void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	__DRIcontext *driContext = radeon->dri.context;
+	void (*old_viewport)(GLcontext *ctx, GLint x, GLint y,
+			     GLsizei w, GLsizei h);
+
+	if (!driContext->driScreenPriv->dri2.enabled)
+		return;
+
+	radeon_update_renderbuffers(driContext, driContext->driDrawablePriv);
+	if (driContext->driDrawablePriv != driContext->driReadablePriv)
+		radeon_update_renderbuffers(driContext, driContext->driReadablePriv);
+
+	old_viewport = ctx->Driver.Viewport;
+	ctx->Driver.Viewport = NULL;
+	radeon_window_moved(radeon);
+	radeon_draw_buffer(ctx, radeon->glCtx->DrawBuffer);
+	ctx->Driver.Viewport = old_viewport;
+}
+
+static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state_atom *state)
+{
+	int i, j, reg;
+	int dwords = (*state->check) (radeon->glCtx, state);
+	drm_r300_cmd_header_t cmd;
+
+	fprintf(stderr, "  emit %s %d/%d\n", state->name, dwords, state->cmd_size);
+
+	if (RADEON_DEBUG & DEBUG_VERBOSE) {
+		for (i = 0; i < dwords;) {
+			cmd = *((drm_r300_cmd_header_t *) &state->cmd[i]);
+			reg = (cmd.packet0.reghi << 8) | cmd.packet0.reglo;
+			fprintf(stderr, "      %s[%d]: cmdpacket0 (first reg=0x%04x, count=%d)\n",
+					state->name, i, reg, cmd.packet0.count);
+			++i;
+			for (j = 0; j < cmd.packet0.count && i < dwords; j++) {
+				fprintf(stderr, "      %s[%d]: 0x%04x = %08x\n",
+						state->name, i, reg, state->cmd[i]);
+				reg += 4;
+				++i;
+			}
+		}
+	}
+}
+
+static void radeon_print_state_atom_kmm(radeonContextPtr radeon, struct radeon_state_atom *state)
+{
+	int i, j, reg, count;
+	int dwords = (*state->check) (radeon->glCtx, state);
+	uint32_t packet0;
+
+	fprintf(stderr, "  emit %s %d/%d\n", state->name, dwords, state->cmd_size);
+
+	if (RADEON_DEBUG & DEBUG_VERBOSE) {
+		for (i = 0; i < dwords;) {
+			packet0 = state->cmd[i];
+			reg = (packet0 & 0x1FFF) << 2;
+			count = ((packet0 & 0x3FFF0000) >> 16) + 1;
+			fprintf(stderr, "      %s[%d]: cmdpacket0 (first reg=0x%04x, count=%d)\n",
+					state->name, i, reg, count);
+			++i;
+			for (j = 0; j < count && i < dwords; j++) {
+				fprintf(stderr, "      %s[%d]: 0x%04x = %08x\n",
+						state->name, i, reg, state->cmd[i]);
+				reg += 4;
+				++i;
+			}
+		}
+	}
+}
+
+static INLINE void radeonEmitAtoms(radeonContextPtr radeon, GLboolean dirty)
+{
+	BATCH_LOCALS(radeon);
+	struct radeon_state_atom *atom;
+	int dwords;
+
+	if (radeon->vtbl.pre_emit_atoms)
+		radeon->vtbl.pre_emit_atoms(radeon);
+
+	/* Emit actual atoms */
+	foreach(atom, &radeon->hw.atomlist) {
+		if ((atom->dirty || radeon->hw.all_dirty) == dirty) {
+			dwords = (*atom->check) (radeon->glCtx, atom);
+			if (dwords) {
+				if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) {
+					if (radeon->radeonScreen->kernel_mm)
+						radeon_print_state_atom_kmm(radeon, atom);
+					else
+						radeon_print_state_atom(radeon, atom);
+				}
+				if (atom->emit) {
+					(*atom->emit)(radeon->glCtx, atom);
+				} else {
+					BEGIN_BATCH_NO_AUTOSTATE(dwords);
+					OUT_BATCH_TABLE(atom->cmd, dwords);
+					END_BATCH();
+				}
+				atom->dirty = GL_FALSE;
+			} else {
+				if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) {
+					fprintf(stderr, "  skip state %s\n",
+						atom->name);
+				}
+			}
+		}
+	}
+
+	COMMIT_BATCH();
+}
+
+GLboolean radeon_revalidate_bos(GLcontext *ctx)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	int flushed = 0;
+	int ret;
+again:
+	ret = radeon_cs_space_check(radeon->cmdbuf.cs, radeon->state.bos, radeon->state.validated_bo_count);
+	if (ret == RADEON_CS_SPACE_OP_TO_BIG)
+		return GL_FALSE;
+	if (ret == RADEON_CS_SPACE_FLUSH) {
+		radeonFlush(ctx);
+		if (flushed)
+			return GL_FALSE;
+		flushed = 1;
+		goto again;
+	}
+	return GL_TRUE;
+}
+
+void radeon_validate_reset_bos(radeonContextPtr radeon)
+{
+	int i;
+
+	for (i = 0; i < radeon->state.validated_bo_count; i++) {
+		radeon_bo_unref(radeon->state.bos[i].bo);
+		radeon->state.bos[i].bo = NULL;
+		radeon->state.bos[i].read_domains = 0;
+		radeon->state.bos[i].write_domain = 0;
+		radeon->state.bos[i].new_accounted = 0;
+	}
+	radeon->state.validated_bo_count = 0;
+}
+
+void radeon_validate_bo(radeonContextPtr radeon, struct radeon_bo *bo, uint32_t read_domains, uint32_t write_domain)
+{
+	radeon_bo_ref(bo);
+	radeon->state.bos[radeon->state.validated_bo_count].bo = bo;
+	radeon->state.bos[radeon->state.validated_bo_count].read_domains = read_domains;
+	radeon->state.bos[radeon->state.validated_bo_count].write_domain = write_domain;
+	radeon->state.bos[radeon->state.validated_bo_count].new_accounted = 0;
+	radeon->state.validated_bo_count++;
+
+	assert(radeon->state.validated_bo_count < RADEON_MAX_BOS);
+}
+
+void radeonEmitState(radeonContextPtr radeon)
+{
+	if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
+		fprintf(stderr, "%s\n", __FUNCTION__);
+
+	if (radeon->vtbl.pre_emit_state)
+		radeon->vtbl.pre_emit_state(radeon);
+
+	/* this code used to return here but now it emits zbs */
+	if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty)
+		return;
+
+	/* To avoid going across the entire set of states multiple times, just check
+	 * for enough space for the case of emitting all state, and inline the
+	 * radeonAllocCmdBuf code here without all the checks.
+	 */
+	rcommonEnsureCmdBufSpace(radeon, radeon->hw.max_state_size, __FUNCTION__);
+
+	if (!radeon->cmdbuf.cs->cdw) {
+		if (RADEON_DEBUG & DEBUG_STATE)
+			fprintf(stderr, "Begin reemit state\n");
+
+		radeonEmitAtoms(radeon, GL_FALSE);
+	}
+
+	if (RADEON_DEBUG & DEBUG_STATE)
+		fprintf(stderr, "Begin dirty state\n");
+
+	radeonEmitAtoms(radeon, GL_TRUE);
+	radeon->hw.is_dirty = GL_FALSE;
+	radeon->hw.all_dirty = GL_FALSE;
+
+}
+
+
+void radeonFlush(GLcontext *ctx)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s %d\n", __FUNCTION__, radeon->cmdbuf.cs->cdw);
+
+	/* okay if we have no cmds in the buffer &&
+	   we have no DMA flush &&
+	   we have no DMA buffer allocated.
+	   then no point flushing anything at all.
+	*/
+	if (!radeon->dma.flush && !radeon->cmdbuf.cs->cdw && !radeon->dma.current)
+		return;
+
+	if (radeon->dma.flush)
+		radeon->dma.flush( ctx );
+
+	radeonEmitState(radeon);
+
+	if (radeon->cmdbuf.cs->cdw)
+		rcommonFlushCmdBuf(radeon, __FUNCTION__);
+
+	if ((ctx->DrawBuffer->Name == 0) && radeon->front_buffer_dirty) {
+		__DRIscreen *const screen = radeon->radeonScreen->driScreen;
+
+		if (screen->dri2.loader && (screen->dri2.loader->base.version >= 2)
+			&& (screen->dri2.loader->flushFrontBuffer != NULL)) {
+			__DRIdrawablePrivate * drawable = radeon_get_drawable(radeon);
+			(*screen->dri2.loader->flushFrontBuffer)(drawable, drawable->loaderPrivate);
+
+			/* Only clear the dirty bit if front-buffer rendering is no longer
+			 * enabled.  This is done so that the dirty bit can only be set in
+			 * glDrawBuffer.  Otherwise the dirty bit would have to be set at
+			 * each of N places that do rendering.  This has worse performances,
+			 * but it is much easier to get correct.
+			 */
+			if (radeon->is_front_buffer_rendering) {
+				radeon->front_buffer_dirty = GL_FALSE;
+			}
+		}
+	}
+}
+
+/* Make sure all commands have been sent to the hardware and have
+ * completed processing.
+ */
+void radeonFinish(GLcontext * ctx)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	struct gl_framebuffer *fb = ctx->DrawBuffer;
+	int i;
+
+	radeonFlush(ctx);
+
+	if (radeon->radeonScreen->kernel_mm) {
+		for (i = 0; i < fb->_NumColorDrawBuffers; i++) {
+			struct radeon_renderbuffer *rrb;
+			rrb = radeon_renderbuffer(fb->_ColorDrawBuffers[i]);
+			if (rrb && rrb->bo)
+				radeon_bo_wait(rrb->bo);
+		}
+		{
+			struct radeon_renderbuffer *rrb;
+			rrb = radeon_get_depthbuffer(radeon);
+			if (rrb && rrb->bo)
+				radeon_bo_wait(rrb->bo);
+		}
+	} else if (radeon->do_irqs) {
+		LOCK_HARDWARE(radeon);
+		radeonEmitIrqLocked(radeon);
+		UNLOCK_HARDWARE(radeon);
+		radeonWaitIrq(radeon);
+	} else {
+		radeonWaitForIdle(radeon);
+	}
+}
+
+/* cmdbuffer */
+/**
+ * Send the current command buffer via ioctl to the hardware.
+ */
+int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller)
+{
+	int ret = 0;
+
+	if (rmesa->cmdbuf.flushing) {
+		fprintf(stderr, "Recursive call into r300FlushCmdBufLocked!\n");
+		exit(-1);
+	}
+	rmesa->cmdbuf.flushing = 1;
+
+	if (RADEON_DEBUG & DEBUG_IOCTL) {
+		fprintf(stderr, "%s from %s - %i cliprects\n",
+			__FUNCTION__, caller, rmesa->numClipRects);
+	}
+
+	if (rmesa->cmdbuf.cs->cdw) {
+		ret = radeon_cs_emit(rmesa->cmdbuf.cs);
+		rmesa->hw.all_dirty = GL_TRUE;
+	}
+	radeon_cs_erase(rmesa->cmdbuf.cs);
+	rmesa->cmdbuf.flushing = 0;
+
+	if (radeon_revalidate_bos(rmesa->glCtx) == GL_FALSE) {
+		fprintf(stderr,"failed to revalidate buffers\n");
+	}
+
+	return ret;
+}
+
+int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller)
+{
+	int ret;
+
+	radeonReleaseDmaRegion(rmesa);
+
+	LOCK_HARDWARE(rmesa);
+	ret = rcommonFlushCmdBufLocked(rmesa, caller);
+	UNLOCK_HARDWARE(rmesa);
+
+	if (ret) {
+		fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret);
+		_mesa_exit(ret);
+	}
+
+	return ret;
+}
+
+/**
+ * Make sure that enough space is available in the command buffer
+ * by flushing if necessary.
+ *
+ * \param dwords The number of dwords we need to be free on the command buffer
+ */
+void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller)
+{
+	if ((rmesa->cmdbuf.cs->cdw + dwords + 128) > rmesa->cmdbuf.size ||
+	    radeon_cs_need_flush(rmesa->cmdbuf.cs)) {
+		rcommonFlushCmdBuf(rmesa, caller);
+	}
+}
+
+void rcommonInitCmdBuf(radeonContextPtr rmesa)
+{
+	GLuint size;
+	/* Initialize command buffer */
+	size = 256 * driQueryOptioni(&rmesa->optionCache,
+				     "command_buffer_size");
+	if (size < 2 * rmesa->hw.max_state_size) {
+		size = 2 * rmesa->hw.max_state_size + 65535;
+	}
+	if (size > 64 * 256)
+		size = 64 * 256;
+
+	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) {
+		fprintf(stderr, "sizeof(drm_r300_cmd_header_t)=%zd\n",
+			sizeof(drm_r300_cmd_header_t));
+		fprintf(stderr, "sizeof(drm_radeon_cmd_buffer_t)=%zd\n",
+			sizeof(drm_radeon_cmd_buffer_t));
+		fprintf(stderr,
+			"Allocating %d bytes command buffer (max state is %d bytes)\n",
+			size * 4, rmesa->hw.max_state_size * 4);
+	}
+
+	if (rmesa->radeonScreen->kernel_mm) {
+		int fd = rmesa->radeonScreen->driScreen->fd;
+		rmesa->cmdbuf.csm = radeon_cs_manager_gem_ctor(fd);
+	} else {
+		rmesa->cmdbuf.csm = radeon_cs_manager_legacy_ctor(rmesa);
+	}
+	if (rmesa->cmdbuf.csm == NULL) {
+		/* FIXME: fatal error */
+		return;
+	}
+	rmesa->cmdbuf.cs = radeon_cs_create(rmesa->cmdbuf.csm, size);
+	assert(rmesa->cmdbuf.cs != NULL);
+	rmesa->cmdbuf.size = size;
+
+	if (!rmesa->radeonScreen->kernel_mm) {
+		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]);
+		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, rmesa->radeonScreen->gartTextures.size);
+	} else {
+		struct drm_radeon_gem_info mminfo = { 0 };
+
+		if (!drmCommandWriteRead(rmesa->dri.fd, DRM_RADEON_GEM_INFO, &mminfo, sizeof(mminfo)))
+		{
+			radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, mminfo.vram_visible);
+			radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, mminfo.gart_size);
+		}
+	}
+
+}
+/**
+ * Destroy the command buffer
+ */
+void rcommonDestroyCmdBuf(radeonContextPtr rmesa)
+{
+	radeon_cs_destroy(rmesa->cmdbuf.cs);
+	if (rmesa->radeonScreen->driScreen->dri2.enabled || rmesa->radeonScreen->kernel_mm) {
+		radeon_cs_manager_gem_dtor(rmesa->cmdbuf.csm);
+	} else {
+		radeon_cs_manager_legacy_dtor(rmesa->cmdbuf.csm);
+	}
+}
+
+void rcommonBeginBatch(radeonContextPtr rmesa, int n,
+		       int dostate,
+		       const char *file,
+		       const char *function,
+		       int line)
+{
+	rcommonEnsureCmdBufSpace(rmesa, n, function);
+	if (!rmesa->cmdbuf.cs->cdw && dostate) {
+		if (RADEON_DEBUG & DEBUG_IOCTL)
+			fprintf(stderr, "Reemit state after flush (from %s)\n", function);
+		radeonEmitState(rmesa);
+	}
+	radeon_cs_begin(rmesa->cmdbuf.cs, n, file, function, line);
+
+        if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_IOCTL)
+                fprintf(stderr, "BEGIN_BATCH(%d) at %d, from %s:%i\n",
+                        n, rmesa->cmdbuf.cs->cdw, function, line);
+
+}
+
+
+
+static void
+radeon_meta_set_passthrough_transform(radeonContextPtr radeon)
+{
+   GLcontext *ctx = radeon->glCtx;
+
+   radeon->meta.saved_vp_x = ctx->Viewport.X;
+   radeon->meta.saved_vp_y = ctx->Viewport.Y;
+   radeon->meta.saved_vp_width = ctx->Viewport.Width;
+   radeon->meta.saved_vp_height = ctx->Viewport.Height;
+   radeon->meta.saved_matrix_mode = ctx->Transform.MatrixMode;
+
+   _mesa_Viewport(0, 0, ctx->DrawBuffer->Width, ctx->DrawBuffer->Height);
+
+   _mesa_MatrixMode(GL_PROJECTION);
+   _mesa_PushMatrix();
+   _mesa_LoadIdentity();
+   _mesa_Ortho(0, ctx->DrawBuffer->Width, 0, ctx->DrawBuffer->Height, 1, -1);
+
+   _mesa_MatrixMode(GL_MODELVIEW);
+   _mesa_PushMatrix();
+   _mesa_LoadIdentity();
+}
+
+static void
+radeon_meta_restore_transform(radeonContextPtr radeon)
+{
+   _mesa_MatrixMode(GL_PROJECTION);
+   _mesa_PopMatrix();
+   _mesa_MatrixMode(GL_MODELVIEW);
+   _mesa_PopMatrix();
+
+   _mesa_MatrixMode(radeon->meta.saved_matrix_mode);
+
+   _mesa_Viewport(radeon->meta.saved_vp_x, radeon->meta.saved_vp_y,
+		  radeon->meta.saved_vp_width, radeon->meta.saved_vp_height);
+}
+
+
+/**
+ * Perform glClear where mask contains only color, depth, and/or stencil.
+ *
+ * The implementation is based on calling into Mesa to set GL state and
+ * performing normal triangle rendering.  The intent of this path is to
+ * have as generic a path as possible, so that any driver could make use of
+ * it.
+ */
+
+
+void radeon_clear_tris(GLcontext *ctx, GLbitfield mask)
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat vertices[4][3];
+   GLfloat color[4][4];
+   GLfloat dst_z;
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
+   int i;
+   GLboolean saved_fp_enable = GL_FALSE, saved_vp_enable = GL_FALSE;
+   GLboolean saved_shader_program = 0;
+   unsigned int saved_active_texture;
+
+   assert((mask & ~(TRI_CLEAR_COLOR_BITS | BUFFER_BIT_DEPTH |
+		    BUFFER_BIT_STENCIL)) == 0);
+
+   _mesa_PushAttrib(GL_COLOR_BUFFER_BIT |
+		    GL_CURRENT_BIT |
+		    GL_DEPTH_BUFFER_BIT |
+		    GL_ENABLE_BIT |
+		    GL_POLYGON_BIT |
+		    GL_STENCIL_BUFFER_BIT |
+		    GL_TRANSFORM_BIT |
+		    GL_CURRENT_BIT);
+   _mesa_PushClientAttrib(GL_CLIENT_VERTEX_ARRAY_BIT);
+   saved_active_texture = ctx->Texture.CurrentUnit;
+
+  /* Disable existing GL state we don't want to apply to a clear. */
+   _mesa_Disable(GL_ALPHA_TEST);
+   _mesa_Disable(GL_BLEND);
+   _mesa_Disable(GL_CULL_FACE);
+   _mesa_Disable(GL_FOG);
+   _mesa_Disable(GL_POLYGON_SMOOTH);
+   _mesa_Disable(GL_POLYGON_STIPPLE);
+   _mesa_Disable(GL_POLYGON_OFFSET_FILL);
+   _mesa_Disable(GL_LIGHTING);
+   _mesa_Disable(GL_CLIP_PLANE0);
+   _mesa_Disable(GL_CLIP_PLANE1);
+   _mesa_Disable(GL_CLIP_PLANE2);
+   _mesa_Disable(GL_CLIP_PLANE3);
+   _mesa_Disable(GL_CLIP_PLANE4);
+   _mesa_Disable(GL_CLIP_PLANE5);
+   _mesa_PolygonMode(GL_FRONT_AND_BACK, GL_FILL);
+   if (ctx->Extensions.ARB_fragment_program && ctx->FragmentProgram.Enabled) {
+      saved_fp_enable = GL_TRUE;
+      _mesa_Disable(GL_FRAGMENT_PROGRAM_ARB);
+   }
+   if (ctx->Extensions.ARB_vertex_program && ctx->VertexProgram.Enabled) {
+      saved_vp_enable = GL_TRUE;
+      _mesa_Disable(GL_VERTEX_PROGRAM_ARB);
+   }
+   if (ctx->Extensions.ARB_shader_objects && ctx->Shader.CurrentProgram) {
+      saved_shader_program = ctx->Shader.CurrentProgram->Name;
+      _mesa_UseProgramObjectARB(0);
+   }
+
+   if (ctx->Texture._EnabledUnits != 0) {
+      int i;
+
+      for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
+	 _mesa_ActiveTextureARB(GL_TEXTURE0 + i);
+	 _mesa_Disable(GL_TEXTURE_1D);
+	 _mesa_Disable(GL_TEXTURE_2D);
+	 _mesa_Disable(GL_TEXTURE_3D);
+	 if (ctx->Extensions.ARB_texture_cube_map)
+	    _mesa_Disable(GL_TEXTURE_CUBE_MAP_ARB);
+	 if (ctx->Extensions.NV_texture_rectangle)
+	    _mesa_Disable(GL_TEXTURE_RECTANGLE_NV);
+	 if (ctx->Extensions.MESA_texture_array) {
+	    _mesa_Disable(GL_TEXTURE_1D_ARRAY_EXT);
+	    _mesa_Disable(GL_TEXTURE_2D_ARRAY_EXT);
+	 }
+      }
+   }
+
+#if FEATURE_ARB_vertex_buffer_object
+   _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, 0);
+   _mesa_BindBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, 0);
+#endif
+
+   radeon_meta_set_passthrough_transform(rmesa);
+
+   for (i = 0; i < 4; i++) {
+      color[i][0] = ctx->Color.ClearColor[0];
+      color[i][1] = ctx->Color.ClearColor[1];
+      color[i][2] = ctx->Color.ClearColor[2];
+      color[i][3] = ctx->Color.ClearColor[3];
+   }
+
+   /* convert clear Z from [0,1] to NDC coord in [-1,1] */
+
+   dst_z = -1.0 + 2.0 * ctx->Depth.Clear;
+   /* Prepare the vertices, which are the same regardless of which buffer we're
+    * drawing to.
+    */
+   vertices[0][0] = fb->_Xmin;
+   vertices[0][1] = fb->_Ymin;
+   vertices[0][2] = dst_z;
+   vertices[1][0] = fb->_Xmax;
+   vertices[1][1] = fb->_Ymin;
+   vertices[1][2] = dst_z;
+   vertices[2][0] = fb->_Xmax;
+   vertices[2][1] = fb->_Ymax;
+   vertices[2][2] = dst_z;
+   vertices[3][0] = fb->_Xmin;
+   vertices[3][1] = fb->_Ymax;
+   vertices[3][2] = dst_z;
+
+   _mesa_ColorPointer(4, GL_FLOAT, 4 * sizeof(GLfloat), &color);
+   _mesa_VertexPointer(3, GL_FLOAT, 3 * sizeof(GLfloat), &vertices);
+   _mesa_Enable(GL_COLOR_ARRAY);
+   _mesa_Enable(GL_VERTEX_ARRAY);
+
+   while (mask != 0) {
+      GLuint this_mask = 0;
+      GLuint color_bit;
+
+      color_bit = _mesa_ffs(mask & TRI_CLEAR_COLOR_BITS);
+      if (color_bit != 0)
+	 this_mask |= (1 << (color_bit - 1));
+
+      /* Clear depth/stencil in the same pass as color. */
+      this_mask |= (mask & (BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL));
+
+      /* Select the current color buffer and use the color write mask if
+       * we have one, otherwise don't write any color channels.
+       */
+      if (this_mask & BUFFER_BIT_FRONT_LEFT)
+	 _mesa_DrawBuffer(GL_FRONT_LEFT);
+      else if (this_mask & BUFFER_BIT_BACK_LEFT)
+	 _mesa_DrawBuffer(GL_BACK_LEFT);
+      else if (color_bit != 0)
+	 _mesa_DrawBuffer(GL_COLOR_ATTACHMENT0 +
+			  (color_bit - BUFFER_COLOR0 - 1));
+      else
+	 _mesa_ColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
+
+      /* Control writing of the depth clear value to depth. */
+      if (this_mask & BUFFER_BIT_DEPTH) {
+	 _mesa_DepthFunc(GL_ALWAYS);
+	 _mesa_DepthMask(GL_TRUE);
+	 _mesa_Enable(GL_DEPTH_TEST);
+      } else {
+	 _mesa_Disable(GL_DEPTH_TEST);
+	 _mesa_DepthMask(GL_FALSE);
+      }
+
+      /* Control writing of the stencil clear value to stencil. */
+      if (this_mask & BUFFER_BIT_STENCIL) {
+	 _mesa_Enable(GL_STENCIL_TEST);
+	 _mesa_StencilOp(GL_REPLACE, GL_REPLACE, GL_REPLACE);
+	 _mesa_StencilFuncSeparate(GL_FRONT_AND_BACK, GL_ALWAYS, ctx->Stencil.Clear,
+				   ctx->Stencil.WriteMask[0]);
+      } else {
+	 _mesa_Disable(GL_STENCIL_TEST);
+      }
+
+      CALL_DrawArrays(ctx->Exec, (GL_TRIANGLE_FAN, 0, 4));
+
+      mask &= ~this_mask;
+   }
+
+   radeon_meta_restore_transform(rmesa);
+
+   _mesa_ActiveTextureARB(GL_TEXTURE0 + saved_active_texture);
+   if (saved_fp_enable)
+      _mesa_Enable(GL_FRAGMENT_PROGRAM_ARB);
+   if (saved_vp_enable)
+      _mesa_Enable(GL_VERTEX_PROGRAM_ARB);
+
+   if (saved_shader_program)
+      _mesa_UseProgramObjectARB(saved_shader_program);
+
+   _mesa_PopClientAttrib();
+   _mesa_PopAttrib();
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_common.h b/src/mesa/drivers/dri/radeon/radeon_common.h
new file mode 100644
index 00000000000..b60792df0be
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_common.h
@@ -0,0 +1,100 @@
+#ifndef COMMON_MISC_H
+#define COMMON_MISC_H
+
+#include "radeon_common_context.h"
+#include "radeon_dma.h"
+#include "radeon_texture.h"
+
+
+#define TRI_CLEAR_COLOR_BITS (BUFFER_BIT_BACK_LEFT |			\
+			      BUFFER_BIT_FRONT_LEFT |			\
+			      BUFFER_BIT_COLOR0 |			\
+			      BUFFER_BIT_COLOR1 |			\
+			      BUFFER_BIT_COLOR2 |			\
+			      BUFFER_BIT_COLOR3 |			\
+			      BUFFER_BIT_COLOR4 |			\
+			      BUFFER_BIT_COLOR5 |			\
+			      BUFFER_BIT_COLOR6 |			\
+			      BUFFER_BIT_COLOR7)
+
+void radeonRecalcScissorRects(radeonContextPtr radeon);
+void radeonSetCliprects(radeonContextPtr radeon);
+void radeonUpdateScissor( GLcontext *ctx );
+void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h);
+
+void radeonWaitForIdleLocked(radeonContextPtr radeon);
+extern uint32_t radeonGetAge(radeonContextPtr radeon);
+void radeonCopyBuffer( __DRIdrawablePrivate *dPriv,
+		       const drm_clip_rect_t	  *rect);
+void radeonSwapBuffers(__DRIdrawablePrivate * dPriv);
+void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
+			 int x, int y, int w, int h );
+
+void radeonUpdatePageFlipping(radeonContextPtr rmesa);
+
+void radeonFlush(GLcontext *ctx);
+void radeonFinish(GLcontext * ctx);
+void radeonEmitState(radeonContextPtr radeon);
+
+void radeon_clear_tris(GLcontext *ctx, GLbitfield mask);
+
+void radeon_window_moved(radeonContextPtr radeon);
+void radeon_draw_buffer(GLcontext *ctx, struct gl_framebuffer *fb);
+void radeonDrawBuffer( GLcontext *ctx, GLenum mode );
+void radeonReadBuffer( GLcontext *ctx, GLenum mode );
+void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height);
+void radeon_get_cliprects(radeonContextPtr radeon,
+			  struct drm_clip_rect **cliprects,
+			  unsigned int *num_cliprects,
+			  int *x_off, int *y_off);
+GLboolean radeon_revalidate_bos(GLcontext *ctx);
+void radeon_validate_bo(radeonContextPtr radeon, struct radeon_bo *bo, uint32_t read_domains, uint32_t write_domain);
+void radeon_validate_reset_bos(radeonContextPtr radeon);
+
+void radeon_fbo_init(struct radeon_context *radeon);
+void
+radeon_renderbuffer_set_bo(struct radeon_renderbuffer *rb,
+			   struct radeon_bo *bo);
+struct radeon_renderbuffer *
+radeon_create_renderbuffer(GLenum format, __DRIdrawablePrivate *driDrawPriv);
+static inline struct radeon_renderbuffer *radeon_renderbuffer(struct gl_renderbuffer *rb)
+{
+	struct radeon_renderbuffer *rrb = (struct radeon_renderbuffer *)rb;
+	if (rrb && rrb->base.ClassID == RADEON_RB_CLASS)
+		return rrb;
+	else
+		return NULL;
+}
+
+static inline struct radeon_renderbuffer *radeon_get_renderbuffer(struct gl_framebuffer *fb, int att_index)
+{
+	if (att_index >= 0)
+		return radeon_renderbuffer(fb->Attachment[att_index].Renderbuffer);
+	else
+		return NULL;
+}
+
+static inline struct radeon_renderbuffer *radeon_get_depthbuffer(radeonContextPtr rmesa)
+{
+	struct radeon_renderbuffer *rrb;
+	rrb = radeon_renderbuffer(rmesa->state.depth.rb);
+	if (!rrb)
+		return NULL;
+
+	return rrb;
+}
+
+static inline struct radeon_renderbuffer *radeon_get_colorbuffer(radeonContextPtr rmesa)
+{
+	struct radeon_renderbuffer *rrb;
+
+	rrb = radeon_renderbuffer(rmesa->state.color.rb);
+	if (!rrb)
+		return NULL;
+	return rrb;
+}
+
+#include "radeon_cmdbuf.h"
+
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.c b/src/mesa/drivers/dri/radeon/radeon_common_context.c
new file mode 100644
index 00000000000..eb0e5b35e5e
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_common_context.c
@@ -0,0 +1,738 @@
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+#include "radeon_common.h"
+#include "xmlpool.h"		/* for symbolic values of enum-type options */
+#include "utils.h"
+#include "vblank.h"
+#include "drirenderbuffer.h"
+#include "main/context.h"
+#include "main/framebuffer.h"
+#include "main/renderbuffer.h"
+#include "main/state.h"
+#include "main/simple_list.h"
+#include "swrast/swrast.h"
+#include "swrast_setup/swrast_setup.h"
+#include "tnl/tnl.h"
+
+#define DRIVER_DATE "20090101"
+
+#ifndef RADEON_DEBUG
+int RADEON_DEBUG = (0);
+#endif
+
+
+static const char* get_chip_family_name(int chip_family)
+{
+	switch(chip_family) {
+	case CHIP_FAMILY_R100: return "R100";
+	case CHIP_FAMILY_RV100: return "RV100";
+	case CHIP_FAMILY_RS100: return "RS100";
+	case CHIP_FAMILY_RV200: return "RV200";
+	case CHIP_FAMILY_RS200: return "RS200";
+	case CHIP_FAMILY_R200: return "R200";
+	case CHIP_FAMILY_RV250: return "RV250";
+	case CHIP_FAMILY_RS300: return "RS300";
+	case CHIP_FAMILY_RV280: return "RV280";
+	case CHIP_FAMILY_R300: return "R300";
+	case CHIP_FAMILY_R350: return "R350";
+	case CHIP_FAMILY_RV350: return "RV350";
+	case CHIP_FAMILY_RV380: return "RV380";
+	case CHIP_FAMILY_R420: return "R420";
+	case CHIP_FAMILY_RV410: return "RV410";
+	case CHIP_FAMILY_RS400: return "RS400";
+	case CHIP_FAMILY_RS600: return "RS600";
+	case CHIP_FAMILY_RS690: return "RS690";
+	case CHIP_FAMILY_RS740: return "RS740";
+	case CHIP_FAMILY_RV515: return "RV515";
+	case CHIP_FAMILY_R520: return "R520";
+	case CHIP_FAMILY_RV530: return "RV530";
+	case CHIP_FAMILY_R580: return "R580";
+	case CHIP_FAMILY_RV560: return "RV560";
+	case CHIP_FAMILY_RV570: return "RV570";
+	default: return "unknown";
+	}
+}
+
+
+/* Return various strings for glGetString().
+ */
+static const GLubyte *radeonGetString(GLcontext * ctx, GLenum name)
+{
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	static char buffer[128];
+
+	switch (name) {
+	case GL_VENDOR:
+		if (IS_R300_CLASS(radeon->radeonScreen))
+			return (GLubyte *) "DRI R300 Project";
+		else
+			return (GLubyte *) "Tungsten Graphics, Inc.";
+
+	case GL_RENDERER:
+	{
+		unsigned offset;
+		GLuint agp_mode = (radeon->radeonScreen->card_type==RADEON_CARD_PCI) ? 0 :
+			radeon->radeonScreen->AGPMode;
+		const char* chipclass;
+		char hardwarename[32];
+
+		if (IS_R300_CLASS(radeon->radeonScreen))
+			chipclass = "R300";
+		else if (IS_R200_CLASS(radeon->radeonScreen))
+			chipclass = "R200";
+		else
+			chipclass = "R100";
+
+		sprintf(hardwarename, "%s (%s %04X)",
+		        chipclass,
+		        get_chip_family_name(radeon->radeonScreen->chip_family),
+		        radeon->radeonScreen->device_id);
+
+		offset = driGetRendererString(buffer, hardwarename, DRIVER_DATE,
+					      agp_mode);
+
+		if (IS_R300_CLASS(radeon->radeonScreen)) {
+			sprintf(&buffer[offset], " %sTCL",
+				(radeon->radeonScreen->chip_flags & RADEON_CHIPSET_TCL)
+				? "" : "NO-");
+		} else {
+			sprintf(&buffer[offset], " %sTCL",
+				!(radeon->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE)
+				? "" : "NO-");
+		}
+
+		if (radeon->radeonScreen->driScreen->dri2.enabled)
+			strcat(buffer, " DRI2");
+
+		return (GLubyte *) buffer;
+	}
+
+	default:
+		return NULL;
+	}
+}
+
+/* Initialize the driver's misc functions.
+ */
+static void radeonInitDriverFuncs(struct dd_function_table *functions)
+{
+	functions->GetString = radeonGetString;
+}
+
+/**
+ * Create and initialize all common fields of the context,
+ * including the Mesa context itself.
+ */
+GLboolean radeonInitContext(radeonContextPtr radeon,
+			    struct dd_function_table* functions,
+			    const __GLcontextModes * glVisual,
+			    __DRIcontextPrivate * driContextPriv,
+			    void *sharedContextPrivate)
+{
+	__DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
+	radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private);
+	GLcontext* ctx;
+	GLcontext* shareCtx;
+	int fthrottle_mode;
+
+	/* Fill in additional standard functions. */
+	radeonInitDriverFuncs(functions);
+
+	radeon->radeonScreen = screen;
+	/* Allocate and initialize the Mesa context */
+	if (sharedContextPrivate)
+		shareCtx = ((radeonContextPtr)sharedContextPrivate)->glCtx;
+	else
+		shareCtx = NULL;
+	radeon->glCtx = _mesa_create_context(glVisual, shareCtx,
+					    functions, (void *)radeon);
+	if (!radeon->glCtx)
+		return GL_FALSE;
+
+	ctx = radeon->glCtx;
+	driContextPriv->driverPrivate = radeon;
+
+	/* DRI fields */
+	radeon->dri.context = driContextPriv;
+	radeon->dri.screen = sPriv;
+	radeon->dri.hwContext = driContextPriv->hHWContext;
+	radeon->dri.hwLock = &sPriv->pSAREA->lock;
+	radeon->dri.fd = sPriv->fd;
+	radeon->dri.drmMinor = sPriv->drm_version.minor;
+
+	radeon->sarea = (drm_radeon_sarea_t *) ((GLubyte *) sPriv->pSAREA +
+					       screen->sarea_priv_offset);
+
+	/* Setup IRQs */
+	fthrottle_mode = driQueryOptioni(&radeon->optionCache, "fthrottle_mode");
+	radeon->iw.irq_seq = -1;
+	radeon->irqsEmitted = 0;
+	radeon->do_irqs = (fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS &&
+			  radeon->radeonScreen->irq);
+
+	radeon->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS);
+
+	if (!radeon->do_irqs)
+		fprintf(stderr,
+			"IRQ's not enabled, falling back to %s: %d %d\n",
+			radeon->do_usleeps ? "usleeps" : "busy waits",
+			fthrottle_mode, radeon->radeonScreen->irq);
+
+        radeon->texture_depth = driQueryOptioni (&radeon->optionCache,
+					        "texture_depth");
+        if (radeon->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
+                radeon->texture_depth = ( glVisual->rgbBits > 16 ) ?
+	        DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16;
+
+	radeon->texture_row_align = 32;
+
+	return GL_TRUE;
+}
+
+
+
+/**
+ * Destroy the command buffer and state atoms.
+ */
+static void radeon_destroy_atom_list(radeonContextPtr radeon)
+{
+	struct radeon_state_atom *atom;
+
+	foreach(atom, &radeon->hw.atomlist) {
+		FREE(atom->cmd);
+		if (atom->lastcmd)
+			FREE(atom->lastcmd);
+	}
+
+}
+
+/**
+ * Cleanup common context fields.
+ * Called by r200DestroyContext/r300DestroyContext
+ */
+void radeonDestroyContext(__DRIcontextPrivate *driContextPriv )
+{
+#ifdef RADEON_BO_TRACK
+	FILE *track;
+#endif
+	GET_CURRENT_CONTEXT(ctx);
+	radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate;
+	radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL;
+
+	if (radeon == current) {
+		radeon_firevertices(radeon);
+		_mesa_make_current(NULL, NULL, NULL);
+	}
+
+	assert(radeon);
+	if (radeon) {
+
+		if (radeon->dma.current) {
+			rcommonFlushCmdBuf( radeon, __FUNCTION__ );
+		}
+
+		radeonReleaseArrays(radeon->glCtx, ~0);
+
+		if (radeon->vtbl.free_context)
+			radeon->vtbl.free_context(radeon->glCtx);
+		_swsetup_DestroyContext( radeon->glCtx );
+		_tnl_DestroyContext( radeon->glCtx );
+		_vbo_DestroyContext( radeon->glCtx );
+		_swrast_DestroyContext( radeon->glCtx );
+
+		/* free atom list */
+		/* free the Mesa context */
+		_mesa_destroy_context(radeon->glCtx);
+
+		/* _mesa_destroy_context() might result in calls to functions that
+		 * depend on the DriverCtx, so don't set it to NULL before.
+		 *
+		 * radeon->glCtx->DriverCtx = NULL;
+		 */
+		/* free the option cache */
+		driDestroyOptionCache(&radeon->optionCache);
+
+		rcommonDestroyCmdBuf(radeon);
+
+		radeon_destroy_atom_list(radeon);
+
+		if (radeon->state.scissor.pClipRects) {
+			FREE(radeon->state.scissor.pClipRects);
+			radeon->state.scissor.pClipRects = 0;
+		}
+	}
+#ifdef RADEON_BO_TRACK
+	track = fopen("/tmp/tracklog", "w");
+	if (track) {
+		radeon_tracker_print(&radeon->radeonScreen->bom->tracker, track);
+		fclose(track);
+	}
+#endif
+	FREE(radeon);
+}
+
+/* Force the context `c' to be unbound from its buffer.
+ */
+GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv)
+{
+	radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate;
+
+	if (RADEON_DEBUG & DEBUG_DRI)
+		fprintf(stderr, "%s ctx %p\n", __FUNCTION__,
+			radeon->glCtx);
+
+	return GL_TRUE;
+}
+
+
+static void
+radeon_make_kernel_renderbuffer_current(radeonContextPtr radeon,
+					struct radeon_framebuffer *draw)
+{
+	/* if radeon->fake */
+	struct radeon_renderbuffer *rb;
+
+	if ((rb = (void *)draw->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->frontOffset,
+						0,
+						0,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->frontPitch * rb->cpp;
+	}
+	if ((rb = (void *)draw->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->backOffset,
+						0,
+						0,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->backPitch * rb->cpp;
+	}
+	if ((rb = (void *)draw->base.Attachment[BUFFER_DEPTH].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->depthOffset,
+						0,
+						0,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp;
+	}
+	if ((rb = (void *)draw->base.Attachment[BUFFER_STENCIL].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->depthOffset,
+						0,
+						0,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp;
+	}
+}
+
+static void
+radeon_make_renderbuffer_current(radeonContextPtr radeon,
+				 struct radeon_framebuffer *draw)
+{
+	int size = 4096*4096*4;
+	/* if radeon->fake */
+	struct radeon_renderbuffer *rb;
+
+	if (radeon->radeonScreen->kernel_mm) {
+		radeon_make_kernel_renderbuffer_current(radeon, draw);
+		return;
+	}
+
+
+	if ((rb = (void *)draw->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->frontOffset +
+						radeon->radeonScreen->fbLocation,
+						size,
+						4096,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->frontPitch * rb->cpp;
+	}
+	if ((rb = (void *)draw->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->backOffset +
+						radeon->radeonScreen->fbLocation,
+						size,
+						4096,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->backPitch * rb->cpp;
+	}
+	if ((rb = (void *)draw->base.Attachment[BUFFER_DEPTH].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->depthOffset +
+						radeon->radeonScreen->fbLocation,
+						size,
+						4096,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp;
+	}
+	if ((rb = (void *)draw->base.Attachment[BUFFER_STENCIL].Renderbuffer)) {
+		if (!rb->bo) {
+			rb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+						radeon->radeonScreen->depthOffset +
+						radeon->radeonScreen->fbLocation,
+						size,
+						4096,
+						RADEON_GEM_DOMAIN_VRAM,
+						0);
+		}
+		rb->cpp = radeon->radeonScreen->cpp;
+		rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp;
+	}
+}
+
+static unsigned
+radeon_bits_per_pixel(const struct radeon_renderbuffer *rb)
+{
+   switch (rb->base._ActualFormat) {
+   case GL_RGB5:
+   case GL_DEPTH_COMPONENT16:
+      return 16;
+   case GL_RGB8:
+   case GL_RGBA8:
+   case GL_DEPTH_COMPONENT24:
+   case GL_DEPTH24_STENCIL8_EXT:
+   case GL_STENCIL_INDEX8_EXT:
+      return 32;
+   default:
+      return 0;
+   }
+}
+
+void
+radeon_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable)
+{
+	unsigned int attachments[10];
+	__DRIbuffer *buffers = NULL;
+	__DRIscreen *screen;
+	struct radeon_renderbuffer *rb;
+	int i, count;
+	struct radeon_framebuffer *draw;
+	radeonContextPtr radeon;
+	char *regname;
+	struct radeon_bo *depth_bo = NULL, *bo;
+
+	if (RADEON_DEBUG & DEBUG_DRI)
+	    fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
+
+	draw = drawable->driverPrivate;
+	screen = context->driScreenPriv;
+	radeon = (radeonContextPtr) context->driverPrivate;
+
+	if (screen->dri2.loader
+	   && (screen->dri2.loader->base.version > 2)
+	   && (screen->dri2.loader->getBuffersWithFormat != NULL)) {
+		struct radeon_renderbuffer *depth_rb;
+		struct radeon_renderbuffer *stencil_rb;
+
+		i = 0;
+		if ((radeon->is_front_buffer_rendering || !draw->color_rb[1])
+			&& draw->color_rb[0]) {
+			attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
+			attachments[i++] = radeon_bits_per_pixel(draw->color_rb[0]);
+		}
+
+		if (draw->color_rb[1]) {
+			attachments[i++] = __DRI_BUFFER_BACK_LEFT;
+			attachments[i++] = radeon_bits_per_pixel(draw->color_rb[1]);
+		}
+
+		depth_rb = radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH);
+		stencil_rb = radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL);
+
+		if ((depth_rb != NULL) && (stencil_rb != NULL)) {
+			attachments[i++] = __DRI_BUFFER_DEPTH_STENCIL;
+			attachments[i++] = radeon_bits_per_pixel(depth_rb);
+		} else if (depth_rb != NULL) {
+			attachments[i++] = __DRI_BUFFER_DEPTH;
+			attachments[i++] = radeon_bits_per_pixel(depth_rb);
+		} else if (stencil_rb != NULL) {
+			attachments[i++] = __DRI_BUFFER_STENCIL;
+			attachments[i++] = radeon_bits_per_pixel(stencil_rb);
+		}
+
+		buffers = (*screen->dri2.loader->getBuffersWithFormat)(drawable,
+								&drawable->w,
+								&drawable->h,
+								attachments, i / 2,
+								&count,
+								drawable->loaderPrivate);
+	} else if (screen->dri2.loader) {
+		i = 0;
+		if (draw->color_rb[0])
+			attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
+		if (draw->color_rb[1])
+			attachments[i++] = __DRI_BUFFER_BACK_LEFT;
+		if (radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH))
+			attachments[i++] = __DRI_BUFFER_DEPTH;
+		if (radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL))
+			attachments[i++] = __DRI_BUFFER_STENCIL;
+
+		buffers = (*screen->dri2.loader->getBuffers)(drawable,
+								 &drawable->w,
+								 &drawable->h,
+								 attachments, i,
+								 &count,
+								 drawable->loaderPrivate);
+	}
+
+	if (buffers == NULL)
+		return;
+
+	/* set one cliprect to cover the whole drawable */
+	drawable->x = 0;
+	drawable->y = 0;
+	drawable->backX = 0;
+	drawable->backY = 0;
+	drawable->numClipRects = 1;
+	drawable->pClipRects[0].x1 = 0;
+	drawable->pClipRects[0].y1 = 0;
+	drawable->pClipRects[0].x2 = drawable->w;
+	drawable->pClipRects[0].y2 = drawable->h;
+	drawable->numBackClipRects = 1;
+	drawable->pBackClipRects[0].x1 = 0;
+	drawable->pBackClipRects[0].y1 = 0;
+	drawable->pBackClipRects[0].x2 = drawable->w;
+	drawable->pBackClipRects[0].y2 = drawable->h;
+	for (i = 0; i < count; i++) {
+		switch (buffers[i].attachment) {
+		case __DRI_BUFFER_FRONT_LEFT:
+			rb = draw->color_rb[0];
+			regname = "dri2 front buffer";
+			break;
+		case __DRI_BUFFER_FAKE_FRONT_LEFT:
+			rb = draw->color_rb[0];
+			regname = "dri2 fake front buffer";
+			break;
+		case __DRI_BUFFER_BACK_LEFT:
+			rb = draw->color_rb[1];
+			regname = "dri2 back buffer";
+			break;
+		case __DRI_BUFFER_DEPTH:
+			rb = radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH);
+			regname = "dri2 depth buffer";
+			break;
+		case __DRI_BUFFER_DEPTH_STENCIL:
+			rb = radeon_get_renderbuffer(&draw->base, BUFFER_DEPTH);
+			regname = "dri2 depth / stencil buffer";
+			break;
+		case __DRI_BUFFER_STENCIL:
+			rb = radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL);
+			regname = "dri2 stencil buffer";
+			break;
+		case __DRI_BUFFER_ACCUM:
+		default:
+			fprintf(stderr,
+				"unhandled buffer attach event, attacment type %d\n",
+				buffers[i].attachment);
+			return;
+		}
+
+		if (rb == NULL)
+			continue;
+
+		if (rb->bo) {
+			uint32_t name = radeon_gem_name_bo(rb->bo);
+			if (name == buffers[i].name)
+				continue;
+		}
+
+		if (RADEON_DEBUG & DEBUG_DRI)
+			fprintf(stderr,
+				"attaching buffer %s, %d, at %d, cpp %d, pitch %d\n",
+				regname, buffers[i].name, buffers[i].attachment,
+				buffers[i].cpp, buffers[i].pitch);
+
+		rb->cpp = buffers[i].cpp;
+		rb->pitch = buffers[i].pitch;
+		rb->width = drawable->w;
+		rb->height = drawable->h;
+		rb->has_surface = 0;
+
+		if (buffers[i].attachment == __DRI_BUFFER_STENCIL && depth_bo) {
+			if (RADEON_DEBUG & DEBUG_DRI)
+				fprintf(stderr, "(reusing depth buffer as stencil)\n");
+			bo = depth_bo;
+			radeon_bo_ref(bo);
+		} else {
+			bo = radeon_bo_open(radeon->radeonScreen->bom,
+						buffers[i].name,
+						0,
+						0,
+						RADEON_GEM_DOMAIN_VRAM,
+						buffers[i].flags);
+			if (bo == NULL) {
+
+				fprintf(stderr, "failed to attach %s %d\n",
+					regname, buffers[i].name);
+
+			}
+		}
+
+		if (buffers[i].attachment == __DRI_BUFFER_DEPTH) {
+			if (draw->base.Visual.depthBits == 16)
+				rb->cpp = 2;
+			depth_bo = bo;
+		}
+
+		radeon_renderbuffer_set_bo(rb, bo);
+		radeon_bo_unref(bo);
+
+		if (buffers[i].attachment == __DRI_BUFFER_DEPTH_STENCIL) {
+			rb = radeon_get_renderbuffer(&draw->base, BUFFER_STENCIL);
+			if (rb != NULL) {
+				struct radeon_bo *stencil_bo = NULL;
+
+				if (rb->bo) {
+					uint32_t name = radeon_gem_name_bo(rb->bo);
+					if (name == buffers[i].name)
+						continue;
+				}
+
+				stencil_bo = bo;
+				radeon_bo_ref(stencil_bo);
+				radeon_renderbuffer_set_bo(rb, stencil_bo);
+				radeon_bo_unref(stencil_bo);
+			}
+		}
+	}
+
+	driUpdateFramebufferSize(radeon->glCtx, drawable);
+}
+
+/* Force the context `c' to be the current context and associate with it
+ * buffer `b'.
+ */
+GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv,
+			    __DRIdrawablePrivate * driDrawPriv,
+			    __DRIdrawablePrivate * driReadPriv)
+{
+	radeonContextPtr radeon;
+	struct radeon_framebuffer *drfb;
+	struct gl_framebuffer *readfb;
+
+	if (!driContextPriv) {
+		if (RADEON_DEBUG & DEBUG_DRI)
+			fprintf(stderr, "%s ctx is null\n", __FUNCTION__);
+		_mesa_make_current(NULL, NULL, NULL);
+		return GL_TRUE;
+	}
+
+	radeon = (radeonContextPtr) driContextPriv->driverPrivate;
+	drfb = driDrawPriv->driverPrivate;
+	readfb = driReadPriv->driverPrivate;
+
+	if (driContextPriv->driScreenPriv->dri2.enabled) {
+		radeon_update_renderbuffers(driContextPriv, driDrawPriv);
+		if (driDrawPriv != driReadPriv)
+			radeon_update_renderbuffers(driContextPriv, driReadPriv);
+		_mesa_reference_renderbuffer(&radeon->state.color.rb,
+			&(radeon_get_renderbuffer(&drfb->base, BUFFER_BACK_LEFT)->base));
+		_mesa_reference_renderbuffer(&radeon->state.depth.rb,
+			&(radeon_get_renderbuffer(&drfb->base, BUFFER_DEPTH)->base));
+	} else {
+		radeon_make_renderbuffer_current(radeon, drfb);
+	}
+
+
+	if (RADEON_DEBUG & DEBUG_DRI)
+	     fprintf(stderr, "%s ctx %p dfb %p rfb %p\n", __FUNCTION__, radeon->glCtx, drfb, readfb);
+
+	driUpdateFramebufferSize(radeon->glCtx, driDrawPriv);
+	if (driReadPriv != driDrawPriv)
+		driUpdateFramebufferSize(radeon->glCtx, driReadPriv);
+
+	_mesa_make_current(radeon->glCtx, &drfb->base, readfb);
+
+	_mesa_update_state(radeon->glCtx);
+
+	if (radeon->glCtx->DrawBuffer == &drfb->base) {
+		if (driDrawPriv->swap_interval == (unsigned)-1) {
+			int i;
+			driDrawPriv->vblFlags =
+				(radeon->radeonScreen->irq != 0)
+				? driGetDefaultVBlankFlags(&radeon->
+							   optionCache)
+				: VBLANK_FLAG_NO_IRQ;
+
+			driDrawableInitVBlank(driDrawPriv);
+			drfb->vbl_waited = driDrawPriv->vblSeq;
+
+			for (i = 0; i < 2; i++) {
+				if (drfb->color_rb[i])
+					drfb->color_rb[i]->vbl_pending = driDrawPriv->vblSeq;
+			}
+
+		}
+
+		radeon_window_moved(radeon);
+		radeon_draw_buffer(radeon->glCtx, &drfb->base);
+	}
+
+
+	if (RADEON_DEBUG & DEBUG_DRI)
+		fprintf(stderr, "End %s\n", __FUNCTION__);
+	return GL_TRUE;
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.h b/src/mesa/drivers/dri/radeon/radeon_common_context.h
new file mode 100644
index 00000000000..061168fe96b
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_common_context.h
@@ -0,0 +1,581 @@
+
+#ifndef COMMON_CONTEXT_H
+#define COMMON_CONTEXT_H
+
+#include "main/mm.h"
+#include "math/m_vector.h"
+#include "texmem.h"
+#include "tnl/t_context.h"
+#include "main/colormac.h"
+
+#include "radeon_screen.h"
+#include "radeon_drm.h"
+#include "dri_util.h"
+#include "tnl/t_vertex.h"
+
+struct radeon_context;
+
+#include "radeon_bocs_wrapper.h"
+
+/* This union is used to avoid warnings/miscompilation
+   with float to uint32_t casts due to strict-aliasing */
+typedef union { GLfloat f; uint32_t ui32; } float_ui32_type;
+
+struct radeon_context;
+typedef struct radeon_context radeonContextRec;
+typedef struct radeon_context *radeonContextPtr;
+
+
+#define TEX_0   0x1
+#define TEX_1   0x2
+#define TEX_2   0x4
+#define TEX_3	0x8
+#define TEX_4	0x10
+#define TEX_5	0x20
+
+/* Rasterizing fallbacks */
+/* See correponding strings in r200_swtcl.c */
+#define RADEON_FALLBACK_TEXTURE		0x0001
+#define RADEON_FALLBACK_DRAW_BUFFER	0x0002
+#define RADEON_FALLBACK_STENCIL		0x0004
+#define RADEON_FALLBACK_RENDER_MODE	0x0008
+#define RADEON_FALLBACK_BLEND_EQ	0x0010
+#define RADEON_FALLBACK_BLEND_FUNC	0x0020
+#define RADEON_FALLBACK_DISABLE 	0x0040
+#define RADEON_FALLBACK_BORDER_MODE	0x0080
+#define RADEON_FALLBACK_DEPTH_BUFFER	0x0100
+#define RADEON_FALLBACK_STENCIL_BUFFER  0x0200
+
+#define R200_FALLBACK_TEXTURE           0x01
+#define R200_FALLBACK_DRAW_BUFFER       0x02
+#define R200_FALLBACK_STENCIL           0x04
+#define R200_FALLBACK_RENDER_MODE       0x08
+#define R200_FALLBACK_DISABLE           0x10
+#define R200_FALLBACK_BORDER_MODE       0x20
+
+#define RADEON_TCL_FALLBACK_RASTER            0x1 /* rasterization */
+#define RADEON_TCL_FALLBACK_UNFILLED          0x2 /* unfilled tris */
+#define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE     0x4 /* twoside tris */
+#define RADEON_TCL_FALLBACK_MATERIAL          0x8 /* material in vb */
+#define RADEON_TCL_FALLBACK_TEXGEN_0          0x10 /* texgen, unit 0 */
+#define RADEON_TCL_FALLBACK_TEXGEN_1          0x20 /* texgen, unit 1 */
+#define RADEON_TCL_FALLBACK_TEXGEN_2          0x40 /* texgen, unit 2 */
+#define RADEON_TCL_FALLBACK_TCL_DISABLE       0x80 /* user disable */
+#define RADEON_TCL_FALLBACK_FOGCOORDSPEC      0x100 /* fogcoord, sep. spec light */
+
+/* The blit width for texture uploads
+ */
+#define BLIT_WIDTH_BYTES 1024
+
+/* Use the templated vertex format:
+ */
+#define COLOR_IS_RGBA
+#define TAG(x) radeon##x
+#include "tnl_dd/t_dd_vertex.h"
+#undef TAG
+
+#define RADEON_RB_CLASS 0xdeadbeef
+
+struct radeon_renderbuffer
+{
+	struct gl_renderbuffer base;
+	struct radeon_bo *bo;
+	unsigned int cpp;
+	/* unsigned int offset; */
+	unsigned int pitch;
+	unsigned int width;
+	unsigned int height;
+
+	uint32_t draw_offset; /* FBO */
+	/* boo Xorg 6.8.2 compat */
+	int has_surface;
+
+	GLuint pf_pending;  /**< sequence number of pending flip */
+	GLuint vbl_pending;   /**< vblank sequence number of pending flip */
+	__DRIdrawablePrivate *dPriv;
+};
+
+struct radeon_framebuffer
+{
+	struct gl_framebuffer base;
+
+	struct radeon_renderbuffer *color_rb[2];
+
+	GLuint vbl_waited;
+
+	/* buffer swap */
+	int64_t swap_ust;
+	int64_t swap_missed_ust;
+
+	GLuint swap_count;
+	GLuint swap_missed_count;
+
+	/* Drawable page flipping state */
+	GLboolean pf_active;
+	GLint pf_current_page;
+	GLint pf_num_pages;
+
+};
+
+
+struct radeon_colorbuffer_state {
+	GLuint clear;
+	int roundEnable;
+	struct gl_renderbuffer *rb;
+	uint32_t draw_offset; /* offset into color renderbuffer - FBOs */
+};
+
+struct radeon_depthbuffer_state {
+	GLuint clear;
+	struct gl_renderbuffer *rb;
+};
+
+struct radeon_scissor_state {
+	drm_clip_rect_t rect;
+	GLboolean enabled;
+
+	GLuint numClipRects;	/* Cliprects active */
+	GLuint numAllocedClipRects;	/* Cliprects available */
+	drm_clip_rect_t *pClipRects;
+};
+
+struct radeon_stencilbuffer_state {
+	GLuint clear;		/* rb3d_stencilrefmask value */
+};
+
+struct radeon_stipple_state {
+	GLuint mask[32];
+};
+
+struct radeon_state_atom {
+	struct radeon_state_atom *next, *prev;
+	const char *name;	/* for debug */
+	int cmd_size;		/* size in bytes */
+        GLuint idx;
+	GLuint is_tcl;
+        GLuint *cmd;		/* one or more cmd's */
+	GLuint *lastcmd;		/* one or more cmd's */
+	GLboolean dirty;	/* dirty-mark in emit_state_list */
+        int (*check) (GLcontext *, struct radeon_state_atom *atom); /* is this state active? */
+        void (*emit) (GLcontext *, struct radeon_state_atom *atom);
+};
+
+struct radeon_hw_state {
+  	/* Head of the linked list of state atoms. */
+	struct radeon_state_atom atomlist;
+	int max_state_size;	/* Number of bytes necessary for a full state emit. */
+	GLboolean is_dirty, all_dirty;
+};
+
+
+/* Texture related */
+typedef struct _radeon_texture_image radeon_texture_image;
+
+struct _radeon_texture_image {
+	struct gl_texture_image base;
+
+	/**
+	 * If mt != 0, the image is stored in hardware format in the
+	 * given mipmap tree. In this case, base.Data may point into the
+	 * mapping of the buffer object that contains the mipmap tree.
+	 *
+	 * If mt == 0, the image is stored in normal memory pointed to
+	 * by base.Data.
+	 */
+	struct _radeon_mipmap_tree *mt;
+	struct radeon_bo *bo;
+
+	int mtlevel; /** if mt != 0, this is the image's level in the mipmap tree */
+	int mtface; /** if mt != 0, this is the image's face in the mipmap tree */
+};
+
+
+static INLINE radeon_texture_image *get_radeon_texture_image(struct gl_texture_image *image)
+{
+	return (radeon_texture_image*)image;
+}
+
+
+typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
+
+#define RADEON_TXO_MICRO_TILE               (1 << 3)
+
+/* Texture object in locally shared texture space.
+ */
+struct radeon_tex_obj {
+	struct gl_texture_object base;
+	struct _radeon_mipmap_tree *mt;
+
+	/**
+	 * This is true if we've verified that the mipmap tree above is complete
+	 * and so on.
+	 */
+	GLboolean validated;
+
+	GLuint override_offset;
+	GLboolean image_override; /* Image overridden by GLX_EXT_tfp */
+	GLuint tile_bits;	/* hw texture tile bits used on this texture */
+        struct radeon_bo *bo;
+
+	GLuint pp_txfilter;	/* hardware register values */
+	GLuint pp_txformat;
+	GLuint pp_txformat_x;
+	GLuint pp_txsize;	/* npot only */
+	GLuint pp_txpitch;	/* npot only */
+	GLuint pp_border_color;
+	GLuint pp_cubic_faces;	/* cube face 1,2,3,4 log2 sizes */
+
+        GLuint pp_txfilter_1;	/*  r300 */
+
+	GLboolean border_fallback;
+
+
+};
+
+static INLINE radeonTexObj* radeon_tex_obj(struct gl_texture_object *texObj)
+{
+	return (radeonTexObj*)texObj;
+}
+
+/* Need refcounting on dma buffers:
+ */
+struct radeon_dma_buffer {
+	int refcount;		/* the number of retained regions in buf */
+	drmBufPtr buf;
+};
+
+struct radeon_aos {
+	struct radeon_bo *bo; /** Buffer object where vertex data is stored */
+	int offset; /** Offset into buffer object, in bytes */
+	int components; /** Number of components per vertex */
+	int stride; /** Stride in dwords (may be 0 for repeating) */
+	int count; /** Number of vertices */
+};
+
+struct radeon_dma {
+        /* Active dma region.  Allocations for vertices and retained
+         * regions come from here.  Also used for emitting random vertices,
+         * these may be flushed by calling flush_current();
+         */
+        struct radeon_bo *current; /** Buffer that DMA memory is allocated from */
+        int current_used; /** Number of bytes allocated and forgotten about */
+        int current_vertexptr; /** End of active vertex region */
+
+        /**
+         * If current_vertexptr != current_used then flush must be non-zero.
+         * flush must be called before non-active vertex allocations can be
+         * performed.
+         */
+        void (*flush) (GLcontext *);
+
+        /* Number of "in-flight" DMA buffers, i.e. the number of buffers
+         * for which a DISCARD command is currently queued in the command buffer
+.
+         */
+        GLuint nr_released_bufs;
+};
+
+/* radeon_swtcl.c
+ */
+struct radeon_swtcl_info {
+
+	GLuint RenderIndex;
+	GLuint vertex_size;
+	GLubyte *verts;
+
+	/* Fallback rasterization functions
+	 */
+	GLuint hw_primitive;
+	GLenum render_primitive;
+	GLuint numverts;
+
+	struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
+	GLuint vertex_attr_count;
+
+};
+
+#define RADEON_MAX_AOS_ARRAYS		16
+struct radeon_tcl_info {
+	struct radeon_aos aos[RADEON_MAX_AOS_ARRAYS];
+	GLuint aos_count;
+	struct radeon_bo *elt_dma_bo; /** Buffer object that contains element indices */
+	int elt_dma_offset; /** Offset into this buffer object, in bytes */
+};
+
+struct radeon_ioctl {
+	GLuint vertex_offset;
+        struct radeon_bo *bo;
+	GLuint vertex_size;
+};
+
+#define RADEON_MAX_PRIMS 64
+
+struct radeon_prim {
+	GLuint start;
+	GLuint end;
+	GLuint prim;
+};
+
+static INLINE GLuint radeonPackColor(GLuint cpp,
+                                     GLubyte r, GLubyte g,
+                                     GLubyte b, GLubyte a)
+{
+	switch (cpp) {
+	case 2:
+		return PACK_COLOR_565(r, g, b);
+	case 4:
+		return PACK_COLOR_8888(a, r, g, b);
+	default:
+		return 0;
+	}
+}
+
+#define MAX_CMD_BUF_SZ (16*1024)
+
+#define MAX_DMA_BUF_SZ (64*1024)
+
+struct radeon_store {
+	GLuint statenr;
+	GLuint primnr;
+	char cmd_buf[MAX_CMD_BUF_SZ];
+	int cmd_used;
+	int elts_start;
+};
+
+struct radeon_dri_mirror {
+	__DRIcontextPrivate *context;	/* DRI context */
+	__DRIscreenPrivate *screen;	/* DRI screen */
+
+	drm_context_t hwContext;
+	drm_hw_lock_t *hwLock;
+	int fd;
+	int drmMinor;
+};
+
+#define DEBUG_TEXTURE	0x001
+#define DEBUG_STATE	0x002
+#define DEBUG_IOCTL	0x004
+#define DEBUG_PRIMS	0x008
+#define DEBUG_VERTS	0x010
+#define DEBUG_FALLBACKS	0x020
+#define DEBUG_VFMT	0x040
+#define DEBUG_CODEGEN	0x080
+#define DEBUG_VERBOSE	0x100
+#define DEBUG_DRI       0x200
+#define DEBUG_DMA       0x400
+#define DEBUG_SANITY    0x800
+#define DEBUG_SYNC      0x1000
+#define DEBUG_PIXEL     0x2000
+#define DEBUG_MEMORY    0x4000
+
+
+typedef void (*radeon_tri_func) (radeonContextPtr,
+				 radeonVertex *,
+				 radeonVertex *, radeonVertex *);
+
+typedef void (*radeon_line_func) (radeonContextPtr,
+				  radeonVertex *, radeonVertex *);
+
+typedef void (*radeon_point_func) (radeonContextPtr, radeonVertex *);
+
+#define RADEON_MAX_BOS 32
+struct radeon_state {
+	struct radeon_colorbuffer_state color;
+	struct radeon_depthbuffer_state depth;
+	struct radeon_scissor_state scissor;
+	struct radeon_stencilbuffer_state stencil;
+
+	struct radeon_cs_space_check bos[RADEON_MAX_BOS];
+	int validated_bo_count;
+};
+
+/**
+ * This structure holds the command buffer while it is being constructed.
+ *
+ * The first batch of commands in the buffer is always the state that needs
+ * to be re-emitted when the context is lost. This batch can be skipped
+ * otherwise.
+ */
+struct radeon_cmdbuf {
+	struct radeon_cs_manager    *csm;
+	struct radeon_cs            *cs;
+	int size; /** # of dwords total */
+	unsigned int flushing:1; /** whether we're currently in FlushCmdBufLocked */
+};
+
+struct radeon_context {
+   GLcontext *glCtx;
+   radeonScreenPtr radeonScreen;	/* Screen private DRI data */
+
+   /* Texture object bookkeeping
+    */
+   int                   texture_depth;
+   float                 initialMaxAnisotropy;
+   uint32_t              texture_row_align;
+
+  struct radeon_dma dma;
+  struct radeon_hw_state hw;
+   /* Rasterization and vertex state:
+    */
+   GLuint TclFallback;
+   GLuint Fallback;
+   GLuint NewGLState;
+   DECLARE_RENDERINPUTS(tnl_index_bitset);	/* index of bits for last tnl_install_attrs */
+
+   /* Drawable, cliprect and scissor information */
+   GLuint numClipRects;	/* Cliprects for the draw buffer */
+   drm_clip_rect_t *pClipRects;
+   unsigned int lastStamp;
+   GLboolean lost_context;
+   drm_radeon_sarea_t *sarea;	/* Private SAREA data */
+
+   /* Mirrors of some DRI state */
+   struct radeon_dri_mirror dri;
+
+   /* Busy waiting */
+   GLuint do_usleeps;
+   GLuint do_irqs;
+   GLuint irqsEmitted;
+   drm_radeon_irq_wait_t iw;
+
+   /* Derived state - for r300 only */
+   struct radeon_state state;
+
+   struct radeon_swtcl_info swtcl;
+   struct radeon_tcl_info tcl;
+   /* Configuration cache
+    */
+   driOptionCache optionCache;
+
+   struct radeon_cmdbuf cmdbuf;
+
+  drm_clip_rect_t fboRect;
+  GLboolean constant_cliprect; /* use for FBO or DRI2 rendering */
+  GLboolean front_cliprects;
+
+   /**
+    * Set if rendering has occured to the drawable's front buffer.
+    *
+    * This is used in the DRI2 case to detect that glFlush should also copy
+    * the contents of the fake front buffer to the real front buffer.
+    */
+   GLboolean front_buffer_dirty;
+
+   /**
+    * Track whether front-buffer rendering is currently enabled
+    *
+    * A separate flag is used to track this in order to support MRT more
+    * easily.
+    */
+   GLboolean is_front_buffer_rendering;
+
+  struct {
+      struct gl_fragment_program *bitmap_fp;
+      struct gl_vertex_program *passthrough_vp;
+
+      struct gl_fragment_program *saved_fp;
+      GLboolean saved_fp_enable;
+      struct gl_vertex_program *saved_vp;
+      GLboolean saved_vp_enable;
+
+      GLint saved_vp_x, saved_vp_y;
+      GLsizei saved_vp_width, saved_vp_height;
+      GLenum saved_matrix_mode;
+   } meta;
+
+   struct {
+	   void (*get_lock)(radeonContextPtr radeon);
+	   void (*update_viewport_offset)(GLcontext *ctx);
+	   void (*emit_cs_header)(struct radeon_cs *cs, radeonContextPtr rmesa);
+	   void (*swtcl_flush)(GLcontext *ctx, uint32_t offset);
+	   void (*pre_emit_atoms)(radeonContextPtr rmesa);
+	   void (*pre_emit_state)(radeonContextPtr rmesa);
+	   void (*fallback)(GLcontext *ctx, GLuint bit, GLboolean mode);
+	   void (*free_context)(GLcontext *ctx);
+   } vtbl;
+};
+
+#define RADEON_CONTEXT(glctx) ((radeonContextPtr)(ctx->DriverCtx))
+
+static inline __DRIdrawablePrivate* radeon_get_drawable(radeonContextPtr radeon)
+{
+	return radeon->dri.context->driDrawablePriv;
+}
+
+static inline __DRIdrawablePrivate* radeon_get_readable(radeonContextPtr radeon)
+{
+	return radeon->dri.context->driReadablePriv;
+}
+
+
+/**
+ * This function takes a float and packs it into a uint32_t
+ */
+static INLINE uint32_t radeonPackFloat32(float fl)
+{
+	union {
+		float fl;
+		uint32_t u;
+	} u;
+
+	u.fl = fl;
+	return u.u;
+}
+
+/* This is probably wrong for some values, I need to test this
+ * some more.  Range checking would be a good idea also..
+ *
+ * But it works for most things.  I'll fix it later if someone
+ * else with a better clue doesn't
+ */
+static INLINE uint32_t radeonPackFloat24(float f)
+{
+	float mantissa;
+	int exponent;
+	uint32_t float24 = 0;
+
+	if (f == 0.0)
+		return 0;
+
+	mantissa = frexpf(f, &exponent);
+
+	/* Handle -ve */
+	if (mantissa < 0) {
+		float24 |= (1 << 23);
+		mantissa = mantissa * -1.0;
+	}
+	/* Handle exponent, bias of 63 */
+	exponent += 62;
+	float24 |= (exponent << 16);
+	/* Kill 7 LSB of mantissa */
+	float24 |= (radeonPackFloat32(mantissa) & 0x7FFFFF) >> 7;
+
+	return float24;
+}
+
+GLboolean radeonInitContext(radeonContextPtr radeon,
+			    struct dd_function_table* functions,
+			    const __GLcontextModes * glVisual,
+			    __DRIcontextPrivate * driContextPriv,
+			    void *sharedContextPrivate);
+
+void radeonCleanupContext(radeonContextPtr radeon);
+GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv);
+void radeon_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable);
+GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv,
+			    __DRIdrawablePrivate * driDrawPriv,
+			    __DRIdrawablePrivate * driReadPriv);
+extern void radeonDestroyContext(__DRIcontextPrivate * driContextPriv);
+
+/* ================================================================
+ * Debugging:
+ */
+#define DO_DEBUG		1
+
+#if DO_DEBUG
+extern int RADEON_DEBUG;
+#else
+#define RADEON_DEBUG		0
+#endif
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_compat.c b/src/mesa/drivers/dri/radeon/radeon_compat.c
deleted file mode 100644
index 46b490d61f7..00000000000
--- a/src/mesa/drivers/dri/radeon/radeon_compat.c
+++ /dev/null
@@ -1,301 +0,0 @@
-/**************************************************************************
-
-Copyright 2002 ATI Technologies Inc., Ontario, Canada, and
-               Tungsten Graphics Inc., Austin, Texas.
-
-All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining a
-copy of this software and associated documentation files (the "Software"),
-to deal in the Software without restriction, including without limitation
-on the rights to use, copy, modify, merge, publish, distribute, sub
-license, and/or sell copies of the Software, and to permit persons to whom
-the Software is furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice (including the next
-paragraph) shall be included in all copies or substantial portions of the
-Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
-ATI, TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
-DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
-OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
-USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell <[email protected]>
- *
- */
-
-#include "main/glheader.h"
-#include "main/imports.h"
-
-#include "radeon_context.h"
-#include "radeon_state.h"
-#include "radeon_ioctl.h"
-
-
-static struct { 
-	int start; 
-	int len; 
-	const char *name;
-} packet[RADEON_MAX_STATE_PACKETS] = {
-	{ RADEON_PP_MISC,7,"RADEON_PP_MISC" },
-	{ RADEON_PP_CNTL,3,"RADEON_PP_CNTL" },
-	{ RADEON_RB3D_COLORPITCH,1,"RADEON_RB3D_COLORPITCH" },
-	{ RADEON_RE_LINE_PATTERN,2,"RADEON_RE_LINE_PATTERN" },
-	{ RADEON_SE_LINE_WIDTH,1,"RADEON_SE_LINE_WIDTH" },
-	{ RADEON_PP_LUM_MATRIX,1,"RADEON_PP_LUM_MATRIX" },
-	{ RADEON_PP_ROT_MATRIX_0,2,"RADEON_PP_ROT_MATRIX_0" },
-	{ RADEON_RB3D_STENCILREFMASK,3,"RADEON_RB3D_STENCILREFMASK" },
-	{ RADEON_SE_VPORT_XSCALE,6,"RADEON_SE_VPORT_XSCALE" },
-	{ RADEON_SE_CNTL,2,"RADEON_SE_CNTL" },
-	{ RADEON_SE_CNTL_STATUS,1,"RADEON_SE_CNTL_STATUS" },
-	{ RADEON_RE_MISC,1,"RADEON_RE_MISC" },
-	{ RADEON_PP_TXFILTER_0,6,"RADEON_PP_TXFILTER_0" },
-	{ RADEON_PP_BORDER_COLOR_0,1,"RADEON_PP_BORDER_COLOR_0" },
-	{ RADEON_PP_TXFILTER_1,6,"RADEON_PP_TXFILTER_1" },
-	{ RADEON_PP_BORDER_COLOR_1,1,"RADEON_PP_BORDER_COLOR_1" },
-	{ RADEON_PP_TXFILTER_2,6,"RADEON_PP_TXFILTER_2" },
-	{ RADEON_PP_BORDER_COLOR_2,1,"RADEON_PP_BORDER_COLOR_2" },
-	{ RADEON_SE_ZBIAS_FACTOR,2,"RADEON_SE_ZBIAS_FACTOR" },
-	{ RADEON_SE_TCL_OUTPUT_VTX_FMT,11,"RADEON_SE_TCL_OUTPUT_VTX_FMT" },
-	{ RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED,17,"RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED" },
-};
-
-
-static void radeonCompatEmitPacket( radeonContextPtr rmesa, 
-				    struct radeon_state_atom *state )
-{
-   drm_radeon_sarea_t *sarea = rmesa->sarea;
-   drm_radeon_context_regs_t *ctx = &sarea->context_state;
-   drm_radeon_texture_regs_t *tex0 = &sarea->tex_state[0];
-   drm_radeon_texture_regs_t *tex1 = &sarea->tex_state[1];
-   int i;
-   int *buf = state->cmd;
-
-   for ( i = 0 ; i < state->cmd_size ; ) {
-      drm_radeon_cmd_header_t *header = (drm_radeon_cmd_header_t *)&buf[i++];
-
-      if (RADEON_DEBUG & DEBUG_STATE)
-	 fprintf(stderr, "%s %d: %s\n", __FUNCTION__, header->packet.packet_id,
-		 packet[(int)header->packet.packet_id].name);
-
-      switch (header->packet.packet_id) {
-      case RADEON_EMIT_PP_MISC:
-	 ctx->pp_misc = buf[i++]; 
-	 ctx->pp_fog_color = buf[i++];
-	 ctx->re_solid_color = buf[i++];
-	 ctx->rb3d_blendcntl = buf[i++];
-	 ctx->rb3d_depthoffset = buf[i++];
-	 ctx->rb3d_depthpitch = buf[i++];
-	 ctx->rb3d_zstencilcntl = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_CONTEXT;
-	 break;
-      case RADEON_EMIT_PP_CNTL:
-	 ctx->pp_cntl = buf[i++];
-	 ctx->rb3d_cntl = buf[i++];
-	 ctx->rb3d_coloroffset = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_CONTEXT;
-	 break;
-      case RADEON_EMIT_RB3D_COLORPITCH:
-	 ctx->rb3d_colorpitch = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_CONTEXT;
-	 break;
-      case RADEON_EMIT_RE_LINE_PATTERN:
-	 ctx->re_line_pattern = buf[i++];
-	 ctx->re_line_state = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_LINE;
-	 break;
-      case RADEON_EMIT_SE_LINE_WIDTH:
-	 ctx->se_line_width = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_LINE;
-	 break;
-      case RADEON_EMIT_PP_LUM_MATRIX:
-	 ctx->pp_lum_matrix = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_BUMPMAP;
-	 break;
-      case RADEON_EMIT_PP_ROT_MATRIX_0:
-	 ctx->pp_rot_matrix_0 = buf[i++];
-	 ctx->pp_rot_matrix_1 = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_BUMPMAP;
-	 break;
-      case RADEON_EMIT_RB3D_STENCILREFMASK:
-	 ctx->rb3d_stencilrefmask = buf[i++];
-	 ctx->rb3d_ropcntl = buf[i++];
-	 ctx->rb3d_planemask = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_MASKS;
-	 break;
-      case RADEON_EMIT_SE_VPORT_XSCALE:
-	 ctx->se_vport_xscale = buf[i++];
-	 ctx->se_vport_xoffset = buf[i++];
-	 ctx->se_vport_yscale = buf[i++];
-	 ctx->se_vport_yoffset = buf[i++];
-	 ctx->se_vport_zscale = buf[i++];
-	 ctx->se_vport_zoffset = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_VIEWPORT;
-	 break;
-      case RADEON_EMIT_SE_CNTL:
-	 ctx->se_cntl = buf[i++];
-	 ctx->se_coord_fmt = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_VERTFMT;
-	 break;
-      case RADEON_EMIT_SE_CNTL_STATUS:
-	 ctx->se_cntl_status = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_SETUP;
-	 break;
-      case RADEON_EMIT_RE_MISC:
-	 ctx->re_misc = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_MISC;
-	 break;
-      case RADEON_EMIT_PP_TXFILTER_0:
-	 tex0->pp_txfilter = buf[i++];
-	 tex0->pp_txformat = buf[i++];
-	 tex0->pp_txoffset = buf[i++];
-	 tex0->pp_txcblend = buf[i++];
-	 tex0->pp_txablend = buf[i++];
-	 tex0->pp_tfactor = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_TEX0;
-	 break;
-      case RADEON_EMIT_PP_BORDER_COLOR_0:
-	 tex0->pp_border_color = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_TEX0;
-	 break;
-      case RADEON_EMIT_PP_TXFILTER_1:
-	 tex1->pp_txfilter = buf[i++];
-	 tex1->pp_txformat = buf[i++];
-	 tex1->pp_txoffset = buf[i++];
-	 tex1->pp_txcblend = buf[i++];
-	 tex1->pp_txablend = buf[i++];
-	 tex1->pp_tfactor = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_TEX1;
-	 break;
-      case RADEON_EMIT_PP_BORDER_COLOR_1:
-	 tex1->pp_border_color = buf[i++];
-	 sarea->dirty |= RADEON_UPLOAD_TEX1;
-	 break;
-
-      case RADEON_EMIT_SE_ZBIAS_FACTOR:
-	 i++;
-	 i++;
-	 break;
-
-      case RADEON_EMIT_PP_TXFILTER_2:
-      case RADEON_EMIT_PP_BORDER_COLOR_2:
-      case RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT:
-      case RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED:
-      default:
-	 /* These states aren't understood by radeon drm 1.1 */
-	 fprintf(stderr, "Tried to emit unsupported state\n");
-	 return;
-      }
-   }
-}
-
-
-
-static void radeonCompatEmitStateLocked( radeonContextPtr rmesa )
-{
-   struct radeon_state_atom *atom;
-
-   if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!rmesa->hw.is_dirty && !rmesa->hw.all_dirty)
-      return;
-
-   foreach(atom, &rmesa->hw.atomlist) {
-      if (rmesa->hw.all_dirty)
-	 atom->dirty = GL_TRUE;
-      if (atom->is_tcl)
-	 atom->dirty = GL_FALSE;
-      if (atom->dirty)
-	 radeonCompatEmitPacket(rmesa, atom);
-   }
- 
-   rmesa->hw.is_dirty = GL_FALSE;
-   rmesa->hw.all_dirty = GL_FALSE;
-}
-
-
-static void radeonCompatEmitPrimitiveLocked( radeonContextPtr rmesa,
-					     GLuint hw_primitive,
-					     GLuint nverts,
-					     drm_clip_rect_t *pbox,
-					     GLuint nbox )
-{
-   int i;
-
-   for ( i = 0 ; i < nbox ; ) {
-      int nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, nbox );
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
-      drm_radeon_vertex_t vtx;
-      
-      rmesa->sarea->dirty |= RADEON_UPLOAD_CLIPRECTS;
-      rmesa->sarea->nbox = nr - i;
-
-      for ( ; i < nr ; i++) 
-	 *b++ = pbox[i];
-      
-      if (RADEON_DEBUG & DEBUG_IOCTL)
-	 fprintf(stderr, 
-		 "RadeonFlushVertexBuffer: prim %x buf %d verts %d "
-		 "disc %d nbox %d\n",
-		 hw_primitive, 
-		 rmesa->dma.current.buf->buf->idx, 
-		 nverts, 
-		 nr == nbox,
-		 rmesa->sarea->nbox );
-
-      vtx.prim = hw_primitive;
-      vtx.idx = rmesa->dma.current.buf->buf->idx;
-      vtx.count = nverts;
-      vtx.discard = (nr == nbox);      
-
-      drmCommandWrite( rmesa->dri.fd, 
-		       DRM_RADEON_VERTEX,
-		       &vtx, sizeof(vtx));
-   }
-}
-
-
-
-/* No 'start' for 1.1 vertices ioctl: only one vertex prim/buffer!  
- */
-void radeonCompatEmitPrimitive( radeonContextPtr rmesa,
-				GLuint vertex_format,
-				GLuint hw_primitive,
-				GLuint nrverts )
-{
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   LOCK_HARDWARE( rmesa );
-
-   radeonCompatEmitStateLocked( rmesa );
-   rmesa->sarea->vc_format = vertex_format;
-   
-   if (rmesa->state.scissor.enabled) {
-      radeonCompatEmitPrimitiveLocked( rmesa, 
-				       hw_primitive,
-				       nrverts,
-				       rmesa->state.scissor.pClipRects,
-				       rmesa->state.scissor.numClipRects );
-   }
-   else {
-      radeonCompatEmitPrimitiveLocked( rmesa, 
-				       hw_primitive,
-				       nrverts,
-				       rmesa->pClipRects,
-				       rmesa->numClipRects );
-   }
-
-
-   UNLOCK_HARDWARE( rmesa );
-}
-
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.c b/src/mesa/drivers/dri/radeon/radeon_context.c
index ea81a3250b7..8f780c443c9 100644
--- a/src/mesa/drivers/dri/radeon/radeon_context.c
+++ b/src/mesa/drivers/dri/radeon/radeon_context.c
@@ -53,6 +53,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "drivers/common/driverfuncs.h"
 
+#include "radeon_common.h"
 #include "radeon_context.h"
 #include "radeon_ioctl.h"
 #include "radeon_state.h"
@@ -65,6 +66,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define need_GL_EXT_blend_minmax
 #define need_GL_EXT_fog_coord
 #define need_GL_EXT_secondary_color
+#define need_GL_EXT_framebuffer_object
 #include "extension_helper.h"
 
 #define DRIVER_DATE	"20061018"
@@ -72,40 +74,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "vblank.h"
 #include "utils.h"
 #include "xmlpool.h" /* for symbolic values of enum-type options */
-#ifndef RADEON_DEBUG
-int RADEON_DEBUG = (0);
-#endif
-
-
-/* Return various strings for glGetString().
- */
-static const GLubyte *radeonGetString( GLcontext *ctx, GLenum name )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   static char buffer[128];
-   unsigned   offset;
-   GLuint agp_mode = (rmesa->radeonScreen->card_type==RADEON_CARD_PCI) ? 0 :
-      rmesa->radeonScreen->AGPMode;
-
-   switch ( name ) {
-   case GL_VENDOR:
-      return (GLubyte *)"Tungsten Graphics, Inc.";
-
-   case GL_RENDERER:
-      offset = driGetRendererString( buffer, "Radeon", DRIVER_DATE,
-				     agp_mode );
-
-      sprintf( & buffer[ offset ], " %sTCL",
-	       !(rmesa->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE)
-	       ? "" : "NO-" );
-
-      return (GLubyte *)buffer;
-
-   default:
-      return NULL;
-   }
-}
-
 
 /* Extension strings exported by the R100 driver.
  */
@@ -121,6 +89,7 @@ const struct dri_extension card_extensions[] =
     { "GL_EXT_blend_logic_op",             NULL },
     { "GL_EXT_blend_subtract",             GL_EXT_blend_minmax_functions },
     { "GL_EXT_fog_coord",                  GL_EXT_fog_coord_functions },
+    { "GL_EXT_packed_depth_stencil",	   NULL},
     { "GL_EXT_secondary_color",            GL_EXT_secondary_color_functions },
     { "GL_EXT_stencil_wrap",               NULL },
     { "GL_EXT_texture_edge_clamp",         NULL },
@@ -137,6 +106,11 @@ const struct dri_extension card_extensions[] =
     { NULL,                                NULL }
 };
 
+const struct dri_extension mm_extensions[] = {
+  { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions },
+  { NULL, NULL }
+};
+
 extern const struct tnl_pipeline_stage _radeon_render_stage;
 extern const struct tnl_pipeline_stage _radeon_tcl_stage;
 
@@ -160,15 +134,6 @@ static const struct tnl_pipeline_stage *radeon_pipeline[] = {
    NULL,
 };
 
-
-
-/* Initialize the driver's misc functions.
- */
-static void radeonInitDriverFuncs( struct dd_function_table *functions )
-{
-    functions->GetString	= radeonGetString;
-}
-
 static const struct dri_debug_control debug_control[] =
 {
     { "fall",  DEBUG_FALLBACKS },
@@ -188,19 +153,69 @@ static const struct dri_debug_control debug_control[] =
     { NULL,    0 }
 };
 
+static void r100_get_lock(radeonContextPtr radeon)
+{
+   r100ContextPtr rmesa = (r100ContextPtr)radeon;
+   drm_radeon_sarea_t *sarea = radeon->sarea;
+
+   RADEON_STATECHANGE(rmesa, ctx);
+   if (rmesa->radeon.sarea->tiling_enabled) {
+      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |=
+	 RADEON_COLOR_TILE_ENABLE;
+   } else {
+      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] &=
+	 ~RADEON_COLOR_TILE_ENABLE;
+   }
+   
+   if (sarea->ctx_owner != rmesa->radeon.dri.hwContext) {
+      sarea->ctx_owner = rmesa->radeon.dri.hwContext;
+      
+      if (!radeon->radeonScreen->kernel_mm)
+         radeon_bo_legacy_texture_age(radeon->radeonScreen->bom);
+   }
+}
+
+static void r100_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmesa)
+{
+}
+
+static void r100_vtbl_pre_emit_state(radeonContextPtr radeon)
+{
+   r100ContextPtr rmesa = (r100ContextPtr)radeon;
+   
+   /* r100 always needs to emit ZBS to avoid TCL lockups */
+   rmesa->hw.zbs.dirty = 1;
+   radeon->hw.is_dirty = 1;
+}
+
+static void r100_vtbl_free_context(GLcontext *ctx)
+{
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   _mesa_vector4f_free( &rmesa->tcl.ObjClean );
+}
+
+static void r100_init_vtbl(radeonContextPtr radeon)
+{
+   radeon->vtbl.get_lock = r100_get_lock;
+   radeon->vtbl.update_viewport_offset = radeonUpdateViewportOffset;
+   radeon->vtbl.emit_cs_header = r100_vtbl_emit_cs_header;
+   radeon->vtbl.swtcl_flush = r100_swtcl_flush;
+   radeon->vtbl.pre_emit_state = r100_vtbl_pre_emit_state;
+   radeon->vtbl.fallback = radeonFallback;
+}
 
 /* Create the device specific context.
  */
 GLboolean
-radeonCreateContext( const __GLcontextModes *glVisual,
+r100CreateContext( const __GLcontextModes *glVisual,
                      __DRIcontextPrivate *driContextPriv,
                      void *sharedContextPrivate)
 {
    __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
    radeonScreenPtr screen = (radeonScreenPtr)(sPriv->private);
    struct dd_function_table functions;
-   radeonContextPtr rmesa;
-   GLcontext *ctx, *shareCtx;
+   r100ContextPtr rmesa;
+   GLcontext *ctx;
    int i;
    int tcl_mode, fthrottle_mode;
 
@@ -209,10 +224,12 @@ radeonCreateContext( const __GLcontextModes *glVisual,
    assert(screen);
 
    /* Allocate the Radeon context */
-   rmesa = (radeonContextPtr) CALLOC( sizeof(*rmesa) );
+   rmesa = (r100ContextPtr) CALLOC( sizeof(*rmesa) );
    if ( !rmesa )
       return GL_FALSE;
 
+   r100_init_vtbl(&rmesa->radeon);
+
    /* init exp fog table data */
    radeonInitStaticFogData();
    
@@ -220,12 +237,12 @@ radeonCreateContext( const __GLcontextModes *glVisual,
     * Do this here so that initialMaxAnisotropy is set before we create
     * the default textures.
     */
-   driParseConfigFiles (&rmesa->optionCache, &screen->optionCache,
+   driParseConfigFiles (&rmesa->radeon.optionCache, &screen->optionCache,
 			screen->driScreen->myNum, "radeon");
-   rmesa->initialMaxAnisotropy = driQueryOptionf(&rmesa->optionCache,
+   rmesa->radeon.initialMaxAnisotropy = driQueryOptionf(&rmesa->radeon.optionCache,
                                                  "def_max_anisotropy");
 
-   if ( driQueryOptionb( &rmesa->optionCache, "hyperz" ) ) {
+   if ( driQueryOptionb( &rmesa->radeon.optionCache, "hyperz" ) ) {
       if ( sPriv->drm_version.minor < 13 )
 	 fprintf( stderr, "DRM version 1.%d too old to support HyperZ, "
 			  "disabling.\n", sPriv->drm_version.minor );
@@ -240,65 +257,17 @@ radeonCreateContext( const __GLcontextModes *glVisual,
     * (the texture functions are especially important)
     */
    _mesa_init_driver_functions( &functions );
-   radeonInitDriverFuncs( &functions );
    radeonInitTextureFuncs( &functions );
 
-   /* Allocate the Mesa context */
-   if (sharedContextPrivate)
-      shareCtx = ((radeonContextPtr) sharedContextPrivate)->glCtx;
-   else
-      shareCtx = NULL;
-   rmesa->glCtx = _mesa_create_context(glVisual, shareCtx,
-                                       &functions, (void *) rmesa);
-   if (!rmesa->glCtx) {
-      FREE(rmesa);
-      return GL_FALSE;
-   }
-   driContextPriv->driverPrivate = rmesa;
-
-   /* Init radeon context data */
-   rmesa->dri.context = driContextPriv;
-   rmesa->dri.screen = sPriv;
-   rmesa->dri.drawable = NULL;
-   rmesa->dri.readable = NULL;
-   rmesa->dri.hwContext = driContextPriv->hHWContext;
-   rmesa->dri.hwLock = &sPriv->pSAREA->lock;
-   rmesa->dri.fd = sPriv->fd;
-   rmesa->dri.drmMinor = sPriv->drm_version.minor;
-
-   rmesa->radeonScreen = screen;
-   rmesa->sarea = (drm_radeon_sarea_t *)((GLubyte *)sPriv->pSAREA +
-				       screen->sarea_priv_offset);
-
-
-   rmesa->dma.buf0_address = rmesa->radeonScreen->buffers->list[0].address;
-
-   (void) memset( rmesa->texture_heaps, 0, sizeof( rmesa->texture_heaps ) );
-   make_empty_list( & rmesa->swapped );
-
-   rmesa->nr_heaps = screen->numTexHeaps;
-   for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
-      rmesa->texture_heaps[i] = driCreateTextureHeap( i, rmesa,
-	    screen->texSize[i],
-	    12,
-	    RADEON_NR_TEX_REGIONS,
-	    (drmTextureRegionPtr)rmesa->sarea->tex_list[i],
-	    & rmesa->sarea->tex_age[i],
-	    & rmesa->swapped,
-	    sizeof( radeonTexObj ),
-	    (destroy_texture_object_t *) radeonDestroyTexObj );
-
-      driSetTextureSwapCounterLocation( rmesa->texture_heaps[i],
-					& rmesa->c_textureSwaps );
+   if (!radeonInitContext(&rmesa->radeon, &functions,
+			  glVisual, driContextPriv,
+			  sharedContextPrivate)) {
+     FREE(rmesa);
+     return GL_FALSE;
    }
-   rmesa->texture_depth = driQueryOptioni (&rmesa->optionCache,
-					   "texture_depth");
-   if (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
-      rmesa->texture_depth = ( screen->cpp == 4 ) ?
-	 DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16;
 
-   rmesa->swtcl.RenderIndex = ~0;
-   rmesa->hw.all_dirty = GL_TRUE;
+   rmesa->radeon.swtcl.RenderIndex = ~0;
+   rmesa->radeon.hw.all_dirty = GL_TRUE;
 
    /* Set the maximum texture size small enough that we can guarentee that
     * all texture units can bind a maximal texture and have all of them in
@@ -306,26 +275,18 @@ radeonCreateContext( const __GLcontextModes *glVisual,
     * setting allow larger textures.
     */
 
-   ctx = rmesa->glCtx;
-   ctx->Const.MaxTextureUnits = driQueryOptioni (&rmesa->optionCache,
+   ctx = rmesa->radeon.glCtx;
+   ctx->Const.MaxTextureUnits = driQueryOptioni (&rmesa->radeon.optionCache,
 						 "texture_units");
    ctx->Const.MaxTextureImageUnits = ctx->Const.MaxTextureUnits;
    ctx->Const.MaxTextureCoordUnits = ctx->Const.MaxTextureUnits;
 
-   i = driQueryOptioni( &rmesa->optionCache, "allow_large_textures");
-
-   driCalculateMaxTextureLevels( rmesa->texture_heaps,
-				 rmesa->nr_heaps,
-				 & ctx->Const,
-				 4,
-				 11, /* max 2D texture size is 2048x2048 */
-				 8,  /* 256^3 */
-				 9,  /* \todo: max cube texture size seems to be 512x512(x6) */
-				 11, /* max rect texture size is 2048x2048. */
-				 12,
-				 GL_FALSE,
-				 i );
+   i = driQueryOptioni( &rmesa->radeon.optionCache, "allow_large_textures");
 
+   /* FIXME: When no memory manager is available we should set this 
+    * to some reasonable value based on texture memory pool size */
+   /* FIXME: does r100 support 2048x2048 texture ? */
+   ctx->Const.MaxTextureLevels = 12;
 
    ctx->Const.MaxTextureMaxAnisotropy = 16.0;
 
@@ -388,38 +349,39 @@ radeonCreateContext( const __GLcontextModes *glVisual,
    }
 
    driInitExtensions( ctx, card_extensions, GL_TRUE );
-   if (rmesa->radeonScreen->drmSupportsCubeMapsR100)
+   if (rmesa->radeon.radeonScreen->kernel_mm)
+     driInitExtensions(ctx, mm_extensions, GL_FALSE);
+   if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR100)
       _mesa_enable_extension( ctx, "GL_ARB_texture_cube_map" );
-   if (rmesa->glCtx->Mesa_DXTn) {
+   if (rmesa->radeon.glCtx->Mesa_DXTn) {
       _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
       _mesa_enable_extension( ctx, "GL_S3_s3tc" );
    }
-   else if (driQueryOptionb (&rmesa->optionCache, "force_s3tc_enable")) {
+   else if (driQueryOptionb (&rmesa->radeon.optionCache, "force_s3tc_enable")) {
       _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
    }
 
-   if (rmesa->dri.drmMinor >= 9)
+   if (rmesa->radeon.dri.drmMinor >= 9)
       _mesa_enable_extension( ctx, "GL_NV_texture_rectangle");
 
    /* XXX these should really go right after _mesa_init_driver_functions() */
+   radeon_fbo_init(&rmesa->radeon);
+   radeonInitSpanFuncs( ctx );
    radeonInitIoctlFuncs( ctx );
    radeonInitStateFuncs( ctx );
-   radeonInitSpanFuncs( ctx );
    radeonInitState( rmesa );
    radeonInitSwtcl( ctx );
 
    _mesa_vector4f_alloc( &rmesa->tcl.ObjClean, 0, 
 			 ctx->Const.MaxArrayLockSize, 32 );
 
-   fthrottle_mode = driQueryOptioni(&rmesa->optionCache, "fthrottle_mode");
-   rmesa->iw.irq_seq = -1;
-   rmesa->irqsEmitted = 0;
-   rmesa->do_irqs = (rmesa->radeonScreen->irq != 0 &&
-		     fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS);
-
-   rmesa->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS);
+   fthrottle_mode = driQueryOptioni(&rmesa->radeon.optionCache, "fthrottle_mode");
+   rmesa->radeon.iw.irq_seq = -1;
+   rmesa->radeon.irqsEmitted = 0;
+   rmesa->radeon.do_irqs = (rmesa->radeon.radeonScreen->irq != 0 &&
+			    fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS);
 
-   (*sPriv->systemTime->getUST)( & rmesa->swap_ust );
+   rmesa->radeon.do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS);
 
 
 #if DO_DEBUG
@@ -427,206 +389,21 @@ radeonCreateContext( const __GLcontextModes *glVisual,
 				       debug_control );
 #endif
 
-   tcl_mode = driQueryOptioni(&rmesa->optionCache, "tcl_mode");
-   if (driQueryOptionb(&rmesa->optionCache, "no_rast")) {
+   tcl_mode = driQueryOptioni(&rmesa->radeon.optionCache, "tcl_mode");
+   if (driQueryOptionb(&rmesa->radeon.optionCache, "no_rast")) {
       fprintf(stderr, "disabling 3D acceleration\n");
       FALLBACK(rmesa, RADEON_FALLBACK_DISABLE, 1);
    } else if (tcl_mode == DRI_CONF_TCL_SW ||
-	      !(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
-      if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
-	 rmesa->radeonScreen->chip_flags &= ~RADEON_CHIPSET_TCL;
+	      !(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
+      if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
+	 rmesa->radeon.radeonScreen->chip_flags &= ~RADEON_CHIPSET_TCL;
 	 fprintf(stderr, "Disabling HW TCL support\n");
       }
-      TCL_FALLBACK(rmesa->glCtx, RADEON_TCL_FALLBACK_TCL_DISABLE, 1);
+      TCL_FALLBACK(rmesa->radeon.glCtx, RADEON_TCL_FALLBACK_TCL_DISABLE, 1);
    }
 
-   if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
+   if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
 /*       _tnl_need_dlist_norm_lengths( ctx, GL_FALSE ); */
    }
    return GL_TRUE;
 }
-
-
-/* Destroy the device specific context.
- */
-/* Destroy the Mesa and driver specific context data.
- */
-void radeonDestroyContext( __DRIcontextPrivate *driContextPriv )
-{
-   GET_CURRENT_CONTEXT(ctx);
-   radeonContextPtr rmesa = (radeonContextPtr) driContextPriv->driverPrivate;
-   radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL;
-
-   /* check if we're deleting the currently bound context */
-   if (rmesa == current) {
-      RADEON_FIREVERTICES( rmesa );
-      _mesa_make_current(NULL, NULL, NULL);
-   }
-
-   /* Free radeon context resources */
-   assert(rmesa); /* should never be null */
-   if ( rmesa ) {
-      GLboolean   release_texture_heaps;
-
-
-      release_texture_heaps = (rmesa->glCtx->Shared->RefCount == 1);
-      _swsetup_DestroyContext( rmesa->glCtx );
-      _tnl_DestroyContext( rmesa->glCtx );
-      _vbo_DestroyContext( rmesa->glCtx );
-      _swrast_DestroyContext( rmesa->glCtx );
-
-      radeonDestroySwtcl( rmesa->glCtx );
-      radeonReleaseArrays( rmesa->glCtx, ~0 );
-      if (rmesa->dma.current.buf) {
-	 radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ );
-	 radeonFlushCmdBuf( rmesa, __FUNCTION__ );
-      }
-
-      _mesa_vector4f_free( &rmesa->tcl.ObjClean );
-
-      if (rmesa->state.scissor.pClipRects) {
-	 FREE(rmesa->state.scissor.pClipRects);
-	 rmesa->state.scissor.pClipRects = NULL;
-      }
-
-      if ( release_texture_heaps ) {
-         /* This share group is about to go away, free our private
-          * texture object data.
-          */
-         int i;
-
-         for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
-	    driDestroyTextureHeap( rmesa->texture_heaps[ i ] );
-	    rmesa->texture_heaps[ i ] = NULL;
-         }
-
-	 assert( is_empty_list( & rmesa->swapped ) );
-      }
-
-      /* free the Mesa context */
-      rmesa->glCtx->DriverCtx = NULL;
-      _mesa_destroy_context( rmesa->glCtx );
-
-      /* free the option cache */
-      driDestroyOptionCache (&rmesa->optionCache);
-
-      FREE( rmesa );
-   }
-}
-
-
-
-
-void
-radeonSwapBuffers( __DRIdrawablePrivate *dPriv )
-{
-
-   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
-      radeonContextPtr rmesa;
-      GLcontext *ctx;
-      rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-      ctx = rmesa->glCtx;
-      if (ctx->Visual.doubleBufferMode) {
-         _mesa_notifySwapBuffers( ctx );  /* flush pending rendering comands */
-
-         if ( rmesa->doPageFlip ) {
-            radeonPageFlip( dPriv );
-         }
-         else {
-	     radeonCopyBuffer( dPriv, NULL );
-         }
-      }
-   }
-   else {
-      /* XXX this shouldn't be an error but we can't handle it for now */
-      _mesa_problem(NULL, "%s: drawable has no context!", __FUNCTION__);
-   }
-}
-
-void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
-			 int x, int y, int w, int h )
-{
-    if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
-	radeonContextPtr radeon;
-	GLcontext *ctx;
-
-	radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-	ctx = radeon->glCtx;
-
-	if (ctx->Visual.doubleBufferMode) {
-	    drm_clip_rect_t rect;
-	    rect.x1 = x + dPriv->x;
-	    rect.y1 = (dPriv->h - y - h) + dPriv->y;
-	    rect.x2 = rect.x1 + w;
-	    rect.y2 = rect.y1 + h;
-	    _mesa_notifySwapBuffers(ctx);	/* flush pending rendering comands */
-	    radeonCopyBuffer(dPriv, &rect);
-	}
-    } else {
-	/* XXX this shouldn't be an error but we can't handle it for now */
-	_mesa_problem(NULL, "%s: drawable has no context!",
-		      __FUNCTION__);
-    }
-}
-
-/* Make context `c' the current context and bind it to the given
- * drawing and reading surfaces.
- */
-GLboolean
-radeonMakeCurrent( __DRIcontextPrivate *driContextPriv,
-                   __DRIdrawablePrivate *driDrawPriv,
-                   __DRIdrawablePrivate *driReadPriv )
-{
-   if ( driContextPriv ) {
-      radeonContextPtr newCtx = 
-	 (radeonContextPtr) driContextPriv->driverPrivate;
-
-      if (RADEON_DEBUG & DEBUG_DRI)
-	 fprintf(stderr, "%s ctx %p\n", __FUNCTION__, (void *) newCtx->glCtx);
-
-      newCtx->dri.readable = driReadPriv;
-
-      if ( (newCtx->dri.drawable != driDrawPriv) ||
-           newCtx->lastStamp != driDrawPriv->lastStamp ) {
-	 if (driDrawPriv->swap_interval == (unsigned)-1) {
-	    driDrawPriv->vblFlags = (newCtx->radeonScreen->irq != 0)
-	       ? driGetDefaultVBlankFlags(&newCtx->optionCache)
-	       : VBLANK_FLAG_NO_IRQ;
-
-	    driDrawableInitVBlank( driDrawPriv );
-	 }
-
-	 newCtx->dri.drawable = driDrawPriv;
-
-	 radeonSetCliprects(newCtx);
-	 radeonUpdateViewportOffset( newCtx->glCtx );
-      }
-
-      _mesa_make_current( newCtx->glCtx,
-			  (GLframebuffer *) driDrawPriv->driverPrivate,
-			  (GLframebuffer *) driReadPriv->driverPrivate );
-
-      _mesa_update_state( newCtx->glCtx );
-   } else {
-      if (RADEON_DEBUG & DEBUG_DRI)
-	 fprintf(stderr, "%s ctx is null\n", __FUNCTION__);
-      _mesa_make_current( NULL, NULL, NULL );
-   }
-
-   if (RADEON_DEBUG & DEBUG_DRI)
-      fprintf(stderr, "End %s\n", __FUNCTION__);
-   return GL_TRUE;
-}
-
-/* Force the context `c' to be unbound from its buffer.
- */
-GLboolean
-radeonUnbindContext( __DRIcontextPrivate *driContextPriv )
-{
-   radeonContextPtr rmesa = (radeonContextPtr) driContextPriv->driverPrivate;
-
-   if (RADEON_DEBUG & DEBUG_DRI)
-      fprintf(stderr, "%s ctx %p\n", __FUNCTION__, (void *) rmesa->glCtx);
-
-   return GL_TRUE;
-}
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.h b/src/mesa/drivers/dri/radeon/radeon_context.h
index 53df766f8cb..1795d8bdb6d 100644
--- a/src/mesa/drivers/dri/radeon/radeon_context.h
+++ b/src/mesa/drivers/dri/radeon/radeon_context.h
@@ -48,91 +48,23 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "drm.h"
 #include "radeon_drm.h"
 #include "texmem.h"
-
 #include "main/macros.h"
 #include "main/mtypes.h"
 #include "main/colormac.h"
-
-struct radeon_context;
-typedef struct radeon_context radeonContextRec;
-typedef struct radeon_context *radeonContextPtr;
-
-/* This union is used to avoid warnings/miscompilation
-   with float to uint32_t casts due to strict-aliasing */
-typedef union {
-	GLfloat f;
-	uint32_t ui32;
-} float_ui32_type;
-
-#include "radeon_lock.h"
 #include "radeon_screen.h"
-#include "main/mm.h"
-
-#include "math/m_vector.h"
-
-#define TEX_0   0x1
-#define TEX_1   0x2
-#define TEX_2   0x4
-#define TEX_ALL 0x7
-
-/* Rasterizing fallbacks */
-/* See correponding strings in r200_swtcl.c */
-#define RADEON_FALLBACK_TEXTURE		0x0001
-#define RADEON_FALLBACK_DRAW_BUFFER	0x0002
-#define RADEON_FALLBACK_STENCIL		0x0004
-#define RADEON_FALLBACK_RENDER_MODE	0x0008
-#define RADEON_FALLBACK_BLEND_EQ	0x0010
-#define RADEON_FALLBACK_BLEND_FUNC	0x0020
-#define RADEON_FALLBACK_DISABLE 	0x0040
-#define RADEON_FALLBACK_BORDER_MODE	0x0080
-
-/* The blit width for texture uploads
- */
-#define BLIT_WIDTH_BYTES 1024
 
-/* Use the templated vertex format:
- */
-#define COLOR_IS_RGBA
-#define TAG(x) radeon##x
-#include "tnl_dd/t_dd_vertex.h"
-#undef TAG
-
-typedef void (*radeon_tri_func) (radeonContextPtr,
-				 radeonVertex *,
-				 radeonVertex *, radeonVertex *);
-
-typedef void (*radeon_line_func) (radeonContextPtr,
-				  radeonVertex *, radeonVertex *);
+#include "radeon_common.h"
 
-typedef void (*radeon_point_func) (radeonContextPtr, radeonVertex *);
-
-struct radeon_colorbuffer_state {
-	GLuint clear;
-	int roundEnable;
-};
 
-struct radeon_depthbuffer_state {
-	GLuint clear;
-	GLfloat scale;
-};
+struct r100_context;
+typedef struct r100_context r100ContextRec;
+typedef struct r100_context *r100ContextPtr;
 
-struct radeon_scissor_state {
-	drm_clip_rect_t rect;
-	GLboolean enabled;
+#include "radeon_lock.h"
 
-	GLuint numClipRects;	/* Cliprects active */
-	GLuint numAllocedClipRects;	/* Cliprects available */
-	drm_clip_rect_t *pClipRects;
-};
 
-struct radeon_stencilbuffer_state {
-	GLboolean hwBuffer;
-	GLuint clear;		/* rb3d_stencilrefmask value */
-};
 
-struct radeon_stipple_state {
-	GLuint mask[32];
-};
+#define R100_TEX_ALL 0x7
 
 /* used for both tcl_vtx and vc_frmt tex bits (they are identical) */
 #define RADEON_ST_BIT(unit) \
@@ -141,42 +73,6 @@ struct radeon_stipple_state {
 #define RADEON_Q_BIT(unit) \
 (unit == 0 ? RADEON_CP_VC_FRMT_Q0 : (RADEON_CP_VC_FRMT_Q1 >> 2) << (2 * unit))
 
-typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
-
-/* Texture object in locally shared texture space.
- */
-struct radeon_tex_obj {
-	driTextureObject base;
-
-	GLuint bufAddr;		/* Offset to start of locally
-				   shared texture block */
-
-	GLuint dirty_state;	/* Flags (1 per texunit) for
-				   whether or not this texobj
-				   has dirty hardware state
-				   (pp_*) that needs to be
-				   brought into the
-				   texunit. */
-
-	drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS];
-	/* Six, for the cube faces */
-
-	GLboolean image_override; /* Image overridden by GLX_EXT_tfp */
-
-	GLuint pp_txfilter;	/* hardware register values */
-	GLuint pp_txformat;
-	GLuint pp_txoffset;	/* Image location in texmem.
-				   All cube faces follow. */
-	GLuint pp_txsize;	/* npot only */
-	GLuint pp_txpitch;	/* npot only */
-	GLuint pp_border_color;
-	GLuint pp_cubic_faces;	/* cube face 1,2,3,4 log2 sizes */
-
-	GLboolean border_fallback;
-
-	GLuint tile_bits;	/* hw texture tile bits used on this texture */
-};
-
 struct radeon_texture_env_state {
 	radeonTexObjPtr texobj;
 	GLenum format;
@@ -187,17 +83,6 @@ struct radeon_texture_state {
 	struct radeon_texture_env_state unit[RADEON_MAX_TEXTURE_UNITS];
 };
 
-struct radeon_state_atom {
-	struct radeon_state_atom *next, *prev;
-	const char *name;	/* for debug */
-	int cmd_size;		/* size in bytes */
-	GLuint is_tcl;
-	int *cmd;		/* one or more cmd's */
-	int *lastcmd;		/* one or more cmd's */
-	GLboolean dirty;	/* dirty-mark in emit_state_list */
-	 GLboolean(*check) (GLcontext *);	/* is this state active? */
-};
-
 /* Trying to keep these relatively short as the variables are becoming
  * extravagently long.  Drop the driver name prefix off the front of
  * everything - I think we know which driver we're in by now, and keep the
@@ -410,10 +295,7 @@ struct radeon_state_atom {
 #define SHN_SHININESS      1
 #define SHN_STATE_SIZE     2
 
-struct radeon_hw_state {
-	/* Head of the linked list of state atoms. */
-	struct radeon_state_atom atomlist;
-
+struct r100_hw_state {
 	/* Hardware state, stored as cmdbuf commands:  
 	 *   -- Need to doublebuffer for
 	 *           - eliding noop statechange loops? (except line stipple count)
@@ -438,89 +320,19 @@ struct radeon_hw_state {
 	struct radeon_state_atom glt;
 	struct radeon_state_atom txr[3];	/* for NPOT */
 
-	int max_state_size;	/* Number of bytes necessary for a full state emit. */
-	GLboolean is_dirty, all_dirty;
 };
 
-struct radeon_state {
-	/* Derived state for internal purposes:
-	 */
-	struct radeon_colorbuffer_state color;
-	struct radeon_depthbuffer_state depth;
-	struct radeon_scissor_state scissor;
-	struct radeon_stencilbuffer_state stencil;
+
+struct r100_state {
 	struct radeon_stipple_state stipple;
 	struct radeon_texture_state texture;
 };
 
-/* Need refcounting on dma buffers:
- */
-struct radeon_dma_buffer {
-	int refcount;		/* the number of retained regions in buf */
-	drmBufPtr buf;
-};
-
-#define GET_START(rvb) (rmesa->radeonScreen->gart_buffer_offset +			\
-			(rvb)->address - rmesa->dma.buf0_address +	\
-			(rvb)->start)
-
-/* A retained region, eg vertices for indexed vertices.
- */
-struct radeon_dma_region {
-	struct radeon_dma_buffer *buf;
-	char *address;		/* == buf->address */
-	int start, end, ptr;	/* offsets from start of buf */
-	int aos_start;
-	int aos_stride;
-	int aos_size;
-};
-
-struct radeon_dma {
-	/* Active dma region.  Allocations for vertices and retained
-	 * regions come from here.  Also used for emitting random vertices,
-	 * these may be flushed by calling flush_current();
-	 */
-	struct radeon_dma_region current;
-
-	void (*flush) (radeonContextPtr);
-
-	char *buf0_address;	/* start of buf[0], for index calcs */
-	GLuint nr_released_bufs;	/* flush after so many buffers released */
-};
-
-struct radeon_dri_mirror {
-	__DRIcontextPrivate *context;	/* DRI context */
-	__DRIscreenPrivate *screen;	/* DRI screen */
-
-   /**
-    * DRI drawable bound to this context for drawing.
-    */
-	__DRIdrawablePrivate *drawable;
-
-   /**
-    * DRI drawable bound to this context for reading.
-    */
-	__DRIdrawablePrivate *readable;
-
-	drm_context_t hwContext;
-	drm_hw_lock_t *hwLock;
-	int fd;
-	int drmMinor;
-};
-
 #define RADEON_CMD_BUF_SZ  (8*1024)
-
-struct radeon_store {
-	GLuint statenr;
-	GLuint primnr;
-	char cmd_buf[RADEON_CMD_BUF_SZ];
-	int cmd_used;
-	int elts_start;
-};
-
+#define R200_ELT_BUF_SZ  (8*1024)
 /* radeon_tcl.c
  */
-struct radeon_tcl_info {
+struct r100_tcl_info {
 	GLuint vertex_format;
 	GLuint hw_primitive;
 
@@ -529,30 +341,18 @@ struct radeon_tcl_info {
 	 */
 	GLvector4f ObjClean;
 
-	struct radeon_dma_region *aos_components[8];
-	GLuint nr_aos_components;
-
 	GLuint *Elts;
 
-	struct radeon_dma_region indexed_verts;
-	struct radeon_dma_region obj;
-	struct radeon_dma_region rgba;
-	struct radeon_dma_region spec;
-	struct radeon_dma_region fog;
-	struct radeon_dma_region tex[RADEON_MAX_TEXTURE_UNITS];
-	struct radeon_dma_region norm;
+        int elt_cmd_offset;
+	int elt_cmd_start;
+        int elt_used;
 };
 
 /* radeon_swtcl.c
  */
-struct radeon_swtcl_info {
-	GLuint RenderIndex;
-	GLuint vertex_size;
+struct r100_swtcl_info {
 	GLuint vertex_format;
 
-	struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
-	GLuint vertex_attr_count;
-
 	GLubyte *verts;
 
 	/* Fallback rasterization functions
@@ -561,10 +361,6 @@ struct radeon_swtcl_info {
 	radeon_line_func draw_line;
 	radeon_tri_func draw_tri;
 
-	GLuint hw_primitive;
-	GLenum render_primitive;
-	GLuint numverts;
-
    /**
     * Offset of the 4UB color data within a hardware (swtcl) vertex.
     */
@@ -576,22 +372,9 @@ struct radeon_swtcl_info {
 	GLuint specoffset;
 
 	GLboolean needproj;
-
-	struct radeon_dma_region indexed_verts;
 };
 
-struct radeon_ioctl {
-	GLuint vertex_offset;
-	GLuint vertex_size;
-};
-
-#define RADEON_MAX_PRIMS 64
 
-struct radeon_prim {
-	GLuint start;
-	GLuint end;
-	GLuint prim;
-};
 
 /* A maximum total of 20 elements per vertex:  3 floats for position, 3
  * floats for normal, 4 floats for color, 4 bytes for secondary color,
@@ -602,59 +385,18 @@ struct radeon_prim {
  */
 #define RADEON_MAX_VERTEX_SIZE 20
 
-struct radeon_context {
-	GLcontext *glCtx;	/* Mesa context */
+struct r100_context {
+        struct radeon_context radeon;
 
 	/* Driver and hardware state management
 	 */
-	struct radeon_hw_state hw;
-	struct radeon_state state;
-
-	/* Texture object bookkeeping
-	 */
-	unsigned nr_heaps;
-	driTexHeap *texture_heaps[RADEON_NR_TEX_HEAPS];
-	driTextureObject swapped;
-	int texture_depth;
-	float initialMaxAnisotropy;
-
-	/* Rasterization and vertex state:
-	 */
-	GLuint TclFallback;
-	GLuint Fallback;
-	GLuint NewGLState;
-	 DECLARE_RENDERINPUTS(tnl_index_bitset);	/* index of bits for last tnl_install_attrs */
+	struct r100_hw_state hw;
+	struct r100_state state;
 
 	/* Vertex buffers
 	 */
 	struct radeon_ioctl ioctl;
-	struct radeon_dma dma;
 	struct radeon_store store;
-	/* A full state emit as of the first state emit in the main store, in case
-	 * the context is lost.
-	 */
-	struct radeon_store backup_store;
-
-	/* Page flipping
-	 */
-	GLuint doPageFlip;
-
-	/* Busy waiting
-	 */
-	GLuint do_usleeps;
-	GLuint do_irqs;
-	GLuint irqsEmitted;
-	drm_radeon_irq_wait_t iw;
-
-	/* Drawable, cliprect and scissor information
-	 */
-	GLuint numClipRects;	/* Cliprects for the draw buffer */
-	drm_clip_rect_t *pClipRects;
-	unsigned int lastStamp;
-	GLboolean lost_context;
-	GLboolean save_on_next_emit;
-	radeonScreenPtr radeonScreen;	/* Screen private DRI data */
-	drm_radeon_sarea_t *sarea;	/* Private SAREA data */
 
 	/* TCL stuff
 	 */
@@ -667,29 +409,13 @@ struct radeon_context {
 	GLmatrix tmpmat[RADEON_MAX_TEXTURE_UNITS];
 	GLuint last_ReallyEnabled;
 
-	/* VBI
-	 */
-	int64_t swap_ust;
-	int64_t swap_missed_ust;
-
-	GLuint swap_count;
-	GLuint swap_missed_count;
-
 	/* radeon_tcl.c
 	 */
-	struct radeon_tcl_info tcl;
+	struct r100_tcl_info tcl;
 
 	/* radeon_swtcl.c
 	 */
-	struct radeon_swtcl_info swtcl;
-
-	/* Mirrors of some DRI state
-	 */
-	struct radeon_dri_mirror dri;
-
-	/* Configuration cache
-	 */
-	driOptionCache optionCache;
+	struct r100_swtcl_info swtcl;
 
 	GLboolean using_hyperz;
 	GLboolean texmicrotile;
@@ -703,61 +429,19 @@ struct radeon_context {
 	GLuint c_textureSwaps;
 	GLuint c_textureBytes;
 	GLuint c_vertexBuffers;
+
 };
 
-#define RADEON_CONTEXT(ctx)		((radeonContextPtr)(ctx->DriverCtx))
-
-static INLINE GLuint radeonPackColor(GLuint cpp,
-                                     GLubyte r, GLubyte g,
-                                     GLubyte b, GLubyte a)
-{
-	switch (cpp) {
-	case 2:
-		return PACK_COLOR_565(r, g, b);
-	case 4:
-		return PACK_COLOR_8888(a, r, g, b);
-	default:
-		return 0;
-	}
-}
+
+#define R100_CONTEXT(ctx)		((r100ContextPtr)(ctx->DriverCtx))
+
 
 #define RADEON_OLD_PACKETS 1
 
-extern void radeonDestroyContext(__DRIcontextPrivate * driContextPriv);
-extern GLboolean radeonCreateContext(const __GLcontextModes * glVisual,
-				     __DRIcontextPrivate * driContextPriv,
-				     void *sharedContextPrivate);
-extern void radeonSwapBuffers(__DRIdrawablePrivate * dPriv);
-extern void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
-				int x, int y, int w, int h);
-extern GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv,
-				   __DRIdrawablePrivate * driDrawPriv,
-				   __DRIdrawablePrivate * driReadPriv);
-extern GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv);
-
-/* ================================================================
- * Debugging:
- */
-#define DO_DEBUG		1
-
-#if DO_DEBUG
-extern int RADEON_DEBUG;
-#else
-#define RADEON_DEBUG		0
-#endif
-
-#define DEBUG_TEXTURE	0x0001
-#define DEBUG_STATE	0x0002
-#define DEBUG_IOCTL	0x0004
-#define DEBUG_PRIMS	0x0008
-#define DEBUG_VERTS	0x0010
-#define DEBUG_FALLBACKS	0x0020
-#define DEBUG_VFMT	0x0040
-#define DEBUG_CODEGEN	0x0080
-#define DEBUG_VERBOSE	0x0100
-#define DEBUG_DRI       0x0200
-#define DEBUG_DMA       0x0400
-#define DEBUG_SANITY    0x0800
-#define DEBUG_SYNC      0x1000
+extern GLboolean r100CreateContext( const __GLcontextModes *glVisual,
+				    __DRIcontextPrivate *driContextPriv,
+				    void *sharedContextPrivate);
+  
+
 
 #endif				/* __RADEON_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_drm.h b/src/mesa/drivers/dri/radeon/radeon_cs_drm.h
new file mode 100644
index 00000000000..984725a6c9d
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_cs_drm.h
@@ -0,0 +1,207 @@
+/* 
+ * Copyright © 2008 Nicolai Haehnle
+ * Copyright © 2008 Jérôme Glisse
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ */
+/*
+ * Authors:
+ *      Aapo Tahkola <[email protected]>
+ *      Nicolai Haehnle <[email protected]>
+ *      Jérôme Glisse <[email protected]>
+ */
+#ifndef RADEON_CS_H
+#define RADEON_CS_H
+
+#include <stdint.h>
+#include <string.h>
+#include "drm.h"
+#include "radeon_drm.h"
+
+struct radeon_cs_reloc {
+    struct radeon_bo    *bo;
+    uint32_t            read_domain;
+    uint32_t            write_domain;
+    uint32_t            flags;
+};
+
+
+#define RADEON_CS_SPACE_OK 0
+#define RADEON_CS_SPACE_OP_TO_BIG 1
+#define RADEON_CS_SPACE_FLUSH 2
+
+struct radeon_cs_space_check {
+    struct radeon_bo *bo;
+    uint32_t read_domains;
+    uint32_t write_domain;
+    uint32_t new_accounted;
+};
+
+struct radeon_cs_manager;
+
+struct radeon_cs {
+    struct radeon_cs_manager    *csm;
+    void                        *relocs;
+    uint32_t                    *packets;
+    unsigned                    crelocs;
+    unsigned                    relocs_total_size;
+    unsigned                    cdw;
+    unsigned                    ndw;
+    int                         section;
+    unsigned                    section_ndw;
+    unsigned                    section_cdw;
+    const char                  *section_file;
+    const char                  *section_func;
+    int                         section_line;
+
+};
+
+/* cs functions */
+struct radeon_cs_funcs {
+    struct radeon_cs *(*cs_create)(struct radeon_cs_manager *csm,
+                                   uint32_t ndw);
+    int (*cs_write_reloc)(struct radeon_cs *cs,
+                          struct radeon_bo *bo,
+                          uint32_t read_domain,
+                          uint32_t write_domain,
+                          uint32_t flags);
+    int (*cs_begin)(struct radeon_cs *cs,
+                    uint32_t ndw,
+                    const char *file,
+                    const char *func,
+                    int line);
+    int (*cs_end)(struct radeon_cs *cs,
+                  const char *file,
+                  const char *func,
+                  int line);
+    int (*cs_emit)(struct radeon_cs *cs);
+    int (*cs_destroy)(struct radeon_cs *cs);
+    int (*cs_erase)(struct radeon_cs *cs);
+    int (*cs_need_flush)(struct radeon_cs *cs);
+    void (*cs_print)(struct radeon_cs *cs, FILE *file);
+    int (*cs_space_check)(struct radeon_cs *cs, struct radeon_cs_space_check *bos,
+			  int num_bo);
+};
+
+struct radeon_cs_manager {
+    struct radeon_cs_funcs  *funcs;
+    int                     fd;
+    uint32_t vram_limit, gart_limit;
+    uint32_t vram_write_used, gart_write_used;
+    uint32_t read_used;
+};
+
+static inline struct radeon_cs *radeon_cs_create(struct radeon_cs_manager *csm,
+                                                 uint32_t ndw)
+{
+    return csm->funcs->cs_create(csm, ndw);
+}
+
+static inline int radeon_cs_write_reloc(struct radeon_cs *cs,
+                                        struct radeon_bo *bo,
+                                        uint32_t read_domain,
+                                        uint32_t write_domain,
+                                        uint32_t flags)
+{
+    return cs->csm->funcs->cs_write_reloc(cs,
+                                          bo,
+                                          read_domain,
+                                          write_domain,
+                                          flags);
+}
+
+static inline int radeon_cs_begin(struct radeon_cs *cs,
+                                  uint32_t ndw,
+                                  const char *file,
+                                  const char *func,
+                                  int line)
+{
+    return cs->csm->funcs->cs_begin(cs, ndw, file, func, line);
+}
+
+static inline int radeon_cs_end(struct radeon_cs *cs,
+                                const char *file,
+                                const char *func,
+                                int line)
+{
+    return cs->csm->funcs->cs_end(cs, file, func, line);
+}
+
+static inline int radeon_cs_emit(struct radeon_cs *cs)
+{
+    return cs->csm->funcs->cs_emit(cs);
+}
+
+static inline int radeon_cs_destroy(struct radeon_cs *cs)
+{
+    return cs->csm->funcs->cs_destroy(cs);
+}
+
+static inline int radeon_cs_erase(struct radeon_cs *cs)
+{
+    return cs->csm->funcs->cs_erase(cs);
+}
+
+static inline int radeon_cs_need_flush(struct radeon_cs *cs)
+{
+    return cs->csm->funcs->cs_need_flush(cs);
+}
+
+static inline void radeon_cs_print(struct radeon_cs *cs, FILE *file)
+{
+    cs->csm->funcs->cs_print(cs, file);
+}
+
+static inline int radeon_cs_space_check(struct radeon_cs *cs,
+					    struct radeon_cs_space_check *bos,
+					    int num_bo)
+{
+    return cs->csm->funcs->cs_space_check(cs, bos, num_bo);
+}
+
+static inline void radeon_cs_set_limit(struct radeon_cs *cs, uint32_t domain, uint32_t limit)
+{
+    
+    if (domain == RADEON_GEM_DOMAIN_VRAM)
+	cs->csm->vram_limit = limit;
+    else
+	cs->csm->gart_limit = limit;
+}
+
+static inline void radeon_cs_write_dword(struct radeon_cs *cs, uint32_t dword)
+{
+    cs->packets[cs->cdw++] = dword;
+    if (cs->section) {
+        cs->section_cdw++;
+    }
+}
+
+static inline void radeon_cs_write_qword(struct radeon_cs *cs, uint64_t qword)
+{
+
+    memcpy(cs->packets + cs->cdw, &qword, sizeof(qword));
+    cs->cdw+=2;
+    if (cs->section) {
+        cs->section_cdw+=2;
+    }
+}
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
new file mode 100644
index 00000000000..e4ee2b99151
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
@@ -0,0 +1,496 @@
+/* 
+ * Copyright © 2008 Nicolai Haehnle
+ * Copyright © 2008 Jérôme Glisse
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ */
+/*
+ * Authors:
+ *      Aapo Tahkola <[email protected]>
+ *      Nicolai Haehnle <[email protected]>
+ *      Jérôme Glisse <[email protected]>
+ */
+#include <errno.h>
+
+#include "radeon_bocs_wrapper.h"
+
+struct cs_manager_legacy {
+    struct radeon_cs_manager    base;
+    struct radeon_context       *ctx;
+    /* hack for scratch stuff */
+    uint32_t                    pending_age;
+    uint32_t                    pending_count;
+
+
+};
+
+struct cs_reloc_legacy {
+    struct radeon_cs_reloc  base;
+    uint32_t                cindices;
+    uint32_t                *indices;
+};
+
+
+static struct radeon_cs *cs_create(struct radeon_cs_manager *csm,
+                                   uint32_t ndw)
+{
+    struct radeon_cs *cs;
+
+    cs = (struct radeon_cs*)calloc(1, sizeof(struct radeon_cs));
+    if (cs == NULL) {
+        return NULL;
+    }
+    cs->csm = csm;
+    cs->ndw = (ndw + 0x3FF) & (~0x3FF);
+    cs->packets = (uint32_t*)malloc(4*cs->ndw);
+    if (cs->packets == NULL) {
+        free(cs);
+        return NULL;
+    }
+    cs->relocs_total_size = 0;
+    return cs;
+}
+
+static int cs_write_reloc(struct radeon_cs *cs,
+                          struct radeon_bo *bo,
+                          uint32_t read_domain,
+                          uint32_t write_domain,
+                          uint32_t flags)
+{
+    struct cs_reloc_legacy *relocs;
+    int i;
+
+    relocs = (struct cs_reloc_legacy *)cs->relocs;
+    /* check domains */
+    if ((read_domain && write_domain) || (!read_domain && !write_domain)) {
+        /* in one CS a bo can only be in read or write domain but not
+         * in read & write domain at the same sime
+         */
+        return -EINVAL;
+    }
+    if (read_domain == RADEON_GEM_DOMAIN_CPU) {
+        return -EINVAL;
+    }
+    if (write_domain == RADEON_GEM_DOMAIN_CPU) {
+        return -EINVAL;
+    }
+    /* check if bo is already referenced */
+    for(i = 0; i < cs->crelocs; i++) {
+        uint32_t *indices;
+
+        if (relocs[i].base.bo->handle == bo->handle) {
+            /* Check domains must be in read or write. As we check already
+             * checked that in argument one of the read or write domain was
+             * set we only need to check that if previous reloc as the read
+             * domain set then the read_domain should also be set for this
+             * new relocation.
+             */
+            if (relocs[i].base.read_domain && !read_domain) {
+                return -EINVAL;
+            }
+            if (relocs[i].base.write_domain && !write_domain) {
+                return -EINVAL;
+            }
+            relocs[i].base.read_domain |= read_domain;
+            relocs[i].base.write_domain |= write_domain;
+            /* save indice */
+            relocs[i].cindices++;
+            indices = (uint32_t*)realloc(relocs[i].indices,
+                                         relocs[i].cindices * 4);
+            if (indices == NULL) {
+                relocs[i].cindices -= 1;
+                return -ENOMEM;
+            }
+            relocs[i].indices = indices;
+            relocs[i].indices[relocs[i].cindices - 1] = cs->cdw - 1;
+            return 0;
+        }
+    }
+    /* add bo to reloc */
+    relocs = (struct cs_reloc_legacy*)
+             realloc(cs->relocs,
+                     sizeof(struct cs_reloc_legacy) * (cs->crelocs + 1));
+    if (relocs == NULL) {
+        return -ENOMEM;
+    }
+    cs->relocs = relocs;
+    relocs[cs->crelocs].base.bo = bo;
+    relocs[cs->crelocs].base.read_domain = read_domain;
+    relocs[cs->crelocs].base.write_domain = write_domain;
+    relocs[cs->crelocs].base.flags = flags;
+    relocs[cs->crelocs].indices = (uint32_t*)malloc(4);
+    if (relocs[cs->crelocs].indices == NULL) {
+        return -ENOMEM;
+    }
+    relocs[cs->crelocs].indices[0] = cs->cdw - 1;
+    relocs[cs->crelocs].cindices = 1;
+    cs->relocs_total_size += radeon_bo_legacy_relocs_size(bo);
+    cs->crelocs++;
+    radeon_bo_ref(bo);
+    return 0;
+}
+
+static int cs_begin(struct radeon_cs *cs,
+                    uint32_t ndw,
+                    const char *file,
+                    const char *func,
+                    int line)
+{
+    if (cs->section) {
+        fprintf(stderr, "CS already in a section(%s,%s,%d)\n",
+                cs->section_file, cs->section_func, cs->section_line);
+        fprintf(stderr, "CS can't start section(%s,%s,%d)\n",
+                file, func, line);
+        return -EPIPE;
+    }
+    cs->section = 1;
+    cs->section_ndw = ndw;
+    cs->section_cdw = 0;
+    cs->section_file = file;
+    cs->section_func = func;
+    cs->section_line = line;
+
+
+    if (cs->cdw + ndw > cs->ndw) {
+        uint32_t tmp, *ptr;
+	int num = (ndw > 0x3FF) ? ndw : 0x3FF;
+
+        tmp = (cs->cdw + 1 + num) & (~num);
+        ptr = (uint32_t*)realloc(cs->packets, 4 * tmp);
+        if (ptr == NULL) {
+            return -ENOMEM;
+        }
+        cs->packets = ptr;
+        cs->ndw = tmp;
+    }
+
+    return 0;
+}
+
+static int cs_end(struct radeon_cs *cs,
+                  const char *file,
+                  const char *func,
+                  int line)
+
+{
+    if (!cs->section) {
+        fprintf(stderr, "CS no section to end at (%s,%s,%d)\n",
+                file, func, line);
+        return -EPIPE;
+    }
+    cs->section = 0;
+    if (cs->section_ndw != cs->section_cdw) {
+        fprintf(stderr, "CS section size missmatch start at (%s,%s,%d) %d vs %d\n",
+                cs->section_file, cs->section_func, cs->section_line, cs->section_ndw, cs->section_cdw);
+        fprintf(stderr, "CS section end at (%s,%s,%d)\n",
+                file, func, line);
+        return -EPIPE;
+    }
+    return 0;
+}
+
+static int cs_process_relocs(struct radeon_cs *cs)
+{
+    struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm;
+    struct cs_reloc_legacy *relocs;
+    int i, j, r;
+
+    csm = (struct cs_manager_legacy*)cs->csm;
+    relocs = (struct cs_reloc_legacy *)cs->relocs;
+ restart:
+    for (i = 0; i < cs->crelocs; i++) {
+        for (j = 0; j < relocs[i].cindices; j++) {
+            uint32_t soffset, eoffset;
+
+            r = radeon_bo_legacy_validate(relocs[i].base.bo,
+                                           &soffset, &eoffset);
+	    if (r == -EAGAIN)
+	      goto restart;
+            if (r) {
+                fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n",
+                        relocs[i].base.bo, soffset, eoffset);
+                return r;
+            }
+            cs->packets[relocs[i].indices[j]] += soffset;
+            if (cs->packets[relocs[i].indices[j]] >= eoffset) {
+	      /*                radeon_bo_debug(relocs[i].base.bo, 12); */
+                fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n",
+                        relocs[i].base.bo, soffset, eoffset);
+                fprintf(stderr, "above end: %p 0x%08X 0x%08X\n",
+                        relocs[i].base.bo,
+                        cs->packets[relocs[i].indices[j]],
+                        eoffset);
+                exit(0);
+                return -EINVAL;
+            }
+        }
+    }
+    return 0;
+}
+
+static int cs_set_age(struct radeon_cs *cs)
+{
+    struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm;
+    struct cs_reloc_legacy *relocs;
+    int i;
+
+    relocs = (struct cs_reloc_legacy *)cs->relocs;
+    for (i = 0; i < cs->crelocs; i++) {
+        radeon_bo_legacy_pending(relocs[i].base.bo, csm->pending_age);
+        radeon_bo_unref(relocs[i].base.bo);
+    }
+    return 0;
+}
+
+static int cs_emit(struct radeon_cs *cs)
+{
+    struct cs_manager_legacy *csm = (struct cs_manager_legacy*)cs->csm;
+    drm_radeon_cmd_buffer_t cmd;
+    drm_r300_cmd_header_t age;
+    uint64_t ull;
+    int r;
+
+    csm->ctx->vtbl.emit_cs_header(cs, csm->ctx);
+
+    /* append buffer age */
+    if (IS_R300_CLASS(csm->ctx->radeonScreen)) {
+      age.scratch.cmd_type = R300_CMD_SCRATCH;
+      /* Scratch register 2 corresponds to what radeonGetAge polls */
+      csm->pending_age = 0;
+      csm->pending_count = 1;
+      ull = (uint64_t) (intptr_t) &csm->pending_age;
+      age.scratch.reg = 2;
+      age.scratch.n_bufs = 1;
+      age.scratch.flags = 0;
+      radeon_cs_write_dword(cs, age.u);
+      radeon_cs_write_qword(cs, ull);
+      radeon_cs_write_dword(cs, 0);
+    }
+
+    r = cs_process_relocs(cs);
+    if (r) {
+        return 0;
+    }
+
+    cmd.buf = (char *)cs->packets;
+    cmd.bufsz = cs->cdw * 4;
+    if (csm->ctx->state.scissor.enabled) {
+        cmd.nbox = csm->ctx->state.scissor.numClipRects;
+        cmd.boxes = (drm_clip_rect_t *) csm->ctx->state.scissor.pClipRects;
+    } else {
+        cmd.nbox = csm->ctx->numClipRects;
+        cmd.boxes = (drm_clip_rect_t *) csm->ctx->pClipRects;
+    }
+
+    //dump_cmdbuf(cs);
+
+    r = drmCommandWrite(cs->csm->fd, DRM_RADEON_CMDBUF, &cmd, sizeof(cmd));
+    if (r) {
+        return r;
+    }
+    if (!IS_R300_CLASS(csm->ctx->radeonScreen)) {
+	drm_radeon_irq_emit_t emit_cmd;
+	emit_cmd.irq_seq = &csm->pending_age;
+	r = drmCommandWrite(cs->csm->fd, DRM_RADEON_IRQ_EMIT, &emit_cmd, sizeof(emit_cmd));
+	if (r) {
+		return r;
+	}
+    }
+    cs_set_age(cs);
+
+    cs->csm->read_used = 0;
+    cs->csm->vram_write_used = 0;
+    cs->csm->gart_write_used = 0;
+    return 0;
+}
+
+static void inline cs_free_reloc(void *relocs_p, int crelocs)
+{
+    struct cs_reloc_legacy *relocs = relocs_p;
+    int i;
+    if (!relocs_p)
+      return;
+    for (i = 0; i < crelocs; i++)
+      free(relocs[i].indices);
+}
+
+static int cs_destroy(struct radeon_cs *cs)
+{
+    cs_free_reloc(cs->relocs, cs->crelocs);
+    free(cs->relocs);
+    free(cs->packets);
+    free(cs);
+    return 0;
+}
+
+static int cs_erase(struct radeon_cs *cs)
+{
+    cs_free_reloc(cs->relocs, cs->crelocs);
+    free(cs->relocs);
+    cs->relocs_total_size = 0;
+    cs->relocs = NULL;
+    cs->crelocs = 0;
+    cs->cdw = 0;
+    cs->section = 0;
+    return 0;
+}
+
+static int cs_need_flush(struct radeon_cs *cs)
+{
+    /* this function used to flush when the BO usage got to
+     * a certain size, now the higher levels handle this better */
+    return 0;
+}
+
+static void cs_print(struct radeon_cs *cs, FILE *file)
+{
+}
+
+static int cs_check_space(struct radeon_cs *cs, struct radeon_cs_space_check *bos, int num_bo)
+{
+    struct radeon_cs_manager *csm = cs->csm;
+    int this_op_read = 0, this_op_gart_write = 0, this_op_vram_write = 0;
+    uint32_t read_domains, write_domain;
+    int i;
+    struct radeon_bo *bo;
+
+    /* check the totals for this operation */
+
+    if (num_bo == 0)
+        return 0;
+
+    /* prepare */
+    for (i = 0; i < num_bo; i++) {
+      bo = bos[i].bo;
+
+      bos[i].new_accounted = 0;
+      read_domains = bos[i].read_domains;
+      write_domain = bos[i].write_domain;
+		
+      /* pinned bos don't count */
+      if (radeon_legacy_bo_is_static(bo))
+	  continue;
+ 
+      /* already accounted this bo */
+      if (write_domain && (write_domain == bo->space_accounted))
+	  continue;
+
+      if (read_domains && ((read_domains << 16) == bo->space_accounted))
+	  continue;
+      
+      if (bo->space_accounted == 0) {
+	  if (write_domain == RADEON_GEM_DOMAIN_VRAM)
+	      this_op_vram_write += bo->size;
+	  else if (write_domain == RADEON_GEM_DOMAIN_GTT)
+	      this_op_gart_write += bo->size;
+	  else
+	      this_op_read += bo->size;
+	  bos[i].new_accounted = (read_domains << 16) | write_domain;
+      } else {
+	  uint16_t old_read, old_write;
+	  
+	  old_read = bo->space_accounted >> 16;
+	  old_write = bo->space_accounted & 0xffff;
+
+	  if (write_domain && (old_read & write_domain)) {
+	      bos[i].new_accounted = write_domain;
+	      /* moving from read to a write domain */
+	      if (write_domain == RADEON_GEM_DOMAIN_VRAM) {
+		  this_op_read -= bo->size;
+		  this_op_vram_write += bo->size;
+	      } else if (write_domain == RADEON_GEM_DOMAIN_VRAM) {
+		  this_op_read -= bo->size;
+		  this_op_gart_write += bo->size;
+	      }
+	  } else if (read_domains & old_write) {
+	      bos[i].new_accounted = bo->space_accounted & 0xffff;
+	  } else {
+	      /* rewrite the domains */
+	      if (write_domain != old_write)
+		  fprintf(stderr,"WRITE DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, write_domain, old_write);
+	      if (read_domains != old_read)
+		  fprintf(stderr,"READ DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, read_domains, old_read);
+	      return RADEON_CS_SPACE_FLUSH;
+	  }
+      }
+	}
+	
+	if (this_op_read < 0)
+		this_op_read = 0;
+
+	/* check sizes - operation first */
+	if ((this_op_read + this_op_gart_write > csm->gart_limit) ||
+	    (this_op_vram_write > csm->vram_limit)) {
+	    return RADEON_CS_SPACE_OP_TO_BIG;
+	}
+
+	if (((csm->vram_write_used + this_op_vram_write) > csm->vram_limit) ||
+	    ((csm->read_used + csm->gart_write_used + this_op_gart_write + this_op_read) > csm->gart_limit)) {
+		return RADEON_CS_SPACE_FLUSH;
+	}
+
+	csm->gart_write_used += this_op_gart_write;
+	csm->vram_write_used += this_op_vram_write;
+	csm->read_used += this_op_read;
+	/* commit */
+	for (i = 0; i < num_bo; i++) {
+		bo = bos[i].bo;
+		bo->space_accounted = bos[i].new_accounted;
+	}
+
+	return RADEON_CS_SPACE_OK;
+}
+
+static struct radeon_cs_funcs  radeon_cs_legacy_funcs = {
+    cs_create,
+    cs_write_reloc,
+    cs_begin,
+    cs_end,
+    cs_emit,
+    cs_destroy,
+    cs_erase,
+    cs_need_flush,
+    cs_print,
+    cs_check_space
+};
+
+struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx)
+{
+    struct cs_manager_legacy *csm;
+
+    csm = (struct cs_manager_legacy*)
+          calloc(1, sizeof(struct cs_manager_legacy));
+    if (csm == NULL) {
+        return NULL;
+    }
+    csm->base.funcs = &radeon_cs_legacy_funcs;
+    csm->base.fd = ctx->dri.fd;
+    csm->ctx = ctx;
+    csm->pending_age = 1;
+    return (struct radeon_cs_manager*)csm;
+}
+
+void radeon_cs_manager_legacy_dtor(struct radeon_cs_manager *csm)
+{
+    free(csm);
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h
new file mode 100644
index 00000000000..e177b4bafe2
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h
@@ -0,0 +1,40 @@
+/* 
+ * Copyright © 2008 Nicolai Haehnle
+ * Copyright © 2008 Jérôme Glisse
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ */
+/*
+ * Authors:
+ *      Aapo Tahkola <[email protected]>
+ *      Nicolai Haehnle <[email protected]>
+ *      Jérôme Glisse <[email protected]>
+ */
+#ifndef RADEON_CS_LEGACY_H
+#define RADEON_CS_LEGACY_H
+
+#include "radeon_common.h"
+
+struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx);
+void radeon_cs_manager_legacy_dtor(struct radeon_cs_manager *csm);
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c
new file mode 100644
index 00000000000..48b0d638183
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.c
@@ -0,0 +1,341 @@
+/**************************************************************************
+
+Copyright (C) 2004 Nicolai Haehnle.
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+#include "radeon_common.h"
+
+#if defined(USE_X86_ASM)
+#define COPY_DWORDS( dst, src, nr )					\
+do {									\
+	int __tmp;							\
+	__asm__ __volatile__( "rep ; movsl"				\
+			      : "=%c" (__tmp), "=D" (dst), "=S" (__tmp)	\
+			      : "0" (nr),				\
+			        "D" ((long)dst),			\
+			        "S" ((long)src) );			\
+} while (0)
+#else
+#define COPY_DWORDS( dst, src, nr )		\
+do {						\
+   int j;					\
+   for ( j = 0 ; j < nr ; j++ )			\
+      dst[j] = ((int *)src)[j];			\
+   dst += nr;					\
+} while (0)
+#endif
+
+static void radeonEmitVec4(uint32_t *out, GLvoid * data, int stride, int count)
+{
+	int i;
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
+			__FUNCTION__, count, stride, (void *)out, (void *)data);
+
+	if (stride == 4)
+		COPY_DWORDS(out, data, count);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out++;
+			data += stride;
+		}
+}
+
+void radeonEmitVec8(uint32_t *out, GLvoid * data, int stride, int count)
+{
+	int i;
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
+			__FUNCTION__, count, stride, (void *)out, (void *)data);
+
+	if (stride == 8)
+		COPY_DWORDS(out, data, count * 2);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out[1] = *(int *)(data + 4);
+			out += 2;
+			data += stride;
+		}
+}
+
+void radeonEmitVec12(uint32_t *out, GLvoid * data, int stride, int count)
+{
+	int i;
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
+			__FUNCTION__, count, stride, (void *)out, (void *)data);
+
+	if (stride == 12) {
+		COPY_DWORDS(out, data, count * 3);
+    }
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out[1] = *(int *)(data + 4);
+			out[2] = *(int *)(data + 8);
+			out += 3;
+			data += stride;
+		}
+}
+
+static void radeonEmitVec16(uint32_t *out, GLvoid * data, int stride, int count)
+{
+	int i;
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
+			__FUNCTION__, count, stride, (void *)out, (void *)data);
+
+	if (stride == 16)
+		COPY_DWORDS(out, data, count * 4);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out[1] = *(int *)(data + 4);
+			out[2] = *(int *)(data + 8);
+			out[3] = *(int *)(data + 12);
+			out += 4;
+			data += stride;
+		}
+}
+
+void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
+			 GLvoid * data, int size, int stride, int count)
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	uint32_t *out;
+
+	if (stride == 0) {
+		radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32);
+		count = 1;
+		aos->stride = 0;
+	} else {
+		radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * count * 4, 32);
+		aos->stride = size;
+	}
+
+	aos->components = size;
+	aos->count = count;
+
+	out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
+	switch (size) {
+	case 1: radeonEmitVec4(out, data, stride, count); break;
+	case 2: radeonEmitVec8(out, data, stride, count); break;
+	case 3: radeonEmitVec12(out, data, stride, count); break;
+	case 4: radeonEmitVec16(out, data, stride, count); break;
+	default:
+		assert(0);
+		break;
+	}
+}
+
+void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
+{
+
+	size = MAX2(size, MAX_DMA_BUF_SZ);
+
+	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
+		fprintf(stderr, "%s\n", __FUNCTION__);
+
+	if (rmesa->dma.flush) {
+		rmesa->dma.flush(rmesa->glCtx);
+	}
+
+	if (rmesa->dma.nr_released_bufs > 4) {
+		rcommonFlushCmdBuf(rmesa, __FUNCTION__);
+		rmesa->dma.nr_released_bufs = 0;
+	}
+
+	if (rmesa->dma.current) {
+		radeon_bo_unmap(rmesa->dma.current);
+		radeon_bo_unref(rmesa->dma.current);
+		rmesa->dma.current = 0;
+	}
+
+again_alloc:	
+	rmesa->dma.current = radeon_bo_open(rmesa->radeonScreen->bom,
+					    0, size, 4, RADEON_GEM_DOMAIN_GTT,
+					    0);
+
+	if (!rmesa->dma.current) {
+		rcommonFlushCmdBuf(rmesa, __FUNCTION__);
+		rmesa->dma.nr_released_bufs = 0;
+		goto again_alloc;
+	}
+
+	rmesa->dma.current_used = 0;
+	rmesa->dma.current_vertexptr = 0;
+	
+	radeon_validate_bo(rmesa, rmesa->dma.current, RADEON_GEM_DOMAIN_GTT, 0);
+
+	if (radeon_revalidate_bos(rmesa->glCtx) == GL_FALSE)
+	  fprintf(stderr,"failure to revalidate BOs - badness\n");
+
+	if (!rmesa->dma.current) {
+        /* Cmd buff have been flushed in radeon_revalidate_bos */
+		rmesa->dma.nr_released_bufs = 0;
+		goto again_alloc;
+	}
+
+	radeon_bo_map(rmesa->dma.current, 1);
+}
+
+/* Allocates a region from rmesa->dma.current.  If there isn't enough
+ * space in current, grab a new buffer (and discard what was left of current)
+ */
+void radeonAllocDmaRegion(radeonContextPtr rmesa,
+			  struct radeon_bo **pbo, int *poffset,
+			  int bytes, int alignment)
+{
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
+
+	if (rmesa->dma.flush)
+		rmesa->dma.flush(rmesa->glCtx);
+
+	assert(rmesa->dma.current_used == rmesa->dma.current_vertexptr);
+
+	alignment--;
+	rmesa->dma.current_used = (rmesa->dma.current_used + alignment) & ~alignment;
+
+	if (!rmesa->dma.current || rmesa->dma.current_used + bytes > rmesa->dma.current->size)
+		radeonRefillCurrentDmaRegion(rmesa, (bytes + 15) & ~15);
+
+	*poffset = rmesa->dma.current_used;
+	*pbo = rmesa->dma.current;
+	radeon_bo_ref(*pbo);
+
+	/* Always align to at least 16 bytes */
+	rmesa->dma.current_used = (rmesa->dma.current_used + bytes + 15) & ~15;
+	rmesa->dma.current_vertexptr = rmesa->dma.current_used;
+
+	assert(rmesa->dma.current_used <= rmesa->dma.current->size);
+}
+
+void radeonReleaseDmaRegion(radeonContextPtr rmesa)
+{
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s %p\n", __FUNCTION__, rmesa->dma.current);
+	if (rmesa->dma.current) {
+		rmesa->dma.nr_released_bufs++;
+		radeon_bo_unmap(rmesa->dma.current);
+	        radeon_bo_unref(rmesa->dma.current);
+	}
+	rmesa->dma.current = NULL;
+}
+
+
+/* Flush vertices in the current dma region.
+ */
+void rcommon_flush_last_swtcl_prim( GLcontext *ctx  )
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	struct radeon_dma *dma = &rmesa->dma;
+		
+
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s %p\n", __FUNCTION__, dma->current);
+	dma->flush = NULL;
+
+	if (dma->current) {
+	    GLuint current_offset = dma->current_used;
+
+	    assert (dma->current_used +
+		    rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
+		    dma->current_vertexptr);
+
+	    if (dma->current_used != dma->current_vertexptr) {
+		    dma->current_used = dma->current_vertexptr;
+
+		    rmesa->vtbl.swtcl_flush(ctx, current_offset);
+	    }
+	    rmesa->swtcl.numverts = 0;
+	}
+}
+/* Alloc space in the current dma region.
+ */
+void *
+rcommonAllocDmaLowVerts( radeonContextPtr rmesa, int nverts, int vsize )
+{
+	GLuint bytes = vsize * nverts;
+	void *head;
+restart:
+	if (!rmesa->dma.current || rmesa->dma.current_vertexptr + bytes > rmesa->dma.current->size) {
+                radeonRefillCurrentDmaRegion(rmesa, bytes);
+	}
+
+        if (!rmesa->dma.flush) {
+		/* make sure we have enough space to use this in cmdbuf */
+   		rcommonEnsureCmdBufSpace(rmesa,
+			      rmesa->hw.max_state_size + (12*sizeof(int)),
+			      __FUNCTION__);
+		/* if cmdbuf flushed DMA restart */
+		if (!rmesa->dma.current)
+			goto restart;
+                rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+                rmesa->dma.flush = rcommon_flush_last_swtcl_prim;
+        }
+
+	ASSERT( vsize == rmesa->swtcl.vertex_size * 4 );
+        ASSERT( rmesa->dma.flush == rcommon_flush_last_swtcl_prim );
+        ASSERT( rmesa->dma.current_used +
+                rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
+                rmesa->dma.current_vertexptr );
+
+	head = (rmesa->dma.current->ptr + rmesa->dma.current_vertexptr);
+	rmesa->dma.current_vertexptr += bytes;
+	rmesa->swtcl.numverts += nverts;
+	return head;
+}
+
+void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs )
+{
+   radeonContextPtr radeon = RADEON_CONTEXT( ctx );
+   int i;
+
+   if (radeon->dma.flush) {
+       radeon->dma.flush(radeon->glCtx);
+   }
+   if (radeon->tcl.elt_dma_bo) {
+	   radeon_bo_unref(radeon->tcl.elt_dma_bo);
+	   radeon->tcl.elt_dma_bo = NULL;
+   }
+   for (i = 0; i < radeon->tcl.aos_count; i++) {
+      if (radeon->tcl.aos[i].bo) {
+         radeon_bo_unref(radeon->tcl.aos[i].bo);
+         radeon->tcl.aos[i].bo = NULL;
+      }
+   }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.h b/src/mesa/drivers/dri/radeon/radeon_dma.h
new file mode 100644
index 00000000000..06e388fc1de
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.h
@@ -0,0 +1,52 @@
+/**************************************************************************
+
+Copyright (C) 2004 Nicolai Haehnle.
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+#ifndef RADEON_DMA_H
+#define RADEON_DMA_H
+
+void radeonEmitVec8(uint32_t *out, GLvoid * data, int stride, int count);
+void radeonEmitVec12(uint32_t *out, GLvoid * data, int stride, int count);
+
+void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
+			 GLvoid * data, int size, int stride, int count);
+
+void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size);
+void radeonAllocDmaRegion(radeonContextPtr rmesa,
+			  struct radeon_bo **pbo, int *poffset,
+			  int bytes, int alignment);
+void radeonReleaseDmaRegion(radeonContextPtr rmesa);
+
+void rcommon_flush_last_swtcl_prim(GLcontext *ctx);
+
+void *rcommonAllocDmaLowVerts(radeonContextPtr rmesa, int nverts, int vsize);
+void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs );
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_fbo.c b/src/mesa/drivers/dri/radeon/radeon_fbo.c
new file mode 100644
index 00000000000..f62ca7f9eb3
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_fbo.c
@@ -0,0 +1,588 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Red Hat Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "main/imports.h"
+#include "main/macros.h"
+#include "main/mtypes.h"
+#include "main/fbobject.h"
+#include "main/framebuffer.h"
+#include "main/renderbuffer.h"
+#include "main/context.h"
+#include "main/texformat.h"
+#include "main/texrender.h"
+
+#include "radeon_common.h"
+#include "radeon_mipmap_tree.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+#define DBG(...) do {                                           \
+        if (RADEON_DEBUG & FILE_DEBUG_FLAG)                      \
+                _mesa_printf(__VA_ARGS__);                      \
+} while(0)
+
+static struct gl_framebuffer *
+radeon_new_framebuffer(GLcontext *ctx, GLuint name)
+{
+  return _mesa_new_framebuffer(ctx, name);
+}
+
+static void
+radeon_delete_renderbuffer(struct gl_renderbuffer *rb)
+{
+  struct radeon_renderbuffer *rrb = radeon_renderbuffer(rb);
+
+  ASSERT(rrb);
+
+  if (rrb && rrb->bo) {
+    radeon_bo_unref(rrb->bo);
+  }
+  _mesa_free(rrb);
+}
+
+static void *
+radeon_get_pointer(GLcontext *ctx, struct gl_renderbuffer *rb,
+		   GLint x, GLint y)
+{
+  return NULL;
+}
+
+/**
+ * Called via glRenderbufferStorageEXT() to set the format and allocate
+ * storage for a user-created renderbuffer.
+ */
+static GLboolean
+radeon_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                                 GLenum internalFormat,
+                                 GLuint width, GLuint height)
+{
+  struct radeon_context *radeon = RADEON_CONTEXT(ctx);
+  struct radeon_renderbuffer *rrb = radeon_renderbuffer(rb);
+  GLboolean software_buffer = GL_FALSE;
+  int cpp;
+
+   ASSERT(rb->Name != 0);
+  switch (internalFormat) {
+   case GL_R3_G3_B2:
+   case GL_RGB4:
+   case GL_RGB5:
+      rb->_ActualFormat = GL_RGB5;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 5;
+      rb->GreenBits = 6;
+      rb->BlueBits = 5;
+      cpp = 2;
+      break;
+   case GL_RGB:
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      rb->_ActualFormat = GL_RGB8;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 8;
+      rb->GreenBits = 8;
+      rb->BlueBits = 8;
+      rb->AlphaBits = 0;
+      cpp = 4;
+      break;
+   case GL_RGBA:
+   case GL_RGBA2:
+   case GL_RGBA4:
+   case GL_RGB5_A1:
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      rb->_ActualFormat = GL_RGBA8;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 8;
+      rb->GreenBits = 8;
+      rb->BlueBits = 8;
+      rb->AlphaBits = 8;
+      cpp = 4;
+      break;
+   case GL_STENCIL_INDEX:
+   case GL_STENCIL_INDEX1_EXT:
+   case GL_STENCIL_INDEX4_EXT:
+   case GL_STENCIL_INDEX8_EXT:
+   case GL_STENCIL_INDEX16_EXT:
+      /* alloc a depth+stencil buffer */
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->StencilBits = 8;
+      cpp = 4;
+      break;
+   case GL_DEPTH_COMPONENT16:
+      rb->_ActualFormat = GL_DEPTH_COMPONENT16;
+      rb->DataType = GL_UNSIGNED_SHORT;
+      rb->DepthBits = 16;
+      cpp = 2;
+      break;
+   case GL_DEPTH_COMPONENT:
+   case GL_DEPTH_COMPONENT24:
+   case GL_DEPTH_COMPONENT32:
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->DepthBits = 24;
+      cpp = 4;
+      break;
+   case GL_DEPTH_STENCIL_EXT:
+   case GL_DEPTH24_STENCIL8_EXT:
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->DepthBits = 24;
+      rb->StencilBits = 8;
+      cpp = 4;
+      break;
+   default:
+      _mesa_problem(ctx,
+                    "Unexpected format in intel_alloc_renderbuffer_storage");
+      return GL_FALSE;
+   }
+
+  radeonFlush(ctx);
+
+  if (rrb->bo)
+    radeon_bo_unref(rrb->bo);
+  
+    
+   if (software_buffer) {
+      return _mesa_soft_renderbuffer_storage(ctx, rb, internalFormat,
+                                             width, height);
+   }
+   else {
+     uint32_t size = width * height * cpp;
+     uint32_t pitch = ((cpp * width + 63) & ~63) / cpp;
+
+     fprintf(stderr,"Allocating %d x %d radeon RBO (pitch %d)\n", width,
+	  height, pitch);
+
+     rrb->pitch = pitch * cpp;
+     rrb->cpp = cpp;
+     rrb->bo = radeon_bo_open(radeon->radeonScreen->bom,
+			      0,
+			      size,
+			      0,
+			      RADEON_GEM_DOMAIN_VRAM,
+			      0);
+     rb->Width = width;
+     rb->Height = height;
+       return GL_TRUE;
+   }    
+   
+}
+
+
+/**
+ * Called for each hardware renderbuffer when a _window_ is resized.
+ * Just update fields.
+ * Not used for user-created renderbuffers!
+ */
+static GLboolean
+radeon_alloc_window_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                           GLenum internalFormat, GLuint width, GLuint height)
+{
+   ASSERT(rb->Name == 0);
+   rb->Width = width;
+   rb->Height = height;
+   rb->_ActualFormat = internalFormat;
+
+   return GL_TRUE;
+}
+
+
+static void
+radeon_resize_buffers(GLcontext *ctx, struct gl_framebuffer *fb,
+		     GLuint width, GLuint height)
+{
+     struct radeon_framebuffer *radeon_fb = (struct radeon_framebuffer*)fb;
+   int i;
+
+   _mesa_resize_framebuffer(ctx, fb, width, height);
+
+   fb->Initialized = GL_TRUE; /* XXX remove someday */
+
+   if (fb->Name != 0) {
+      return;
+   }
+
+   /* Make sure all window system renderbuffers are up to date */
+   for (i = 0; i < 2; i++) {
+      struct gl_renderbuffer *rb = &radeon_fb->color_rb[i]->base;
+
+      /* only resize if size is changing */
+      if (rb && (rb->Width != width || rb->Height != height)) {
+	 rb->AllocStorage(ctx, rb, rb->InternalFormat, width, height);
+      }
+   }
+}
+
+
+/** Dummy function for gl_renderbuffer::AllocStorage() */
+static GLboolean
+radeon_nop_alloc_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+			 GLenum internalFormat, GLuint width, GLuint height)
+{
+   _mesa_problem(ctx, "radeon_op_alloc_storage should never be called.");
+   return GL_FALSE;
+}
+
+struct radeon_renderbuffer *
+radeon_create_renderbuffer(GLenum format, __DRIdrawablePrivate *driDrawPriv)
+{
+    struct radeon_renderbuffer *rrb;
+
+    rrb = CALLOC_STRUCT(radeon_renderbuffer);
+    if (!rrb)
+	return NULL;
+
+    _mesa_init_renderbuffer(&rrb->base, 0);
+    rrb->base.ClassID = RADEON_RB_CLASS;
+
+    /* XXX format junk */
+    switch (format) {
+	case GL_RGB5:
+	    rrb->base._ActualFormat = GL_RGB5;
+	    rrb->base._BaseFormat = GL_RGBA;
+	    rrb->base.RedBits = 5;
+	    rrb->base.GreenBits = 6;
+	    rrb->base.BlueBits = 5;
+	    rrb->base.DataType = GL_UNSIGNED_BYTE;
+	    break;
+	case GL_RGB8:
+	    rrb->base._ActualFormat = GL_RGB8;
+	    rrb->base._BaseFormat = GL_RGB;
+	    rrb->base.RedBits = 8;
+	    rrb->base.GreenBits = 8;
+	    rrb->base.BlueBits = 8;
+	    rrb->base.AlphaBits = 8;
+	    rrb->base.DataType = GL_UNSIGNED_BYTE;
+	    break;
+	case GL_RGBA8:
+	    rrb->base._ActualFormat = GL_RGBA8;
+	    rrb->base._BaseFormat = GL_RGBA;
+	    rrb->base.RedBits = 8;
+	    rrb->base.GreenBits = 8;
+	    rrb->base.BlueBits = 8;
+	    rrb->base.AlphaBits = 8;
+	    rrb->base.DataType = GL_UNSIGNED_BYTE;
+	    break;
+	case GL_STENCIL_INDEX8_EXT:
+	    rrb->base._ActualFormat = GL_STENCIL_INDEX8_EXT;
+	    rrb->base._BaseFormat = GL_STENCIL_INDEX;
+	    rrb->base.StencilBits = 8;
+	    rrb->base.DataType = GL_UNSIGNED_BYTE;
+	    break;
+	case GL_DEPTH_COMPONENT16:
+	    rrb->base._ActualFormat = GL_DEPTH_COMPONENT16;
+	    rrb->base._BaseFormat = GL_DEPTH_COMPONENT;
+	    rrb->base.DepthBits = 16;
+	    rrb->base.DataType = GL_UNSIGNED_SHORT;
+	    break;
+	case GL_DEPTH_COMPONENT24:
+	    rrb->base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+	    rrb->base._BaseFormat = GL_DEPTH_COMPONENT;
+	    rrb->base.DepthBits = 24;
+	    rrb->base.DataType = GL_UNSIGNED_INT;
+	    break;
+	case GL_DEPTH24_STENCIL8_EXT:
+	    rrb->base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+	    rrb->base._BaseFormat = GL_DEPTH_STENCIL_EXT;
+	    rrb->base.DepthBits = 24;
+	    rrb->base.StencilBits = 8;
+	    rrb->base.DataType = GL_UNSIGNED_INT_24_8_EXT;
+	    break;
+	default:
+	    fprintf(stderr, "%s: Unknown format 0x%04x\n", __FUNCTION__, format);
+	    _mesa_delete_renderbuffer(&rrb->base);
+	    return NULL;
+    }
+
+    rrb->dPriv = driDrawPriv;
+    rrb->base.InternalFormat = format;
+
+    rrb->base.Delete = radeon_delete_renderbuffer;
+    rrb->base.AllocStorage = radeon_alloc_window_storage;
+    rrb->base.GetPointer = radeon_get_pointer;
+
+    rrb->bo = NULL;
+    return rrb;
+}
+
+static struct gl_renderbuffer *
+radeon_new_renderbuffer(GLcontext * ctx, GLuint name)
+{
+  struct radeon_renderbuffer *rrb;
+
+  rrb = CALLOC_STRUCT(radeon_renderbuffer);
+  if (!rrb)
+    return NULL;
+
+  _mesa_init_renderbuffer(&rrb->base, name);
+  rrb->base.ClassID = RADEON_RB_CLASS;
+
+  rrb->base.Delete = radeon_delete_renderbuffer;
+  rrb->base.AllocStorage = radeon_alloc_renderbuffer_storage;
+  rrb->base.GetPointer = radeon_get_pointer;
+
+  return &rrb->base;
+}
+
+static void
+radeon_bind_framebuffer(GLcontext * ctx, GLenum target,
+                       struct gl_framebuffer *fb, struct gl_framebuffer *fbread)
+{
+   if (target == GL_FRAMEBUFFER_EXT || target == GL_DRAW_FRAMEBUFFER_EXT) {
+      radeon_draw_buffer(ctx, fb);
+   }
+   else {
+      /* don't need to do anything if target == GL_READ_FRAMEBUFFER_EXT */
+   }
+}
+
+static void
+radeon_framebuffer_renderbuffer(GLcontext * ctx,
+                               struct gl_framebuffer *fb,
+                               GLenum attachment, struct gl_renderbuffer *rb)
+{
+
+   radeonFlush(ctx);
+
+   _mesa_framebuffer_renderbuffer(ctx, fb, attachment, rb);
+   radeon_draw_buffer(ctx, fb);
+}
+
+
+static GLboolean
+radeon_update_wrapper(GLcontext *ctx, struct radeon_renderbuffer *rrb, 
+		     struct gl_texture_image *texImage)
+{
+	int retry = 0;
+restart:
+	if (texImage->TexFormat == &_mesa_texformat_argb8888) {
+		rrb->cpp = 4;
+		rrb->base._ActualFormat = GL_RGBA8;
+		rrb->base._BaseFormat = GL_RGBA;
+		rrb->base.DataType = GL_UNSIGNED_BYTE;
+		DBG("Render to RGBA8 texture OK\n");
+	}
+	else if (texImage->TexFormat == &_mesa_texformat_rgb565) {
+		rrb->cpp = 2;
+		rrb->base._ActualFormat = GL_RGB5;
+		rrb->base._BaseFormat = GL_RGB;
+		rrb->base.DataType = GL_UNSIGNED_SHORT;
+		DBG("Render to RGB5 texture OK\n");
+	}
+	else if (texImage->TexFormat == &_mesa_texformat_z16) {
+		rrb->cpp = 2;
+		rrb->base._ActualFormat = GL_DEPTH_COMPONENT16;
+		rrb->base._BaseFormat = GL_DEPTH_COMPONENT;
+		rrb->base.DataType = GL_UNSIGNED_SHORT;
+		DBG("Render to DEPTH16 texture OK\n");
+	}
+	else if (texImage->TexFormat == &_mesa_texformat_s8_z24) {
+		rrb->cpp = 4;
+		rrb->base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+		rrb->base._BaseFormat = GL_DEPTH_STENCIL_EXT;
+		rrb->base.DataType = GL_UNSIGNED_INT_24_8_EXT;
+		DBG("Render to DEPTH_STENCIL texture OK\n");
+	}
+	else {
+		/* try redoing the FBO */
+		if (retry == 1) {
+			DBG("Render to texture BAD FORMAT %d\n",
+			    texImage->TexFormat->MesaFormat);
+			return GL_FALSE;
+		}
+		texImage->TexFormat = radeonChooseTextureFormat(ctx, texImage->InternalFormat, 0,
+								texImage->TexFormat->DataType,
+								1);
+
+		retry++;
+		goto restart;
+	}
+	
+	rrb->pitch = texImage->Width * rrb->cpp;
+	rrb->base.InternalFormat = rrb->base._ActualFormat;
+	rrb->base.Width = texImage->Width;
+	rrb->base.Height = texImage->Height;
+	rrb->base.RedBits = texImage->TexFormat->RedBits;
+	rrb->base.GreenBits = texImage->TexFormat->GreenBits;
+	rrb->base.BlueBits = texImage->TexFormat->BlueBits;
+	rrb->base.AlphaBits = texImage->TexFormat->AlphaBits;
+	rrb->base.DepthBits = texImage->TexFormat->DepthBits;
+	
+	rrb->base.Delete = radeon_delete_renderbuffer;
+	rrb->base.AllocStorage = radeon_nop_alloc_storage;
+	
+	return GL_TRUE;
+}
+
+
+static struct radeon_renderbuffer *
+radeon_wrap_texture(GLcontext * ctx, struct gl_texture_image *texImage)
+{
+  const GLuint name = ~0;   /* not significant, but distinct for debugging */
+  struct radeon_renderbuffer *rrb;
+
+   /* make an radeon_renderbuffer to wrap the texture image */
+   rrb = CALLOC_STRUCT(radeon_renderbuffer);
+   if (!rrb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glFramebufferTexture");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&rrb->base, name);
+   rrb->base.ClassID = RADEON_RB_CLASS;
+
+   if (!radeon_update_wrapper(ctx, rrb, texImage)) {
+      _mesa_free(rrb);
+      return NULL;
+   }
+
+   return rrb;
+  
+}
+static void
+radeon_render_texture(GLcontext * ctx,
+                     struct gl_framebuffer *fb,
+                     struct gl_renderbuffer_attachment *att)
+{
+   struct gl_texture_image *newImage
+      = att->Texture->Image[att->CubeMapFace][att->TextureLevel];
+   struct radeon_renderbuffer *rrb = radeon_renderbuffer(att->Renderbuffer);
+   radeon_texture_image *radeon_image;
+   GLuint imageOffset;
+
+   (void) fb;
+
+   ASSERT(newImage);
+
+   if (newImage->Border != 0) {
+      /* Fallback on drawing to a texture with a border, which won't have a
+       * miptree.
+       */
+      _mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
+      _mesa_render_texture(ctx, fb, att);
+      return;
+   }
+   else if (!rrb) {
+      rrb = radeon_wrap_texture(ctx, newImage);
+      if (rrb) {
+         /* bind the wrapper to the attachment point */
+         _mesa_reference_renderbuffer(&att->Renderbuffer, &rrb->base);
+      }
+      else {
+         /* fallback to software rendering */
+         _mesa_render_texture(ctx, fb, att);
+         return;
+      }
+   }
+
+   if (!radeon_update_wrapper(ctx, rrb, newImage)) {
+       _mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
+       _mesa_render_texture(ctx, fb, att);
+       return;
+   }
+
+   DBG("Begin render texture tid %x tex=%u w=%d h=%d refcount=%d\n",
+       _glthread_GetID(),
+       att->Texture->Name, newImage->Width, newImage->Height,
+       rrb->base.RefCount);
+
+   /* point the renderbufer's region to the texture image region */
+   radeon_image = (radeon_texture_image *)newImage;
+   if (rrb->bo != radeon_image->mt->bo) {
+      if (rrb->bo)
+  	radeon_bo_unref(rrb->bo);
+      rrb->bo = radeon_image->mt->bo;
+      radeon_bo_ref(rrb->bo);
+   }
+
+   /* compute offset of the particular 2D image within the texture region */
+   imageOffset = radeon_miptree_image_offset(radeon_image->mt,
+                                            att->CubeMapFace,
+                                            att->TextureLevel);
+
+   if (att->Texture->Target == GL_TEXTURE_3D) {
+      GLuint offsets[6];
+      radeon_miptree_depth_offsets(radeon_image->mt, att->TextureLevel,
+				   offsets);
+      imageOffset += offsets[att->Zoffset];
+   }
+
+   /* store that offset in the region */
+   rrb->draw_offset = imageOffset;
+
+   /* update drawing region, etc */
+   radeon_draw_buffer(ctx, fb);
+}
+
+static void
+radeon_finish_render_texture(GLcontext * ctx,
+                            struct gl_renderbuffer_attachment *att)
+{
+
+}
+static void
+radeon_validate_framebuffer(GLcontext *ctx, struct gl_framebuffer *fb)
+{
+}
+
+static void
+radeon_blit_framebuffer(GLcontext *ctx,
+                       GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
+                       GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+                       GLbitfield mask, GLenum filter)
+{
+}
+
+void radeon_fbo_init(struct radeon_context *radeon)
+{
+  radeon->glCtx->Driver.NewFramebuffer = radeon_new_framebuffer;
+  radeon->glCtx->Driver.NewRenderbuffer = radeon_new_renderbuffer;
+  radeon->glCtx->Driver.BindFramebuffer = radeon_bind_framebuffer;
+  radeon->glCtx->Driver.FramebufferRenderbuffer = radeon_framebuffer_renderbuffer;
+  radeon->glCtx->Driver.RenderTexture = radeon_render_texture;
+  radeon->glCtx->Driver.FinishRenderTexture = radeon_finish_render_texture;
+  radeon->glCtx->Driver.ResizeBuffers = radeon_resize_buffers;
+  radeon->glCtx->Driver.ValidateFramebuffer = radeon_validate_framebuffer;
+  radeon->glCtx->Driver.BlitFramebuffer = radeon_blit_framebuffer;
+}
+
+  
+void radeon_renderbuffer_set_bo(struct radeon_renderbuffer *rb,
+				struct radeon_bo *bo)
+{
+  struct radeon_bo *old;
+  old = rb->bo;
+  rb->bo = bo;
+  radeon_bo_ref(bo);
+  if (old)
+    radeon_bo_unref(old);
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.c b/src/mesa/drivers/dri/radeon/radeon_ioctl.c
index 09acf6b4f85..01c45df2dfa 100644
--- a/src/mesa/drivers/dri/radeon/radeon_ioctl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.c
@@ -35,7 +35,21 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
 
 #include <sched.h>
-#include <errno.h> 
+#include <errno.h>
+
+#include "main/attrib.h"
+#include "main/enable.h"
+#include "main/blend.h"
+#include "main/bufferobj.h"
+#include "main/buffers.h"
+#include "main/depth.h"
+#include "main/shaders.h"
+#include "main/texstate.h"
+#include "main/varray.h"
+#include "glapi/dispatch.h"
+#include "swrast/swrast.h"
+#include "main/stencil.h"
+#include "main/matrix.h"
 
 #include "main/glheader.h"
 #include "main/imports.h"
@@ -43,6 +57,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "swrast/swrast.h"
 
 #include "radeon_context.h"
+#include "radeon_common.h"
 #include "radeon_state.h"
 #include "radeon_ioctl.h"
 #include "radeon_tcl.h"
@@ -58,75 +73,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define RADEON_IDLE_RETRY           16
 
 
-static void radeonWaitForIdle( radeonContextPtr rmesa );
-static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, 
-				    const char * caller );
-
-static void print_state_atom( struct radeon_state_atom *state )
-{
-   int i;
-
-   fprintf(stderr, "emit %s/%d\n", state->name, state->cmd_size);
-
-   if (RADEON_DEBUG & DEBUG_VERBOSE) 
-      for (i = 0 ; i < state->cmd_size ; i++) 
-	 fprintf(stderr, "\t%s[%d]: %x\n", state->name, i, state->cmd[i]);
-
-}
-
-static void radeonSaveHwState( radeonContextPtr rmesa )
-{
-   struct radeon_state_atom *atom;
-   char * dest = rmesa->backup_store.cmd_buf;
-
-   if (RADEON_DEBUG & DEBUG_STATE)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-   
-   rmesa->backup_store.cmd_used = 0;
-
-   foreach( atom, &rmesa->hw.atomlist ) {
-      if ( atom->check( rmesa->glCtx ) ) {
-	 int size = atom->cmd_size * 4;
-	 memcpy( dest, atom->cmd, size);
-	 dest += size;
-	 rmesa->backup_store.cmd_used += size;
-	 if (RADEON_DEBUG & DEBUG_STATE)
-	    print_state_atom( atom );
-      }
-   }
-
-   assert( rmesa->backup_store.cmd_used <= RADEON_CMD_BUF_SZ );
-   if (RADEON_DEBUG & DEBUG_STATE)
-      fprintf(stderr, "Returning to radeonEmitState\n");
-}
-
-/* At this point we were in FlushCmdBufLocked but we had lost our context, so
- * we need to unwire our current cmdbuf, hook the one with the saved state in
- * it, flush it, and then put the current one back.  This is so commands at the
- * start of a cmdbuf can rely on the state being kept from the previous one.
- */
-static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa )
-{
-   GLuint nr_released_bufs;
-   struct radeon_store saved_store;
-
-   if (rmesa->backup_store.cmd_used == 0)
-      return;
-
-   if (RADEON_DEBUG & DEBUG_STATE)
-      fprintf(stderr, "Emitting backup state on lost context\n");
-
-   rmesa->lost_context = GL_FALSE;
-
-   nr_released_bufs = rmesa->dma.nr_released_bufs;
-   saved_store = rmesa->store;
-   rmesa->dma.nr_released_bufs = 0;
-   rmesa->store = rmesa->backup_store;
-   radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
-   rmesa->dma.nr_released_bufs = nr_released_bufs;
-   rmesa->store = saved_store;
-}
-
 /* =============================================================
  * Kernel command buffer handling
  */
@@ -134,965 +80,386 @@ static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa )
 /* The state atoms will be emitted in the order they appear in the atom list,
  * so this step is important.
  */
-void radeonSetUpAtomList( radeonContextPtr rmesa )
+void radeonSetUpAtomList( r100ContextPtr rmesa )
 {
-   int i, mtu = rmesa->glCtx->Const.MaxTextureUnits;
-
-   make_empty_list(&rmesa->hw.atomlist);
-   rmesa->hw.atomlist.name = "atom-list";
-
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.ctx);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.set);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.lin);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.msk);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.vpt);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.tcl);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.msc);
+   int i, mtu = rmesa->radeon.glCtx->Const.MaxTextureUnits;
+
+   make_empty_list(&rmesa->radeon.hw.atomlist);
+   rmesa->radeon.hw.atomlist.name = "atom-list";
+
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.ctx);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.set);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.lin);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.msk);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.vpt);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.tcl);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.msc);
    for (i = 0; i < mtu; ++i) {
-       insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.tex[i]);
-       insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.txr[i]);
-       insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.cube[i]);
+       insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.tex[i]);
+       insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.txr[i]);
+       insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.cube[i]);
    }
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.zbs);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.mtl);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.zbs);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.mtl);
    for (i = 0; i < 3 + mtu; ++i)
-      insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.mat[i]);
+      insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.mat[i]);
    for (i = 0; i < 8; ++i)
-      insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.lit[i]);
+      insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.lit[i]);
    for (i = 0; i < 6; ++i)
-      insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.ucp[i]);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.eye);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.grd);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.fog);
-   insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.glt);
+      insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.ucp[i]);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.eye);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.grd);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.fog);
+   insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.glt);
 }
 
-void radeonEmitState( radeonContextPtr rmesa )
+void radeonEmitScissor(r100ContextPtr rmesa)
 {
-   struct radeon_state_atom *atom;
-   char *dest;
-
-   if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (rmesa->save_on_next_emit) {
-      radeonSaveHwState(rmesa);
-      rmesa->save_on_next_emit = GL_FALSE;
-   }
-
-   /* this code used to return here but now it emits zbs */
-
-   /* To avoid going across the entire set of states multiple times, just check
-    * for enough space for the case of emitting all state, and inline the
-    * radeonAllocCmdBuf code here without all the checks.
-    */
-   radeonEnsureCmdBufSpace(rmesa, rmesa->hw.max_state_size);
-   dest = rmesa->store.cmd_buf + rmesa->store.cmd_used;
-
-   /* We always always emit zbs, this is due to a bug found by keithw in
-      the hardware and rediscovered after Erics changes by me.
-      if you ever touch this code make sure you emit zbs otherwise
-      you get tcl lockups on at least M7/7500 class of chips - airlied */
-   rmesa->hw.zbs.dirty=1;
-
-   if (RADEON_DEBUG & DEBUG_STATE) {
-      foreach(atom, &rmesa->hw.atomlist) {
-	 if (atom->dirty || rmesa->hw.all_dirty) {
-	    if (atom->check(rmesa->glCtx))
-	       print_state_atom(atom);
-	    else
-	       fprintf(stderr, "skip state %s\n", atom->name);
-	 }
-      }
-   }
-
-   foreach(atom, &rmesa->hw.atomlist) {
-      if (rmesa->hw.all_dirty)
-	 atom->dirty = GL_TRUE;
-      if (!(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) &&
-	   atom->is_tcl)
-	 atom->dirty = GL_FALSE;
-      if (atom->dirty) {
-	 if (atom->check(rmesa->glCtx)) {
-	    int size = atom->cmd_size * 4;
-	    memcpy(dest, atom->cmd, size);
-	    dest += size;
-	    rmesa->store.cmd_used += size;
-	    atom->dirty = GL_FALSE;
-	 }
-      }
-   }
-
-   assert(rmesa->store.cmd_used <= RADEON_CMD_BUF_SZ);
- 
-   rmesa->hw.is_dirty = GL_FALSE;
-   rmesa->hw.all_dirty = GL_FALSE;
+    BATCH_LOCALS(&rmesa->radeon);
+    if (!rmesa->radeon.radeonScreen->kernel_mm) {
+       return;
+    }
+    if (rmesa->radeon.state.scissor.enabled) {
+        BEGIN_BATCH(6);
+        OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 0));
+        OUT_BATCH(rmesa->hw.ctx.cmd[CTX_PP_CNTL] | RADEON_SCISSOR_ENABLE);
+        OUT_BATCH(CP_PACKET0(RADEON_RE_TOP_LEFT, 0));
+        OUT_BATCH((rmesa->radeon.state.scissor.rect.y1 << 16) |
+                  rmesa->radeon.state.scissor.rect.x1);
+        OUT_BATCH(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0));
+        OUT_BATCH(((rmesa->radeon.state.scissor.rect.y2 - 1) << 16) |
+                  (rmesa->radeon.state.scissor.rect.x2 - 1));
+        END_BATCH();
+    } else {
+        BEGIN_BATCH(2);
+        OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 0));
+        OUT_BATCH(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & ~RADEON_SCISSOR_ENABLE);
+        END_BATCH();
+    }
 }
 
 /* Fire a section of the retained (indexed_verts) buffer as a regular
- * primtive.  
+ * primtive.
  */
-extern void radeonEmitVbufPrim( radeonContextPtr rmesa,
+extern void radeonEmitVbufPrim( r100ContextPtr rmesa,
 				GLuint vertex_format,
 				GLuint primitive,
 				GLuint vertex_nr )
 {
-   drm_radeon_cmd_header_t *cmd;
-
+   BATCH_LOCALS(&rmesa->radeon);
 
    assert(!(primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
-   
-   radeonEmitState( rmesa );
 
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s cmd_used/4: %d\n", __FUNCTION__,
-	      rmesa->store.cmd_used/4);
-   
-   cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, VBUF_BUFSZ,
-						       __FUNCTION__ );
+   radeonEmitState(&rmesa->radeon);
+   radeonEmitScissor(rmesa);
+
 #if RADEON_OLD_PACKETS
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
-   cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM | (3 << 16);
-   cmd[2].i = rmesa->ioctl.vertex_offset;
-   cmd[3].i = vertex_nr;
-   cmd[4].i = vertex_format;
-   cmd[5].i = (primitive | 
-	       RADEON_CP_VC_CNTL_PRIM_WALK_LIST |
-	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
-	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
-	       (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT));
+   BEGIN_BATCH(8);
+   OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 3);
+   if (!rmesa->radeon.radeonScreen->kernel_mm) {
+     OUT_BATCH_RELOC(rmesa->ioctl.vertex_offset, rmesa->ioctl.bo, rmesa->ioctl.vertex_offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+   } else {
+     OUT_BATCH(rmesa->ioctl.vertex_offset);
+   }
 
-   if (RADEON_DEBUG & DEBUG_PRIMS)
-      fprintf(stderr, "%s: header 0x%x offt 0x%x vfmt 0x%x vfcntl %x \n",
-	      __FUNCTION__,
-	      cmd[1].i, cmd[2].i, cmd[4].i, cmd[5].i);
-#else
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
-   cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_VBUF | (1 << 16);
-   cmd[2].i = vertex_format;
-   cmd[3].i = (primitive | 
-	       RADEON_CP_VC_CNTL_PRIM_WALK_LIST |
-	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
-	       RADEON_CP_VC_CNTL_MAOS_ENABLE |
-	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
-	       (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT));
+   OUT_BATCH(vertex_nr);
+   OUT_BATCH(vertex_format);
+   OUT_BATCH(primitive |  RADEON_CP_VC_CNTL_PRIM_WALK_LIST |
+	     RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	     RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
+	     (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT));
+
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+     radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			   rmesa->ioctl.bo,
+			   RADEON_GEM_DOMAIN_GTT,
+			   0, 0);
+   }
 
+   END_BATCH();
 
-   if (RADEON_DEBUG & DEBUG_PRIMS)
-      fprintf(stderr, "%s: header 0x%x vfmt 0x%x vfcntl %x \n",
-	      __FUNCTION__,
-	      cmd[1].i, cmd[2].i, cmd[3].i);
+#else
+   BEGIN_BATCH(4);
+   OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_DRAW_VBUF, 1);
+   OUT_BATCH(vertex_format);
+   OUT_BATCH(primitive |
+	     RADEON_CP_VC_CNTL_PRIM_WALK_LIST |
+	     RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	     RADEON_CP_VC_CNTL_MAOS_ENABLE |
+	     RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
+	     (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT));
+   END_BATCH();
 #endif
 }
 
-
-void radeonFlushElts( radeonContextPtr rmesa )
+void radeonFlushElts( GLcontext *ctx )
 {
-   int *cmd = (int *)(rmesa->store.cmd_buf + rmesa->store.elts_start);
-   int dwords;
-#if RADEON_OLD_PACKETS
-   int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 24)) / 2;
-#else
-   int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 16)) / 2;
-#endif
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&rmesa->radeon);
+   int nr;
+   uint32_t *cmd = (uint32_t *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_start);
+   int dwords = (rmesa->radeon.cmdbuf.cs->section_ndw - rmesa->radeon.cmdbuf.cs->section_cdw);
 
    if (RADEON_DEBUG & DEBUG_IOCTL)
       fprintf(stderr, "%s\n", __FUNCTION__);
 
-   assert( rmesa->dma.flush == radeonFlushElts );
-   rmesa->dma.flush = NULL;
+   assert( rmesa->radeon.dma.flush == radeonFlushElts );
+   rmesa->radeon.dma.flush = NULL;
 
-   /* Cope with odd number of elts:
-    */
-   rmesa->store.cmd_used = (rmesa->store.cmd_used + 2) & ~2;
-   dwords = (rmesa->store.cmd_used - rmesa->store.elts_start) / 4;
+   nr = rmesa->tcl.elt_used;
+
+#if RADEON_OLD_PACKETS
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+     dwords -= 2;
+   }
+#endif
 
 #if RADEON_OLD_PACKETS
-   cmd[1] |= (dwords - 3) << 16;
+   cmd[1] |= (dwords + 3) << 16;
    cmd[5] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT;
 #else
-   cmd[1] |= (dwords - 3) << 16;
+   cmd[1] |= (dwords + 2) << 16;
    cmd[3] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT;
 #endif
 
+   rmesa->radeon.cmdbuf.cs->cdw += dwords;
+   rmesa->radeon.cmdbuf.cs->section_cdw += dwords;
+
+#if RADEON_OLD_PACKETS
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+      radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			    rmesa->ioctl.bo,
+			    RADEON_GEM_DOMAIN_GTT,
+			    0, 0);
+   }
+#endif
+
+   END_BATCH();
+
    if (RADEON_DEBUG & DEBUG_SYNC) {
       fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
-      radeonFinish( rmesa->glCtx );
+      radeonFinish( rmesa->radeon.glCtx );
    }
-}
 
+}
 
-GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa,
+GLushort *radeonAllocEltsOpenEnded( r100ContextPtr rmesa,
 				    GLuint vertex_format,
 				    GLuint primitive,
 				    GLuint min_nr )
 {
-   drm_radeon_cmd_header_t *cmd;
    GLushort *retval;
+   int align_min_nr;
+   BATCH_LOCALS(&rmesa->radeon);
 
    if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s %d\n", __FUNCTION__, min_nr);
+      fprintf(stderr, "%s %d prim %x\n", __FUNCTION__, min_nr, primitive);
 
    assert((primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
-   
-   radeonEmitState( rmesa );
-   
-   cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa,
-						       ELTS_BUFSZ(min_nr),
-						       __FUNCTION__ );
+
+   radeonEmitState(&rmesa->radeon);
+   radeonEmitScissor(rmesa);
+
+   rmesa->tcl.elt_cmd_start = rmesa->radeon.cmdbuf.cs->cdw;
+
+   /* round up min_nr to align the state */
+   align_min_nr = (min_nr + 1) & ~1;
+
 #if RADEON_OLD_PACKETS
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
-   cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM;
-   cmd[2].i = rmesa->ioctl.vertex_offset;
-   cmd[3].i = 0xffff;
-   cmd[4].i = vertex_format;
-   cmd[5].i = (primitive | 
-	       RADEON_CP_VC_CNTL_PRIM_WALK_IND |
-	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
-	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
-
-   retval = (GLushort *)(cmd+6);
-#else   
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
-   cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_INDX;
-   cmd[2].i = vertex_format;
-   cmd[3].i = (primitive | 
-	       RADEON_CP_VC_CNTL_PRIM_WALK_IND |
-	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
-	       RADEON_CP_VC_CNTL_MAOS_ENABLE |
-	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
-
-   retval = (GLushort *)(cmd+4);
+   BEGIN_BATCH_NO_AUTOSTATE(2+ELTS_BUFSZ(align_min_nr)/4);
+   OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 0);
+   if (!rmesa->radeon.radeonScreen->kernel_mm) {
+     OUT_BATCH_RELOC(rmesa->ioctl.vertex_offset, rmesa->ioctl.bo, rmesa->ioctl.vertex_offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+   } else {
+     OUT_BATCH(rmesa->ioctl.vertex_offset);
+   }
+   OUT_BATCH(0xffff);
+   OUT_BATCH(vertex_format);
+   OUT_BATCH(primitive |
+	     RADEON_CP_VC_CNTL_PRIM_WALK_IND |
+	     RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	     RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
+#else
+   BEGIN_BATCH_NO_AUTOSTATE(ELTS_BUFSZ(align_min_nr)/4);
+   OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_DRAW_INDX, 0);
+   OUT_BATCH(vertex_format);
+   OUT_BATCH(primitive |
+	     RADEON_CP_VC_CNTL_PRIM_WALK_IND |
+	     RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	     RADEON_CP_VC_CNTL_MAOS_ENABLE |
+	     RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
 #endif
 
-   if (RADEON_DEBUG & DEBUG_PRIMS)
-      fprintf(stderr, "%s: header 0x%x vfmt 0x%x prim %x \n",
-	      __FUNCTION__,
-	      cmd[1].i, vertex_format, primitive);
 
-   assert(!rmesa->dma.flush);
-   rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
-   rmesa->dma.flush = radeonFlushElts;
+   rmesa->tcl.elt_cmd_offset = rmesa->radeon.cmdbuf.cs->cdw;
+   rmesa->tcl.elt_used = min_nr;
 
-   rmesa->store.elts_start = ((char *)cmd) - rmesa->store.cmd_buf;
+   retval = (GLushort *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_offset);
 
-   return retval;
-}
+   if (RADEON_DEBUG & DEBUG_PRIMS)
+      fprintf(stderr, "%s: header prim %x \n",
+	      __FUNCTION__, primitive);
 
+   assert(!rmesa->radeon.dma.flush);
+   rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+   rmesa->radeon.dma.flush = radeonFlushElts;
 
+   return retval;
+}
 
-void radeonEmitVertexAOS( radeonContextPtr rmesa,
+void radeonEmitVertexAOS( r100ContextPtr rmesa,
 			  GLuint vertex_size,
+			  struct radeon_bo *bo,
 			  GLuint offset )
 {
 #if RADEON_OLD_PACKETS
-   rmesa->ioctl.vertex_size = vertex_size;
    rmesa->ioctl.vertex_offset = offset;
+   rmesa->ioctl.bo = bo;
 #else
-   drm_radeon_cmd_header_t *cmd;
+   BATCH_LOCALS(&rmesa->radeon);
 
    if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_IOCTL))
       fprintf(stderr, "%s:  vertex_size 0x%x offset 0x%x \n",
 	      __FUNCTION__, vertex_size, offset);
 
-   cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, VERT_AOS_BUFSZ,
-						  __FUNCTION__ );
+   BEGIN_BATCH(7);
+   OUT_BATCH_PACKET3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, 2);
+   OUT_BATCH(1);
+   OUT_BATCH(vertex_size | (vertex_size << 8));
+   OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+   END_BATCH();
 
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
-   cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (2 << 16);
-   cmd[2].i = 1;
-   cmd[3].i = vertex_size | (vertex_size << 8);
-   cmd[4].i = offset;
 #endif
 }
-		       
 
-void radeonEmitAOS( radeonContextPtr rmesa,
-		    struct radeon_dma_region **component,
+
+void radeonEmitAOS( r100ContextPtr rmesa,
 		    GLuint nr,
 		    GLuint offset )
 {
 #if RADEON_OLD_PACKETS
    assert( nr == 1 );
-   assert( component[0]->aos_size == component[0]->aos_stride );
-   rmesa->ioctl.vertex_size = component[0]->aos_size;
-   rmesa->ioctl.vertex_offset = 
-      (component[0]->aos_start + offset * component[0]->aos_stride * 4);
+   rmesa->ioctl.bo = rmesa->radeon.tcl.aos[0].bo;
+   rmesa->ioctl.vertex_offset =
+     (rmesa->radeon.tcl.aos[0].offset + offset * rmesa->radeon.tcl.aos[0].stride * 4);
 #else
-   drm_radeon_cmd_header_t *cmd;
-   int sz = AOS_BUFSZ(nr);
+   BATCH_LOCALS(&rmesa->radeon);
+   uint32_t voffset;
+   //   int sz = AOS_BUFSZ(nr);
+   int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2;
    int i;
-   int *tmp;
 
    if (RADEON_DEBUG & DEBUG_IOCTL)
       fprintf(stderr, "%s\n", __FUNCTION__);
 
-
-   cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, sz,
-						  __FUNCTION__ );
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
-   cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (((sz / sizeof(int))-3) << 16);
-   cmd[2].i = nr;
-   tmp = &cmd[0].i;
-   cmd += 3;
-
-   for (i = 0 ; i < nr ; i++) {
-      if (i & 1) {
-	 cmd[0].i |= ((component[i]->aos_stride << 24) | 
-		      (component[i]->aos_size << 16));
-	 cmd[2].i = (component[i]->aos_start + 
-		     offset * component[i]->aos_stride * 4);
-	 cmd += 3;
+   BEGIN_BATCH(sz+2+(nr * 2));
+   OUT_BATCH_PACKET3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, sz - 1);
+   OUT_BATCH(nr);
+
+   if (!rmesa->radeon.radeonScreen->kernel_mm) {
+      for (i = 0; i + 1 < nr; i += 2) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) |
+		   (rmesa->radeon.tcl.aos[i].stride << 8) |
+		   (rmesa->radeon.tcl.aos[i + 1].components << 16) |
+		   (rmesa->radeon.tcl.aos[i + 1].stride << 24));
+
+	 voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+	 OUT_BATCH_RELOC(voffset,
+			 rmesa->radeon.tcl.aos[i].bo,
+			 voffset,
+			 RADEON_GEM_DOMAIN_GTT,
+			 0, 0);
+	 voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+	 OUT_BATCH_RELOC(voffset,
+			 rmesa->radeon.tcl.aos[i+1].bo,
+			 voffset,
+			 RADEON_GEM_DOMAIN_GTT,
+			 0, 0);
       }
-      else {
-	 cmd[0].i = ((component[i]->aos_stride << 8) | 
-		     (component[i]->aos_size << 0));
-	 cmd[1].i = (component[i]->aos_start + 
-		     offset * component[i]->aos_stride * 4);
-      }
-   }
-
-   if (RADEON_DEBUG & DEBUG_VERTS) {
-      fprintf(stderr, "%s:\n", __FUNCTION__);
-      for (i = 0 ; i < sz ; i++)
-	 fprintf(stderr, "   %d: %x\n", i, tmp[i]);
-   }
-#endif
-}
-
-/* using already shifted color_fmt! */
-void radeonEmitBlit( radeonContextPtr rmesa, /* FIXME: which drmMinor is required? */
-		   GLuint color_fmt,
-		   GLuint src_pitch,
-		   GLuint src_offset,
-		   GLuint dst_pitch,
-		   GLuint dst_offset,
-		   GLint srcx, GLint srcy,
-		   GLint dstx, GLint dsty,
-		   GLuint w, GLuint h )
-{
-   drm_radeon_cmd_header_t *cmd;
-
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s src %x/%x %d,%d dst: %x/%x %d,%d sz: %dx%d\n",
-	      __FUNCTION__, 
-	      src_pitch, src_offset, srcx, srcy,
-	      dst_pitch, dst_offset, dstx, dsty,
-	      w, h);
-
-   assert( (src_pitch & 63) == 0 );
-   assert( (dst_pitch & 63) == 0 );
-   assert( (src_offset & 1023) == 0 ); 
-   assert( (dst_offset & 1023) == 0 ); 
-   assert( w < (1<<16) );
-   assert( h < (1<<16) );
-
-   cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, 8 * sizeof(int),
-						  __FUNCTION__ );
-
-
-   cmd[0].i = 0;
-   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
-   cmd[1].i = RADEON_CP_PACKET3_CNTL_BITBLT_MULTI | (5 << 16);
-   cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
-	       RADEON_GMC_DST_PITCH_OFFSET_CNTL |
-	       RADEON_GMC_BRUSH_NONE |
-	       color_fmt |
-	       RADEON_GMC_SRC_DATATYPE_COLOR |
-	       RADEON_ROP3_S |
-	       RADEON_DP_SRC_SOURCE_MEMORY |
-	       RADEON_GMC_CLR_CMP_CNTL_DIS |
-	       RADEON_GMC_WR_MSK_DIS );
-
-   cmd[3].i = ((src_pitch/64)<<22) | (src_offset >> 10);
-   cmd[4].i = ((dst_pitch/64)<<22) | (dst_offset >> 10);
-   cmd[5].i = (srcx << 16) | srcy;
-   cmd[6].i = (dstx << 16) | dsty; /* dst */
-   cmd[7].i = (w << 16) | h;
-}
-
-
-void radeonEmitWait( radeonContextPtr rmesa, GLuint flags )
-{
-   drm_radeon_cmd_header_t *cmd;
-
-   assert( !(flags & ~(RADEON_WAIT_2D|RADEON_WAIT_3D)) );
-
-   cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, 1 * sizeof(int),
-					   __FUNCTION__ );
-   cmd[0].i = 0;
-   cmd[0].wait.cmd_type = RADEON_CMD_WAIT;
-   cmd[0].wait.flags = flags;
-}
 
-
-static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, 
-				    const char * caller )
-{
-   int ret, i;
-   drm_radeon_cmd_buffer_t cmd;
-
-   if (rmesa->lost_context)
-      radeonBackUpAndEmitLostStateLocked(rmesa);
-
-   if (RADEON_DEBUG & DEBUG_IOCTL) {
-      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); 
-
-      if (RADEON_DEBUG & DEBUG_VERBOSE) 
-	 for (i = 0 ; i < rmesa->store.cmd_used ; i += 4 )
-	    fprintf(stderr, "%d: %x\n", i/4, 
-		    *(int *)(&rmesa->store.cmd_buf[i]));
-   }
-
-   if (RADEON_DEBUG & DEBUG_DMA)
-      fprintf(stderr, "%s: Releasing %d buffers\n", __FUNCTION__,
-	      rmesa->dma.nr_released_bufs);
-
-
-   if (RADEON_DEBUG & DEBUG_SANITY) {
-      if (rmesa->state.scissor.enabled) 
-	 ret = radeonSanityCmdBuffer( rmesa, 
-				      rmesa->state.scissor.numClipRects,
-				      rmesa->state.scissor.pClipRects);
-      else
-	 ret = radeonSanityCmdBuffer( rmesa, 
-				      rmesa->numClipRects,
-				      rmesa->pClipRects);
-      if (ret) {
-	 fprintf(stderr, "drmSanityCommandWrite: %d\n", ret);	 
-	 goto out;
+      if (nr & 1) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) |
+		   (rmesa->radeon.tcl.aos[nr - 1].stride << 8));
+	 voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+	 OUT_BATCH_RELOC(voffset,
+			 rmesa->radeon.tcl.aos[nr - 1].bo,
+			 voffset,
+			 RADEON_GEM_DOMAIN_GTT,
+			 0, 0);
       }
-   }
-
-
-   cmd.bufsz = rmesa->store.cmd_used;
-   cmd.buf = rmesa->store.cmd_buf;
-
-   if (rmesa->state.scissor.enabled) {
-      cmd.nbox = rmesa->state.scissor.numClipRects;
-      cmd.boxes = rmesa->state.scissor.pClipRects;
    } else {
-      cmd.nbox = rmesa->numClipRects;
-      cmd.boxes = rmesa->pClipRects;
-   }
-
-   ret = drmCommandWrite( rmesa->dri.fd,
-			  DRM_RADEON_CMDBUF,
-			  &cmd, sizeof(cmd) );
-
-   if (ret)
-      fprintf(stderr, "drmCommandWrite: %d\n", ret);
-
-   if (RADEON_DEBUG & DEBUG_SYNC) {
-      fprintf(stderr, "\nSyncing in %s\n\n", __FUNCTION__);
-      radeonWaitForIdleLocked( rmesa );
-   }
-
- out:
-   rmesa->store.primnr = 0;
-   rmesa->store.statenr = 0;
-   rmesa->store.cmd_used = 0;
-   rmesa->dma.nr_released_bufs = 0;
-   rmesa->save_on_next_emit = 1;
-
-   return ret;
-}
-
-
-/* Note: does not emit any commands to avoid recursion on
- * radeonAllocCmdBuf.
- */
-void radeonFlushCmdBuf( radeonContextPtr rmesa, const char *caller )
-{
-   int ret;
-
-	      
-   LOCK_HARDWARE( rmesa );
-
-   ret = radeonFlushCmdBufLocked( rmesa, caller );
-
-   UNLOCK_HARDWARE( rmesa );
-
-   if (ret) {
-      fprintf(stderr, "drm_radeon_cmd_buffer_t: %d (exiting)\n", ret);
-      exit(ret);
-   }
-}
-
-/* =============================================================
- * Hardware vertex buffer handling
- */
-
-
-void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa )
-{
-   struct radeon_dma_buffer *dmabuf;
-   int fd = rmesa->dri.fd;
-   int index = 0;
-   int size = 0;
-   drmDMAReq dma;
-   int ret;
-
-   if (RADEON_DEBUG & (DEBUG_IOCTL|DEBUG_DMA))
-      fprintf(stderr, "%s\n", __FUNCTION__);  
-
-   if (rmesa->dma.flush) {
-      rmesa->dma.flush( rmesa );
-   }
-
-   if (rmesa->dma.current.buf)
-      radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ );
-
-   if (rmesa->dma.nr_released_bufs > 4)
-      radeonFlushCmdBuf( rmesa, __FUNCTION__ );
-
-   dma.context = rmesa->dri.hwContext;
-   dma.send_count = 0;
-   dma.send_list = NULL;
-   dma.send_sizes = NULL;
-   dma.flags = 0;
-   dma.request_count = 1;
-   dma.request_size = RADEON_BUFFER_SIZE;
-   dma.request_list = &index;
-   dma.request_sizes = &size;
-   dma.granted_count = 0;
-
-   LOCK_HARDWARE(rmesa);	/* no need to validate */
-
-   ret = drmDMA( fd, &dma );
-      
-   if (ret != 0) {
-      /* Free some up this way?
-       */
-      if (rmesa->dma.nr_released_bufs) {
-	 radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
-      }
-      
-      if (RADEON_DEBUG & DEBUG_DMA)
-	 fprintf(stderr, "Waiting for buffers\n");
-
-      radeonWaitForIdleLocked( rmesa );
-      ret = drmDMA( fd, &dma );
-
-      if ( ret != 0 ) {
-	 UNLOCK_HARDWARE( rmesa );
-	 fprintf( stderr, "Error: Could not get dma buffer... exiting\n" );
-	 exit( -1 );
-      }
-   }
-
-   UNLOCK_HARDWARE(rmesa);
-
-   if (RADEON_DEBUG & DEBUG_DMA)
-      fprintf(stderr, "Allocated buffer %d\n", index);
-
-   dmabuf = CALLOC_STRUCT( radeon_dma_buffer );
-   dmabuf->buf = &rmesa->radeonScreen->buffers->list[index];
-   dmabuf->refcount = 1;
-
-   rmesa->dma.current.buf = dmabuf;
-   rmesa->dma.current.address = dmabuf->buf->address;
-   rmesa->dma.current.end = dmabuf->buf->total;
-   rmesa->dma.current.start = 0;
-   rmesa->dma.current.ptr = 0;
-
-   rmesa->c_vertexBuffers++;
-}
-
-void radeonReleaseDmaRegion( radeonContextPtr rmesa,
-			     struct radeon_dma_region *region,
-			     const char *caller )
-{
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); 
-   
-   if (!region->buf)
-      return;
-
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
-
-   if (--region->buf->refcount == 0) {
-      drm_radeon_cmd_header_t *cmd;
-
-      if (RADEON_DEBUG & (DEBUG_IOCTL|DEBUG_DMA))
-	 fprintf(stderr, "%s -- DISCARD BUF %d\n", __FUNCTION__,
-		 region->buf->buf->idx);  
-      
-      cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, sizeof(*cmd), 
-						     __FUNCTION__ );
-      cmd->dma.cmd_type = RADEON_CMD_DMA_DISCARD;
-      cmd->dma.buf_idx = region->buf->buf->idx;
-      FREE(region->buf);
-      rmesa->dma.nr_released_bufs++;
-   }
-
-   region->buf = NULL;
-   region->start = 0;
-}
-
-/* Allocates a region from rmesa->dma.current.  If there isn't enough
- * space in current, grab a new buffer (and discard what was left of current)
- */
-void radeonAllocDmaRegion( radeonContextPtr rmesa, 
-			   struct radeon_dma_region *region,
-			   int bytes,
-			   int alignment )
-{
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
-
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
-
-   if (region->buf)
-      radeonReleaseDmaRegion( rmesa, region, __FUNCTION__ );
-
-   alignment--;
-   rmesa->dma.current.start = rmesa->dma.current.ptr = 
-      (rmesa->dma.current.ptr + alignment) & ~alignment;
-
-   if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) 
-      radeonRefillCurrentDmaRegion( rmesa );
-
-   region->start = rmesa->dma.current.start;
-   region->ptr = rmesa->dma.current.start;
-   region->end = rmesa->dma.current.start + bytes;
-   region->address = rmesa->dma.current.address;
-   region->buf = rmesa->dma.current.buf;
-   region->buf->refcount++;
-
-   rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */
-   rmesa->dma.current.start = 
-      rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;  
-}
-
-/* ================================================================
- * SwapBuffers with client-side throttling
- */
-
-static uint32_t radeonGetLastFrame (radeonContextPtr rmesa) 
-{
-   drm_radeon_getparam_t gp;
-   int ret;
-   uint32_t frame;
-
-   gp.param = RADEON_PARAM_LAST_FRAME;
-   gp.value = (int *)&frame;
-   ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_GETPARAM,
-			      &gp, sizeof(gp) );
-
-   if ( ret ) {
-      fprintf( stderr, "%s: drm_radeon_getparam_t: %d\n", __FUNCTION__, ret );
-      exit(1);
-   }
-
-   return frame;
-}
-
-static void radeonEmitIrqLocked( radeonContextPtr rmesa )
-{
-   drm_radeon_irq_emit_t ie;
-   int ret;
-
-   ie.irq_seq = &rmesa->iw.irq_seq;
-   ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_IRQ_EMIT, 
-			      &ie, sizeof(ie) );
-   if ( ret ) {
-      fprintf( stderr, "%s: drm_radeon_irq_emit_t: %d\n", __FUNCTION__, ret );
-      exit(1);
-   }
-}
-
-
-static void radeonWaitIrq( radeonContextPtr rmesa )
-{
-   int ret;
-
-   do {
-      ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_IRQ_WAIT,
-			     &rmesa->iw, sizeof(rmesa->iw) );
-   } while (ret && (errno == EINTR || errno == EBUSY));
-
-   if ( ret ) {
-      fprintf( stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, ret );
-      exit(1);
-   }
-}
-
-
-static void radeonWaitForFrameCompletion( radeonContextPtr rmesa )
-{
-   drm_radeon_sarea_t *sarea = rmesa->sarea;
-
-   if (rmesa->do_irqs) {
-      if (radeonGetLastFrame(rmesa) < sarea->last_frame) {
-	 if (!rmesa->irqsEmitted) {
-	    while (radeonGetLastFrame (rmesa) < sarea->last_frame)
-	       ;
-	 }
-	 else {
-	    UNLOCK_HARDWARE( rmesa ); 
-	    radeonWaitIrq( rmesa );	
-	    LOCK_HARDWARE( rmesa ); 
-	 }
-	 rmesa->irqsEmitted = 10;
+      for (i = 0; i + 1 < nr; i += 2) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) |
+		   (rmesa->radeon.tcl.aos[i].stride << 8) |
+		   (rmesa->radeon.tcl.aos[i + 1].components << 16) |
+		   (rmesa->radeon.tcl.aos[i + 1].stride << 24));
+
+	 voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+	 OUT_BATCH(voffset);
+	 voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+	 OUT_BATCH(voffset);
       }
 
-      if (rmesa->irqsEmitted) {
-	 radeonEmitIrqLocked( rmesa );
-	 rmesa->irqsEmitted--;
+      if (nr & 1) {
+	 OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) |
+		   (rmesa->radeon.tcl.aos[nr - 1].stride << 8));
+	 voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+	 OUT_BATCH(voffset);
       }
-   } 
-   else {
-      while (radeonGetLastFrame (rmesa) < sarea->last_frame) {
-	 UNLOCK_HARDWARE( rmesa ); 
-	 if (rmesa->do_usleeps) 
-	    DO_USLEEP( 1 );
-	 LOCK_HARDWARE( rmesa ); 
+      for (i = 0; i + 1 < nr; i += 2) {
+	 voffset =  rmesa->radeon.tcl.aos[i + 0].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride;
+	 radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			       rmesa->radeon.tcl.aos[i+0].bo,
+			       RADEON_GEM_DOMAIN_GTT,
+			       0, 0);
+	 voffset =  rmesa->radeon.tcl.aos[i + 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride;
+	 radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			       rmesa->radeon.tcl.aos[i+1].bo,
+			       RADEON_GEM_DOMAIN_GTT,
+			       0, 0);
       }
-   }
-}
-
-/* Copy the back color buffer to the front color buffer.
- */
-void radeonCopyBuffer( __DRIdrawablePrivate *dPriv,
-		       const drm_clip_rect_t	  *rect)
-{
-   radeonContextPtr rmesa;
-   GLint nbox, i, ret;
-   GLboolean   missed_target;
-   int64_t ust;
-   __DRIscreenPrivate *psp;
-
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
-      fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *) rmesa->glCtx );
-   }
-
-   RADEON_FIREVERTICES( rmesa );
-   LOCK_HARDWARE( rmesa );
-
-   /* Throttle the frame rate -- only allow one pending swap buffers
-    * request at a time.
-    */
-   radeonWaitForFrameCompletion( rmesa );
-   if (!rect)
-   {
-       UNLOCK_HARDWARE( rmesa );
-       driWaitForVBlank( dPriv, & missed_target );
-       LOCK_HARDWARE( rmesa );
-   }
-
-   nbox = dPriv->numClipRects; /* must be in locked region */
-
-   for ( i = 0 ; i < nbox ; ) {
-      GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox );
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
-      GLint n = 0;
-
-      for ( ; i < nr ; i++ ) {
-
-	  *b = box[i];
-
-	  if (rect)
-	  {
-	      if (rect->x1 > b->x1)
-		  b->x1 = rect->x1;
-	      if (rect->y1 > b->y1)
-		  b->y1 = rect->y1;
-	      if (rect->x2 < b->x2)
-		  b->x2 = rect->x2;
-	      if (rect->y2 < b->y2)
-		  b->y2 = rect->y2;
-
-	      if (b->x1 >= b->x2 || b->y1 >= b->y2)
-		  continue;
-	  }
-
-	  b++;
-	  n++;
-      }
-      rmesa->sarea->nbox = n;
-
-      if (!n)
-	 continue;
-
-      ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP );
-
-      if ( ret ) {
-	 fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret );
-	 UNLOCK_HARDWARE( rmesa );
-	 exit( 1 );
+      if (nr & 1) {
+	 voffset =  rmesa->radeon.tcl.aos[nr - 1].offset +
+	    offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride;
+	 radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
+			       rmesa->radeon.tcl.aos[nr-1].bo,
+			       RADEON_GEM_DOMAIN_GTT,
+			       0, 0);
       }
    }
+   END_BATCH();
 
-   UNLOCK_HARDWARE( rmesa );
-   if (!rect)
-   {
-       psp = dPriv->driScreenPriv;
-       rmesa->swap_count++;
-       (*psp->systemTime->getUST)( & ust );
-       if ( missed_target ) {
-	   rmesa->swap_missed_count++;
-	   rmesa->swap_missed_ust = ust - rmesa->swap_ust;
-       }
-
-       rmesa->swap_ust = ust;
-       rmesa->hw.all_dirty = GL_TRUE;
-   }
-}
-
-void radeonPageFlip( __DRIdrawablePrivate *dPriv )
-{
-   radeonContextPtr rmesa;
-   GLint ret;
-   GLboolean   missed_target;
-   __DRIscreenPrivate *psp;
-
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
-   psp = dPriv->driScreenPriv;
-
-   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
-      fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__,
-	      rmesa->sarea->pfCurrentPage);
-   }
-
-   RADEON_FIREVERTICES( rmesa );
-   LOCK_HARDWARE( rmesa );
-
-   /* Need to do this for the perf box placement:
-    */
-   if (dPriv->numClipRects)
-   {
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
-      b[0] = box[0];
-      rmesa->sarea->nbox = 1;
-   }
-
-   /* Throttle the frame rate -- only allow a few pending swap buffers
-    * request at a time.
-    */
-   radeonWaitForFrameCompletion( rmesa );
-   UNLOCK_HARDWARE( rmesa );
-   driWaitForVBlank( dPriv, & missed_target );
-   if ( missed_target ) {
-      rmesa->swap_missed_count++;
-      (void) (*psp->systemTime->getUST)( & rmesa->swap_missed_ust );
-   }
-   LOCK_HARDWARE( rmesa );
-
-   ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_FLIP );
-
-   UNLOCK_HARDWARE( rmesa );
-
-   if ( ret ) {
-      fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret );
-      exit( 1 );
-   }
-
-   rmesa->swap_count++;
-   (void) (*psp->systemTime->getUST)( & rmesa->swap_ust );
-
-   /* Get ready for drawing next frame.  Update the renderbuffers'
-    * flippedOffset/Pitch fields so we draw into the right place.
-    */
-   driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer,
-                        rmesa->sarea->pfCurrentPage);
-
-   radeonUpdateDrawBuffer(rmesa->glCtx);
+#endif
 }
 
-
 /* ================================================================
  * Buffer clear
  */
 #define RADEON_MAX_CLEARS	256
 
-static void radeonClear( GLcontext *ctx, GLbitfield mask )
+static void radeonUserClear(GLcontext *ctx, GLuint mask)
+{
+   radeon_clear_tris(ctx, mask);
+}
+
+static void radeonKernelClear(GLcontext *ctx, GLuint flags)
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
-   drm_radeon_sarea_t *sarea = rmesa->sarea;
+     r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
+   drm_radeon_sarea_t *sarea = rmesa->radeon.sarea;
    uint32_t clear;
-   GLuint flags = 0;
-   GLuint color_mask = 0;
    GLint ret, i;
    GLint cx, cy, cw, ch;
 
-   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
-      fprintf( stderr, "radeonClear\n");
-   }
-
-   {
-      LOCK_HARDWARE( rmesa );
-      UNLOCK_HARDWARE( rmesa );
-      if ( dPriv->numClipRects == 0 ) 
-	 return;
-   }
-   
-   radeonFlush( ctx ); 
-
-   if ( mask & BUFFER_BIT_FRONT_LEFT ) {
-      flags |= RADEON_FRONT;
-      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
-      mask &= ~BUFFER_BIT_FRONT_LEFT;
-   }
-
-   if ( mask & BUFFER_BIT_BACK_LEFT ) {
-      flags |= RADEON_BACK;
-      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
-      mask &= ~BUFFER_BIT_BACK_LEFT;
-   }
-
-   if ( mask & BUFFER_BIT_DEPTH ) {
-      flags |= RADEON_DEPTH;
-      mask &= ~BUFFER_BIT_DEPTH;
-   }
-
-   if ( (mask & BUFFER_BIT_STENCIL) && rmesa->state.stencil.hwBuffer ) {
-      flags |= RADEON_STENCIL;
-      mask &= ~BUFFER_BIT_STENCIL;
-   }
-
-   if ( mask ) {
-      if (RADEON_DEBUG & DEBUG_FALLBACKS)
-	 fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask);
-      _swrast_Clear( ctx, mask );
-   }
-
-   if ( !flags ) 
-      return;
-
-   if (rmesa->using_hyperz) {
-      flags |= RADEON_USE_COMP_ZBUF;
-/*      if (rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL) 
-         flags |= RADEON_USE_HIERZ; */
-      if (!(rmesa->state.stencil.hwBuffer) ||
-	 ((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) &&
-	    ((rmesa->state.stencil.clear & RADEON_STENCIL_WRITE_MASK) == RADEON_STENCIL_WRITE_MASK))) {
-	  flags |= RADEON_CLEAR_FASTZ;
-      }
-   }
-
-   LOCK_HARDWARE( rmesa );
+   LOCK_HARDWARE( &rmesa->radeon );
 
    /* compute region after locking: */
    cx = ctx->DrawBuffer->_Xmin;
@@ -1112,7 +479,7 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask )
 
       gp.param = RADEON_PARAM_LAST_CLEAR;
       gp.value = (int *)&clear;
-      ret = drmCommandWriteRead( rmesa->dri.fd,
+      ret = drmCommandWriteRead( rmesa->radeon.dri.fd,
 				 DRM_RADEON_GETPARAM, &gp, sizeof(gp) );
 
       if ( ret ) {
@@ -1124,20 +491,20 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask )
 	 break;
       }
 
-      if ( rmesa->do_usleeps ) {
-	 UNLOCK_HARDWARE( rmesa );
+      if ( rmesa->radeon.do_usleeps ) {
+	 UNLOCK_HARDWARE( &rmesa->radeon );
 	 DO_USLEEP( 1 );
-	 LOCK_HARDWARE( rmesa );
+	 LOCK_HARDWARE( &rmesa->radeon );
       }
    }
 
    /* Send current state to the hardware */
-   radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
+   rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
 
    for ( i = 0 ; i < dPriv->numClipRects ; ) {
       GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, dPriv->numClipRects );
       drm_clip_rect_t *box = dPriv->pClipRects;
-      drm_clip_rect_t *b = rmesa->sarea->boxes;
+      drm_clip_rect_t *b = rmesa->radeon.sarea->boxes;
       drm_radeon_clear_t clear;
       drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
       GLint n = 0;
@@ -1172,105 +539,107 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask )
 	 }
       }
 
-      rmesa->sarea->nbox = n;
+      rmesa->radeon.sarea->nbox = n;
 
       clear.flags       = flags;
-      clear.clear_color = rmesa->state.color.clear;
-      clear.clear_depth = rmesa->state.depth.clear;
+      clear.clear_color = rmesa->radeon.state.color.clear;
+      clear.clear_depth = rmesa->radeon.state.depth.clear;
       clear.color_mask  = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
-      clear.depth_mask  = rmesa->state.stencil.clear;
+      clear.depth_mask  = rmesa->radeon.state.stencil.clear;
       clear.depth_boxes = depth_boxes;
 
       n--;
-      b = rmesa->sarea->boxes;
+      b = rmesa->radeon.sarea->boxes;
       for ( ; n >= 0 ; n-- ) {
 	 depth_boxes[n].f[CLEAR_X1] = (float)b[n].x1;
 	 depth_boxes[n].f[CLEAR_Y1] = (float)b[n].y1;
 	 depth_boxes[n].f[CLEAR_X2] = (float)b[n].x2;
 	 depth_boxes[n].f[CLEAR_Y2] = (float)b[n].y2;
-	 depth_boxes[n].f[CLEAR_DEPTH] = 
-	    (float)rmesa->state.depth.clear;
+	 depth_boxes[n].f[CLEAR_DEPTH] =
+	    (float)rmesa->radeon.state.depth.clear;
       }
 
-      ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_CLEAR,
+      ret = drmCommandWrite( rmesa->radeon.dri.fd, DRM_RADEON_CLEAR,
 			     &clear, sizeof(drm_radeon_clear_t));
 
       if ( ret ) {
-	 UNLOCK_HARDWARE( rmesa );
+	 UNLOCK_HARDWARE( &rmesa->radeon );
 	 fprintf( stderr, "DRM_RADEON_CLEAR: return = %d\n", ret );
 	 exit( 1 );
       }
    }
-
-   UNLOCK_HARDWARE( rmesa );
-   rmesa->hw.all_dirty = GL_TRUE;
+   UNLOCK_HARDWARE( &rmesa->radeon );
 }
 
-
-void radeonWaitForIdleLocked( radeonContextPtr rmesa )
+static void radeonClear( GLcontext *ctx, GLbitfield mask )
 {
-    int fd = rmesa->dri.fd;
-    int to = 0;
-    int ret, i = 0;
-
-    rmesa->c_drawWaits++;
-
-    do {
-        do {
-            ret = drmCommandNone( fd, DRM_RADEON_CP_IDLE);
-        } while ( ret && errno == EBUSY && i++ < RADEON_IDLE_RETRY );
-    } while ( ( ret == -EBUSY ) && ( to++ < RADEON_TIMEOUT ) );
-
-    if ( ret < 0 ) {
-	UNLOCK_HARDWARE( rmesa );
-	fprintf( stderr, "Error: Radeon timed out... exiting\n" );
-	exit( -1 );
-    }
-}
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
+   GLuint flags = 0;
+   GLuint color_mask = 0;
+   GLuint orig_mask = mask;
+
+   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+      fprintf( stderr, "radeonClear\n");
+   }
 
+   {
+      LOCK_HARDWARE( &rmesa->radeon );
+      UNLOCK_HARDWARE( &rmesa->radeon );
+      if ( dPriv->numClipRects == 0 )
+	 return;
+   }
 
-static void radeonWaitForIdle( radeonContextPtr rmesa )
-{
-   LOCK_HARDWARE(rmesa);
-   radeonWaitForIdleLocked( rmesa );
-   UNLOCK_HARDWARE(rmesa);
-}
+   radeon_firevertices(&rmesa->radeon);
 
+   if ( mask & BUFFER_BIT_FRONT_LEFT ) {
+      flags |= RADEON_FRONT;
+      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      mask &= ~BUFFER_BIT_FRONT_LEFT;
+   }
 
-void radeonFlush( GLcontext *ctx )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   if ( mask & BUFFER_BIT_BACK_LEFT ) {
+      flags |= RADEON_BACK;
+      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      mask &= ~BUFFER_BIT_BACK_LEFT;
+   }
 
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
+   if ( mask & BUFFER_BIT_DEPTH ) {
+      flags |= RADEON_DEPTH;
+      mask &= ~BUFFER_BIT_DEPTH;
+   }
 
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
+   if ( (mask & BUFFER_BIT_STENCIL) ) {
+      flags |= RADEON_STENCIL;
+      mask &= ~BUFFER_BIT_STENCIL;
+   }
 
-   radeonEmitState( rmesa );
-   
-   if (rmesa->store.cmd_used)
-      radeonFlushCmdBuf( rmesa, __FUNCTION__ );
-}
+   if ( mask ) {
+      if (RADEON_DEBUG & DEBUG_FALLBACKS)
+	 fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask);
+      _swrast_Clear( ctx, mask );
+   }
 
-/* Make sure all commands have been sent to the hardware and have
- * completed processing.
- */
-void radeonFinish( GLcontext *ctx )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   radeonFlush( ctx );
-
-   if (rmesa->do_irqs) {
-      LOCK_HARDWARE( rmesa );
-      radeonEmitIrqLocked( rmesa );
-      UNLOCK_HARDWARE( rmesa );
-      radeonWaitIrq( rmesa );
+   if ( !flags )
+      return;
+
+   if (rmesa->using_hyperz) {
+      flags |= RADEON_USE_COMP_ZBUF;
+/*      if (rmesa->radeon.radeonScreen->chipset & RADEON_CHIPSET_TCL)
+         flags |= RADEON_USE_HIERZ; */
+      if (((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) &&
+	    ((rmesa->radeon.state.stencil.clear & RADEON_STENCIL_WRITE_MASK) == RADEON_STENCIL_WRITE_MASK))) {
+	  flags |= RADEON_CLEAR_FASTZ;
+      }
    }
-   else
-      radeonWaitForIdle( rmesa );
-}
 
+   if (rmesa->radeon.radeonScreen->kernel_mm)
+     radeonUserClear(ctx, orig_mask);
+   else {
+      radeonKernelClear(ctx, flags);
+      rmesa->radeon.hw.all_dirty = GL_TRUE;
+   }
+}
 
 void radeonInitIoctlFuncs( GLcontext *ctx )
 {
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.h b/src/mesa/drivers/dri/radeon/radeon_ioctl.h
index 4e3a44df075..18805d4c571 100644
--- a/src/mesa/drivers/dri/radeon/radeon_ioctl.h
+++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.h
@@ -38,31 +38,32 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "main/simple_list.h"
 #include "radeon_lock.h"
+#include "radeon_bocs_wrapper.h"
 
-
-extern void radeonEmitState( radeonContextPtr rmesa );
-extern void radeonEmitVertexAOS( radeonContextPtr rmesa,
+extern void radeonEmitVertexAOS( r100ContextPtr rmesa,
 				 GLuint vertex_size,
+				 struct radeon_bo *bo,
 				 GLuint offset );
 
-extern void radeonEmitVbufPrim( radeonContextPtr rmesa,
+extern void radeonEmitVbufPrim( r100ContextPtr rmesa,
 				GLuint vertex_format,
 				GLuint primitive,
 				GLuint vertex_nr );
 
-extern void radeonFlushElts( radeonContextPtr rmesa );
+extern void radeonFlushElts( GLcontext *ctx );
+			    
 
-extern GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa,
+extern GLushort *radeonAllocEltsOpenEnded( r100ContextPtr rmesa,
 					   GLuint vertex_format,
 					   GLuint primitive,
 					   GLuint min_nr );
 
-extern void radeonEmitAOS( radeonContextPtr rmesa,
-			   struct radeon_dma_region **regions,
+
+extern void radeonEmitAOS( r100ContextPtr rmesa,
 			   GLuint n,
 			   GLuint offset );
 
-extern void radeonEmitBlit( radeonContextPtr rmesa,
+extern void radeonEmitBlit( r100ContextPtr rmesa,
 			    GLuint color_fmt,
 			    GLuint src_pitch,
 			    GLuint src_offset,
@@ -72,30 +73,15 @@ extern void radeonEmitBlit( radeonContextPtr rmesa,
 			    GLint dstx, GLint dsty,
 			    GLuint w, GLuint h );
 
-extern void radeonEmitWait( radeonContextPtr rmesa, GLuint flags );
-
-extern void radeonFlushCmdBuf( radeonContextPtr rmesa, const char * );
-extern void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa );
+extern void radeonEmitWait( r100ContextPtr rmesa, GLuint flags );
 
-extern void radeonAllocDmaRegion( radeonContextPtr rmesa,
-				  struct radeon_dma_region *region,
-				  int bytes, 
-				  int alignment );
+extern void radeonFlushCmdBuf( r100ContextPtr rmesa, const char * );
 
-extern void radeonReleaseDmaRegion( radeonContextPtr rmesa,
-				    struct radeon_dma_region *region,
-				    const char *caller );
-
-extern void radeonCopyBuffer( __DRIdrawablePrivate *drawable,
-			      const drm_clip_rect_t	 *rect);
-extern void radeonPageFlip( __DRIdrawablePrivate *drawable );
 extern void radeonFlush( GLcontext *ctx );
 extern void radeonFinish( GLcontext *ctx );
-extern void radeonWaitForIdleLocked( radeonContextPtr rmesa );
-extern void radeonWaitForVBlank( radeonContextPtr rmesa );
 extern void radeonInitIoctlFuncs( GLcontext *ctx );
-extern void radeonGetAllParams( radeonContextPtr rmesa );
-extern void radeonSetUpAtomList( radeonContextPtr rmesa );
+extern void radeonGetAllParams( r100ContextPtr rmesa );
+extern void radeonSetUpAtomList( r100ContextPtr rmesa );
 
 /* ================================================================
  * Helper macros:
@@ -105,33 +91,33 @@ extern void radeonSetUpAtomList( radeonContextPtr rmesa );
  */
 #define RADEON_NEWPRIM( rmesa )			\
 do {						\
-   if ( rmesa->dma.flush )			\
-      rmesa->dma.flush( rmesa );	\
+   if ( rmesa->radeon.dma.flush )			\
+      rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	\
 } while (0)
 
 /* Can accomodate several state changes and primitive changes without
  * actually firing the buffer.
  */
+
 #define RADEON_STATECHANGE( rmesa, ATOM )			\
 do {								\
    RADEON_NEWPRIM( rmesa );					\
    rmesa->hw.ATOM.dirty = GL_TRUE;				\
-   rmesa->hw.is_dirty = GL_TRUE;				\
+   rmesa->radeon.hw.is_dirty = GL_TRUE;				\
 } while (0)
 
-#define RADEON_DB_STATE( ATOM )			        \
+#define RADEON_DB_STATE( ATOM )				\
    memcpy( rmesa->hw.ATOM.lastcmd, rmesa->hw.ATOM.cmd,	\
 	   rmesa->hw.ATOM.cmd_size * 4)
 
-static INLINE int RADEON_DB_STATECHANGE( 
-   radeonContextPtr rmesa,
-   struct radeon_state_atom *atom )
+static INLINE int RADEON_DB_STATECHANGE(r100ContextPtr rmesa,
+					struct radeon_state_atom *atom )
 {
    if (memcmp(atom->cmd, atom->lastcmd, atom->cmd_size*4)) {
-      int *tmp;
+      GLuint *tmp;
       RADEON_NEWPRIM( rmesa );
       atom->dirty = GL_TRUE;
-      rmesa->hw.is_dirty = GL_TRUE;
+      rmesa->radeon.hw.is_dirty = GL_TRUE;
       tmp = atom->cmd; 
       atom->cmd = atom->lastcmd;
       atom->lastcmd = tmp;
@@ -141,16 +127,6 @@ static INLINE int RADEON_DB_STATECHANGE(
       return 0;
 }
 
-
-/* Fire the buffered vertices no matter what.
- */
-#define RADEON_FIREVERTICES( rmesa )			\
-do {							\
-   if ( rmesa->store.cmd_used || rmesa->dma.flush ) {	\
-      radeonFlush( rmesa->glCtx );			\
-   }							\
-} while (0)
-
 /* Command lengths.  Note that any time you ensure ELTS_BUFSZ or VBUF_BUFSZ
  * are available, you will also be adding an rmesa->state.max_state_size because
  * r200EmitState is called from within r200EmitVbufPrim and r200FlushElts.
@@ -167,36 +143,37 @@ do {							\
 #define VBUF_BUFSZ	(4 * sizeof(int))
 #endif
 
-/* Ensure that a minimum amount of space is available in the command buffer.
- * This is used to ensure atomicity of state updates with the rendering requests
- * that rely on them.
- *
- * An alternative would be to implement a "soft lock" such that when the buffer
- * wraps at an inopportune time, we grab the lock, flush the current buffer,
- * and hang on to the lock until the critical section is finished and we flush
- * the buffer again and unlock.
- */
-static INLINE void radeonEnsureCmdBufSpace( radeonContextPtr rmesa,
-					      int bytes )
-{
-   if (rmesa->store.cmd_used + bytes > RADEON_CMD_BUF_SZ)
-      radeonFlushCmdBuf( rmesa, __FUNCTION__ );
-   assert( bytes <= RADEON_CMD_BUF_SZ );
-}
 
-/* Alloc space in the command buffer
- */
-static INLINE char *radeonAllocCmdBuf( radeonContextPtr rmesa,
-					 int bytes, const char *where )
+static inline uint32_t cmdpacket3(int cmd_type)
 {
-   if (rmesa->store.cmd_used + bytes > RADEON_CMD_BUF_SZ)
-      radeonFlushCmdBuf( rmesa, __FUNCTION__ );
+  drm_radeon_cmd_header_t cmd;
+
+  cmd.i = 0;
+  cmd.header.cmd_type = cmd_type;
+
+  return (uint32_t)cmd.i;
 
-   {
-      char *head = rmesa->store.cmd_buf + rmesa->store.cmd_used;
-      rmesa->store.cmd_used += bytes;
-      return head;
-   }
 }
 
+#define OUT_BATCH_PACKET3(packet, num_extra) do {	      \
+    if (!b_l_rmesa->radeonScreen->kernel_mm) {		      \
+      OUT_BATCH(cmdpacket3(RADEON_CMD_PACKET3));				      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    } else {						      \
+      OUT_BATCH(CP_PACKET2);				      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    }							      \
+  } while(0)
+
+#define OUT_BATCH_PACKET3_CLIP(packet, num_extra) do {	      \
+    if (!b_l_rmesa->radeonScreen->kernel_mm) {		      \
+      OUT_BATCH(cmdpacket3(RADEON_CMD_PACKET3_CLIP));	      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    } else {						      \
+      OUT_BATCH(CP_PACKET2);				      \
+      OUT_BATCH(CP_PACKET3((packet), (num_extra)));	      \
+    }							      \
+  } while(0)
+
+
 #endif /* __RADEON_IOCTL_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.c b/src/mesa/drivers/dri/radeon/radeon_lock.c
index 64bb3ca103f..5774f7ebcf7 100644
--- a/src/mesa/drivers/dri/radeon/radeon_lock.c
+++ b/src/mesa/drivers/dri/radeon/radeon_lock.c
@@ -41,30 +41,13 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "main/glheader.h"
 #include "main/mtypes.h"
-#include "radeon_context.h"
+#include "main/colormac.h"
+#include "dri_util.h"
+#include "radeon_screen.h"
+#include "radeon_common.h"
 #include "radeon_lock.h"
-#include "radeon_tex.h"
-#include "radeon_state.h"
-#include "radeon_ioctl.h"
-
 #include "drirenderbuffer.h"
 
-#if DEBUG_LOCKING
-char *prevLockFile = NULL;
-int prevLockLine = 0;
-#endif
-
-/* Turn on/off page flipping according to the flags in the sarea:
- */
-static void radeonUpdatePageFlipping(radeonContextPtr rmesa)
-{
-	rmesa->doPageFlip = rmesa->sarea->pfState;
-	if (rmesa->glCtx->WinSysDrawBuffer) {
-		driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer,
-				     rmesa->sarea->pfCurrentPage);
-	}
-}
-
 /* Update the hardware state.  This is called if another context has
  * grabbed the hardware lock, which includes the X server.  This
  * function also updates the driver's window state after the X server
@@ -75,10 +58,11 @@ static void radeonUpdatePageFlipping(radeonContextPtr rmesa)
  */
 void radeonGetLock(radeonContextPtr rmesa, GLuint flags)
 {
-	__DRIdrawablePrivate *const drawable = rmesa->dri.drawable;
-	__DRIdrawablePrivate *const readable = rmesa->dri.readable;
+	__DRIdrawablePrivate *const drawable = radeon_get_drawable(rmesa);
+	__DRIdrawablePrivate *const readable = radeon_get_readable(rmesa);
 	__DRIscreenPrivate *sPriv = rmesa->dri.screen;
-	drm_radeon_sarea_t *sarea = rmesa->sarea;
+
+	assert(drawable != NULL);
 
 	drmGetLock(rmesa->dri.fd, rmesa->dri.hwContext, flags);
 
@@ -96,29 +80,42 @@ void radeonGetLock(radeonContextPtr rmesa, GLuint flags)
 	}
 
 	if (rmesa->lastStamp != drawable->lastStamp) {
-		radeonUpdatePageFlipping(rmesa);
-		radeonSetCliprects(rmesa);
-		radeonUpdateViewportOffset(rmesa->glCtx);
-		driUpdateFramebufferSize(rmesa->glCtx, drawable);
+		radeon_window_moved(rmesa);
+		rmesa->lastStamp = drawable->lastStamp;
 	}
 
-	RADEON_STATECHANGE(rmesa, ctx);
-	if (rmesa->sarea->tiling_enabled) {
-		rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |=
-		    RADEON_COLOR_TILE_ENABLE;
-	} else {
-		rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] &=
-		    ~RADEON_COLOR_TILE_ENABLE;
-	}
+	rmesa->vtbl.get_lock(rmesa);
+
+	rmesa->lost_context = GL_TRUE;
+}
 
-	if (sarea->ctx_owner != rmesa->dri.hwContext) {
-		int i;
-		sarea->ctx_owner = rmesa->dri.hwContext;
+void radeon_lock_hardware(radeonContextPtr radeon)
+{
+	char ret = 0;
+	struct radeon_framebuffer *rfb = NULL;
+	struct radeon_renderbuffer *rrb = NULL;
 
-		for (i = 0; i < rmesa->nr_heaps; i++) {
-			DRI_AGE_TEXTURES(rmesa->texture_heaps[i]);
-		}
+	if (radeon_get_drawable(radeon)) {
+		rfb = radeon_get_drawable(radeon)->driverPrivate;
+
+		if (rfb)
+			rrb = radeon_get_renderbuffer(&rfb->base,
+						      rfb->base._ColorDrawBufferIndexes[0]);
 	}
 
-	rmesa->lost_context = GL_TRUE;
+	if (!radeon->radeonScreen->driScreen->dri2.enabled) {
+		DRM_CAS(radeon->dri.hwLock, radeon->dri.hwContext,
+			 (DRM_LOCK_HELD | radeon->dri.hwContext), ret );
+		if (ret)
+			radeonGetLock(radeon, 0);
+	}
+}
+
+void radeon_unlock_hardware(radeonContextPtr radeon)
+{
+	if (!radeon->radeonScreen->driScreen->dri2.enabled) {
+		DRM_UNLOCK( radeon->dri.fd,
+			    radeon->dri.hwLock,
+			    radeon->dri.hwContext );
+	}
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.h b/src/mesa/drivers/dri/radeon/radeon_lock.h
index 86e96aa7d2c..2817709eed6 100644
--- a/src/mesa/drivers/dri/radeon/radeon_lock.h
+++ b/src/mesa/drivers/dri/radeon/radeon_lock.h
@@ -39,74 +39,21 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  *   Kevin E. Martin <[email protected]>
  */
 
-#ifndef __RADEON_LOCK_H__
-#define __RADEON_LOCK_H__
+#ifndef COMMON_LOCK_H
+#define COMMON_LOCK_H
 
-extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags);
-
-/* Turn DEBUG_LOCKING on to find locking conflicts.
- */
-#define DEBUG_LOCKING	0
-
-#if DEBUG_LOCKING
-extern char *prevLockFile;
-extern int prevLockLine;
-
-#define DEBUG_LOCK()							\
-   do {									\
-      prevLockFile = (__FILE__);					\
-      prevLockLine = (__LINE__);					\
-   } while (0)
-
-#define DEBUG_RESET()							\
-   do {									\
-      prevLockFile = 0;							\
-      prevLockLine = 0;							\
-   } while (0)
-
-#define DEBUG_CHECK_LOCK()						\
-   do {									\
-      if ( prevLockFile ) {						\
-	 fprintf( stderr,						\
-		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
-		  prevLockFile, prevLockLine, __FILE__, __LINE__ );	\
-	 exit( 1 );							\
-      }									\
-   } while (0)
-
-#else
+#include "main/colormac.h"
+#include "radeon_screen.h"
+#include "radeon_common.h"
 
-#define DEBUG_LOCK()
-#define DEBUG_RESET()
-#define DEBUG_CHECK_LOCK()
+extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags);
 
-#endif
-
-/*
- * !!! We may want to separate locks from locks with validation.  This
- * could be used to improve performance for those things commands that
- * do not do any drawing !!!
- */
+void radeon_lock_hardware(radeonContextPtr rmesa);
+void radeon_unlock_hardware(radeonContextPtr rmesa);
 
 /* Lock the hardware and validate our state.
  */
-#define LOCK_HARDWARE( rmesa )					\
-   do {								\
-      char __ret = 0;						\
-      DEBUG_CHECK_LOCK();					\
-      DRM_CAS( (rmesa)->dri.hwLock, (rmesa)->dri.hwContext,		\
-	       (DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret );	\
-      if ( __ret )						\
-	 radeonGetLock( (rmesa), 0 );				\
-      DEBUG_LOCK();						\
-   } while (0)
-
-#define UNLOCK_HARDWARE( rmesa )					\
-   do {									\
-      DRM_UNLOCK( (rmesa)->dri.fd,					\
-		  (rmesa)->dri.hwLock,					\
-		  (rmesa)->dri.hwContext );				\
-      DEBUG_RESET();							\
-   } while (0)
+#define LOCK_HARDWARE( rmesa )	radeon_lock_hardware(rmesa)
+#define UNLOCK_HARDWARE( rmesa )  radeon_unlock_hardware(rmesa)
 
-#endif				/* __RADEON_LOCK_H__ */
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos.h b/src/mesa/drivers/dri/radeon/radeon_maos.h
index b8935e84a05..b88eb198d57 100644
--- a/src/mesa/drivers/dri/radeon/radeon_maos.h
+++ b/src/mesa/drivers/dri/radeon/radeon_maos.h
@@ -38,6 +38,5 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_context.h"
 
 extern void radeonEmitArrays( GLcontext *ctx, GLuint inputs );
-extern void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs );
 
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
index 31eea13f4ef..7c6ea0530e0 100644
--- a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
@@ -48,160 +48,35 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_maos.h"
 #include "radeon_tcl.h"
 
-#if 0
-/* Usage:
- *   - from radeon_tcl_render
- *   - call radeonEmitArrays to ensure uptodate arrays in dma
- *   - emit primitives (new type?) which reference the data
- *       -- need to use elts for lineloop, quads, quadstrip/flat
- *       -- other primitives are all well-formed (need tristrip-1,fake-poly)
- *
- */
-static void emit_ubyte_rgba3( GLcontext *ctx,
-		       struct radeon_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
+static void emit_vecfog(GLcontext *ctx, struct radeon_aos *aos,
+			GLvoid *data, int stride, int count)
 {
    int i;
-   radeon_color_t *out = (radeon_color_t *)(rvb->start + rvb->address);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d out %p\n",
-	      __FUNCTION__, count, stride, (void *)out);
-
-   for (i = 0; i < count; i++) {
-      out->red   = *data;
-      out->green = *(data+1);
-      out->blue  = *(data+2);
-      out->alpha = 0xFF;
-      out++;
-      data += stride;
-   }
-}
-
-static void emit_ubyte_rgba4( GLcontext *ctx,
-			      struct radeon_dma_region *rvb,
-			      char *data,
-			      int stride,
-			      int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
+   uint32_t *out;
+   int size = 1;
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 
    if (RADEON_DEBUG & DEBUG_VERTS)
       fprintf(stderr, "%s count %d stride %d\n",
 	      __FUNCTION__, count, stride);
 
-   if (stride == 4)
-       COPY_DWORDS( out, data, count );
-   else
-      for (i = 0; i < count; i++) {
-	 *out++ = LE32_TO_CPU(*(int *)data);
-	 data += stride;
-      }
-}
-
-
-static void emit_ubyte_rgba( GLcontext *ctx,
-			     struct radeon_dma_region *rvb,
-			     char *data,
-			     int size,
-			     int stride,
-			     int count )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size);
-
-   assert (!rvb->buf);
-
    if (stride == 0) {
-      radeonAllocDmaRegion( rmesa, rvb, 4, 4 );
+      radeonAllocDmaRegion( rmesa, &aos->bo, &aos->offset, size * 4, 32 );
       count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = 1;
+      aos->stride = 0;
    }
    else {
-      radeonAllocDmaRegion( rmesa, rvb, 4 * count, 4 );	/* alignment? */
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 1;
-      rvb->aos_size = 1;
+      radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32);
+      aos->stride = size;
    }
 
-   /* Emit the data
-    */
-   switch (size) {
-   case 3:
-      emit_ubyte_rgba3( ctx, rvb, data, stride, count );
-      break;
-   case 4:
-      emit_ubyte_rgba4( ctx, rvb, data, stride, count );
-      break;
-   default:
-      assert(0);
-      exit(1);
-      break;
-   }
-}
-#endif
-
-#if defined(USE_X86_ASM)
-#define COPY_DWORDS( dst, src, nr )					\
-do {									\
-	int __tmp;							\
-	__asm__ __volatile__( "rep ; movsl"				\
-			      : "=%c" (__tmp), "=D" (dst), "=S" (__tmp)	\
-			      : "0" (nr),				\
-			        "D" ((long)dst),			\
-			        "S" ((long)src) );			\
-} while (0)
-#else
-#define COPY_DWORDS( dst, src, nr )		\
-do {						\
-   int j;					\
-   for ( j = 0 ; j < nr ; j++ )			\
-      dst[j] = ((int *)src)[j];			\
-   dst += nr;					\
-} while (0)
-#endif
-
-static void emit_vecfog( GLcontext *ctx,
-			 struct radeon_dma_region *rvb,
-			 char *data,
-			 int stride,
-			 int count )
-{
-   int i;
-   GLfloat *out;
+   aos->components = size;
+   aos->count = count;
 
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   assert (!rvb->buf);
-
-   if (stride == 0) {
-      radeonAllocDmaRegion( rmesa, rvb, 4, 4 );
-      count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = 1;
-   }
-   else {
-      radeonAllocDmaRegion( rmesa, rvb, count * 4, 4 );	/* alignment? */
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 1;
-      rvb->aos_size = 1;
-   }
 
    /* Emit the data
     */
-   out = (GLfloat *)(rvb->address + rvb->start);
+   out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
    for (i = 0; i < count; i++) {
       out[0] = radeonComputeFogBlendFactor( ctx, *(GLfloat *)data );
       out++;
@@ -209,169 +84,9 @@ static void emit_vecfog( GLcontext *ctx,
    }
 }
 
-static void emit_vec4( GLcontext *ctx,
-		       struct radeon_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 4)
-      COPY_DWORDS( out, data, count );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out++;
-	 data += stride;
-      }
-}
-
-
-static void emit_vec8( GLcontext *ctx,
-		       struct radeon_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 8)
-      COPY_DWORDS( out, data, count*2 );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out[1] = *(int *)(data+4);
-	 out += 2;
-	 data += stride;
-      }
-}
-
-static void emit_vec12( GLcontext *ctx,
-		       struct radeon_dma_region *rvb,
-		       char *data,
-		       int stride,
-		       int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d out %p data %p\n",
-	      __FUNCTION__, count, stride, (void *)out, (void *)data);
-
-   if (stride == 12)
-      COPY_DWORDS( out, data, count*3 );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out[1] = *(int *)(data+4);
-	 out[2] = *(int *)(data+8);
-	 out += 3;
-	 data += stride;
-      }
-}
-
-static void emit_vec16( GLcontext *ctx,
-			struct radeon_dma_region *rvb,
-			char *data,
-			int stride,
-			int count )
-{
-   int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d stride %d\n",
-	      __FUNCTION__, count, stride);
-
-   if (stride == 16)
-      COPY_DWORDS( out, data, count*4 );
-   else
-      for (i = 0; i < count; i++) {
-	 out[0] = *(int *)data;
-	 out[1] = *(int *)(data+4);
-	 out[2] = *(int *)(data+8);
-	 out[3] = *(int *)(data+12);
-	 out += 4;
-	 data += stride;
-      }
-}
-
-
-static void emit_vector( GLcontext *ctx,
-			 struct radeon_dma_region *rvb,
-			 char *data,
-			 int size,
-			 int stride,
-			 int count )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if (RADEON_DEBUG & DEBUG_VERTS)
-      fprintf(stderr, "%s count %d size %d stride %d\n",
-	      __FUNCTION__, count, size, stride);
-
-   assert (!rvb->buf);
-
-   if (stride == 0) {
-      radeonAllocDmaRegion( rmesa, rvb, size * 4, 4 );
-      count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = size;
-   }
-   else {
-      radeonAllocDmaRegion( rmesa, rvb, size * count * 4, 4 );	/* alignment? */
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = size;
-      rvb->aos_size = size;
-   }
-
-   /* Emit the data
-    */
-   switch (size) {
-   case 1:
-      emit_vec4( ctx, rvb, data, stride, count );
-      break;
-   case 2:
-      emit_vec8( ctx, rvb, data, stride, count );
-      break;
-   case 3:
-      emit_vec12( ctx, rvb, data, stride, count );
-      break;
-   case 4:
-      emit_vec16( ctx, rvb, data, stride, count );
-      break;
-   default:
-      assert(0);
-      exit(1);
-      break;
-   }
-
-}
-
-
-
-static void emit_s0_vec( GLcontext *ctx,
-			 struct radeon_dma_region *rvb,
-			 char *data,
-			 int stride,
-			 int count )
+static void emit_s0_vec(uint32_t *out, GLvoid *data, int stride, int count)
 {
    int i;
-   int *out = (int *)(rvb->address + rvb->start);
-
    if (RADEON_DEBUG & DEBUG_VERTS)
       fprintf(stderr, "%s count %d stride %d\n",
 	      __FUNCTION__, count, stride);
@@ -384,14 +99,9 @@ static void emit_s0_vec( GLcontext *ctx,
    }
 }
 
-static void emit_stq_vec( GLcontext *ctx,
-			 struct radeon_dma_region *rvb,
-			 char *data,
-			 int stride,
-			 int count )
+static void emit_stq_vec(uint32_t *out, GLvoid *data, int stride, int count)
 {
    int i;
-   int *out = (int *)(rvb->address + rvb->start);
 
    if (RADEON_DEBUG & DEBUG_VERTS)
       fprintf(stderr, "%s count %d stride %d\n",
@@ -409,21 +119,16 @@ static void emit_stq_vec( GLcontext *ctx,
 
 
 
-static void emit_tex_vector( GLcontext *ctx,
-			     struct radeon_dma_region *rvb,
-			     char *data,
-			     int size,
-			     int stride,
-			     int count )
+static void emit_tex_vector(GLcontext *ctx, struct radeon_aos *aos,
+			    GLvoid *data, int size, int stride, int count)
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
    int emitsize;
+   uint32_t *out;
 
    if (RADEON_DEBUG & DEBUG_VERTS)
       fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size);
 
-   assert (!rvb->buf);
-
    switch (size) {
    case 4: emitsize = 3; break;
    case 3: emitsize = 3; break;
@@ -432,34 +137,33 @@ static void emit_tex_vector( GLcontext *ctx,
 
 
    if (stride == 0) {
-      radeonAllocDmaRegion( rmesa, rvb, 4 * emitsize, 4 );
+      radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, emitsize * 4, 32);
       count = 1;
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = 0;
-      rvb->aos_size = emitsize;
+      aos->stride = 0;
    }
    else {
-      radeonAllocDmaRegion( rmesa, rvb, 4 * emitsize * count, 4 );
-      rvb->aos_start = GET_START(rvb);
-      rvb->aos_stride = emitsize;
-      rvb->aos_size = emitsize;
+      radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, emitsize * count * 4, 32);
+      aos->stride = emitsize;
    }
 
+   aos->components = emitsize;
+   aos->count = count;
 
    /* Emit the data
     */
+   out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
    switch (size) {
    case 1:
-      emit_s0_vec( ctx, rvb, data, stride, count ); 
+      emit_s0_vec( out, data, stride, count );
       break;
    case 2:
-      emit_vec8( ctx, rvb, data, stride, count );
+      radeonEmitVec8( out, data, stride, count );
       break;
    case 3:
-      emit_vec12( ctx, rvb, data, stride, count );
+      radeonEmitVec12( out, data, stride, count );
       break;
    case 4:
-      emit_stq_vec( ctx, rvb, data, stride, count );
+      emit_stq_vec( out, data, stride, count );
       break;
    default:
       assert(0);
@@ -476,9 +180,8 @@ static void emit_tex_vector( GLcontext *ctx,
  */
 void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   r100ContextPtr rmesa = R100_CONTEXT( ctx );
    struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb;
-   struct radeon_dma_region **component = rmesa->tcl.aos_components;
    GLuint nr = 0;
    GLuint vfmt = 0;
    GLuint count = VB->Count;
@@ -491,12 +194,12 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
 
    if (1) {
       if (!rmesa->tcl.obj.buf) 
-	 emit_vector( ctx, 
-		      &rmesa->tcl.obj, 
-		      (char *)VB->ObjPtr->data,
-		      VB->ObjPtr->size,
-		      VB->ObjPtr->stride,
-		      count);
+	rcommon_emit_vector( ctx, 
+			     &(rmesa->tcl.aos[nr]),
+			     (char *)VB->ObjPtr->data,
+			     VB->ObjPtr->size,
+			     VB->ObjPtr->stride,
+			     count);
 
       switch( VB->ObjPtr->size ) {
       case 4: vfmt |= RADEON_CP_VC_FRMT_W0;
@@ -505,21 +208,21 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
       default:
          break;
       }
-      component[nr++] = &rmesa->tcl.obj;
+      nr++;
    }
    
 
    if (inputs & VERT_BIT_NORMAL) {
       if (!rmesa->tcl.norm.buf)
-	 emit_vector( ctx, 
-		      &(rmesa->tcl.norm), 
-		      (char *)VB->NormalPtr->data,
-		      3,
-		      VB->NormalPtr->stride,
-		      count);
+	 rcommon_emit_vector( ctx, 
+			      &(rmesa->tcl.aos[nr]),
+			      (char *)VB->NormalPtr->data,
+			      3,
+			      VB->NormalPtr->stride,
+			      count);
 
       vfmt |= RADEON_CP_VC_FRMT_N0;
-      component[nr++] = &rmesa->tcl.norm;
+      nr++;
    }
 
    if (inputs & VERT_BIT_COLOR0) {
@@ -537,31 +240,30 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
       }
 
       if (!rmesa->tcl.rgba.buf)
-	 emit_vector( ctx,
-		      &(rmesa->tcl.rgba),
-		      (char *)VB->ColorPtr[0]->data,
-		      emitsize,
-		      VB->ColorPtr[0]->stride,
-		      count);
-
-
-      component[nr++] = &rmesa->tcl.rgba;
+	rcommon_emit_vector( ctx,
+			     &(rmesa->tcl.aos[nr]),
+			     (char *)VB->ColorPtr[0]->data,
+			     emitsize,
+			     VB->ColorPtr[0]->stride,
+			     count);
+
+      nr++;
    }
 
 
    if (inputs & VERT_BIT_COLOR1) {
       if (!rmesa->tcl.spec.buf) {
 
-	 emit_vector( ctx,
-		      &rmesa->tcl.spec,
-		      (char *)VB->SecondaryColorPtr[0]->data,
-		      3,
-		      VB->SecondaryColorPtr[0]->stride,
-		      count);
+	rcommon_emit_vector( ctx,
+			     &(rmesa->tcl.aos[nr]),
+			     (char *)VB->SecondaryColorPtr[0]->data,
+			     3,
+			     VB->SecondaryColorPtr[0]->stride,
+			     count);
       }
 
       vfmt |= RADEON_CP_VC_FRMT_FPSPEC;
-      component[nr++] = &rmesa->tcl.spec;
+      nr++;
    }
 
 /* FIXME: not sure if this is correct. May need to stitch this together with
@@ -570,13 +272,13 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
    if (inputs & VERT_BIT_FOG) {
       if (!rmesa->tcl.fog.buf)
 	 emit_vecfog( ctx,
-		      &(rmesa->tcl.fog),
+		      &(rmesa->tcl.aos[nr]),
 		      (char *)VB->FogCoordPtr->data,
 		      VB->FogCoordPtr->stride,
 		      count);
 
       vfmt |= RADEON_CP_VC_FRMT_FPFOG;
-      component[nr++] = &rmesa->tcl.fog;
+      nr++;
    }
 
 
@@ -587,11 +289,12 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
       if (inputs & VERT_BIT_TEX(unit)) {
 	 if (!rmesa->tcl.tex[unit].buf)
 	    emit_tex_vector( ctx,
-			     &(rmesa->tcl.tex[unit]),
+			     &(rmesa->tcl.aos[nr]),
 			     (char *)VB->TexCoordPtr[unit]->data,
 			     VB->TexCoordPtr[unit]->size,
 			     VB->TexCoordPtr[unit]->stride,
 			     count );
+	 nr++;
 
 	 vfmt |= RADEON_ST_BIT(unit);
          /* assume we need the 3rd coord if texgen is active for r/q OR at least
@@ -609,7 +312,6 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
 		 (swaptexmatcol != ((rmesa->TexMatColSwap >> unit) & 1)))
 	       radeonUploadTexMatrix( rmesa, unit, swaptexmatcol ) ;
 	 }
-	 component[nr++] = &rmesa->tcl.tex[unit];
       }
    }
 
@@ -622,34 +324,3 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
    rmesa->tcl.vertex_format = vfmt;
 }
 
-
-void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
-   GLuint unit;
-
-#if 0
-   if (RADEON_DEBUG & DEBUG_VERTS) 
-      _tnl_print_vert_flags( __FUNCTION__, newinputs );
-#endif
-
-   if (newinputs & VERT_BIT_POS) 
-     radeonReleaseDmaRegion( rmesa, &rmesa->tcl.obj, __FUNCTION__ );
-
-   if (newinputs & VERT_BIT_NORMAL) 
-      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.norm, __FUNCTION__ );
-
-   if (newinputs & VERT_BIT_COLOR0) 
-      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.rgba, __FUNCTION__ );
-
-   if (newinputs & VERT_BIT_COLOR1) 
-      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.spec, __FUNCTION__ );
-      
-   if (newinputs & VERT_BIT_FOG)
-      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.fog, __FUNCTION__ );
-
-   for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++) {
-      if (newinputs & VERT_BIT_TEX(unit))
-         radeonReleaseDmaRegion( rmesa, &rmesa->tcl.tex[unit], __FUNCTION__ );
-   }
-}
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
index 126d0727c63..78ec1193026 100644
--- a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
@@ -310,7 +310,7 @@ static void init_tcl_verts( void )
 
 void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
    GLuint req = 0;
    GLuint unit;
@@ -374,14 +374,15 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
 	 break;
 
    if (rmesa->tcl.vertex_format == setup_tab[i].vertex_format &&
-       rmesa->tcl.indexed_verts.buf)
+       rmesa->radeon.tcl.aos[0].bo)
       return;
 
-   if (rmesa->tcl.indexed_verts.buf)
+   if (rmesa->radeon.tcl.aos[0].bo)
       radeonReleaseArrays( ctx, ~0 );
 
-   radeonAllocDmaRegion( rmesa,
-			 &rmesa->tcl.indexed_verts, 
+   radeonAllocDmaRegion( &rmesa->radeon,
+			 &rmesa->radeon.tcl.aos[0].bo,
+			 &rmesa->radeon.tcl.aos[0].offset,
 			 VB->Count * setup_tab[i].vertex_size * 4, 
 			 4);
 
@@ -421,29 +422,12 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
 
 
    setup_tab[i].emit( ctx, 0, VB->Count, 
-		      rmesa->tcl.indexed_verts.address + 
-		      rmesa->tcl.indexed_verts.start );
+		      rmesa->radeon.tcl.aos[0].bo->ptr + rmesa->radeon.tcl.aos[0].offset);
 
+   //   rmesa->radeon.tcl.aos[0].size = setup_tab[i].vertex_size;
+   rmesa->radeon.tcl.aos[0].stride = setup_tab[i].vertex_size;
    rmesa->tcl.vertex_format = setup_tab[i].vertex_format;
-   rmesa->tcl.indexed_verts.aos_start = GET_START( &rmesa->tcl.indexed_verts );
-   rmesa->tcl.indexed_verts.aos_size = setup_tab[i].vertex_size;
-   rmesa->tcl.indexed_verts.aos_stride = setup_tab[i].vertex_size;
-
-   rmesa->tcl.aos_components[0] = &rmesa->tcl.indexed_verts;
-   rmesa->tcl.nr_aos_components = 1;
+   rmesa->radeon.tcl.aos_count = 1;
 }
 
 
-
-void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
-
-#if 0
-   if (RADEON_DEBUG & DEBUG_VERTS) 
-      _tnl_print_vert_flags( __FUNCTION__, newinputs );
-#endif
-
-   if (newinputs) 
-     radeonReleaseDmaRegion( rmesa, &rmesa->tcl.indexed_verts, __FUNCTION__ );
-}
diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
new file mode 100644
index 00000000000..8d1ba1cdbab
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
@@ -0,0 +1,415 @@
+/*
+ * Copyright (C) 2008 Nicolai Haehnle.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "radeon_mipmap_tree.h"
+
+#include <errno.h>
+#include <unistd.h>
+
+#include "main/simple_list.h"
+#include "main/texcompress.h"
+#include "main/texformat.h"
+
+static GLuint radeon_compressed_texture_size(GLcontext *ctx,
+		GLsizei width, GLsizei height, GLsizei depth,
+		GLuint mesaFormat)
+{
+	GLuint size = _mesa_compressed_texture_size(ctx, width, height, depth, mesaFormat);
+
+	if (mesaFormat == MESA_FORMAT_RGB_DXT1 ||
+	    mesaFormat == MESA_FORMAT_RGBA_DXT1) {
+		if (width + 3 < 8)	/* width one block */
+			size = size * 4;
+		else if (width + 3 < 16)
+			size = size * 2;
+	} else {
+		/* DXT3/5, 16 bytes per block */
+	  //		WARN_ONCE("DXT 3/5 suffers from multitexturing problems!\n");
+		if (width + 3 < 8)
+			size = size * 2;
+	}
+
+	return size;
+}
+
+
+static int radeon_compressed_num_bytes(GLuint mesaFormat)
+{
+   int bytes = 0;
+   switch(mesaFormat) {
+     
+   case MESA_FORMAT_RGB_FXT1:
+   case MESA_FORMAT_RGBA_FXT1:
+   case MESA_FORMAT_RGB_DXT1:
+   case MESA_FORMAT_RGBA_DXT1:
+     bytes = 2;
+     break;
+     
+   case MESA_FORMAT_RGBA_DXT3:
+   case MESA_FORMAT_RGBA_DXT5:
+     bytes = 4;
+   default:
+     break;
+   }
+   
+   return bytes;
+}
+
+/**
+ * Compute sizes and fill in offset and blit information for the given
+ * image (determined by \p face and \p level).
+ *
+ * \param curOffset points to the offset at which the image is to be stored
+ * and is updated by this function according to the size of the image.
+ */
+static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree *mt,
+	GLuint face, GLuint level, GLuint* curOffset)
+{
+	radeon_mipmap_level *lvl = &mt->levels[level];
+	uint32_t row_align = rmesa->texture_row_align - 1;
+
+	/* Find image size in bytes */
+	if (mt->compressed) {
+		/* TODO: Is this correct? Need test cases for compressed textures! */
+		lvl->rowstride = (lvl->width * mt->bpp + 63) & ~63;
+		lvl->size = radeon_compressed_texture_size(mt->radeon->glCtx,
+							   lvl->width, lvl->height, lvl->depth, mt->compressed);
+	} else if (mt->target == GL_TEXTURE_RECTANGLE_NV) {
+		lvl->rowstride = (lvl->width * mt->bpp + 63) & ~63;
+		lvl->size = lvl->rowstride * lvl->height;
+	} else if (mt->tilebits & RADEON_TXO_MICRO_TILE) {
+		/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
+		 * though the actual offset may be different (if texture is less than
+		 * 32 bytes width) to the untiled case */
+		lvl->rowstride = (lvl->width * mt->bpp * 2 + 31) & ~31;
+		lvl->size = lvl->rowstride * ((lvl->height + 1) / 2) * lvl->depth;
+	} else {
+		lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align;
+		lvl->size = lvl->rowstride * lvl->height * lvl->depth;
+	}
+	assert(lvl->size > 0);
+
+	/* All images are aligned to a 32-byte offset */
+	*curOffset = (*curOffset + 0x1f) & ~0x1f;
+	lvl->faces[face].offset = *curOffset;
+	*curOffset += lvl->size;
+
+	if (RADEON_DEBUG & DEBUG_TEXTURE)
+	  fprintf(stderr,
+		  "level %d, face %d: rs:%d %dx%d at %d\n",
+		  level, face, lvl->rowstride, lvl->width, lvl->height, lvl->faces[face].offset);
+}
+
+static GLuint minify(GLuint size, GLuint levels)
+{
+	size = size >> levels;
+	if (size < 1)
+		size = 1;
+	return size;
+}
+
+
+static void calculate_miptree_layout_r100(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
+{
+	GLuint curOffset;
+	GLuint numLevels;
+	GLuint i;
+	GLuint face;
+
+	numLevels = mt->lastLevel - mt->firstLevel + 1;
+	assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS);
+
+	curOffset = 0;
+	for(face = 0; face < mt->faces; face++) {
+
+		for(i = 0; i < numLevels; i++) {
+			mt->levels[i].width = minify(mt->width0, i);
+			mt->levels[i].height = minify(mt->height0, i);
+			mt->levels[i].depth = minify(mt->depth0, i);
+			compute_tex_image_offset(rmesa, mt, face, i, &curOffset);
+		}
+	}
+
+	/* Note the required size in memory */
+	mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
+}
+
+static void calculate_miptree_layout_r300(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
+{
+	GLuint curOffset;
+	GLuint numLevels;
+	GLuint i;
+
+	numLevels = mt->lastLevel - mt->firstLevel + 1;
+	assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS);
+
+	curOffset = 0;
+	for(i = 0; i < numLevels; i++) {
+		GLuint face;
+
+		mt->levels[i].width = minify(mt->width0, i);
+		mt->levels[i].height = minify(mt->height0, i);
+		mt->levels[i].depth = minify(mt->depth0, i);
+
+		for(face = 0; face < mt->faces; face++)
+			compute_tex_image_offset(rmesa, mt, face, i, &curOffset);
+	}
+
+	/* Note the required size in memory */
+	mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
+}
+
+/**
+ * Create a new mipmap tree, calculate its layout and allocate memory.
+ */
+radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *t,
+		GLenum target, GLuint firstLevel, GLuint lastLevel,
+		GLuint width0, GLuint height0, GLuint depth0,
+		GLuint bpp, GLuint tilebits, GLuint compressed)
+{
+	radeon_mipmap_tree *mt = CALLOC_STRUCT(_radeon_mipmap_tree);
+
+	mt->radeon = rmesa;
+	mt->refcount = 1;
+	mt->t = t;
+	mt->target = target;
+	mt->faces = (target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+	mt->firstLevel = firstLevel;
+	mt->lastLevel = lastLevel;
+	mt->width0 = width0;
+	mt->height0 = height0;
+	mt->depth0 = depth0;
+	mt->bpp = compressed ? radeon_compressed_num_bytes(compressed) : bpp;
+	mt->tilebits = tilebits;
+	mt->compressed = compressed;
+
+	if (rmesa->radeonScreen->chip_family >= CHIP_FAMILY_R300)
+		calculate_miptree_layout_r300(rmesa, mt);
+	else
+		calculate_miptree_layout_r100(rmesa, mt);
+
+	mt->bo = radeon_bo_open(rmesa->radeonScreen->bom,
+                            0, mt->totalsize, 1024,
+                            RADEON_GEM_DOMAIN_VRAM,
+                            0);
+
+	return mt;
+}
+
+void radeon_miptree_reference(radeon_mipmap_tree *mt)
+{
+	mt->refcount++;
+	assert(mt->refcount > 0);
+}
+
+void radeon_miptree_unreference(radeon_mipmap_tree *mt)
+{
+	if (!mt)
+		return;
+
+	assert(mt->refcount > 0);
+	mt->refcount--;
+	if (!mt->refcount) {
+		radeon_bo_unref(mt->bo);
+		free(mt);
+	}
+}
+
+
+/**
+ * Calculate first and last mip levels for the given texture object,
+ * where the dimensions are taken from the given texture image at
+ * the given level.
+ *
+ * Note: level is the OpenGL level number, which is not necessarily the same
+ * as the first level that is actually present.
+ *
+ * The base level image of the given texture face must be non-null,
+ * or this will fail.
+ */
+static void calculate_first_last_level(struct gl_texture_object *tObj,
+				       GLuint *pfirstLevel, GLuint *plastLevel,
+				       GLuint face, GLuint level)
+{
+	const struct gl_texture_image * const baseImage =
+		tObj->Image[face][level];
+
+	assert(baseImage);
+	
+	/* These must be signed values.  MinLod and MaxLod can be negative numbers,
+	* and having firstLevel and lastLevel as signed prevents the need for
+	* extra sign checks.
+	*/
+	int   firstLevel;
+	int   lastLevel;
+
+	/* Yes, this looks overly complicated, but it's all needed.
+	*/
+	switch (tObj->Target) {
+	case GL_TEXTURE_1D:
+	case GL_TEXTURE_2D:
+	case GL_TEXTURE_3D:
+	case GL_TEXTURE_CUBE_MAP:
+		if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) {
+			/* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
+			*/
+			firstLevel = lastLevel = tObj->BaseLevel;
+		} else {
+			firstLevel = tObj->BaseLevel + (GLint)(tObj->MinLod + 0.5);
+			firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+			firstLevel = MIN2(firstLevel, level + baseImage->MaxLog2);
+			lastLevel = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5);
+			lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+			lastLevel = MIN2(lastLevel, level + baseImage->MaxLog2);
+			lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+			lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
+		}
+		break;
+	case GL_TEXTURE_RECTANGLE_NV:
+	case GL_TEXTURE_4D_SGIS:
+		firstLevel = lastLevel = 0;
+		break;
+	default:
+		return;
+	}
+
+	/* save these values */
+	*pfirstLevel = firstLevel;
+	*plastLevel = lastLevel;
+}
+
+
+/**
+ * Checks whether the given miptree can hold the given texture image at the
+ * given face and level.
+ */
+GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
+		struct gl_texture_image *texImage, GLuint face, GLuint level)
+{
+	radeon_mipmap_level *lvl;
+
+	if (face >= mt->faces || level < mt->firstLevel || level > mt->lastLevel)
+		return GL_FALSE;
+
+	if (texImage->IsCompressed != mt->compressed)
+		return GL_FALSE;
+
+	if (!texImage->IsCompressed &&
+	    !mt->compressed &&
+	    texImage->TexFormat->TexelBytes != mt->bpp)
+		return GL_FALSE;
+
+	lvl = &mt->levels[level - mt->firstLevel];
+	if (lvl->width != texImage->Width ||
+	    lvl->height != texImage->Height ||
+	    lvl->depth != texImage->Depth)
+		return GL_FALSE;
+
+	return GL_TRUE;
+}
+
+
+/**
+ * Checks whether the given miptree has the right format to store the given texture object.
+ */
+GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj)
+{
+	struct gl_texture_image *firstImage;
+	GLuint compressed;
+	GLuint numfaces = 1;
+	GLuint firstLevel, lastLevel;
+
+	calculate_first_last_level(texObj, &firstLevel, &lastLevel, 0, texObj->BaseLevel);
+	if (texObj->Target == GL_TEXTURE_CUBE_MAP)
+		numfaces = 6;
+
+	firstImage = texObj->Image[0][firstLevel];
+	compressed = firstImage->IsCompressed ? firstImage->TexFormat->MesaFormat : 0;
+
+	return (mt->firstLevel == firstLevel &&
+	        mt->lastLevel == lastLevel &&
+	        mt->width0 == firstImage->Width &&
+	        mt->height0 == firstImage->Height &&
+	        mt->depth0 == firstImage->Depth &&
+	        mt->bpp == firstImage->TexFormat->TexelBytes &&
+	        mt->compressed == compressed);
+}
+
+
+/**
+ * Try to allocate a mipmap tree for the given texture that will fit the
+ * given image in the given position.
+ */
+void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t,
+		struct gl_texture_image *texImage, GLuint face, GLuint level)
+{
+	GLuint compressed = texImage->IsCompressed ? texImage->TexFormat->MesaFormat : 0;
+	GLuint numfaces = 1;
+	GLuint firstLevel, lastLevel;
+
+	assert(!t->mt);
+
+	calculate_first_last_level(&t->base, &firstLevel, &lastLevel, face, level);
+	if (t->base.Target == GL_TEXTURE_CUBE_MAP)
+		numfaces = 6;
+
+	if (level != firstLevel || face >= numfaces)
+		return;
+
+	t->mt = radeon_miptree_create(rmesa, t, t->base.Target,
+		firstLevel, lastLevel,
+		texImage->Width, texImage->Height, texImage->Depth,
+		texImage->TexFormat->TexelBytes, t->tile_bits, compressed);
+}
+
+/* Although we use the image_offset[] array to store relative offsets
+ * to cube faces, Mesa doesn't know anything about this and expects
+ * each cube face to be treated as a separate image.
+ *
+ * These functions present that view to mesa:
+ */
+void
+radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets)
+{
+     if (mt->target != GL_TEXTURE_3D || mt->faces == 1)
+        offsets[0] = 0;
+     else {
+	int i;
+	for (i = 0; i < 6; i++)
+		offsets[i] = mt->levels[level].faces[i].offset;
+     }
+}
+
+GLuint
+radeon_miptree_image_offset(radeon_mipmap_tree *mt,
+			    GLuint face, GLuint level)
+{
+   if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
+      return (mt->levels[level].faces[face].offset);
+   else
+      return mt->levels[level].faces[0].offset;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
new file mode 100644
index 00000000000..697010bc022
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
@@ -0,0 +1,98 @@
+/*
+ * Copyright (C) 2008 Nicolai Haehnle.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __RADEON_MIPMAP_TREE_H_
+#define __RADEON_MIPMAP_TREE_H_
+
+#include "radeon_common.h"
+
+typedef struct _radeon_mipmap_tree radeon_mipmap_tree;
+typedef struct _radeon_mipmap_level radeon_mipmap_level;
+typedef struct _radeon_mipmap_image radeon_mipmap_image;
+
+struct _radeon_mipmap_image {
+	GLuint offset; /** Offset of this image from the start of mipmap tree buffer, in bytes */
+};
+
+struct _radeon_mipmap_level {
+	GLuint width;
+	GLuint height;
+	GLuint depth;
+	GLuint size; /** Size of each image, in bytes */
+	GLuint rowstride; /** in bytes */
+	radeon_mipmap_image faces[6];
+};
+
+
+/**
+ * A mipmap tree contains texture images in the layout that the hardware
+ * expects.
+ *
+ * The meta-data of mipmap trees is immutable, i.e. you cannot change the
+ * layout on-the-fly; however, the texture contents (i.e. texels) can be
+ * changed.
+ */
+struct _radeon_mipmap_tree {
+	radeonContextPtr radeon;
+	radeonTexObj *t;
+	struct radeon_bo *bo;
+	GLuint refcount;
+
+	GLuint totalsize; /** total size of the miptree, in bytes */
+
+	GLenum target; /** GL_TEXTURE_xxx */
+	GLuint faces; /** # of faces: 6 for cubemaps, 1 otherwise */
+	GLuint firstLevel; /** First mip level stored in this mipmap tree */
+	GLuint lastLevel; /** Last mip level stored in this mipmap tree */
+
+	GLuint width0; /** Width of firstLevel image */
+	GLuint height0; /** Height of firstLevel image */
+	GLuint depth0; /** Depth of firstLevel image */
+
+	GLuint bpp; /** Bytes per texel */
+	GLuint tilebits; /** RADEON_TXO_xxx_TILE */
+	GLuint compressed; /** MESA_FORMAT_xxx indicating a compressed format, or 0 if uncompressed */
+
+	radeon_mipmap_level levels[RADEON_MAX_TEXTURE_LEVELS];
+};
+
+radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *t,
+		GLenum target, GLuint firstLevel, GLuint lastLevel,
+		GLuint width0, GLuint height0, GLuint depth0,
+		GLuint bpp, GLuint tilebits, GLuint compressed);
+void radeon_miptree_reference(radeon_mipmap_tree *mt);
+void radeon_miptree_unreference(radeon_mipmap_tree *mt);
+
+GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
+		struct gl_texture_image *texImage, GLuint face, GLuint level);
+GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj);
+void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t,
+			      struct gl_texture_image *texImage, GLuint face, GLuint level);
+GLuint radeon_miptree_image_offset(radeon_mipmap_tree *mt,
+				   GLuint face, GLuint level);
+void radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets);
+#endif /* __RADEON_MIPMAP_TREE_H_ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.c b/src/mesa/drivers/dri/radeon/radeon_sanity.c
index 6613757fcea..bbed838b592 100644
--- a/src/mesa/drivers/dri/radeon/radeon_sanity.c
+++ b/src/mesa/drivers/dri/radeon/radeon_sanity.c
@@ -973,7 +973,7 @@ static int radeon_emit_packet3_cliprect( drm_radeon_cmd_buffer_t *cmdbuf )
 }
 
 
-int radeonSanityCmdBuffer( radeonContextPtr rmesa,
+int radeonSanityCmdBuffer( r100ContextPtr rmesa,
 			   int nbox,
 			   drm_clip_rect_t *boxes )
 {
diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.h b/src/mesa/drivers/dri/radeon/radeon_sanity.h
index 1ec06bc586b..f30eb1c4f15 100644
--- a/src/mesa/drivers/dri/radeon/radeon_sanity.h
+++ b/src/mesa/drivers/dri/radeon/radeon_sanity.h
@@ -1,7 +1,7 @@
 #ifndef RADEON_SANITY_H
 #define RADEON_SANITY_H
 
-extern int radeonSanityCmdBuffer( radeonContextPtr rmesa,
+extern int radeonSanityCmdBuffer( r100ContextPtr rmesa,
 				  int nbox,
 				  drm_clip_rect_t *boxes );
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c
index 791f59826bd..12ae4ada5db 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.c
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.c
@@ -35,6 +35,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * \author  Gareth Hughes <[email protected]>
  */
 
+#include <errno.h>
 #include "main/glheader.h"
 #include "main/imports.h"
 #include "main/mtypes.h"
@@ -45,32 +46,39 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_chipset.h"
 #include "radeon_macros.h"
 #include "radeon_screen.h"
+#include "radeon_common.h"
+#include "radeon_span.h"
 #if !RADEON_COMMON
 #include "radeon_context.h"
-#include "radeon_span.h"
 #include "radeon_tex.h"
 #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
 #include "r200_context.h"
 #include "r200_ioctl.h"
-#include "r200_span.h"
 #include "r200_tex.h"
 #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
 #include "r300_context.h"
 #include "r300_fragprog.h"
 #include "r300_tex.h"
-#include "radeon_span.h"
 #endif
 
 #include "utils.h"
 #include "vblank.h"
 #include "drirenderbuffer.h"
 
+#include "radeon_bocs_wrapper.h"
+
 #include "GL/internal/dri_interface.h"
 
 /* Radeon configuration
  */
 #include "xmlpool.h"
 
+#define DRI_CONF_COMMAND_BUFFER_SIZE(def,min,max) \
+DRI_CONF_OPT_BEGIN_V(command_buffer_size,int,def, # min ":" # max ) \
+        DRI_CONF_DESC(en,"Size of command buffer (in KB)") \
+        DRI_CONF_DESC(de,"Grösse des Befehlspuffers (in KB)") \
+DRI_CONF_OPT_END
+
 #if !RADEON_COMMON	/* R100 */
 PUBLIC const char __driConfigOptions[] =
 DRI_CONF_BEGIN
@@ -80,6 +88,7 @@ DRI_CONF_BEGIN
         DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0)
         DRI_CONF_MAX_TEXTURE_UNITS(3,2,3)
         DRI_CONF_HYPERZ(false)
+        DRI_CONF_COMMAND_BUFFER_SIZE(8, 8, 32)
     DRI_CONF_SECTION_END
     DRI_CONF_SECTION_QUALITY
         DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB)
@@ -95,7 +104,7 @@ DRI_CONF_BEGIN
         DRI_CONF_NO_RAST(false)
     DRI_CONF_SECTION_END
 DRI_CONF_END;
-static const GLuint __driNConfigOptions = 14;
+static const GLuint __driNConfigOptions = 15;
 
 #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
 
@@ -107,6 +116,7 @@ DRI_CONF_BEGIN
         DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0)
         DRI_CONF_MAX_TEXTURE_UNITS(6,2,6)
         DRI_CONF_HYPERZ(false)
+        DRI_CONF_COMMAND_BUFFER_SIZE(8, 8, 32)
     DRI_CONF_SECTION_END
     DRI_CONF_SECTION_QUALITY
         DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB)
@@ -126,7 +136,7 @@ DRI_CONF_BEGIN
         DRI_CONF_NV_VERTEX_PROGRAM(false)
     DRI_CONF_SECTION_END
 DRI_CONF_END;
-static const GLuint __driNConfigOptions = 16;
+static const GLuint __driNConfigOptions = 17;
 
 extern const struct dri_extension blend_extensions[];
 extern const struct dri_extension ARB_vp_extension[];
@@ -149,11 +159,7 @@ DRI_CONF_OPT_BEGIN_V(texture_coord_units,int,def, # min ":" # max ) \
         DRI_CONF_DESC(de,"Anzahl der Texturkoordinateneinheiten") \
 DRI_CONF_OPT_END
 
-#define DRI_CONF_COMMAND_BUFFER_SIZE(def,min,max) \
-DRI_CONF_OPT_BEGIN_V(command_buffer_size,int,def, # min ":" # max ) \
-        DRI_CONF_DESC(en,"Size of command buffer (in KB)") \
-        DRI_CONF_DESC(de,"Grösse des Befehlspuffers (in KB)") \
-DRI_CONF_OPT_END
+
 
 #define DRI_CONF_DISABLE_S3TC(def) \
 DRI_CONF_OPT_BEGIN(disable_s3tc,bool,def) \
@@ -209,7 +215,6 @@ static const GLuint __driNConfigOptions = 17;
 extern const struct dri_extension gl_20_extension[];
 
 #ifndef RADEON_DEBUG
-int RADEON_DEBUG = 0;
 
 static const struct dri_debug_control debug_control[] = {
 	{"fall", DEBUG_FALLBACKS},
@@ -236,19 +241,36 @@ static const struct dri_debug_control debug_control[] = {
 #endif /* RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) */
 
 extern const struct dri_extension card_extensions[];
+extern const struct dri_extension mm_extensions[];
 
 static int getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo );
 
 static int
-radeonGetParam(int fd, int param, void *value)
+radeonGetParam(__DRIscreenPrivate *sPriv, int param, void *value)
 {
   int ret;
-  drm_radeon_getparam_t gp;
-
-  gp.param = param;
-  gp.value = value;
+  drm_radeon_getparam_t gp = { 0 };
+  struct drm_radeon_info info = { 0 };
+
+  if (sPriv->drm_version.major >= 2) {
+      info.value = (uint64_t)value;
+      switch (param) {
+      case RADEON_PARAM_DEVICE_ID:
+          info.request = RADEON_INFO_DEVICE_ID;
+          break;
+      case RADEON_PARAM_NUM_GB_PIPES:
+          info.request = RADEON_INFO_NUM_GB_PIPES;
+          break;
+      default:
+          return -EINVAL;
+      }
+      ret = drmCommandWriteRead(sPriv->fd, DRM_RADEON_INFO, &info, sizeof(info));
+  } else {
+      gp.param = param;
+      gp.value = value;
 
-  ret = drmCommandWriteRead( fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp));
+      ret = drmCommandWriteRead(sPriv->fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp));
+  }
   return ret;
 }
 
@@ -335,6 +357,12 @@ static const __DRItexOffsetExtension radeonTexOffsetExtension = {
     { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION },
     radeonSetTexOffset,
 };
+
+static const __DRItexBufferExtension radeonTexBufferExtension = {
+    { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION },
+   radeonSetTexBuffer,
+   radeonSetTexBuffer2,
+};
 #endif
 
 #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
@@ -349,6 +377,12 @@ static const __DRItexOffsetExtension r200texOffsetExtension = {
     { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION },
    r200SetTexOffset,
 };
+
+static const __DRItexBufferExtension r200TexBufferExtension = {
+    { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION },
+   r200SetTexBuffer,
+   r200SetTexBuffer2,
+};
 #endif
 
 #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
@@ -356,137 +390,19 @@ static const __DRItexOffsetExtension r300texOffsetExtension = {
     { __DRI_TEX_OFFSET, __DRI_TEX_OFFSET_VERSION },
    r300SetTexOffset,
 };
-#endif
 
-/* Create the device specific screen private data struct.
- */
-static radeonScreenPtr
-radeonCreateScreen( __DRIscreenPrivate *sPriv )
-{
-   radeonScreenPtr screen;
-   RADEONDRIPtr dri_priv = (RADEONDRIPtr)sPriv->pDevPriv;
-   unsigned char *RADEONMMIO;
-   int i;
-   int ret;
-   uint32_t temp = 0;
-
-   if (sPriv->devPrivSize != sizeof(RADEONDRIRec)) {
-      fprintf(stderr,"\nERROR!  sizeof(RADEONDRIRec) does not match passed size from device driver\n");
-      return GL_FALSE;
-   }
-
-   /* Allocate the private area */
-   screen = (radeonScreenPtr) CALLOC( sizeof(*screen) );
-   if ( !screen ) {
-      __driUtilMessage("%s: Could not allocate memory for screen structure",
-		       __FUNCTION__);
-      return NULL;
-   }
-
-#if DO_DEBUG && RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
-	RADEON_DEBUG = driParseDebugString(getenv("RADEON_DEBUG"), debug_control);
+static const __DRItexBufferExtension r300TexBufferExtension = {
+    { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION },
+   r300SetTexBuffer,
+   r300SetTexBuffer2,
+};
 #endif
 
-   /* parse information in __driConfigOptions */
-   driParseOptionInfo (&screen->optionCache,
-		       __driConfigOptions, __driNConfigOptions);
-
-   /* This is first since which regions we map depends on whether or
-    * not we are using a PCI card.
-    */
-   screen->card_type = (dri_priv->IsPCI ? RADEON_CARD_PCI : RADEON_CARD_AGP);
-   {
-      int ret;
-      ret = radeonGetParam( sPriv->fd, RADEON_PARAM_GART_BUFFER_OFFSET,
-			    &screen->gart_buffer_offset);
-
-      if (ret) {
-	 FREE( screen );
-	 fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BUFFER_OFFSET): %d\n", ret);
-	 return NULL;
-      }
-
-      ret = radeonGetParam( sPriv->fd, RADEON_PARAM_GART_BASE,
-			    &screen->gart_base);
-      if (ret) {
-	 FREE( screen );
-	 fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BASE): %d\n", ret);
-	 return NULL;
-      }
-
-      ret = radeonGetParam( sPriv->fd, RADEON_PARAM_IRQ_NR,
-			    &screen->irq);
-      if (ret) {
-	 FREE( screen );
-	 fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_IRQ_NR): %d\n", ret);
-	 return NULL;
-      }
-      screen->drmSupportsCubeMapsR200 = (sPriv->drm_version.minor >= 7);
-      screen->drmSupportsBlendColor = (sPriv->drm_version.minor >= 11);
-      screen->drmSupportsTriPerf = (sPriv->drm_version.minor >= 16);
-      screen->drmSupportsFragShader = (sPriv->drm_version.minor >= 18);
-      screen->drmSupportsPointSprites = (sPriv->drm_version.minor >= 13);
-      screen->drmSupportsCubeMapsR100 = (sPriv->drm_version.minor >= 15);
-      screen->drmSupportsVertexProgram = (sPriv->drm_version.minor >= 25);
-   }
-
-   screen->mmio.handle = dri_priv->registerHandle;
-   screen->mmio.size   = dri_priv->registerSize;
-   if ( drmMap( sPriv->fd,
-		screen->mmio.handle,
-		screen->mmio.size,
-		&screen->mmio.map ) ) {
-      FREE( screen );
-      __driUtilMessage("%s: drmMap failed\n", __FUNCTION__ );
-      return NULL;
-   }
-
-   RADEONMMIO = screen->mmio.map;
-
-   screen->status.handle = dri_priv->statusHandle;
-   screen->status.size   = dri_priv->statusSize;
-   if ( drmMap( sPriv->fd,
-		screen->status.handle,
-		screen->status.size,
-		&screen->status.map ) ) {
-      drmUnmap( screen->mmio.map, screen->mmio.size );
-      FREE( screen );
-      __driUtilMessage("%s: drmMap (2) failed\n", __FUNCTION__ );
-      return NULL;
-   }
-   screen->scratch = (__volatile__ uint32_t *)
-      ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET);
-
-   screen->buffers = drmMapBufs( sPriv->fd );
-   if ( !screen->buffers ) {
-      drmUnmap( screen->status.map, screen->status.size );
-      drmUnmap( screen->mmio.map, screen->mmio.size );
-      FREE( screen );
-      __driUtilMessage("%s: drmMapBufs failed\n", __FUNCTION__ );
-      return NULL;
-   }
-
-   if ( dri_priv->gartTexHandle && dri_priv->gartTexMapSize ) {
-      screen->gartTextures.handle = dri_priv->gartTexHandle;
-      screen->gartTextures.size   = dri_priv->gartTexMapSize;
-      if ( drmMap( sPriv->fd,
-		   screen->gartTextures.handle,
-		   screen->gartTextures.size,
-		   (drmAddressPtr)&screen->gartTextures.map ) ) {
-	 drmUnmapBufs( screen->buffers );
-	 drmUnmap( screen->status.map, screen->status.size );
-	 drmUnmap( screen->mmio.map, screen->mmio.size );
-	 FREE( screen );
-	 __driUtilMessage("%s: drmMap failed for GART texture area\n", __FUNCTION__);
-	 return NULL;
-      }
-
-      screen->gart_texture_offset = dri_priv->gartTexOffset + screen->gart_base;
-   }
-
+static int radeon_set_screen_flags(radeonScreenPtr screen, int device_id)
+{
+   screen->device_id = device_id;
    screen->chip_flags = 0;
-   /* XXX: add more chipsets */
-   switch ( dri_priv->deviceID ) {
+   switch ( device_id ) {
    case PCI_CHIP_RADEON_LY:
    case PCI_CHIP_RADEON_LZ:
    case PCI_CHIP_RADEON_QY:
@@ -561,7 +477,6 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
       screen->chip_family = CHIP_FAMILY_RS300;
       break;
 
-
    case PCI_CHIP_R300_AD:
    case PCI_CHIP_R300_AE:
    case PCI_CHIP_R300_AF:
@@ -821,9 +736,145 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
 
    default:
       fprintf(stderr, "unknown chip id 0x%x, can't guess.\n",
-	      dri_priv->deviceID);
+	      device_id);
+      return -1;
+   }
+
+   return 0;
+}
+
+
+/* Create the device specific screen private data struct.
+ */
+static radeonScreenPtr
+radeonCreateScreen( __DRIscreenPrivate *sPriv )
+{
+   radeonScreenPtr screen;
+   RADEONDRIPtr dri_priv = (RADEONDRIPtr)sPriv->pDevPriv;
+   unsigned char *RADEONMMIO = NULL;
+   int i;
+   int ret;
+   uint32_t temp = 0;
+
+   if (sPriv->devPrivSize != sizeof(RADEONDRIRec)) {
+      fprintf(stderr,"\nERROR!  sizeof(RADEONDRIRec) does not match passed size from device driver\n");
+      return GL_FALSE;
+   }
+
+   /* Allocate the private area */
+   screen = (radeonScreenPtr) CALLOC( sizeof(*screen) );
+   if ( !screen ) {
+      __driUtilMessage("%s: Could not allocate memory for screen structure",
+		       __FUNCTION__);
       return NULL;
    }
+
+#if DO_DEBUG && RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
+	RADEON_DEBUG = driParseDebugString(getenv("RADEON_DEBUG"), debug_control);
+#endif
+
+   /* parse information in __driConfigOptions */
+   driParseOptionInfo (&screen->optionCache,
+		       __driConfigOptions, __driNConfigOptions);
+
+   /* This is first since which regions we map depends on whether or
+    * not we are using a PCI card.
+    */
+   screen->card_type = (dri_priv->IsPCI ? RADEON_CARD_PCI : RADEON_CARD_AGP);
+   {
+      int ret;
+
+      ret = radeonGetParam(sPriv, RADEON_PARAM_GART_BUFFER_OFFSET,
+			    &screen->gart_buffer_offset);
+
+      if (ret) {
+	 FREE( screen );
+	 fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BUFFER_OFFSET): %d\n", ret);
+	 return NULL;
+      }
+
+      ret = radeonGetParam(sPriv, RADEON_PARAM_GART_BASE,
+			    &screen->gart_base);
+      if (ret) {
+	 FREE( screen );
+	 fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BASE): %d\n", ret);
+	 return NULL;
+      }
+
+      ret = radeonGetParam(sPriv, RADEON_PARAM_IRQ_NR,
+			    &screen->irq);
+      if (ret) {
+	 FREE( screen );
+	 fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_IRQ_NR): %d\n", ret);
+	 return NULL;
+      }
+      screen->drmSupportsCubeMapsR200 = (sPriv->drm_version.minor >= 7);
+      screen->drmSupportsBlendColor = (sPriv->drm_version.minor >= 11);
+      screen->drmSupportsTriPerf = (sPriv->drm_version.minor >= 16);
+      screen->drmSupportsFragShader = (sPriv->drm_version.minor >= 18);
+      screen->drmSupportsPointSprites = (sPriv->drm_version.minor >= 13);
+      screen->drmSupportsCubeMapsR100 = (sPriv->drm_version.minor >= 15);
+      screen->drmSupportsVertexProgram = (sPriv->drm_version.minor >= 25);
+   }
+
+   screen->mmio.handle = dri_priv->registerHandle;
+   screen->mmio.size   = dri_priv->registerSize;
+   if ( drmMap( sPriv->fd,
+		screen->mmio.handle,
+		screen->mmio.size,
+		&screen->mmio.map ) ) {
+     FREE( screen );
+     __driUtilMessage("%s: drmMap failed\n", __FUNCTION__ );
+     return NULL;
+   }
+
+   RADEONMMIO = screen->mmio.map;
+
+   screen->status.handle = dri_priv->statusHandle;
+   screen->status.size   = dri_priv->statusSize;
+   if ( drmMap( sPriv->fd,
+		screen->status.handle,
+		screen->status.size,
+		&screen->status.map ) ) {
+     drmUnmap( screen->mmio.map, screen->mmio.size );
+     FREE( screen );
+     __driUtilMessage("%s: drmMap (2) failed\n", __FUNCTION__ );
+     return NULL;
+   }
+   screen->scratch = (__volatile__ uint32_t *)
+     ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET);
+
+   screen->buffers = drmMapBufs( sPriv->fd );
+   if ( !screen->buffers ) {
+     drmUnmap( screen->status.map, screen->status.size );
+     drmUnmap( screen->mmio.map, screen->mmio.size );
+     FREE( screen );
+     __driUtilMessage("%s: drmMapBufs failed\n", __FUNCTION__ );
+     return NULL;
+   }
+
+   if ( dri_priv->gartTexHandle && dri_priv->gartTexMapSize ) {
+     screen->gartTextures.handle = dri_priv->gartTexHandle;
+     screen->gartTextures.size   = dri_priv->gartTexMapSize;
+     if ( drmMap( sPriv->fd,
+		  screen->gartTextures.handle,
+		  screen->gartTextures.size,
+		  (drmAddressPtr)&screen->gartTextures.map ) ) {
+       drmUnmapBufs( screen->buffers );
+       drmUnmap( screen->status.map, screen->status.size );
+       drmUnmap( screen->mmio.map, screen->mmio.size );
+       FREE( screen );
+       __driUtilMessage("%s: drmMap failed for GART texture area\n", __FUNCTION__);
+       return NULL;
+    }
+
+     screen->gart_texture_offset = dri_priv->gartTexOffset + screen->gart_base;
+   }
+
+   ret = radeon_set_screen_flags(screen, dri_priv->deviceID);
+   if (ret == -1)
+     return NULL;
+
    if ((screen->chip_family == CHIP_FAMILY_R350 || screen->chip_family == CHIP_FAMILY_R300) &&
        sPriv->ddx_version.minor < 2) {
       fprintf(stderr, "xf86-video-ati-6.6.2 or newer needed for Radeon 9500/9700/9800 cards.\n");
@@ -848,8 +899,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    screen->cpp = dri_priv->bpp / 8;
    screen->AGPMode = dri_priv->AGPMode;
 
-   ret = radeonGetParam( sPriv->fd, RADEON_PARAM_FB_LOCATION,
-                         &temp);
+   ret = radeonGetParam(sPriv, RADEON_PARAM_FB_LOCATION, &temp);
    if (ret) {
        if (screen->chip_family < CHIP_FAMILY_RS600)
 	   screen->fbLocation      = ( INREG( RADEON_MC_FB_LOCATION ) & 0xffff) << 16;
@@ -863,8 +913,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
    }
 
    if (screen->chip_family >= CHIP_FAMILY_R300) {
-       ret = radeonGetParam( sPriv->fd, RADEON_PARAM_NUM_GB_PIPES,
-			     &temp);
+       ret = radeonGetParam(sPriv, RADEON_PARAM_NUM_GB_PIPES, &temp);
        if (ret) {
 	   fprintf(stderr, "Unable to get num_pipes, need newer drm\n");
 	   switch (screen->chip_family) {
@@ -901,7 +950,6 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
        default:
 	   break;
        }
-
    }
 
    if ( sPriv->drm_version.minor >= 10 ) {
@@ -971,7 +1019,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
 
 #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
    if (IS_R200_CLASS(screen))
-       screen->extensions[i++] = &r200AllocateExtension.base;
+      screen->extensions[i++] = &r200AllocateExtension.base;
 
    screen->extensions[i++] = &r200texOffsetExtension.base;
 #endif
@@ -985,6 +1033,153 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
 
    screen->driScreen = sPriv;
    screen->sarea_priv_offset = dri_priv->sarea_priv_offset;
+   screen->sarea = (drm_radeon_sarea_t *) ((GLubyte *) sPriv->pSAREA +
+					       screen->sarea_priv_offset);
+
+   screen->bom = radeon_bo_manager_legacy_ctor(screen);
+   if (screen->bom == NULL) {
+     free(screen);
+     return NULL;
+   }
+
+   return screen;
+}
+
+static radeonScreenPtr
+radeonCreateScreen2(__DRIscreenPrivate *sPriv)
+{
+   radeonScreenPtr screen;
+   int i;
+   int ret;
+   uint32_t device_id = 0;
+   uint32_t temp = 0;
+
+   /* Allocate the private area */
+   screen = (radeonScreenPtr) CALLOC( sizeof(*screen) );
+   if ( !screen ) {
+      __driUtilMessage("%s: Could not allocate memory for screen structure",
+		       __FUNCTION__);
+      fprintf(stderr, "leaving here\n");
+      return NULL;
+   }
+
+#if DO_DEBUG && RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
+	RADEON_DEBUG = driParseDebugString(getenv("RADEON_DEBUG"), debug_control);
+#endif
+
+   /* parse information in __driConfigOptions */
+   driParseOptionInfo (&screen->optionCache,
+		       __driConfigOptions, __driNConfigOptions);
+
+   screen->kernel_mm = 1;
+   screen->chip_flags = 0;
+
+   /* if we have kms we can support all of these */
+   screen->drmSupportsCubeMapsR200 = 1;
+   screen->drmSupportsBlendColor = 1;
+   screen->drmSupportsTriPerf = 1;
+   screen->drmSupportsFragShader = 1;
+   screen->drmSupportsPointSprites = 1;
+   screen->drmSupportsCubeMapsR100 = 1;
+   screen->drmSupportsVertexProgram = 1;
+   screen->irq = 1;
+
+   ret = radeonGetParam(sPriv, RADEON_PARAM_DEVICE_ID, &device_id);
+   if (ret) {
+     FREE( screen );
+     fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_DEVICE_ID): %d\n", ret);
+     return NULL;
+   }
+
+   ret = radeon_set_screen_flags(screen, device_id);
+   if (ret == -1)
+     return NULL;
+
+   if (screen->chip_family >= CHIP_FAMILY_R300) {
+       ret = radeonGetParam(sPriv, RADEON_PARAM_NUM_GB_PIPES, &temp);
+       if (ret) {
+	   fprintf(stderr, "Unable to get num_pipes, need newer drm\n");
+	   switch (screen->chip_family) {
+	   case CHIP_FAMILY_R300:
+	   case CHIP_FAMILY_R350:
+	       screen->num_gb_pipes = 2;
+	       break;
+	   case CHIP_FAMILY_R420:
+	   case CHIP_FAMILY_R520:
+	   case CHIP_FAMILY_R580:
+	   case CHIP_FAMILY_RV560:
+	   case CHIP_FAMILY_RV570:
+	       screen->num_gb_pipes = 4;
+	       break;
+	   case CHIP_FAMILY_RV350:
+	   case CHIP_FAMILY_RV515:
+	   case CHIP_FAMILY_RV530:
+	   case CHIP_FAMILY_RV410:
+	   default:
+	       screen->num_gb_pipes = 1;
+	       break;
+	   }
+       } else {
+	   screen->num_gb_pipes = temp;
+       }
+
+       /* pipe overrides */
+       switch (device_id) {
+       case PCI_CHIP_R300_AD: /* 9500 with 1 quadpipe verified by: Reid Linnemann <[email protected]> */
+       case PCI_CHIP_RV410_5E4C: /* RV410 SE only have 1 quadpipe */
+       case PCI_CHIP_RV410_5E4F: /* RV410 SE only have 1 quadpipe */
+	   screen->num_gb_pipes = 1;
+	   break;
+       default:
+	   break;
+       }
+
+   }
+
+   if (screen->chip_family <= CHIP_FAMILY_RS200)
+      screen->chip_flags |= RADEON_CLASS_R100;
+   else if (screen->chip_family <= CHIP_FAMILY_RV280)
+      screen->chip_flags |= RADEON_CLASS_R200;
+   else
+      screen->chip_flags |= RADEON_CLASS_R300;
+
+   if (getenv("R300_NO_TCL"))
+     screen->chip_flags &= ~RADEON_CHIPSET_TCL;
+
+   i = 0;
+   screen->extensions[i++] = &driCopySubBufferExtension.base;
+   screen->extensions[i++] = &driFrameTrackingExtension.base;
+   screen->extensions[i++] = &driReadDrawableExtension;
+
+   if ( screen->irq != 0 ) {
+       screen->extensions[i++] = &driSwapControlExtension.base;
+       screen->extensions[i++] = &driMediaStreamCounterExtension.base;
+   }
+
+#if !RADEON_COMMON
+   screen->extensions[i++] = &radeonTexBufferExtension.base;
+#endif
+
+#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
+   if (IS_R200_CLASS(screen))
+       screen->extensions[i++] = &r200AllocateExtension.base;
+
+   screen->extensions[i++] = &r200TexBufferExtension.base;
+#endif
+
+#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
+   screen->extensions[i++] = &r300TexBufferExtension.base;
+#endif
+
+   screen->extensions[i++] = NULL;
+   sPriv->extensions = screen->extensions;
+
+   screen->driScreen = sPriv;
+   screen->bom = radeon_bo_manager_gem_ctor(sPriv->fd);
+   if (screen->bom == NULL) {
+       free(screen);
+       return NULL;
+   }
    return screen;
 }
 
@@ -993,23 +1188,32 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv )
 static void
 radeonDestroyScreen( __DRIscreenPrivate *sPriv )
 {
-   radeonScreenPtr screen = (radeonScreenPtr)sPriv->private;
+    radeonScreenPtr screen = (radeonScreenPtr)sPriv->private;
 
-   if (!screen)
-      return;
+    if (!screen)
+        return;
 
-   if ( screen->gartTextures.map ) {
-      drmUnmap( screen->gartTextures.map, screen->gartTextures.size );
-   }
-   drmUnmapBufs( screen->buffers );
-   drmUnmap( screen->status.map, screen->status.size );
-   drmUnmap( screen->mmio.map, screen->mmio.size );
+    if (screen->kernel_mm) {
+#ifdef RADEON_BO_TRACK
+        radeon_tracker_print(&screen->bom->tracker, stderr);
+#endif
+        radeon_bo_manager_gem_dtor(screen->bom);
+    } else {
+        radeon_bo_manager_legacy_dtor(screen->bom);
+
+        if ( screen->gartTextures.map ) {
+            drmUnmap( screen->gartTextures.map, screen->gartTextures.size );
+        }
+        drmUnmapBufs( screen->buffers );
+        drmUnmap( screen->status.map, screen->status.size );
+        drmUnmap( screen->mmio.map, screen->mmio.size );
+    }
 
-   /* free all option information */
-   driDestroyOptionInfo (&screen->optionCache);
+    /* free all option information */
+    driDestroyOptionInfo (&screen->optionCache);
 
-   FREE( screen );
-   sPriv->private = NULL;
+    FREE( screen );
+    sPriv->private = NULL;
 }
 
 
@@ -1018,16 +1222,21 @@ radeonDestroyScreen( __DRIscreenPrivate *sPriv )
 static GLboolean
 radeonInitDriver( __DRIscreenPrivate *sPriv )
 {
-   sPriv->private = (void *) radeonCreateScreen( sPriv );
-   if ( !sPriv->private ) {
-      radeonDestroyScreen( sPriv );
-      return GL_FALSE;
-   }
+    if (sPriv->dri2.enabled) {
+        sPriv->private = (void *) radeonCreateScreen2( sPriv );
+    } else {
+        sPriv->private = (void *) radeonCreateScreen( sPriv );
+    }
+    if ( !sPriv->private ) {
+        radeonDestroyScreen( sPriv );
+        return GL_FALSE;
+    }
 
-   return GL_TRUE;
+    return GL_TRUE;
 }
 
 
+
 /**
  * Create the Mesa framebuffer and renderbuffers for a given window/drawable.
  *
@@ -1040,101 +1249,111 @@ radeonCreateBuffer( __DRIscreenPrivate *driScrnPriv,
                     const __GLcontextModes *mesaVis,
                     GLboolean isPixmap )
 {
-   radeonScreenPtr screen = (radeonScreenPtr) driScrnPriv->private;
+    radeonScreenPtr screen = (radeonScreenPtr) driScrnPriv->private;
+
+    const GLboolean swDepth = GL_FALSE;
+    const GLboolean swAlpha = GL_FALSE;
+    const GLboolean swAccum = mesaVis->accumRedBits > 0;
+    const GLboolean swStencil = mesaVis->stencilBits > 0 &&
+	mesaVis->depthBits != 24;
+    GLenum rgbFormat;
+    struct radeon_framebuffer *rfb;
 
-   if (isPixmap) {
+    if (isPixmap)
       return GL_FALSE; /* not implemented */
-   }
-   else {
-      const GLboolean swDepth = GL_FALSE;
-      const GLboolean swAlpha = GL_FALSE;
-      const GLboolean swAccum = mesaVis->accumRedBits > 0;
-      const GLboolean swStencil = mesaVis->stencilBits > 0 &&
-         mesaVis->depthBits != 24;
-      struct gl_framebuffer *fb = _mesa_create_framebuffer(mesaVis);
-
-      /* front color renderbuffer */
-      {
-         driRenderbuffer *frontRb
-            = driNewRenderbuffer(GL_RGBA,
-                                 driScrnPriv->pFB + screen->frontOffset,
-                                 screen->cpp,
-                                 screen->frontOffset, screen->frontPitch,
-                                 driDrawPriv);
-         radeonSetSpanFunctions(frontRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_FRONT_LEFT, &frontRb->Base);
-      }
 
-      /* back color renderbuffer */
-      if (mesaVis->doubleBufferMode) {
-         driRenderbuffer *backRb
-            = driNewRenderbuffer(GL_RGBA,
-                                 driScrnPriv->pFB + screen->backOffset,
-                                 screen->cpp,
-                                 screen->backOffset, screen->backPitch,
-                                 driDrawPriv);
-         radeonSetSpanFunctions(backRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_BACK_LEFT, &backRb->Base);
-      }
+    rfb = CALLOC_STRUCT(radeon_framebuffer);
+    if (!rfb)
+      return GL_FALSE;
 
-      /* depth renderbuffer */
-      if (mesaVis->depthBits == 16) {
-         driRenderbuffer *depthRb
-            = driNewRenderbuffer(GL_DEPTH_COMPONENT16,
-                                 driScrnPriv->pFB + screen->depthOffset,
-                                 screen->cpp,
-                                 screen->depthOffset, screen->depthPitch,
-                                 driDrawPriv);
-         radeonSetSpanFunctions(depthRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
-	 depthRb->depthHasSurface = screen->depthHasSurface;
-      }
-      else if (mesaVis->depthBits == 24) {
-         driRenderbuffer *depthRb
-            = driNewRenderbuffer(GL_DEPTH_COMPONENT24,
-                                 driScrnPriv->pFB + screen->depthOffset,
-                                 screen->cpp,
-                                 screen->depthOffset, screen->depthPitch,
-                                 driDrawPriv);
-         radeonSetSpanFunctions(depthRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
-	 depthRb->depthHasSurface = screen->depthHasSurface;
-      }
+    _mesa_initialize_framebuffer(&rfb->base, mesaVis);
+
+    if (mesaVis->redBits == 5)
+        rgbFormat = GL_RGB5;
+    else if (mesaVis->alphaBits == 0)
+        rgbFormat = GL_RGB8;
+    else
+        rgbFormat = GL_RGBA8;
+
+    /* front color renderbuffer */
+    rfb->color_rb[0] = radeon_create_renderbuffer(rgbFormat, driDrawPriv);
+    _mesa_add_renderbuffer(&rfb->base, BUFFER_FRONT_LEFT, &rfb->color_rb[0]->base);
+    rfb->color_rb[0]->has_surface = 1;
+
+    /* back color renderbuffer */
+    if (mesaVis->doubleBufferMode) {
+      rfb->color_rb[1] = radeon_create_renderbuffer(rgbFormat, driDrawPriv);
+	_mesa_add_renderbuffer(&rfb->base, BUFFER_BACK_LEFT, &rfb->color_rb[1]->base);
+	rfb->color_rb[1]->has_surface = 1;
+    }
 
-      /* stencil renderbuffer */
-      if (mesaVis->stencilBits > 0 && !swStencil) {
-         driRenderbuffer *stencilRb
-            = driNewRenderbuffer(GL_STENCIL_INDEX8_EXT,
-                                 driScrnPriv->pFB + screen->depthOffset,
-                                 screen->cpp,
-                                 screen->depthOffset, screen->depthPitch,
-                                 driDrawPriv);
-         radeonSetSpanFunctions(stencilRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &stencilRb->Base);
-	 stencilRb->depthHasSurface = screen->depthHasSurface;
+    if (mesaVis->depthBits == 24) {
+      if (mesaVis->stencilBits == 8) {
+	struct radeon_renderbuffer *depthStencilRb = radeon_create_renderbuffer(GL_DEPTH24_STENCIL8_EXT, driDrawPriv);
+	_mesa_add_renderbuffer(&rfb->base, BUFFER_DEPTH, &depthStencilRb->base);
+	_mesa_add_renderbuffer(&rfb->base, BUFFER_STENCIL, &depthStencilRb->base);
+	depthStencilRb->has_surface = screen->depthHasSurface;
+      } else {
+	/* depth renderbuffer */
+	struct radeon_renderbuffer *depth = radeon_create_renderbuffer(GL_DEPTH_COMPONENT24, driDrawPriv);
+	_mesa_add_renderbuffer(&rfb->base, BUFFER_DEPTH, &depth->base);
+	depth->has_surface = screen->depthHasSurface;
       }
+    } else if (mesaVis->depthBits == 16) {
+      /* just 16-bit depth buffer, no hw stencil */
+	struct radeon_renderbuffer *depth = radeon_create_renderbuffer(GL_DEPTH_COMPONENT16, driDrawPriv);
+	_mesa_add_renderbuffer(&rfb->base, BUFFER_DEPTH, &depth->base);
+	depth->has_surface = screen->depthHasSurface;
+    }
 
-      _mesa_add_soft_renderbuffers(fb,
-                                   GL_FALSE, /* color */
-                                   swDepth,
-                                   swStencil,
-                                   swAccum,
-                                   swAlpha,
-                                   GL_FALSE /* aux */);
-      driDrawPriv->driverPrivate = (void *) fb;
+    _mesa_add_soft_renderbuffers(&rfb->base,
+	    GL_FALSE, /* color */
+	    swDepth,
+	    swStencil,
+	    swAccum,
+	    swAlpha,
+	    GL_FALSE /* aux */);
+    driDrawPriv->driverPrivate = (void *) rfb;
 
-      return (driDrawPriv->driverPrivate != NULL);
-   }
+    return (driDrawPriv->driverPrivate != NULL);
 }
 
 
-static void
+static void radeon_cleanup_renderbuffers(struct radeon_framebuffer *rfb)
+{
+	struct radeon_renderbuffer *rb;
+
+	rb = rfb->color_rb[0];
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+		rb->bo = NULL;
+	}
+	rb = rfb->color_rb[1];
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+		rb->bo = NULL;
+	}
+	rb = radeon_get_renderbuffer(&rfb->base, BUFFER_DEPTH);
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+		rb->bo = NULL;
+	}
+}
+
+void
 radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
 {
-   _mesa_reference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)), NULL);
+    struct radeon_framebuffer *rfb;
+    if (!driDrawPriv)
+	return;
+
+    rfb = (void*)driDrawPriv->driverPrivate;
+    if (!rfb)
+	return;
+    radeon_cleanup_renderbuffers(rfb);
+    _mesa_reference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)), NULL);
 }
 
-#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
 /**
  * Choose the appropriate CreateContext function based on the chipset.
  * Eventually, all drivers will go through this process.
@@ -1145,25 +1364,22 @@ static GLboolean radeonCreateContext(const __GLcontextModes * glVisual,
 {
 	__DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
 	radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private);
-
+#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
 	if (IS_R300_CLASS(screen))
 		return r300CreateContext(glVisual, driContextPriv, sharedContextPriv);
-        return GL_FALSE;
-}
-
-/**
- * Choose the appropriate DestroyContext function based on the chipset.
- */
-static void radeonDestroyContext(__DRIcontextPrivate * driContextPriv)
-{
-	radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate;
-
-	if (IS_R300_CLASS(radeon->radeonScreen))
-		return r300DestroyContext(driContextPriv);
-}
+#endif
 
+#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
+	if (IS_R200_CLASS(screen))
+		return r200CreateContext(glVisual, driContextPriv, sharedContextPriv);
+#endif
 
+#if !RADEON_COMMON
+	(void)screen;
+	return r100CreateContext(glVisual, driContextPriv, sharedContextPriv);
 #endif
+	return GL_FALSE;
+}
 
 
 /**
@@ -1225,13 +1441,103 @@ radeonInitScreen(__DRIscreenPrivate *psp)
    if (!radeonInitDriver(psp))
        return NULL;
 
+   /* for now fill in all modes */
    return radeonFillInModes( psp,
 			     dri_priv->bpp,
 			     (dri_priv->bpp == 16) ? 16 : 24,
-			     (dri_priv->bpp == 16) ? 0  : 8,
-			     (dri_priv->backOffset != dri_priv->depthOffset) );
+			     (dri_priv->bpp == 16) ? 0  : 8, 1);
 }
+#define ARRAY_SIZE(a) (sizeof (a) / sizeof ((a)[0]))
+
+/**
+ * This is the driver specific part of the createNewScreen entry point.
+ * Called when using DRI2.
+ *
+ * \return the __GLcontextModes supported by this driver
+ */
+static const
+__DRIconfig **radeonInitScreen2(__DRIscreenPrivate *psp)
+{
+   GLenum fb_format[3];
+   GLenum fb_type[3];
+   /* GLX_SWAP_COPY_OML is only supported because the Intel driver doesn't
+    * support pageflipping at all.
+    */
+   static const GLenum back_buffer_modes[] = {
+     GLX_NONE, GLX_SWAP_UNDEFINED_OML, /*, GLX_SWAP_COPY_OML*/
+   };
+   uint8_t depth_bits[4], stencil_bits[4], msaa_samples_array[1];
+   int color;
+   __DRIconfig **configs = NULL;
+
+   /* Calling driInitExtensions here, with a NULL context pointer,
+    * does not actually enable the extensions.  It just makes sure
+    * that all the dispatch offsets for all the extensions that
+    * *might* be enables are known.  This is needed because the
+    * dispatch offsets need to be known when _mesa_context_create
+    * is called, but we can't enable the extensions until we have a
+    * context pointer.
+    *
+    * Hello chicken.  Hello egg.  How are you two today?
+    */
+   driInitExtensions( NULL, card_extensions, GL_FALSE );
+   driInitExtensions( NULL, mm_extensions, GL_FALSE );
+#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
+   driInitExtensions( NULL, blend_extensions, GL_FALSE );
+   driInitSingleExtension( NULL, ARB_vp_extension );
+   driInitSingleExtension( NULL, NV_vp_extension );
+   driInitSingleExtension( NULL, ATI_fs_extension );
+   driInitExtensions( NULL, point_extensions, GL_FALSE );
+#endif
+
+   if (!radeonInitDriver(psp)) {
+       return NULL;
+    }
+   depth_bits[0] = 0;
+   stencil_bits[0] = 0;
+   depth_bits[1] = 16;
+   stencil_bits[1] = 0;
+   depth_bits[2] = 24;
+   stencil_bits[2] = 0;
+   depth_bits[3] = 24;
+   stencil_bits[3] = 8;
+
+   msaa_samples_array[0] = 0;
+
+   fb_format[0] = GL_RGB;
+   fb_type[0] = GL_UNSIGNED_SHORT_5_6_5;
+
+   fb_format[1] = GL_BGR;
+   fb_type[1] = GL_UNSIGNED_INT_8_8_8_8_REV;
+
+   fb_format[2] = GL_BGRA;
+   fb_type[2] = GL_UNSIGNED_INT_8_8_8_8_REV;
+
+   for (color = 0; color < ARRAY_SIZE(fb_format); color++) {
+      __DRIconfig **new_configs;
+
+      new_configs = driCreateConfigs(fb_format[color], fb_type[color],
+				     depth_bits,
+				     stencil_bits,
+				     ARRAY_SIZE(depth_bits),
+				     back_buffer_modes,
+				     ARRAY_SIZE(back_buffer_modes),
+				     msaa_samples_array,
+				     ARRAY_SIZE(msaa_samples_array));
+      if (configs == NULL)
+	 configs = new_configs;
+      else
+	 configs = driConcatConfigs(configs, new_configs);
+   }
 
+   if (configs == NULL) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
+   }
+
+   return (const __DRIconfig **)configs;
+}
 
 /**
  * Get information about previous buffer swaps.
@@ -1239,31 +1545,26 @@ radeonInitScreen(__DRIscreenPrivate *psp)
 static int
 getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
 {
-#if !RADEON_COMMON || (RADEON_COMMON && defined(RADEON_COMMON_FOR_R300))
-   radeonContextPtr  rmesa;
-#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R200)
-   r200ContextPtr  rmesa;
-#endif
+    struct radeon_framebuffer *rfb;
 
-   if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL)
-	|| (dPriv->driContextPriv->driverPrivate == NULL)
-	|| (sInfo == NULL) ) {
-      return -1;
+    if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL)
+	 || (dPriv->driContextPriv->driverPrivate == NULL)
+	 || (sInfo == NULL) ) {
+	return -1;
    }
 
-   rmesa = dPriv->driContextPriv->driverPrivate;
-   sInfo->swap_count = rmesa->swap_count;
-   sInfo->swap_ust = rmesa->swap_ust;
-   sInfo->swap_missed_count = rmesa->swap_missed_count;
+    rfb = dPriv->driverPrivate;
+    sInfo->swap_count = rfb->swap_count;
+    sInfo->swap_ust = rfb->swap_ust;
+    sInfo->swap_missed_count = rfb->swap_missed_count;
 
    sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0)
-       ? driCalculateSwapUsage( dPriv, 0, rmesa->swap_missed_ust )
+       ? driCalculateSwapUsage( dPriv, 0, rfb->swap_missed_ust )
        : 0.0;
 
    return 0;
 }
 
-#if !RADEON_COMMON || (RADEON_COMMON && defined(RADEON_COMMON_FOR_R300))
 const struct __DriverAPIRec driDriverAPI = {
    .InitScreen      = radeonInitScreen,
    .DestroyScreen   = radeonDestroyScreen,
@@ -1280,23 +1581,7 @@ const struct __DriverAPIRec driDriverAPI = {
    .WaitForSBC      = NULL,
    .SwapBuffersMSC  = NULL,
    .CopySubBuffer   = radeonCopySubBuffer,
+    /* DRI2 */
+   .InitScreen2     = radeonInitScreen2,
 };
-#else
-const struct __DriverAPIRec driDriverAPI = {
-   .InitScreen      = radeonInitScreen,
-   .DestroyScreen   = radeonDestroyScreen,
-   .CreateContext   = r200CreateContext,
-   .DestroyContext  = r200DestroyContext,
-   .CreateBuffer    = radeonCreateBuffer,
-   .DestroyBuffer   = radeonDestroyBuffer,
-   .SwapBuffers     = r200SwapBuffers,
-   .MakeCurrent     = r200MakeCurrent,
-   .UnbindContext   = r200UnbindContext,
-   .GetSwapInfo     = getSwapInfo,
-   .GetDrawableMSC  = driDrawableGetMSC32,
-   .WaitForMSC      = driWaitForMSC32,
-   .WaitForSBC      = NULL,
-   .SwapBuffersMSC  = NULL,
-   .CopySubBuffer   = r200CopySubBuffer,
-};
-#endif
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.h b/src/mesa/drivers/dri/radeon/radeon_screen.h
index b84c70bfae9..fe5c7d875a0 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.h
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.h
@@ -54,11 +54,12 @@ typedef struct {
    drmAddress map;			/* Mapping of the DRM region */
 } radeonRegionRec, *radeonRegionPtr;
 
-typedef struct {
+typedef struct radeon_screen {
    int chip_family;
    int chip_flags;
    int cpp;
    int card_type;
+   int device_id; /* PCI ID */
    int AGPMode;
    unsigned int irq;			/* IRQ number (0 means none) */
 
@@ -103,9 +104,12 @@ typedef struct {
    /* Configuration cache with default values for all contexts */
    driOptionCache optionCache;
 
-   const __DRIextension *extensions[8];
+   const __DRIextension *extensions[16];
 
    int num_gb_pipes;
+   int kernel_mm;
+   drm_radeon_sarea_t *sarea;	/* Private SAREA data */
+   struct radeon_bo_manager *bom;
 } radeonScreenRec, *radeonScreenPtr;
 
 #define IS_R100_CLASS(screen) \
@@ -115,4 +119,5 @@ typedef struct {
 #define IS_R300_CLASS(screen) \
 	((screen->chip_flags & RADEON_CLASS_MASK) == RADEON_CLASS_R300)
 
+extern void radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv);
 #endif /* __RADEON_SCREEN_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_span.c b/src/mesa/drivers/dri/radeon/radeon_span.c
index 12051ff1c81..e28f28662b4 100644
--- a/src/mesa/drivers/dri/radeon/radeon_span.c
+++ b/src/mesa/drivers/dri/radeon/radeon_span.c
@@ -43,46 +43,203 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/glheader.h"
 #include "swrast/swrast.h"
 
-#include "radeon_context.h"
-#include "radeon_ioctl.h"
-#include "radeon_state.h"
+#include "radeon_common.h"
+#include "radeon_lock.h"
 #include "radeon_span.h"
-#include "radeon_tex.h"
-
-#include "drirenderbuffer.h"
 
 #define DBG 0
 
+static void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb);
+
+static GLubyte *radeon_ptr32(const struct radeon_renderbuffer * rrb,
+			     GLint x, GLint y)
+{
+    GLubyte *ptr = rrb->bo->ptr;
+    uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE;
+    GLint offset;
+    GLint nmacroblkpl;
+    GLint nmicroblkpl;
+
+    if (rrb->has_surface || !(rrb->bo->flags & mask)) {
+        offset = x * rrb->cpp + y * rrb->pitch;
+    } else {
+        offset = 0;
+        if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) {
+            if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) {
+                nmacroblkpl = rrb->pitch >> 5;
+                offset += ((y >> 4) * nmacroblkpl) << 11;
+                offset += ((y & 15) >> 1) << 8;
+                offset += (y & 1) << 4;
+                offset += (x >> 5) << 11;
+                offset += ((x & 31) >> 2) << 5;
+                offset += (x & 3) << 2;
+            } else {
+                nmacroblkpl = rrb->pitch >> 6;
+                offset += ((y >> 3) * nmacroblkpl) << 11;
+                offset += (y & 7) << 8;
+                offset += (x >> 6) << 11;
+                offset += ((x & 63) >> 3) << 5;
+                offset += (x & 7) << 2;
+            }
+        } else {
+            nmicroblkpl = ((rrb->pitch + 31) & ~31) >> 5;
+            offset += (y * nmicroblkpl) << 5;
+            offset += (x >> 3) << 5;
+            offset += (x & 7) << 2;
+        }
+    }
+    return &ptr[offset];
+}
+
+static GLubyte *radeon_ptr16(const struct radeon_renderbuffer * rrb,
+			     GLint x, GLint y)
+{
+    GLubyte *ptr = rrb->bo->ptr;
+    uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE;
+    GLint offset;
+    GLint nmacroblkpl;
+    GLint nmicroblkpl;
+
+    if (rrb->has_surface || !(rrb->bo->flags & mask)) {
+        offset = x * rrb->cpp + y * rrb->pitch;
+    } else {
+        offset = 0;
+        if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) {
+            if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) {
+                nmacroblkpl = rrb->pitch >> 6;
+                offset += ((y >> 4) * nmacroblkpl) << 11;
+                offset += ((y & 15) >> 1) << 8;
+                offset += (y & 1) << 4;
+                offset += (x >> 6) << 11;
+                offset += ((x & 63) >> 3) << 5;
+                offset += (x & 7) << 1;
+            } else {
+                nmacroblkpl = rrb->pitch >> 7;
+                offset += ((y >> 3) * nmacroblkpl) << 11;
+                offset += (y & 7) << 8;
+                offset += (x >> 7) << 11;
+                offset += ((x & 127) >> 4) << 5;
+                offset += (x & 15) << 2;
+            }
+        } else {
+            nmicroblkpl = ((rrb->pitch + 31) & ~31) >> 5;
+            offset += (y * nmicroblkpl) << 5;
+            offset += (x >> 4) << 5;
+            offset += (x & 15) << 2;
+        }
+    }
+    return &ptr[offset];
+}
+
+static GLubyte *radeon_ptr(const struct radeon_renderbuffer * rrb,
+			   GLint x, GLint y)
+{
+    GLubyte *ptr = rrb->bo->ptr;
+    uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE;
+    GLint offset;
+    GLint microblkxs;
+    GLint macroblkxs;
+    GLint nmacroblkpl;
+    GLint nmicroblkpl;
+
+    if (rrb->has_surface || !(rrb->bo->flags & mask)) {
+        offset = x * rrb->cpp + y * rrb->pitch;
+    } else {
+        offset = 0;
+        if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) {
+            if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) {
+                microblkxs = 16 / rrb->cpp;
+                macroblkxs = 128 / rrb->cpp;
+                nmacroblkpl = rrb->pitch / macroblkxs;
+                offset += ((y >> 4) * nmacroblkpl) << 11;
+                offset += ((y & 15) >> 1) << 8;
+                offset += (y & 1) << 4;
+                offset += (x / macroblkxs) << 11;
+                offset += ((x & (macroblkxs - 1)) / microblkxs) << 5;
+                offset += (x & (microblkxs - 1)) * rrb->cpp;
+            } else {
+                microblkxs = 32 / rrb->cpp;
+                macroblkxs = 256 / rrb->cpp;
+                nmacroblkpl = rrb->pitch / macroblkxs;
+                offset += ((y >> 3) * nmacroblkpl) << 11;
+                offset += (y & 7) << 8;
+                offset += (x / macroblkxs) << 11;
+                offset += ((x & (macroblkxs - 1)) / microblkxs) << 5;
+                offset += (x & (microblkxs - 1)) * rrb->cpp;
+            }
+        } else {
+            microblkxs = 32 / rrb->cpp;
+            nmicroblkpl = ((rrb->pitch + 31) & ~31) >> 5;
+            offset += (y * nmicroblkpl) << 5;
+            offset += (x / microblkxs) << 5;
+            offset += (x & (microblkxs - 1)) * rrb->cpp;
+        }
+    }
+    return &ptr[offset];
+}
+
+#ifndef COMPILE_R300
+static uint32_t
+z24s8_to_s8z24(uint32_t val)
+{
+   return (val << 24) | (val >> 8);
+}
+
+static uint32_t
+s8z24_to_z24s8(uint32_t val)
+{
+   return (val >> 24) | (val << 8);
+}
+#endif
+
 /*
  * Note that all information needed to access pixels in a renderbuffer
  * should be obtained through the gl_renderbuffer parameter, not per-context
  * information.
  */
 #define LOCAL_VARS						\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   const __DRIdrawablePrivate *dPriv = drb->dPriv;		\
-   const GLuint bottom = dPriv->h - 1;				\
-   GLubyte *buf = (GLubyte *) drb->flippedData			\
-      + (dPriv->y * drb->flippedPitch + dPriv->x) * drb->cpp;	\
-   GLuint p;							\
-   (void) p;
+   struct radeon_context *radeon = RADEON_CONTEXT(ctx);			\
+   struct radeon_renderbuffer *rrb = (void *) rb;		\
+   const GLint yScale = ctx->DrawBuffer->Name ? 1 : -1;			\
+   const GLint yBias = ctx->DrawBuffer->Name ? 0 : rrb->base.Height - 1;\
+   unsigned int num_cliprects;						\
+   struct drm_clip_rect *cliprects;					\
+   int x_off, y_off;							\
+   GLuint p;						\
+   (void)p;						\
+   radeon_get_cliprects(radeon, &cliprects, &num_cliprects, &x_off, &y_off);
 
 #define LOCAL_DEPTH_VARS				\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;	\
-   const __DRIdrawablePrivate *dPriv = drb->dPriv;	\
-   const GLuint bottom = dPriv->h - 1;			\
-   GLuint xo = dPriv->x;				\
-   GLuint yo = dPriv->y;				\
-   GLubyte *buf = (GLubyte *) drb->Base.Data;
+   struct radeon_context *radeon = RADEON_CONTEXT(ctx);			\
+   struct radeon_renderbuffer *rrb = (void *) rb;	\
+   const GLint yScale = ctx->DrawBuffer->Name ? 1 : -1;			\
+   const GLint yBias = ctx->DrawBuffer->Name ? 0 : rrb->base.Height - 1;\
+   unsigned int num_cliprects;						\
+   struct drm_clip_rect *cliprects;					\
+   int x_off, y_off;							\
+  radeon_get_cliprects(radeon, &cliprects, &num_cliprects, &x_off, &y_off);
 
 #define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
 
-#define Y_FLIP(Y) (bottom - (Y))
+#define Y_FLIP(_y) ((_y) * yScale + yBias)
 
 #define HW_LOCK()
 
 #define HW_UNLOCK()
 
+/* XXX FBO: this is identical to the macro in spantmp2.h except we get
+ * the cliprect info from the context, not the driDrawable.
+ * Move this into spantmp2.h someday.
+ */
+#define HW_CLIPLOOP()							\
+   do {									\
+      int _nc = num_cliprects;						\
+      while ( _nc-- ) {							\
+	 int minx = cliprects[_nc].x1 - x_off;				\
+	 int miny = cliprects[_nc].y1 - y_off;				\
+	 int maxx = cliprects[_nc].x2 - x_off;				\
+	 int maxy = cliprects[_nc].y2 - y_off;
+	
 /* ================================================================
  * Color buffer
  */
@@ -94,7 +251,21 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #define TAG(x)    radeon##x##_RGB565
 #define TAG2(x,y) radeon##x##_RGB565##y
-#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 2)
+#define GET_PTR(X,Y) radeon_ptr16(rrb, (X) + x_off, (Y) + y_off)
+#include "spantmp2.h"
+
+/* 32 bit, xRGB8888 color spanline and pixel functions
+ */
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
+
+#define TAG(x)    radeon##x##_xRGB8888
+#define TAG2(x,y) radeon##x##_xRGB8888##y
+#define GET_VALUE(_x, _y) ((*(GLuint*)(radeon_ptr32(rrb, _x + x_off, _y + y_off)) | 0xff000000))
+#define PUT_VALUE(_x, _y, d) { \
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );		\
+   *_ptr = d;								\
+} while (0)
 #include "spantmp2.h"
 
 /* 32 bit, ARGB8888 color spanline and pixel functions
@@ -104,7 +275,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #define TAG(x)    radeon##x##_ARGB8888
 #define TAG2(x,y) radeon##x##_ARGB8888##y
-#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 4)
+#define GET_PTR(X,Y) radeon_ptr32(rrb, (X) + x_off, (Y) + y_off)
 #include "spantmp2.h"
 
 /* ================================================================
@@ -121,70 +292,20 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * too...
  */
 
-static GLuint radeon_mba_z32(const driRenderbuffer * drb, GLint x, GLint y)
-{
-	GLuint pitch = drb->pitch;
-	if (drb->depthHasSurface) {
-		return 4 * (x + y * pitch);
-	} else {
-		GLuint ba, address = 0;	/* a[0..1] = 0           */
-
-#ifdef COMPILE_R300
-		ba = (y / 8) * (pitch / 8) + (x / 8);
-#else
-		ba = (y / 16) * (pitch / 16) + (x / 16);
-#endif
-
-		address |= (x & 0x7) << 2;	/* a[2..4] = x[0..2]     */
-		address |= (y & 0x3) << 5;	/* a[5..6] = y[0..1]     */
-		address |= (((x & 0x10) >> 2) ^ (y & 0x4)) << 5;	/* a[7]    = x[4] ^ y[2] */
-		address |= (ba & 0x3) << 8;	/* a[8..9] = ba[0..1]    */
-
-		address |= (y & 0x8) << 7;	/* a[10]   = y[3]        */
-		address |= (((x & 0x8) << 1) ^ (y & 0x10)) << 7;	/* a[11]   = x[3] ^ y[4] */
-		address |= (ba & ~0x3) << 10;	/* a[12..] = ba[2..]     */
-
-		return address;
-	}
-}
-
-static INLINE GLuint
-radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y)
-{
-	GLuint pitch = drb->pitch;
-	if (drb->depthHasSurface) {
-		return 2 * (x + y * pitch);
-	} else {
-		GLuint ba, address = 0;	/* a[0]    = 0           */
-
-		ba = (y / 16) * (pitch / 32) + (x / 32);
-
-		address |= (x & 0x7) << 1;	/* a[1..3] = x[0..2]     */
-		address |= (y & 0x7) << 4;	/* a[4..6] = y[0..2]     */
-		address |= (x & 0x8) << 4;	/* a[7]    = x[3]        */
-		address |= (ba & 0x3) << 8;	/* a[8..9] = ba[0..1]    */
-		address |= (y & 0x8) << 7;	/* a[10]   = y[3]        */
-		address |= ((x & 0x10) ^ (y & 0x10)) << 7;	/* a[11]   = x[4] ^ y[4] */
-		address |= (ba & ~0x3) << 10;	/* a[12..] = ba[2..]     */
-
-		return address;
-	}
-}
-
 /* 16-bit depth buffer functions
  */
 #define VALUE_TYPE GLushort
 
 #define WRITE_DEPTH( _x, _y, d )					\
-   *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )) = d;
+   *(GLushort *)radeon_ptr(rrb, _x + x_off, _y + y_off) = d
 
 #define READ_DEPTH( d, _x, _y )						\
-   d = *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo ));
+   d = *(GLushort *)radeon_ptr(rrb, _x + x_off, _y + y_off)
 
 #define TAG(x) radeon##x##_z16
 #include "depthtmp.h"
 
-/* 24 bit depth, 8 bit stencil depthbuffer functions
+/* 24 bit depth
  *
  * Careful: It looks like the R300 uses ZZZS byte order while the R200
  * uses SZZZ for 24 bit depth, 8 bit stencil mode.
@@ -194,35 +315,76 @@ radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y)
 #ifdef COMPILE_R300
 #define WRITE_DEPTH( _x, _y, d )					\
 do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );		\
+   GLuint tmp = *_ptr;				\
    tmp &= 0x000000ff;							\
    tmp |= ((d << 8) & 0xffffff00);					\
-   *(GLuint *)(buf + offset) = tmp;					\
+   *_ptr = tmp;					\
 } while (0)
 #else
 #define WRITE_DEPTH( _x, _y, d )					\
 do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );	\
+   GLuint tmp = *_ptr;							\
    tmp &= 0xff000000;							\
    tmp |= ((d) & 0x00ffffff);						\
-   *(GLuint *)(buf + offset) = tmp;					\
+   *_ptr = tmp;					\
 } while (0)
 #endif
 
 #ifdef COMPILE_R300
 #define READ_DEPTH( d, _x, _y )						\
-  do { \
-    d = (*(GLuint *)(buf + radeon_mba_z32( drb, _x + xo,		\
-					 _y + yo )) & 0xffffff00) >> 8; \
+  do {									\
+    d = (*(GLuint*)(radeon_ptr32(rrb, _x + x_off, _y + y_off)) & 0xffffff00) >> 8; \
   }while(0)
 #else
-#define READ_DEPTH( d, _x, _y )						\
-   d = *(GLuint *)(buf + radeon_mba_z32( drb, _x + xo,			\
-					 _y + yo )) & 0x00ffffff;
+#define READ_DEPTH( d, _x, _y )	\
+  d = *(GLuint*)(radeon_ptr32(rrb, _x + x_off,	_y + y_off)) & 0x00ffffff;
+#endif
+/*
+    fprintf(stderr, "dval(%d, %d, %d, %d)=0x%08X\n", _x, xo, _y, yo, d);\
+   d = *(GLuint*)(radeon_ptr(rrb, _x,	_y )) & 0x00ffffff;
+*/
+#define TAG(x) radeon##x##_z24
+#include "depthtmp.h"
+
+/* 24 bit depth, 8 bit stencil depthbuffer functions
+ * EXT_depth_stencil
+ *
+ * Careful: It looks like the R300 uses ZZZS byte order while the R200
+ * uses SZZZ for 24 bit depth, 8 bit stencil mode.
+ */
+#define VALUE_TYPE GLuint
+
+#ifdef COMPILE_R300
+#define WRITE_DEPTH( _x, _y, d )					\
+do {									\
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );		\
+   *_ptr = d;								\
+} while (0)
+#else
+#define WRITE_DEPTH( _x, _y, d )					\
+do {									\
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );	\
+   GLuint tmp = z24s8_to_s8z24(d);					\
+   *_ptr = tmp;					\
+} while (0)
 #endif
 
+#ifdef COMPILE_R300
+#define READ_DEPTH( d, _x, _y )						\
+  do { \
+    d = (*(GLuint*)(radeon_ptr32(rrb, _x + x_off, _y + y_off)));	\
+  }while(0)
+#else
+#define READ_DEPTH( d, _x, _y )	do {					\
+    d = s8z24_to_z24s8(*(GLuint*)(radeon_ptr32(rrb, _x + x_off,	_y + y_off ))); \
+  } while (0)
+#endif
+/*
+    fprintf(stderr, "dval(%d, %d, %d, %d)=0x%08X\n", _x, xo, _y, yo, d);\
+   d = *(GLuint*)(radeon_ptr(rrb, _x,	_y )) & 0x00ffffff;
+*/
 #define TAG(x) radeon##x##_z24_s8
 #include "depthtmp.h"
 
@@ -235,35 +397,35 @@ do {									\
 #ifdef COMPILE_R300
 #define WRITE_STENCIL( _x, _y, d )					\
 do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
+   GLuint *_ptr = (GLuint*)radeon_ptr32(rrb, _x + x_off, _y + y_off);		\
+   GLuint tmp = *_ptr;				\
    tmp &= 0xffffff00;							\
    tmp |= (d) & 0xff;							\
-   *(GLuint *)(buf + offset) = tmp;					\
+   *_ptr = tmp;					\
 } while (0)
 #else
 #define WRITE_STENCIL( _x, _y, d )					\
 do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
+   GLuint *_ptr = (GLuint*)radeon_ptr32(rrb, _x + x_off, _y + y_off);		\
+   GLuint tmp = *_ptr;				\
    tmp &= 0x00ffffff;							\
    tmp |= (((d) & 0xff) << 24);						\
-   *(GLuint *)(buf + offset) = tmp;					\
+   *_ptr = tmp;					\
 } while (0)
 #endif
 
 #ifdef COMPILE_R300
 #define READ_STENCIL( d, _x, _y )					\
 do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );		\
+   GLuint tmp = *_ptr;				\
    d = tmp & 0x000000ff;						\
 } while (0)
 #else
 #define READ_STENCIL( d, _x, _y )					\
 do {									\
-   GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo );		\
-   GLuint tmp = *(GLuint *)(buf + offset);				\
+   GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x + x_off, _y + y_off );		\
+   GLuint tmp = *_ptr;				\
    d = (tmp & 0xff000000) >> 24;					\
 } while (0)
 #endif
@@ -271,29 +433,105 @@ do {									\
 #define TAG(x) radeon##x##_z24_s8
 #include "stenciltmp.h"
 
-/* Move locking out to get reasonable span performance (10x better
- * than doing this in HW_LOCK above).  WaitForIdle() is the main
- * culprit.
- */
 
+static void map_unmap_rb(struct gl_renderbuffer *rb, int flag)
+{
+	struct radeon_renderbuffer *rrb = radeon_renderbuffer(rb);
+	int r;
+	
+	if (rrb == NULL || !rrb->bo)
+		return;
+
+	if (flag) {
+		r = radeon_bo_map(rrb->bo, 1);
+		if (r) {
+			fprintf(stderr, "(%s) error(%d) mapping buffer.\n",
+				__FUNCTION__, r);
+		}
+
+		radeonSetSpanFunctions(rrb);
+	} else {
+		radeon_bo_unmap(rrb->bo);
+		rb->GetRow = NULL;
+		rb->PutRow = NULL;
+	}
+}
+
+static void
+radeon_map_unmap_buffers(GLcontext *ctx, GLboolean map)
+{
+	GLuint i, j;
+
+	/* color draw buffers */
+	for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers; j++)
+		map_unmap_rb(ctx->DrawBuffer->_ColorDrawBuffers[j], map);
+
+	/* check for render to textures */
+	for (i = 0; i < BUFFER_COUNT; i++) {
+		struct gl_renderbuffer_attachment *att =
+			ctx->DrawBuffer->Attachment + i;
+		struct gl_texture_object *tex = att->Texture;
+		if (tex) {
+			/* render to texture */
+			ASSERT(att->Renderbuffer);
+			if (map)
+				ctx->Driver.MapTexture(ctx, tex);
+			else
+				ctx->Driver.UnmapTexture(ctx, tex);
+		}
+	}
+	
+	map_unmap_rb(ctx->ReadBuffer->_ColorReadBuffer, map);
+
+	/* depth buffer (Note wrapper!) */
+	if (ctx->DrawBuffer->_DepthBuffer)
+		map_unmap_rb(ctx->DrawBuffer->_DepthBuffer->Wrapped, map);
+	
+	if (ctx->DrawBuffer->_StencilBuffer)
+		map_unmap_rb(ctx->DrawBuffer->_StencilBuffer->Wrapped, map);
+
+}
 static void radeonSpanRenderStart(GLcontext * ctx)
 {
 	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-#ifdef COMPILE_R300
-	r300ContextPtr r300 = (r300ContextPtr) rmesa;
-	R300_FIREVERTICES(r300);
-#else
-	RADEON_FIREVERTICES(rmesa);
-#endif
-	LOCK_HARDWARE(rmesa);
-	radeonWaitForIdleLocked(rmesa);
+	int i;
+
+	radeon_firevertices(rmesa);
+
+	/* The locking and wait for idle should really only be needed in classic mode.
+	 * In a future memory manager based implementation, this should become
+	 * unnecessary due to the fact that mapping our buffers, textures, etc.
+	 * should implicitly wait for any previous rendering commands that must
+	 * be waited on. */
+	if (!rmesa->radeonScreen->driScreen->dri2.enabled) {
+		LOCK_HARDWARE(rmesa);
+		radeonWaitForIdleLocked(rmesa);
+	}
+	for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
+		if (ctx->Texture.Unit[i]._ReallyEnabled)
+			ctx->Driver.MapTexture(ctx, ctx->Texture.Unit[i]._Current);
+	}
+
+	radeon_map_unmap_buffers(ctx, 1);
+
+
+
 }
 
 static void radeonSpanRenderFinish(GLcontext * ctx)
 {
 	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	int i;
 	_swrast_flush(ctx);
-	UNLOCK_HARDWARE(rmesa);
+	if (!rmesa->radeonScreen->driScreen->dri2.enabled) {
+		UNLOCK_HARDWARE(rmesa);
+	}
+	for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
+		if (ctx->Texture.Unit[i]._ReallyEnabled)
+			ctx->Driver.UnmapTexture(ctx, ctx->Texture.Unit[i]._Current);
+	}
+
+	radeon_map_unmap_buffers(ctx, 0);
 }
 
 void radeonInitSpanFuncs(GLcontext * ctx)
@@ -307,20 +545,21 @@ void radeonInitSpanFuncs(GLcontext * ctx)
 /**
  * Plug in the Get/Put routines for the given driRenderbuffer.
  */
-void radeonSetSpanFunctions(driRenderbuffer * drb, const GLvisual * vis)
+static void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb)
 {
-	if (drb->Base.InternalFormat == GL_RGBA) {
-		if (vis->redBits == 5 && vis->greenBits == 6
-		    && vis->blueBits == 5) {
-			radeonInitPointers_RGB565(&drb->Base);
-		} else {
-			radeonInitPointers_ARGB8888(&drb->Base);
-		}
-	} else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) {
-		radeonInitDepthPointers_z16(&drb->Base);
-	} else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) {
-		radeonInitDepthPointers_z24_s8(&drb->Base);
-	} else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) {
-		radeonInitStencilPointers_z24_s8(&drb->Base);
+	if (rrb->base._ActualFormat == GL_RGB5) {
+		radeonInitPointers_RGB565(&rrb->base);
+	} else if (rrb->base._ActualFormat == GL_RGB8) {
+		radeonInitPointers_xRGB8888(&rrb->base);
+	} else if (rrb->base._ActualFormat == GL_RGBA8) {
+		radeonInitPointers_ARGB8888(&rrb->base);
+	} else if (rrb->base._ActualFormat == GL_DEPTH_COMPONENT16) {
+		radeonInitDepthPointers_z16(&rrb->base);
+	} else if (rrb->base._ActualFormat == GL_DEPTH_COMPONENT24) {
+		radeonInitDepthPointers_z24(&rrb->base);
+	} else if (rrb->base._ActualFormat == GL_DEPTH24_STENCIL8_EXT) {
+		radeonInitDepthPointers_z24_s8(&rrb->base);
+	} else if (rrb->base._ActualFormat == GL_STENCIL_INDEX8_EXT) {
+		radeonInitStencilPointers_z24_s8(&rrb->base);
 	}
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_span.h b/src/mesa/drivers/dri/radeon/radeon_span.h
index 9abe0864b17..ea6a2e7fb4e 100644
--- a/src/mesa/drivers/dri/radeon/radeon_span.h
+++ b/src/mesa/drivers/dri/radeon/radeon_span.h
@@ -42,9 +42,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifndef __RADEON_SPAN_H__
 #define __RADEON_SPAN_H__
 
-#include "drirenderbuffer.h"
-
 extern void radeonInitSpanFuncs(GLcontext * ctx);
-extern void radeonSetSpanFunctions(driRenderbuffer * rb, const GLvisual * vis);
 
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_state.c b/src/mesa/drivers/dri/radeon/radeon_state.c
index 4432f85691e..06b8c299368 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state.c
+++ b/src/mesa/drivers/dri/radeon/radeon_state.c
@@ -47,6 +47,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "swrast_setup/swrast_setup.h"
 
 #include "radeon_context.h"
+#include "radeon_mipmap_tree.h"
 #include "radeon_ioctl.h"
 #include "radeon_state.h"
 #include "radeon_tcl.h"
@@ -62,7 +63,7 @@ static void radeonUpdateSpecular( GLcontext *ctx );
 
 static void radeonAlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    int pp_misc = rmesa->hw.ctx.cmd[CTX_PP_MISC];
    GLubyte refByte;
 
@@ -106,7 +107,7 @@ static void radeonAlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref )
 static void radeonBlendEquationSeparate( GLcontext *ctx,
 					 GLenum modeRGB, GLenum modeA )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & ~RADEON_COMB_FCN_MASK;
    GLboolean fallback = GL_FALSE;
 
@@ -147,8 +148,8 @@ static void radeonBlendFuncSeparate( GLcontext *ctx,
 				     GLenum sfactorRGB, GLenum dfactorRGB,
 				     GLenum sfactorA, GLenum dfactorA )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & 
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] &
       ~(RADEON_SRC_BLEND_MASK | RADEON_DST_BLEND_MASK);
    GLboolean fallback = GL_FALSE;
 
@@ -257,7 +258,7 @@ static void radeonBlendFuncSeparate( GLcontext *ctx,
 
 static void radeonDepthFunc( GLcontext *ctx, GLenum func )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    RADEON_STATECHANGE( rmesa, ctx );
    rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_Z_TEST_MASK;
@@ -293,7 +294,7 @@ static void radeonDepthFunc( GLcontext *ctx, GLenum func )
 
 static void radeonDepthMask( GLcontext *ctx, GLboolean flag )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    RADEON_STATECHANGE( rmesa, ctx );
 
    if ( ctx->Depth.Mask ) {
@@ -305,16 +306,16 @@ static void radeonDepthMask( GLcontext *ctx, GLboolean flag )
 
 static void radeonClearDepth( GLcontext *ctx, GLclampd d )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint format = (rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &
 		    RADEON_DEPTH_FORMAT_MASK);
 
    switch ( format ) {
    case RADEON_DEPTH_FORMAT_16BIT_INT_Z:
-      rmesa->state.depth.clear = d * 0x0000ffff;
+      rmesa->radeon.state.depth.clear = d * 0x0000ffff;
       break;
    case RADEON_DEPTH_FORMAT_24BIT_INT_Z:
-      rmesa->state.depth.clear = d * 0x00ffffff;
+      rmesa->radeon.state.depth.clear = d * 0x00ffffff;
       break;
    }
 }
@@ -327,7 +328,7 @@ static void radeonClearDepth( GLcontext *ctx, GLclampd d )
 
 static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    union { int i; float f; } c, d;
    GLchan col[4];
 
@@ -391,7 +392,7 @@ static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
 	 rmesa->hw.fog.cmd[FOG_D] = d.i;
       }
       break;
-   case GL_FOG_COLOR: 
+   case GL_FOG_COLOR:
       RADEON_STATECHANGE( rmesa, ctx );
       UNCLAMPED_FLOAT_TO_RGB_CHAN( col, ctx->Fog.Color );
       rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] &= ~RADEON_FOG_COLOR_MASK;
@@ -406,109 +407,13 @@ static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
    }
 }
 
-
-/* =============================================================
- * Scissoring
- */
-
-
-static GLboolean intersect_rect( drm_clip_rect_t *out,
-				 drm_clip_rect_t *a,
-				 drm_clip_rect_t *b )
-{
-   *out = *a;
-   if ( b->x1 > out->x1 ) out->x1 = b->x1;
-   if ( b->y1 > out->y1 ) out->y1 = b->y1;
-   if ( b->x2 < out->x2 ) out->x2 = b->x2;
-   if ( b->y2 < out->y2 ) out->y2 = b->y2;
-   if ( out->x1 >= out->x2 ) return GL_FALSE;
-   if ( out->y1 >= out->y2 ) return GL_FALSE;
-   return GL_TRUE;
-}
-
-
-void radeonRecalcScissorRects( radeonContextPtr rmesa )
-{
-   drm_clip_rect_t *out;
-   int i;
-
-   /* Grow cliprect store?
-    */
-   if (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) {
-      while (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) {
-	 rmesa->state.scissor.numAllocedClipRects += 1;	/* zero case */
-	 rmesa->state.scissor.numAllocedClipRects *= 2;
-      }
-
-      if (rmesa->state.scissor.pClipRects)
-	 FREE(rmesa->state.scissor.pClipRects);
-
-      rmesa->state.scissor.pClipRects = 
-	 MALLOC( rmesa->state.scissor.numAllocedClipRects * 
-		 sizeof(drm_clip_rect_t) );
-
-      if ( rmesa->state.scissor.pClipRects == NULL ) {
-	 rmesa->state.scissor.numAllocedClipRects = 0;
-	 return;
-      }
-   }
-   
-   out = rmesa->state.scissor.pClipRects;
-   rmesa->state.scissor.numClipRects = 0;
-
-   for ( i = 0 ; i < rmesa->numClipRects ;  i++ ) {
-      if ( intersect_rect( out, 
-			   &rmesa->pClipRects[i], 
-			   &rmesa->state.scissor.rect ) ) {
-	 rmesa->state.scissor.numClipRects++;
-	 out++;
-      }
-   }
-}
-
-
-static void radeonUpdateScissor( GLcontext *ctx )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if ( rmesa->dri.drawable ) {
-      __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
-
-      int x = ctx->Scissor.X;
-      int y = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height;
-      int w = ctx->Scissor.X + ctx->Scissor.Width - 1;
-      int h = dPriv->h - ctx->Scissor.Y - 1;
-
-      rmesa->state.scissor.rect.x1 = x + dPriv->x;
-      rmesa->state.scissor.rect.y1 = y + dPriv->y;
-      rmesa->state.scissor.rect.x2 = w + dPriv->x + 1;
-      rmesa->state.scissor.rect.y2 = h + dPriv->y + 1;
-
-      radeonRecalcScissorRects( rmesa );
-   }
-}
-
-
-static void radeonScissor( GLcontext *ctx,
-			   GLint x, GLint y, GLsizei w, GLsizei h )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if ( ctx->Scissor.Enabled ) {
-      RADEON_FIREVERTICES( rmesa );	/* don't pipeline cliprect changes */
-      radeonUpdateScissor( ctx );
-   }
-
-}
-
-
 /* =============================================================
  * Culling
  */
 
 static void radeonCullFace( GLcontext *ctx, GLenum unused )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint s = rmesa->hw.set.cmd[SET_SE_CNTL];
    GLuint t = rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL];
 
@@ -545,7 +450,7 @@ static void radeonCullFace( GLcontext *ctx, GLenum unused )
 
 static void radeonFrontFace( GLcontext *ctx, GLenum mode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    RADEON_STATECHANGE( rmesa, set );
    rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_FFACE_CULL_DIR_MASK;
@@ -570,7 +475,7 @@ static void radeonFrontFace( GLcontext *ctx, GLenum mode )
  */
 static void radeonLineWidth( GLcontext *ctx, GLfloat widthf )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    RADEON_STATECHANGE( rmesa, lin );
    RADEON_STATECHANGE( rmesa, set );
@@ -587,10 +492,10 @@ static void radeonLineWidth( GLcontext *ctx, GLfloat widthf )
 
 static void radeonLineStipple( GLcontext *ctx, GLint factor, GLushort pattern )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    RADEON_STATECHANGE( rmesa, lin );
-   rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] = 
+   rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] =
       ((((GLuint)factor & 0xff) << 16) | ((GLuint)pattern));
 }
 
@@ -602,8 +507,8 @@ static void radeonColorMask( GLcontext *ctx,
 			     GLboolean r, GLboolean g,
 			     GLboolean b, GLboolean a )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   GLuint mask = radeonPackColor( rmesa->radeonScreen->cpp,
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   GLuint mask = radeonPackColor( rmesa->radeon.radeonScreen->cpp,
 				  ctx->Color.ColorMask[RCOMP],
 				  ctx->Color.ColorMask[GCOMP],
 				  ctx->Color.ColorMask[BCOMP],
@@ -623,8 +528,9 @@ static void radeonColorMask( GLcontext *ctx,
 static void radeonPolygonOffset( GLcontext *ctx,
 				 GLfloat factor, GLfloat units )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   float_ui32_type constant =  { units * rmesa->state.depth.scale };
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
+   float_ui32_type constant =  { units * depthScale };
    float_ui32_type factoru = { factor };
 
    RADEON_STATECHANGE( rmesa, zbs );
@@ -634,7 +540,7 @@ static void radeonPolygonOffset( GLcontext *ctx,
 
 static void radeonPolygonStipple( GLcontext *ctx, const GLubyte *mask )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint i;
    drm_radeon_stipple_t stipple;
 
@@ -646,27 +552,27 @@ static void radeonPolygonStipple( GLcontext *ctx, const GLubyte *mask )
 
    /* TODO: push this into cmd mechanism
     */
-   RADEON_FIREVERTICES( rmesa );
-   LOCK_HARDWARE( rmesa );
+   radeon_firevertices(&rmesa->radeon);
+   LOCK_HARDWARE( &rmesa->radeon );
 
    /* FIXME: Use window x,y offsets into stipple RAM.
     */
    stipple.mask = rmesa->state.stipple.mask;
-   drmCommandWrite( rmesa->dri.fd, DRM_RADEON_STIPPLE, 
+   drmCommandWrite( rmesa->radeon.dri.fd, DRM_RADEON_STIPPLE,
                     &stipple, sizeof(drm_radeon_stipple_t) );
-   UNLOCK_HARDWARE( rmesa );
+   UNLOCK_HARDWARE( &rmesa->radeon );
 }
 
 static void radeonPolygonMode( GLcontext *ctx, GLenum face, GLenum mode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLboolean flag = (ctx->_TriangleCaps & DD_TRI_UNFILLED) != 0;
 
    /* Can't generally do unfilled via tcl, but some good special
-    * cases work. 
+    * cases work.
     */
    TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_UNFILLED, flag);
-   if (rmesa->TclFallback) {
+   if (rmesa->radeon.TclFallback) {
       radeonChooseRenderState( ctx );
       radeonChooseVertexState( ctx );
    }
@@ -686,7 +592,7 @@ static void radeonPolygonMode( GLcontext *ctx, GLenum face, GLenum mode )
  */
 static void radeonUpdateSpecular( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    uint32_t p = rmesa->hw.ctx.cmd[CTX_PP_CNTL];
    GLuint flag = 0;
 
@@ -711,7 +617,7 @@ static void radeonUpdateSpecular( GLcontext *ctx )
       rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
       rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
       p |=  RADEON_SPECULAR_ENABLE;
-      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= 
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &=
 	 ~RADEON_DIFFUSE_SPECULAR_COMBINE;
    }
    else if (ctx->Light.Enabled) {
@@ -741,7 +647,7 @@ static void radeonUpdateSpecular( GLcontext *ctx )
 	    RADEON_TCL_COMPUTE_SPECULAR) != 0;
       }
    }
- 
+
    TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_FOGCOORDSPEC, flag);
 
    if (NEED_SECONDARY_COLOR(ctx)) {
@@ -757,7 +663,7 @@ static void radeonUpdateSpecular( GLcontext *ctx )
 
    /* Update vertex/render formats
     */
-   if (rmesa->TclFallback) { 
+   if (rmesa->radeon.TclFallback) {
       radeonChooseRenderState( ctx );
       radeonChooseVertexState( ctx );
    }
@@ -769,12 +675,12 @@ static void radeonUpdateSpecular( GLcontext *ctx )
  */
 
 
-/* Update on colormaterial, material emmissive/ambient, 
+/* Update on colormaterial, material emmissive/ambient,
  * lightmodel.globalambient
  */
 static void update_global_ambient( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    float *fcmd = (float *)RADEON_DB_STATE( glt );
 
    /* Need to do more if both emmissive & ambient are PREMULT:
@@ -782,23 +688,23 @@ static void update_global_ambient( GLcontext *ctx )
     */
    if ((rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &
        ((3 << RADEON_EMISSIVE_SOURCE_SHIFT) |
-	(3 << RADEON_AMBIENT_SOURCE_SHIFT))) == 0) 
+	(3 << RADEON_AMBIENT_SOURCE_SHIFT))) == 0)
    {
-      COPY_3V( &fcmd[GLT_RED], 
+      COPY_3V( &fcmd[GLT_RED],
 	       ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_EMISSION]);
       ACC_SCALE_3V( &fcmd[GLT_RED],
 		   ctx->Light.Model.Ambient,
 		   ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT]);
-   } 
+   }
    else
    {
       COPY_3V( &fcmd[GLT_RED], ctx->Light.Model.Ambient );
    }
-   
+
    RADEON_DB_STATECHANGE(rmesa, &rmesa->hw.glt);
 }
 
-/* Update on change to 
+/* Update on change to
  *    - light[p].colors
  *    - light[p].enabled
  */
@@ -809,13 +715,13 @@ static void update_light_colors( GLcontext *ctx, GLuint p )
 /*     fprintf(stderr, "%s\n", __FUNCTION__); */
 
    if (l->Enabled) {
-      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+      r100ContextPtr rmesa = R100_CONTEXT(ctx);
       float *fcmd = (float *)RADEON_DB_STATE( lit[p] );
 
-      COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient );	 
+      COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient );
       COPY_4V( &fcmd[LIT_DIFFUSE_RED], l->Diffuse );
       COPY_4V( &fcmd[LIT_SPECULAR_RED], l->Specular );
-      
+
       RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] );
    }
 }
@@ -829,7 +735,7 @@ static void check_twoside_fallback( GLcontext *ctx )
 
    if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
       if (ctx->Light.ColorMaterialEnabled &&
-	  (ctx->Light.ColorMaterialBitmask & BACK_MATERIAL_BITS) != 
+	  (ctx->Light.ColorMaterialBitmask & BACK_MATERIAL_BITS) !=
 	  ((ctx->Light.ColorMaterialBitmask & FRONT_MATERIAL_BITS)<<1))
 	 fallback = GL_TRUE;
       else {
@@ -837,7 +743,7 @@ static void check_twoside_fallback( GLcontext *ctx )
 	    if (memcmp( ctx->Light.Material.Attrib[i],
 			ctx->Light.Material.Attrib[i+1],
 			sizeof(GLfloat)*4) != 0) {
-	       fallback = GL_TRUE;  
+	       fallback = GL_TRUE;
 	       break;
 	    }
       }
@@ -849,14 +755,14 @@ static void check_twoside_fallback( GLcontext *ctx )
 
 static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
 {
-      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+      r100ContextPtr rmesa = R100_CONTEXT(ctx);
       GLuint light_model_ctl1 = rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL];
 
       light_model_ctl1 &= ~((3 << RADEON_EMISSIVE_SOURCE_SHIFT) |
 			   (3 << RADEON_AMBIENT_SOURCE_SHIFT) |
 			   (3 << RADEON_DIFFUSE_SOURCE_SHIFT) |
-			   (3 << RADEON_SPECULAR_SOURCE_SHIFT)); 
-   
+			   (3 << RADEON_SPECULAR_SOURCE_SHIFT));
+
    if (ctx->Light.ColorMaterialEnabled) {
       GLuint mask = ctx->Light.ColorMaterialBitmask;
 
@@ -877,7 +783,7 @@ static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
 	 light_model_ctl1 |= (RADEON_LM_SOURCE_STATE_MULT <<
 			     RADEON_AMBIENT_SOURCE_SHIFT);
       }
-	 
+
       if (mask & MAT_BIT_FRONT_DIFFUSE) {
 	 light_model_ctl1 |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE <<
 			     RADEON_DIFFUSE_SOURCE_SHIFT);
@@ -886,7 +792,7 @@ static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
 	 light_model_ctl1 |= (RADEON_LM_SOURCE_STATE_MULT <<
 			     RADEON_DIFFUSE_SOURCE_SHIFT);
       }
-   
+
       if (mask & MAT_BIT_FRONT_SPECULAR) {
 	 light_model_ctl1 |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE <<
 			     RADEON_SPECULAR_SOURCE_SHIFT);
@@ -904,27 +810,27 @@ static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
 		   (RADEON_LM_SOURCE_STATE_MULT << RADEON_DIFFUSE_SOURCE_SHIFT) |
 		   (RADEON_LM_SOURCE_STATE_MULT << RADEON_SPECULAR_SOURCE_SHIFT);
    }
-   
+
       if (light_model_ctl1 != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) {
 	 RADEON_STATECHANGE( rmesa, tcl );
-	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = light_model_ctl1;      
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = light_model_ctl1;
    }
 }
 
 void radeonUpdateMaterial( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLfloat (*mat)[4] = ctx->Light.Material.Attrib;
    GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( mtl );
    GLuint mask = ~0;
-   
+
    if (ctx->Light.ColorMaterialEnabled)
       mask &= ~ctx->Light.ColorMaterialBitmask;
 
    if (RADEON_DEBUG & DEBUG_STATE)
       fprintf(stderr, "%s\n", __FUNCTION__);
 
-      
+
    if (mask & MAT_BIT_FRONT_EMISSION) {
       fcmd[MTL_EMMISSIVE_RED]   = mat[MAT_ATTRIB_FRONT_EMISSION][0];
       fcmd[MTL_EMMISSIVE_GREEN] = mat[MAT_ATTRIB_FRONT_EMISSION][1];
@@ -974,11 +880,11 @@ void radeonUpdateMaterial( GLcontext *ctx )
  *
  * which are calculated in light.c and are correct for the current
  * lighting space (model or eye), hence dependencies on _NEW_MODELVIEW
- * and _MESA_NEW_NEED_EYE_COORDS.  
+ * and _MESA_NEW_NEED_EYE_COORDS.
  */
 static void update_light( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    /* Have to check these, or have an automatic shortcircuit mechanism
     * to remove noop statechanges. (Or just do a better job on the
@@ -991,12 +897,12 @@ static void update_light( GLcontext *ctx )
 	 tmp &= ~RADEON_LIGHT_IN_MODELSPACE;
       else
 	 tmp |= RADEON_LIGHT_IN_MODELSPACE;
-      
+
 
       /* Leave this test disabled: (unexplained q3 lockup) (even with
          new packets)
       */
-      if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) 
+      if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL])
       {
 	 RADEON_STATECHANGE( rmesa, tcl );
 	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = tmp;
@@ -1020,10 +926,10 @@ static void update_light( GLcontext *ctx )
 	 if (ctx->Light.Light[p].Enabled) {
 	    struct gl_light *l = &ctx->Light.Light[p];
 	    GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( lit[p] );
-	    
+
 	    if (l->EyePosition[3] == 0.0) {
-	       COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm ); 
-	       COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm ); 
+	       COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm );
+	       COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm );
 	       fcmd[LIT_POSITION_W] = 0;
 	       fcmd[LIT_DIRECTION_W] = 0;
 	    } else {
@@ -1043,30 +949,30 @@ static void update_light( GLcontext *ctx )
 static void radeonLightfv( GLcontext *ctx, GLenum light,
 			   GLenum pname, const GLfloat *params )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLint p = light - GL_LIGHT0;
    struct gl_light *l = &ctx->Light.Light[p];
    GLfloat *fcmd = (GLfloat *)rmesa->hw.lit[p].cmd;
-   
+
 
    switch (pname) {
-   case GL_AMBIENT:		
+   case GL_AMBIENT:
    case GL_DIFFUSE:
    case GL_SPECULAR:
       update_light_colors( ctx, p );
       break;
 
-   case GL_SPOT_DIRECTION: 
-      /* picked up in update_light */	
+   case GL_SPOT_DIRECTION:
+      /* picked up in update_light */
       break;
 
    case GL_POSITION: {
-      /* positions picked up in update_light, but can do flag here */	
+      /* positions picked up in update_light, but can do flag here */
       GLuint flag;
       GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
 
       /* FIXME: Set RANGE_ATTEN only when needed */
-      if (p&1) 
+      if (p&1)
 	 flag = RADEON_LIGHT_1_IS_LOCAL;
       else
 	 flag = RADEON_LIGHT_0_IS_LOCAL;
@@ -1158,16 +1064,16 @@ static void radeonLightfv( GLcontext *ctx, GLenum light,
    }
 }
 
-		  
+
 
 
 static void radeonLightModelfv( GLcontext *ctx, GLenum pname,
 				const GLfloat *param )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    switch (pname) {
-      case GL_LIGHT_MODEL_AMBIENT: 
+      case GL_LIGHT_MODEL_AMBIENT:
 	 update_global_ambient( ctx );
 	 break;
 
@@ -1188,7 +1094,7 @@ static void radeonLightModelfv( GLcontext *ctx, GLenum pname,
 
 	 check_twoside_fallback( ctx );
 
-	 if (rmesa->TclFallback) {
+	 if (rmesa->radeon.TclFallback) {
 	    radeonChooseRenderState( ctx );
 	    radeonChooseVertexState( ctx );
 	 }
@@ -1205,7 +1111,7 @@ static void radeonLightModelfv( GLcontext *ctx, GLenum pname,
 
 static void radeonShadeModel( GLcontext *ctx, GLenum mode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint s = rmesa->hw.set.cmd[SET_SE_CNTL];
 
    s &= ~(RADEON_DIFFUSE_SHADE_MASK |
@@ -1244,7 +1150,7 @@ static void radeonShadeModel( GLcontext *ctx, GLenum mode )
 static void radeonClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq )
 {
    GLint p = (GLint) plane - (GLint) GL_CLIP_PLANE0;
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p];
 
    RADEON_STATECHANGE( rmesa, ucp[p] );
@@ -1256,7 +1162,7 @@ static void radeonClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq )
 
 static void radeonUpdateClipPlanes( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint p;
 
    for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
@@ -1281,7 +1187,7 @@ static void
 radeonStencilFuncSeparate( GLcontext *ctx, GLenum face, GLenum func,
                            GLint ref, GLuint mask )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint refmask = (((ctx->Stencil.Ref[0] & 0xff) << RADEON_STENCIL_REF_SHIFT) |
 		     ((ctx->Stencil.ValueMask[0] & 0xff) << RADEON_STENCIL_MASK_SHIFT));
 
@@ -1325,7 +1231,7 @@ radeonStencilFuncSeparate( GLcontext *ctx, GLenum face, GLenum func,
 static void
 radeonStencilMaskSeparate( GLcontext *ctx, GLenum face, GLuint mask )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    RADEON_STATECHANGE( rmesa, msk );
    rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] &= ~RADEON_STENCIL_WRITE_MASK;
@@ -1336,20 +1242,20 @@ radeonStencilMaskSeparate( GLcontext *ctx, GLenum face, GLuint mask )
 static void radeonStencilOpSeparate( GLcontext *ctx, GLenum face, GLenum fail,
                                      GLenum zfail, GLenum zpass )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
    /* radeon 7200 have stencil bug, DEC and INC_WRAP will actually both do DEC_WRAP,
       and DEC_WRAP (and INVERT) will do INVERT. No way to get correct INC_WRAP and DEC,
       but DEC_WRAP can be fixed by using DEC and INC_WRAP at least use INC. */
-   
+
    GLuint tempRADEON_STENCIL_FAIL_DEC_WRAP;
    GLuint tempRADEON_STENCIL_FAIL_INC_WRAP;
    GLuint tempRADEON_STENCIL_ZFAIL_DEC_WRAP;
    GLuint tempRADEON_STENCIL_ZFAIL_INC_WRAP;
    GLuint tempRADEON_STENCIL_ZPASS_DEC_WRAP;
    GLuint tempRADEON_STENCIL_ZPASS_INC_WRAP;
-   
-   if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_BROKEN_STENCIL) {
+
+   if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_BROKEN_STENCIL) {
       tempRADEON_STENCIL_FAIL_DEC_WRAP = RADEON_STENCIL_FAIL_DEC;
       tempRADEON_STENCIL_FAIL_INC_WRAP = RADEON_STENCIL_FAIL_INC;
       tempRADEON_STENCIL_ZFAIL_DEC_WRAP = RADEON_STENCIL_ZFAIL_DEC;
@@ -1365,7 +1271,7 @@ static void radeonStencilOpSeparate( GLcontext *ctx, GLenum face, GLenum fail,
       tempRADEON_STENCIL_ZPASS_DEC_WRAP = RADEON_STENCIL_ZPASS_DEC_WRAP;
       tempRADEON_STENCIL_ZPASS_INC_WRAP = RADEON_STENCIL_ZPASS_INC_WRAP;
    }
-   
+
    RADEON_STATECHANGE( rmesa, ctx );
    rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~(RADEON_STENCIL_FAIL_MASK |
 					       RADEON_STENCIL_ZFAIL_MASK |
@@ -1455,9 +1361,9 @@ static void radeonStencilOpSeparate( GLcontext *ctx, GLenum face, GLenum fail,
 
 static void radeonClearStencil( GLcontext *ctx, GLint s )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
-   rmesa->state.stencil.clear = 
+   rmesa->radeon.state.stencil.clear =
       ((GLuint) (ctx->Stencil.Clear & 0xff) |
        (0xff << RADEON_STENCIL_MASK_SHIFT) |
        ((ctx->Stencil.WriteMask[0] & 0xff) << RADEON_STENCIL_WRITEMASK_SHIFT));
@@ -1481,20 +1387,30 @@ static void radeonClearStencil( GLcontext *ctx, GLint s )
  */
 void radeonUpdateWindow( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
-   GLfloat xoffset = (GLfloat)dPriv->x;
-   GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
+   GLfloat xoffset = dPriv ? (GLfloat) dPriv->x : 0;
+   GLfloat yoffset = dPriv ? (GLfloat) dPriv->y + dPriv->h : 0;
    const GLfloat *v = ctx->Viewport._WindowMap.m;
+   const GLboolean render_to_fbo = (ctx->DrawBuffer ? (ctx->DrawBuffer->Name != 0) : 0);
+   const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
+   GLfloat y_scale, y_bias;
+
+   if (render_to_fbo) {
+      y_scale = 1.0;
+      y_bias = 0;
+   } else {
+      y_scale = -1.0;
+      y_bias = yoffset;
+   }
 
    float_ui32_type sx = { v[MAT_SX] };
    float_ui32_type tx = { v[MAT_TX] + xoffset + SUBPIXEL_X };
-   float_ui32_type sy = { - v[MAT_SY] };
-   float_ui32_type ty = { (- v[MAT_TY]) + yoffset + SUBPIXEL_Y };
-   float_ui32_type sz = { v[MAT_SZ] * rmesa->state.depth.scale };
-   float_ui32_type tz = { v[MAT_TZ] * rmesa->state.depth.scale };
+   float_ui32_type sy = { v[MAT_SY] * y_scale };
+   float_ui32_type ty = { (v[MAT_TY] * y_scale) + y_bias + SUBPIXEL_Y };
+   float_ui32_type sz = { v[MAT_SZ] * depthScale };
+   float_ui32_type tz = { v[MAT_TZ] * depthScale };
 
-   RADEON_FIREVERTICES( rmesa );
    RADEON_STATECHANGE( rmesa, vpt );
 
    rmesa->hw.vpt.cmd[VPT_SE_VPORT_XSCALE]  = sx.ui32;
@@ -1514,6 +1430,8 @@ static void radeonViewport( GLcontext *ctx, GLint x, GLint y,
     * values, or keep the originals hanging around.
     */
    radeonUpdateWindow( ctx );
+
+   radeon_viewport(ctx, x, y, width, height);
 }
 
 static void radeonDepthRange( GLcontext *ctx, GLclampd nearval,
@@ -1524,8 +1442,8 @@ static void radeonDepthRange( GLcontext *ctx, GLclampd nearval,
 
 void radeonUpdateViewportOffset( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon);
    GLfloat xoffset = (GLfloat)dPriv->x;
    GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h;
    const GLfloat *v = ctx->Viewport._WindowMap.m;
@@ -1555,8 +1473,8 @@ void radeonUpdateViewportOffset( GLcontext *ctx )
                 RADEON_STIPPLE_Y_OFFSET_MASK);
 
          /* add magic offsets, then invert */
-         stx = 31 - ((rmesa->dri.drawable->x - 1) & RADEON_STIPPLE_COORD_MASK);
-         sty = 31 - ((rmesa->dri.drawable->y + rmesa->dri.drawable->h - 1)
+         stx = 31 - ((dPriv->x - 1) & RADEON_STIPPLE_COORD_MASK);
+         sty = 31 - ((dPriv->y + dPriv->h - 1)
                      & RADEON_STIPPLE_COORD_MASK);
 
          m |= ((stx << RADEON_STIPPLE_X_OFFSET_SHIFT) |
@@ -1580,20 +1498,20 @@ void radeonUpdateViewportOffset( GLcontext *ctx )
 
 static void radeonClearColor( GLcontext *ctx, const GLfloat color[4] )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLubyte c[4];
    CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]);
    CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]);
    CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]);
    CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]);
-   rmesa->state.color.clear = radeonPackColor( rmesa->radeonScreen->cpp,
+   rmesa->radeon.state.color.clear = radeonPackColor( rmesa->radeon.radeonScreen->cpp,
 					       c[0], c[1], c[2], c[3] );
 }
 
 
 static void radeonRenderMode( GLcontext *ctx, GLenum mode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    FALLBACK( rmesa, RADEON_FALLBACK_RENDER_MODE, (mode != GL_RENDER) );
 }
 
@@ -1619,7 +1537,7 @@ static GLuint radeon_rop_tab[] = {
 
 static void radeonLogicOpCode( GLcontext *ctx, GLenum opcode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint rop = (GLuint)opcode - GL_CLEAR;
 
    ASSERT( rop < 16 );
@@ -1628,105 +1546,13 @@ static void radeonLogicOpCode( GLcontext *ctx, GLenum opcode )
    rmesa->hw.msk.cmd[MSK_RB3D_ROPCNTL] = radeon_rop_tab[rop];
 }
 
-
-/**
- * Set up the cliprects for either front or back-buffer drawing.
- */
-void radeonSetCliprects( radeonContextPtr rmesa )
-{
-   __DRIdrawablePrivate *const drawable = rmesa->dri.drawable;
-   __DRIdrawablePrivate *const readable = rmesa->dri.readable;
-   GLframebuffer *const draw_fb = (GLframebuffer*) drawable->driverPrivate;
-   GLframebuffer *const read_fb = (GLframebuffer*) readable->driverPrivate;
-
-   if (draw_fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) {
-      /* Can't ignore 2d windows if we are page flipping.
-       */
-      if ( drawable->numBackClipRects == 0 || rmesa->doPageFlip ) {
-	 rmesa->numClipRects = drawable->numClipRects;
-	 rmesa->pClipRects = drawable->pClipRects;
-      }
-      else {
-	 rmesa->numClipRects = drawable->numBackClipRects;
-	 rmesa->pClipRects = drawable->pBackClipRects;
-      }
-   }
-   else {
-      /* front buffer (or none, or multiple buffers */
-      rmesa->numClipRects = drawable->numClipRects;
-      rmesa->pClipRects = drawable->pClipRects;
-   }
-
-   if ((draw_fb->Width != drawable->w) || (draw_fb->Height != drawable->h)) {
-      _mesa_resize_framebuffer(rmesa->glCtx, draw_fb,
-			       drawable->w, drawable->h);
-      draw_fb->Initialized = GL_TRUE;
-   }
-
-   if (drawable != readable) {
-      if ((read_fb->Width != readable->w) || (read_fb->Height != readable->h)) {
-	 _mesa_resize_framebuffer(rmesa->glCtx, read_fb,
-				  readable->w, readable->h);
-	 read_fb->Initialized = GL_TRUE;
-      }
-   }
-
-   if (rmesa->state.scissor.enabled)
-      radeonRecalcScissorRects( rmesa );
-
-   rmesa->lastStamp = drawable->lastStamp;
-}
-
-
-/**
- * Called via glDrawBuffer.
- */
-static void radeonDrawBuffer( GLcontext *ctx, GLenum mode )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if (RADEON_DEBUG & DEBUG_DRI)
-      fprintf(stderr, "%s %s\n", __FUNCTION__,
-	      _mesa_lookup_enum_by_nr( mode ));
-
-   RADEON_FIREVERTICES(rmesa);	/* don't pipeline cliprect changes */
-
-   if (ctx->DrawBuffer->_NumColorDrawBuffers != 1) {
-      /* 0 (GL_NONE) buffers or multiple color drawing buffers */
-      FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE );
-      return;
-   }
-
-   switch ( ctx->DrawBuffer->_ColorDrawBufferIndexes[0] ) {
-   case BUFFER_FRONT_LEFT:
-   case BUFFER_BACK_LEFT:
-      FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE );
-      break;
-   default:
-      FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE );
-      return;
-   }
-
-   radeonSetCliprects( rmesa );
-
-   /* We'll set the drawing engine's offset/pitch parameters later
-    * when we update other state.
-    */
-}
-
-static void radeonReadBuffer( GLcontext *ctx, GLenum mode )
-{
-   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
-}
-
-
 /* =============================================================
  * State enable/disable
  */
 
 static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint p, flag;
 
    if ( RADEON_DEBUG & DEBUG_STATE )
@@ -1787,7 +1613,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_CLIP_PLANE2:
    case GL_CLIP_PLANE3:
    case GL_CLIP_PLANE4:
-   case GL_CLIP_PLANE5: 
+   case GL_CLIP_PLANE5:
       p = cap-GL_CLIP_PLANE0;
       RADEON_STATECHANGE( rmesa, tcl );
       if (state) {
@@ -1821,10 +1647,10 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
       RADEON_STATECHANGE(rmesa, ctx );
       if ( state ) {
 	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_DITHER_ENABLE;
-	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~rmesa->state.color.roundEnable;
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~rmesa->radeon.state.color.roundEnable;
       } else {
 	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_DITHER_ENABLE;
-	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  rmesa->state.color.roundEnable;
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  rmesa->radeon.state.color.roundEnable;
       }
       break;
 
@@ -1852,13 +1678,13 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_LIGHT7:
       RADEON_STATECHANGE(rmesa, tcl);
       p = cap - GL_LIGHT0;
-      if (p&1) 
+      if (p&1)
 	 flag = (RADEON_LIGHT_1_ENABLE |
-		 RADEON_LIGHT_1_ENABLE_AMBIENT | 
+		 RADEON_LIGHT_1_ENABLE_AMBIENT |
 		 RADEON_LIGHT_1_ENABLE_SPECULAR);
       else
 	 flag = (RADEON_LIGHT_0_ENABLE |
-		 RADEON_LIGHT_0_ENABLE_AMBIENT | 
+		 RADEON_LIGHT_0_ENABLE_AMBIENT |
 		 RADEON_LIGHT_0_ENABLE_SPECULAR);
 
       if (state)
@@ -1866,7 +1692,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
       else
 	 rmesa->hw.tcl.cmd[p/2 + TCL_PER_LIGHT_CTL_0] &= ~flag;
 
-      /* 
+      /*
        */
       update_light_colors( ctx, p );
       break;
@@ -1904,7 +1730,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
 	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_ROP_ENABLE;
       }
       break;
-      
+
    case GL_NORMALIZE:
       RADEON_STATECHANGE( rmesa, tcl );
       if ( state ) {
@@ -1971,21 +1797,30 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
    }
 
    case GL_SCISSOR_TEST:
-      RADEON_FIREVERTICES( rmesa );
-      rmesa->state.scissor.enabled = state;
+      radeon_firevertices(&rmesa->radeon);
+      rmesa->radeon.state.scissor.enabled = state;
       radeonUpdateScissor( ctx );
       break;
 
    case GL_STENCIL_TEST:
-      if ( rmesa->state.stencil.hwBuffer ) {
-	 RADEON_STATECHANGE( rmesa, ctx );
-	 if ( state ) {
-	    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_STENCIL_ENABLE;
+      {
+	 GLboolean hw_stencil = GL_FALSE;
+	 if (ctx->DrawBuffer) {
+	    struct radeon_renderbuffer *rrbStencil
+	       = radeon_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL);
+	    hw_stencil = (rrbStencil && rrbStencil->bo);
+	 }
+
+	 if (hw_stencil) {
+	    RADEON_STATECHANGE( rmesa, ctx );
+	    if ( state ) {
+	       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_STENCIL_ENABLE;
+	    } else {
+	       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_STENCIL_ENABLE;
+	    }
 	 } else {
-	    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_STENCIL_ENABLE;
+	    FALLBACK( rmesa, RADEON_FALLBACK_STENCIL, state );
 	 }
-      } else {
-	 FALLBACK( rmesa, RADEON_FALLBACK_STENCIL, state );
       }
       break;
 
@@ -1995,7 +1830,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_TEXTURE_GEN_T:
       /* Picked up in radeonUpdateTextureState.
        */
-      rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE; 
+      rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE;
       break;
 
    case GL_COLOR_SUM_EXT:
@@ -2010,7 +1845,7 @@ static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
 
 static void radeonLightingSpaceChange( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLboolean tmp;
    RADEON_STATECHANGE( rmesa, tcl );
 
@@ -2029,7 +1864,7 @@ static void radeonLightingSpaceChange( GLcontext *ctx )
       rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_RESCALE_NORMALS;
    }
 
-   if (RADEON_DEBUG & DEBUG_STATE) 
+   if (RADEON_DEBUG & DEBUG_STATE)
       fprintf(stderr, "%s %d AFTER %x\n", __FUNCTION__, ctx->_NeedEyeCoords,
 	      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]);
 }
@@ -2039,7 +1874,7 @@ static void radeonLightingSpaceChange( GLcontext *ctx )
  */
 
 
-void radeonUploadTexMatrix( radeonContextPtr rmesa,
+void radeonUploadTexMatrix( r100ContextPtr rmesa,
 			    int unit, GLboolean swapcols )
 {
 /* Here's how this works: on r100, only 3 tex coords can be submitted, so the
@@ -2065,7 +1900,7 @@ void radeonUploadTexMatrix( radeonContextPtr rmesa,
    int idx = TEXMAT_0 + unit;
    float *dest = ((float *)RADEON_DB_STATE( mat[idx] )) + MAT_ELT_0;
    int i;
-   struct gl_texture_unit tUnit = rmesa->glCtx->Texture.Unit[unit];
+   struct gl_texture_unit tUnit = rmesa->radeon.glCtx->Texture.Unit[unit];
    GLfloat *src = rmesa->tmpmat[unit].m;
 
    rmesa->TexMatColSwap &= ~(1 << unit);
@@ -2119,7 +1954,7 @@ void radeonUploadTexMatrix( radeonContextPtr rmesa,
 }
 
 
-static void upload_matrix( radeonContextPtr rmesa, GLfloat *src, int idx )
+static void upload_matrix( r100ContextPtr rmesa, GLfloat *src, int idx )
 {
    float *dest = ((float *)RADEON_DB_STATE( mat[idx] ))+MAT_ELT_0;
    int i;
@@ -2135,7 +1970,7 @@ static void upload_matrix( radeonContextPtr rmesa, GLfloat *src, int idx )
    RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.mat[idx] );
 }
 
-static void upload_matrix_t( radeonContextPtr rmesa, GLfloat *src, int idx )
+static void upload_matrix_t( r100ContextPtr rmesa, GLfloat *src, int idx )
 {
    float *dest = ((float *)RADEON_DB_STATE( mat[idx] ))+MAT_ELT_0;
    memcpy(dest, src, 16*sizeof(float));
@@ -2145,7 +1980,7 @@ static void upload_matrix_t( radeonContextPtr rmesa, GLfloat *src, int idx )
 
 static void update_texturematrix( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   r100ContextPtr rmesa = R100_CONTEXT( ctx );
    GLuint tpc = rmesa->hw.tcl.cmd[TCL_TEXTURE_PROC_CTL];
    GLuint vs = rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL];
    int unit;
@@ -2209,64 +2044,75 @@ static void update_texturematrix( GLcontext *ctx )
    }
 }
 
-
-/**
- * Tell the card where to render (offset, pitch).
- * Effected by glDrawBuffer, etc
- */
-void
-radeonUpdateDrawBuffer(GLcontext *ctx)
+static GLboolean r100ValidateBuffers(GLcontext *ctx)
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   struct gl_framebuffer *fb = ctx->DrawBuffer;
-   driRenderbuffer *drb;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   struct radeon_renderbuffer *rrb;
+   int i;
 
-   if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) {
-      /* draw to front */
-      drb = (driRenderbuffer *) fb->Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
-   }
-   else if (fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) {
-      /* draw to back */
-      drb = (driRenderbuffer *) fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+   radeon_validate_reset_bos(&rmesa->radeon);
+
+   rrb = radeon_get_colorbuffer(&rmesa->radeon);
+   /* color buffer */
+   if (rrb && rrb->bo) {
+     radeon_validate_bo(&rmesa->radeon, rrb->bo,
+			0, RADEON_GEM_DOMAIN_VRAM);
    }
-   else {
-      /* drawing to multiple buffers, or none */
-      return;
+
+   /* depth buffer */
+   rrb = radeon_get_depthbuffer(&rmesa->radeon);
+   /* color buffer */
+   if (rrb && rrb->bo) {
+     radeon_validate_bo(&rmesa->radeon, rrb->bo,
+			0, RADEON_GEM_DOMAIN_VRAM);
    }
 
-   assert(drb);
-   assert(drb->flippedPitch);
+   for (i = 0; i < ctx->Const.MaxTextureImageUnits; ++i) {
+      radeonTexObj *t;
 
-   RADEON_STATECHANGE( rmesa, ctx );
+      if (!ctx->Texture.Unit[i]._ReallyEnabled)
+	 continue;
 
-   /* Note: we used the (possibly) page-flipped values */
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET]
-     = ((drb->flippedOffset + rmesa->radeonScreen->fbLocation)
-	& RADEON_COLOROFFSET_MASK);
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = drb->flippedPitch;
-   if (rmesa->sarea->tiling_enabled) {
-      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= RADEON_COLOR_TILE_ENABLE;
+      t = radeon_tex_obj(ctx->Texture.Unit[i]._Current);
+      if (t->image_override && t->bo)
+	radeon_validate_bo(&rmesa->radeon, t->bo,
+			   RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
+      else if (t->mt->bo)
+	radeon_validate_bo(&rmesa->radeon, t->mt->bo,
+			   RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0);
    }
-}
 
+   if (rmesa->radeon.dma.current)
+       radeon_validate_bo(&rmesa->radeon, rmesa->radeon.dma.current,
+			  RADEON_GEM_DOMAIN_GTT, 0);
+
+   return radeon_revalidate_bos(ctx);
+}
 
-void radeonValidateState( GLcontext *ctx )
+GLboolean radeonValidateState( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   GLuint new_state = rmesa->NewGLState;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   GLuint new_state = rmesa->radeon.NewGLState;
 
    if (new_state & _NEW_BUFFERS) {
-     radeonUpdateDrawBuffer(ctx);
+     _mesa_update_framebuffer(ctx);
+     /* this updates the DrawBuffer's Width/Height if it's a FBO */
+     _mesa_update_draw_buffer_bounds(ctx);
+     RADEON_STATECHANGE(rmesa, ctx);
    }
 
    if (new_state & _NEW_TEXTURE) {
       radeonUpdateTextureState( ctx );
-      new_state |= rmesa->NewGLState; /* may add TEXTURE_MATRIX */
+      new_state |= rmesa->radeon.NewGLState; /* may add TEXTURE_MATRIX */
    }
 
+   /* we need to do a space check here */
+   if (!r100ValidateBuffers(ctx))
+     return GL_FALSE;
+
    /* Need an event driven matrix update?
     */
-   if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION)) 
+   if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION))
       upload_matrix( rmesa, ctx->_ModelProjectMatrix.m, MODEL_PROJ );
 
    /* Need these for lighting (shouldn't upload otherwise)
@@ -2290,12 +2136,14 @@ void radeonValidateState( GLcontext *ctx )
    /* emit all active clip planes if projection matrix changes.
     */
    if (new_state & (_NEW_PROJECTION)) {
-      if (ctx->Transform.ClipPlanesEnabled) 
+      if (ctx->Transform.ClipPlanesEnabled)
 	 radeonUpdateClipPlanes( ctx );
    }
 
 
-   rmesa->NewGLState = 0;
+   rmesa->radeon.NewGLState = 0;
+
+   return GL_TRUE;
 }
 
 
@@ -2306,7 +2154,7 @@ static void radeonInvalidateState( GLcontext *ctx, GLuint new_state )
    _vbo_InvalidateState( ctx, new_state );
    _tnl_InvalidateState( ctx, new_state );
    _ae_invalidate_state( ctx, new_state );
-   RADEON_CONTEXT(ctx)->NewGLState |= new_state;
+   R100_CONTEXT(ctx)->radeon.NewGLState |= new_state;
 }
 
 
@@ -2317,8 +2165,8 @@ static GLboolean check_material( GLcontext *ctx )
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    GLint i;
 
-   for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT; 
-	i < _TNL_ATTRIB_MAT_BACK_INDEXES; 
+   for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT;
+	i < _TNL_ATTRIB_MAT_BACK_INDEXES;
 	i++)
       if (tnl->vb.AttribPtr[i] &&
 	  tnl->vb.AttribPtr[i]->stride)
@@ -2326,20 +2174,21 @@ static GLboolean check_material( GLcontext *ctx )
 
    return GL_FALSE;
 }
-      
+
 
 static void radeonWrapRunPipeline( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLboolean has_material;
 
    if (0)
-      fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->NewGLState);
+      fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->radeon.NewGLState);
 
    /* Validate state:
     */
-   if (rmesa->NewGLState)
-      radeonValidateState( ctx );
+   if (rmesa->radeon.NewGLState)
+      if (!radeonValidateState( ctx ))
+	 FALLBACK(rmesa, RADEON_FALLBACK_TEXTURE, GL_TRUE);
 
    has_material = (ctx->Light.Enabled && check_material( ctx ));
 
@@ -2348,7 +2197,7 @@ static void radeonWrapRunPipeline( GLcontext *ctx )
    }
 
    /* Run the pipeline.
-    */ 
+    */
    _tnl_run_pipeline( ctx );
 
    if (has_material) {
diff --git a/src/mesa/drivers/dri/radeon/radeon_state.h b/src/mesa/drivers/dri/radeon/radeon_state.h
index 2171879f759..a7c8eef32a5 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state.h
+++ b/src/mesa/drivers/dri/radeon/radeon_state.h
@@ -39,30 +39,25 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "radeon_context.h"
 
-extern void radeonInitState( radeonContextPtr rmesa );
+extern void radeonInitState( r100ContextPtr rmesa );
 extern void radeonInitStateFuncs( GLcontext *ctx );
 
 extern void radeonUpdateMaterial( GLcontext *ctx );
 
-extern void radeonSetCliprects( radeonContextPtr rmesa );
-extern void radeonRecalcScissorRects( radeonContextPtr rmesa );
 extern void radeonUpdateViewportOffset( GLcontext *ctx );
 extern void radeonUpdateWindow( GLcontext *ctx );
 extern void radeonUpdateDrawBuffer( GLcontext *ctx );
-extern void radeonUploadTexMatrix( radeonContextPtr rmesa,
+extern void radeonUploadTexMatrix( r100ContextPtr rmesa,
 				   int unit, GLboolean swapcols );
 
-extern void radeonValidateState( GLcontext *ctx );
-
-extern void radeonPrintDirty( radeonContextPtr rmesa,
-			      const char *msg );
+extern GLboolean radeonValidateState( GLcontext *ctx );
 
 
 extern void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode );
 #define FALLBACK( rmesa, bit, mode ) do {				\
    if ( 0 ) fprintf( stderr, "FALLBACK in %s: #%d=%d\n",		\
 		     __FUNCTION__, bit, mode );				\
-   radeonFallback( rmesa->glCtx, bit, mode );				\
+   radeonFallback( rmesa->radeon.glCtx, bit, mode );				\
 } while (0)
 
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c
index 57dc3800501..f5d4189d669 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state_init.c
+++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c
@@ -38,39 +38,140 @@
 #include "swrast_setup/swrast_setup.h"
 
 #include "radeon_context.h"
+#include "radeon_mipmap_tree.h"
 #include "radeon_ioctl.h"
 #include "radeon_state.h"
 #include "radeon_tcl.h"
 #include "radeon_tex.h"
 #include "radeon_swtcl.h"
 
+#include "../r200/r200_reg.h"
+
 #include "xmlpool.h"
 
+/* New (1.3) state mechanism.  3 commands (packet, scalar, vector) in
+ * 1.3 cmdbuffers allow all previous state to be updated as well as
+ * the tcl scalar and vector areas.
+ */
+static struct {
+	int start;
+	int len;
+	const char *name;
+} packet[RADEON_MAX_STATE_PACKETS] = {
+	{RADEON_PP_MISC, 7, "RADEON_PP_MISC"},
+	{RADEON_PP_CNTL, 3, "RADEON_PP_CNTL"},
+	{RADEON_RB3D_COLORPITCH, 1, "RADEON_RB3D_COLORPITCH"},
+	{RADEON_RE_LINE_PATTERN, 2, "RADEON_RE_LINE_PATTERN"},
+	{RADEON_SE_LINE_WIDTH, 1, "RADEON_SE_LINE_WIDTH"},
+	{RADEON_PP_LUM_MATRIX, 1, "RADEON_PP_LUM_MATRIX"},
+	{RADEON_PP_ROT_MATRIX_0, 2, "RADEON_PP_ROT_MATRIX_0"},
+	{RADEON_RB3D_STENCILREFMASK, 3, "RADEON_RB3D_STENCILREFMASK"},
+	{RADEON_SE_VPORT_XSCALE, 6, "RADEON_SE_VPORT_XSCALE"},
+	{RADEON_SE_CNTL, 2, "RADEON_SE_CNTL"},
+	{RADEON_SE_CNTL_STATUS, 1, "RADEON_SE_CNTL_STATUS"},
+	{RADEON_RE_MISC, 1, "RADEON_RE_MISC"},
+	{RADEON_PP_TXFILTER_0, 6, "RADEON_PP_TXFILTER_0"},
+	{RADEON_PP_BORDER_COLOR_0, 1, "RADEON_PP_BORDER_COLOR_0"},
+	{RADEON_PP_TXFILTER_1, 6, "RADEON_PP_TXFILTER_1"},
+	{RADEON_PP_BORDER_COLOR_1, 1, "RADEON_PP_BORDER_COLOR_1"},
+	{RADEON_PP_TXFILTER_2, 6, "RADEON_PP_TXFILTER_2"},
+	{RADEON_PP_BORDER_COLOR_2, 1, "RADEON_PP_BORDER_COLOR_2"},
+	{RADEON_SE_ZBIAS_FACTOR, 2, "RADEON_SE_ZBIAS_FACTOR"},
+	{RADEON_SE_TCL_OUTPUT_VTX_FMT, 11, "RADEON_SE_TCL_OUTPUT_VTX_FMT"},
+	{RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, 17,
+		    "RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED"},
+	{R200_PP_TXCBLEND_0, 4, "R200_PP_TXCBLEND_0"},
+	{R200_PP_TXCBLEND_1, 4, "R200_PP_TXCBLEND_1"},
+	{R200_PP_TXCBLEND_2, 4, "R200_PP_TXCBLEND_2"},
+	{R200_PP_TXCBLEND_3, 4, "R200_PP_TXCBLEND_3"},
+	{R200_PP_TXCBLEND_4, 4, "R200_PP_TXCBLEND_4"},
+	{R200_PP_TXCBLEND_5, 4, "R200_PP_TXCBLEND_5"},
+	{R200_PP_TXCBLEND_6, 4, "R200_PP_TXCBLEND_6"},
+	{R200_PP_TXCBLEND_7, 4, "R200_PP_TXCBLEND_7"},
+	{R200_SE_TCL_LIGHT_MODEL_CTL_0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0"},
+	{R200_PP_TFACTOR_0, 6, "R200_PP_TFACTOR_0"},
+	{R200_SE_VTX_FMT_0, 4, "R200_SE_VTX_FMT_0"},
+	{R200_SE_VAP_CNTL, 1, "R200_SE_VAP_CNTL"},
+	{R200_SE_TCL_MATRIX_SEL_0, 5, "R200_SE_TCL_MATRIX_SEL_0"},
+	{R200_SE_TCL_TEX_PROC_CTL_2, 5, "R200_SE_TCL_TEX_PROC_CTL_2"},
+	{R200_SE_TCL_UCP_VERT_BLEND_CTL, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL"},
+	{R200_PP_TXFILTER_0, 6, "R200_PP_TXFILTER_0"},
+	{R200_PP_TXFILTER_1, 6, "R200_PP_TXFILTER_1"},
+	{R200_PP_TXFILTER_2, 6, "R200_PP_TXFILTER_2"},
+	{R200_PP_TXFILTER_3, 6, "R200_PP_TXFILTER_3"},
+	{R200_PP_TXFILTER_4, 6, "R200_PP_TXFILTER_4"},
+	{R200_PP_TXFILTER_5, 6, "R200_PP_TXFILTER_5"},
+	{R200_PP_TXOFFSET_0, 1, "R200_PP_TXOFFSET_0"},
+	{R200_PP_TXOFFSET_1, 1, "R200_PP_TXOFFSET_1"},
+	{R200_PP_TXOFFSET_2, 1, "R200_PP_TXOFFSET_2"},
+	{R200_PP_TXOFFSET_3, 1, "R200_PP_TXOFFSET_3"},
+	{R200_PP_TXOFFSET_4, 1, "R200_PP_TXOFFSET_4"},
+	{R200_PP_TXOFFSET_5, 1, "R200_PP_TXOFFSET_5"},
+	{R200_SE_VTE_CNTL, 1, "R200_SE_VTE_CNTL"},
+	{R200_SE_TCL_OUTPUT_VTX_COMP_SEL, 1,
+	 "R200_SE_TCL_OUTPUT_VTX_COMP_SEL"},
+	{R200_PP_TAM_DEBUG3, 1, "R200_PP_TAM_DEBUG3"},
+	{R200_PP_CNTL_X, 1, "R200_PP_CNTL_X"},
+	{R200_RB3D_DEPTHXY_OFFSET, 1, "R200_RB3D_DEPTHXY_OFFSET"},
+	{R200_RE_AUX_SCISSOR_CNTL, 1, "R200_RE_AUX_SCISSOR_CNTL"},
+	{R200_RE_SCISSOR_TL_0, 2, "R200_RE_SCISSOR_TL_0"},
+	{R200_RE_SCISSOR_TL_1, 2, "R200_RE_SCISSOR_TL_1"},
+	{R200_RE_SCISSOR_TL_2, 2, "R200_RE_SCISSOR_TL_2"},
+	{R200_SE_VAP_CNTL_STATUS, 1, "R200_SE_VAP_CNTL_STATUS"},
+	{R200_SE_VTX_STATE_CNTL, 1, "R200_SE_VTX_STATE_CNTL"},
+	{R200_RE_POINTSIZE, 1, "R200_RE_POINTSIZE"},
+	{R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0, 4,
+		    "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0"},
+	{R200_PP_CUBIC_FACES_0, 1, "R200_PP_CUBIC_FACES_0"},	/* 61 */
+	{R200_PP_CUBIC_OFFSET_F1_0, 5, "R200_PP_CUBIC_OFFSET_F1_0"}, /* 62 */
+	{R200_PP_CUBIC_FACES_1, 1, "R200_PP_CUBIC_FACES_1"},
+	{R200_PP_CUBIC_OFFSET_F1_1, 5, "R200_PP_CUBIC_OFFSET_F1_1"},
+	{R200_PP_CUBIC_FACES_2, 1, "R200_PP_CUBIC_FACES_2"},
+	{R200_PP_CUBIC_OFFSET_F1_2, 5, "R200_PP_CUBIC_OFFSET_F1_2"},
+	{R200_PP_CUBIC_FACES_3, 1, "R200_PP_CUBIC_FACES_3"},
+	{R200_PP_CUBIC_OFFSET_F1_3, 5, "R200_PP_CUBIC_OFFSET_F1_3"},
+	{R200_PP_CUBIC_FACES_4, 1, "R200_PP_CUBIC_FACES_4"},
+	{R200_PP_CUBIC_OFFSET_F1_4, 5, "R200_PP_CUBIC_OFFSET_F1_4"},
+	{R200_PP_CUBIC_FACES_5, 1, "R200_PP_CUBIC_FACES_5"},
+	{R200_PP_CUBIC_OFFSET_F1_5, 5, "R200_PP_CUBIC_OFFSET_F1_5"},
+	{RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0"},
+	{RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1"},
+	{RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_2"},
+	{R200_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"},
+	{R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"},
+	{RADEON_PP_CUBIC_FACES_0, 1, "RADEON_PP_CUBIC_FACES_0"},
+	{RADEON_PP_CUBIC_OFFSET_T0_0, 5, "RADEON_PP_CUBIC_OFFSET_T0_0"},
+	{RADEON_PP_CUBIC_FACES_1, 1, "RADEON_PP_CUBIC_FACES_1"},
+	{RADEON_PP_CUBIC_OFFSET_T1_0, 5, "RADEON_PP_CUBIC_OFFSET_T1_0"},
+	{RADEON_PP_CUBIC_FACES_2, 1, "RADEON_PP_CUBIC_FACES_2"},
+	{RADEON_PP_CUBIC_OFFSET_T2_0, 5, "RADEON_PP_CUBIC_OFFSET_T2_0"},
+	{R200_PP_TRI_PERF, 2, "R200_PP_TRI_PERF"},
+	{R200_PP_TXCBLEND_8, 32, "R200_PP_AFS_0"},     /* 85 */
+	{R200_PP_TXCBLEND_0, 32, "R200_PP_AFS_1"},
+	{R200_PP_TFACTOR_0, 8, "R200_ATF_TFACTOR"},
+	{R200_PP_TXFILTER_0, 8, "R200_PP_TXCTLALL_0"},
+	{R200_PP_TXFILTER_1, 8, "R200_PP_TXCTLALL_1"},
+	{R200_PP_TXFILTER_2, 8, "R200_PP_TXCTLALL_2"},
+	{R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"},
+	{R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"},
+	{R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"},
+	{R200_VAP_PVS_CNTL_1, 2, "R200_VAP_PVS_CNTL"},
+};
+
 /* =============================================================
  * State initialization
  */
-
-void radeonPrintDirty( radeonContextPtr rmesa, const char *msg )
+static int cmdpkt( r100ContextPtr rmesa, int id ) 
 {
-   struct radeon_state_atom *l;
-
-   fprintf(stderr, msg);
-   fprintf(stderr, ": ");
+   drm_radeon_cmd_header_t h;
 
-   foreach(l, &rmesa->hw.atomlist) {
-      if (l->dirty || rmesa->hw.all_dirty)
-	 fprintf(stderr, "%s, ", l->name);
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+     return CP_PACKET0(packet[id].start, packet[id].len - 1);
+   } else {
+     h.i = 0;
+     h.packet.cmd_type = RADEON_CMD_PACKET;
+     h.packet.packet_id = id;
    }
-
-   fprintf(stderr, "\n");
-}
-
-static int cmdpkt( int id ) 
-{
-   drm_radeon_cmd_header_t h;
-   h.i = 0;
-   h.packet.cmd_type = RADEON_CMD_PACKET;
-   h.packet.packet_id = id;
    return h.i;
 }
 
@@ -96,17 +197,17 @@ static int cmdscl( int offset, int stride, int count )
    return h.i;
 }
 
-#define CHECK( NM, FLAG )			\
-static GLboolean check_##NM( GLcontext *ctx )	\
-{						\
-   return FLAG;					\
+#define CHECK( NM, FLAG )				\
+static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom )	\
+{							\
+   return FLAG ? atom->cmd_size : 0;			\
 }
 
 #define TCL_CHECK( NM, FLAG )				\
-static GLboolean check_##NM( GLcontext *ctx )		\
+static int check_##NM( GLcontext *ctx, struct radeon_state_atom *atom )	\
 {							\
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);	\
-   return !rmesa->TclFallback && (FLAG);		\
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);	\
+   return (!rmesa->radeon.TclFallback && (FLAG)) ? atom->cmd_size : 0;	\
 }
 
 
@@ -146,81 +247,392 @@ CHECK( txr0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT))
 CHECK( txr1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT))
 CHECK( txr2, (ctx->Texture.Unit[2]._ReallyEnabled & TEXTURE_RECT_BIT))
 
+#define OUT_VEC(hdr, data) do {			\
+    drm_radeon_cmd_header_t h;					\
+    h.i = hdr;								\
+    OUT_BATCH(CP_PACKET0(RADEON_SE_TCL_STATE_FLUSH, 0));		\
+    OUT_BATCH(0);							\
+    OUT_BATCH(CP_PACKET0(R200_SE_TCL_VECTOR_INDX_REG, 0));		\
+    OUT_BATCH(h.vectors.offset | (h.vectors.stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT)); \
+    OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_VECTOR_DATA_REG, h.vectors.count - 1));	\
+    OUT_BATCH_TABLE((data), h.vectors.count);				\
+  } while(0)
+
+#define OUT_SCL(hdr, data) do {					\
+    drm_radeon_cmd_header_t h;						\
+    h.i = hdr;								\
+    OUT_BATCH(CP_PACKET0(R200_SE_TCL_SCALAR_INDX_REG, 0));		\
+    OUT_BATCH((h.scalars.offset) | (h.scalars.stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT)); \
+    OUT_BATCH(CP_PACKET0_ONE(R200_SE_TCL_SCALAR_DATA_REG, h.scalars.count - 1));	\
+    OUT_BATCH_TABLE((data), h.scalars.count);				\
+  } while(0)
+
+static void scl_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   uint32_t dwords = atom->cmd_size;
+   
+   dwords += 2;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_SCL(atom->cmd[0], atom->cmd+1);
+   END_BATCH();
+}
 
 
-/* Initialize the context's hardware state.
- */
-void radeonInitState( radeonContextPtr rmesa )
+static void vec_emit(GLcontext *ctx, struct radeon_state_atom *atom)
 {
-   GLcontext *ctx = rmesa->glCtx;
-   GLuint color_fmt, depth_fmt, i;
-   GLint drawPitch, drawOffset;
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 4;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_VEC(atom->cmd[0], atom->cmd+1);
+   END_BATCH();
+}
 
-   switch ( rmesa->radeonScreen->cpp ) {
-   case 2:
-      color_fmt = RADEON_COLOR_FORMAT_RGB565;
-      break;
-   case 4:
-      color_fmt = RADEON_COLOR_FORMAT_ARGB8888;
-      break;
-   default:
-      fprintf( stderr, "Error: Unsupported pixel depth... exiting\n" );
-      exit( -1 );
+
+static void lit_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   uint32_t dwords = atom->cmd_size;
+
+   dwords += 6;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+   OUT_VEC(atom->cmd[LIT_CMD_0], atom->cmd+1);
+   OUT_SCL(atom->cmd[LIT_CMD_1], atom->cmd+LIT_CMD_1+1);
+   END_BATCH();
+}
+
+static void ctx_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   struct radeon_renderbuffer *rrb;
+   uint32_t cbpitch;
+   uint32_t zbpitch, depth_fmt;
+   uint32_t dwords = atom->cmd_size;
+
+   /* output the first 7 bytes of context */
+   BEGIN_BATCH_NO_AUTOSTATE(dwords + 4);
+   OUT_BATCH_TABLE(atom->cmd, 5);
+
+   rrb = radeon_get_depthbuffer(&r100->radeon);
+   if (!rrb) {
+     OUT_BATCH(0);
+     OUT_BATCH(0);
+   } else {
+     zbpitch = (rrb->pitch / rrb->cpp);
+     if (r100->using_hyperz)
+       zbpitch |= RADEON_DEPTH_HYPERZ;
+
+     OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+     OUT_BATCH(zbpitch);
+     if (rrb->cpp == 4)
+        depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
+     else
+        depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK;
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt;
+   }
+     
+   OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]);
+   OUT_BATCH(atom->cmd[CTX_CMD_1]);
+   OUT_BATCH(atom->cmd[CTX_PP_CNTL]);
+
+   rrb = radeon_get_colorbuffer(&r100->radeon);
+   if (!rrb || !rrb->bo) {
+      OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
+      OUT_BATCH(atom->cmd[CTX_RB3D_COLOROFFSET]);
+   } else {
+      atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10);
+      if (rrb->cpp == 4)
+         atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888;
+      else
+         atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565;
+
+      OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
+      OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
    }
 
-   rmesa->state.color.clear = 0x00000000;
+   OUT_BATCH(atom->cmd[CTX_CMD_2]);
+
+   if (!rrb || !rrb->bo) {
+     OUT_BATCH(atom->cmd[CTX_RB3D_COLORPITCH]);
+   } else {
+     cbpitch = (rrb->pitch / rrb->cpp);
+     if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
+       cbpitch |= RADEON_COLOR_TILE_ENABLE;
+     OUT_BATCH(cbpitch);
+   }
+
+   END_BATCH();
+}
+
+static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   struct radeon_renderbuffer *rrb, *drb;
+   uint32_t cbpitch = 0;
+   uint32_t zbpitch = 0;
+   uint32_t dwords = atom->cmd_size;
+   uint32_t depth_fmt;
+
+   rrb = radeon_get_colorbuffer(&r100->radeon);
+   if (!rrb || !rrb->bo) {
+      fprintf(stderr, "no rrb\n");
+      return;
+   }
+
+   atom->cmd[CTX_RB3D_CNTL] &= ~(0xf << 10);
+   if (rrb->cpp == 4)
+	atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_ARGB8888;
+   else
+	atom->cmd[CTX_RB3D_CNTL] |= RADEON_COLOR_FORMAT_RGB565;
+
+   cbpitch = (rrb->pitch / rrb->cpp);
+   if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE)
+       cbpitch |= R200_COLOR_TILE_ENABLE;
+
+   drb = radeon_get_depthbuffer(&r100->radeon);
+   if (drb) {
+     zbpitch = (drb->pitch / drb->cpp);
+     if (drb->cpp == 4)
+        depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
+     else
+        depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_DEPTH_FORMAT_MASK;
+     atom->cmd[CTX_RB3D_ZSTENCILCNTL] |= depth_fmt;
+     
+   }
+
+   /* output the first 7 bytes of context */
+   dwords = 10;
+   if (drb)
+     dwords += 6;
+   if (rrb)
+     dwords += 6;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+
+   /* In the CS case we need to split this up */
+   OUT_BATCH(CP_PACKET0(packet[0].start, 3));
+   OUT_BATCH_TABLE((atom->cmd + 1), 4);
+
+   if (drb) {
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHOFFSET, 0));
+     OUT_BATCH_RELOC(0, drb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_DEPTHPITCH, 0));
+     OUT_BATCH(zbpitch);
+   }
+
+   OUT_BATCH(CP_PACKET0(RADEON_RB3D_ZSTENCILCNTL, 0));
+   OUT_BATCH(atom->cmd[CTX_RB3D_ZSTENCILCNTL]);
+   OUT_BATCH(CP_PACKET0(RADEON_PP_CNTL, 1));
+   OUT_BATCH(atom->cmd[CTX_PP_CNTL]);
+   OUT_BATCH(atom->cmd[CTX_RB3D_CNTL]);
+
+   if (rrb) {
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLOROFFSET, 0));
+     OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+
+     OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0));
+     OUT_BATCH(cbpitch);
+   }
+
+   // if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) {
+   //   OUT_BATCH_TABLE((atom->cmd + 14), 4);
+   // }
+
+   END_BATCH();
+   BEGIN_BATCH_NO_AUTOSTATE(4);
+   OUT_BATCH(CP_PACKET0(RADEON_RE_TOP_LEFT, 0));
+   OUT_BATCH(0);
+   OUT_BATCH(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0));
+   if (rrb) {
+       OUT_BATCH(((rrb->width - 1) << RADEON_RE_WIDTH_SHIFT) |
+                 ((rrb->height - 1) << RADEON_RE_HEIGHT_SHIFT));
+   } else {
+       OUT_BATCH(0);
+   }
+   END_BATCH();
+}
+
+static void cube_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   uint32_t dwords = 2;
+   int i = atom->idx, j;
+   radeonTexObj *t = r100->state.texture.unit[i].texobj;
+   radeon_mipmap_level *lvl;
+   uint32_t base_reg;
+
+   if (!(ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT))
+	return;
+
+   if (!t)
+	return;
+
+   if (!t->mt)
+	return;
+
+   switch(i) {
+	case 1: base_reg = RADEON_PP_CUBIC_OFFSET_T1_0; break;
+	case 2: base_reg = RADEON_PP_CUBIC_OFFSET_T2_0; break;
+	default:
+	case 0: base_reg = RADEON_PP_CUBIC_OFFSET_T0_0; break;
+   };
+   BEGIN_BATCH_NO_AUTOSTATE(dwords + (5 * 4));
+   OUT_BATCH_TABLE(atom->cmd, 2);
+   lvl = &t->mt->levels[0];
+   for (j = 0; j < 5; j++) {
+	OUT_BATCH(CP_PACKET0(base_reg + (4 * (j-1)), 0));
+	OUT_BATCH_RELOC(lvl->faces[j].offset, t->mt->bo, lvl->faces[j].offset,
+			RADEON_GEM_DOMAIN_VRAM, 0, 0);
+   }
+   END_BATCH();
+}
+
+static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   uint32_t dwords = atom->cmd_size;
+   int i = atom->idx;
+   radeonTexObj *t = r100->state.texture.unit[i].texobj;
+   radeon_mipmap_level *lvl;
+
+   if (t && t->mt && !t->image_override)
+     dwords += 2;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+
+   OUT_BATCH_TABLE(atom->cmd, 3);
+   if (t && t->mt && !t->image_override) {
+     if ((ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) {
+   	lvl = &t->mt->levels[0];
+	OUT_BATCH_RELOC(lvl->faces[5].offset, t->mt->bo, lvl->faces[5].offset,
+			RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+     } else {
+        OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+		     RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+     }
+   } else if (!t) {
+     /* workaround for old CS mechanism */
+     OUT_BATCH(r100->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP]);
+     //     OUT_BATCH(r100->radeon.radeonScreen);
+   } else {
+     OUT_BATCH(t->override_offset);
+   }
+
+   OUT_BATCH_TABLE((atom->cmd+4), 5);
+   END_BATCH();
+}
+
+static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+   r100ContextPtr r100 = R100_CONTEXT(ctx);
+   BATCH_LOCALS(&r100->radeon);
+   uint32_t dwords = atom->cmd_size;
+   int i = atom->idx;
+   radeonTexObj *t = r100->state.texture.unit[i].texobj;
+   radeon_mipmap_level *lvl;
+   int hastexture = 1;
+
+   if (!t)
+	hastexture = 0;
+   else {
+	if (!t->mt && !t->bo)
+		hastexture = 0;
+   }
+   dwords += 1;
+   if (hastexture)
+     dwords += 2;
+   else
+     dwords -= 2;
+   BEGIN_BATCH_NO_AUTOSTATE(dwords);
+
+   OUT_BATCH(CP_PACKET0(RADEON_PP_TXFILTER_0 + (24 * i), 1));
+   OUT_BATCH_TABLE((atom->cmd + 1), 2);
+
+   if (hastexture) {
+     OUT_BATCH(CP_PACKET0(RADEON_PP_TXOFFSET_0 + (24 * i), 0));
+     if (t->mt && !t->image_override) {
+        if ((ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) {
+            lvl = &t->mt->levels[0];
+	    OUT_BATCH_RELOC(lvl->faces[5].offset, t->mt->bo, lvl->faces[5].offset,
+			RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+        } else {
+           OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+		     RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+        }
+      } else {
+	if (t->bo)
+            OUT_BATCH_RELOC(t->tile_bits, t->bo, 0,
+                            RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+      }
+   }
+
+   OUT_BATCH(CP_PACKET0(RADEON_PP_TXCBLEND_0 + (i * 24), 1));
+   OUT_BATCH_TABLE((atom->cmd+4), 2);
+   OUT_BATCH(CP_PACKET0(RADEON_PP_BORDER_COLOR_0 + (i * 4), 0));
+   OUT_BATCH((atom->cmd[TEX_PP_BORDER_COLOR]));
+   END_BATCH();
+}
+
+/* Initialize the context's hardware state.
+ */
+void radeonInitState( r100ContextPtr rmesa )
+{
+   GLcontext *ctx = rmesa->radeon.glCtx;
+   GLuint i;
+
+   rmesa->radeon.state.color.clear = 0x00000000;
 
    switch ( ctx->Visual.depthBits ) {
    case 16:
-      rmesa->state.depth.clear = 0x0000ffff;
-      rmesa->state.depth.scale = 1.0 / (GLfloat)0xffff;
-      depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
-      rmesa->state.stencil.clear = 0x00000000;
+      rmesa->radeon.state.depth.clear = 0x0000ffff;
+      rmesa->radeon.state.stencil.clear = 0x00000000;
       break;
    case 24:
-      rmesa->state.depth.clear = 0x00ffffff;
-      rmesa->state.depth.scale = 1.0 / (GLfloat)0xffffff;
-      depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
-      rmesa->state.stencil.clear = 0xffff0000;
+      rmesa->radeon.state.depth.clear = 0x00ffffff;
+      rmesa->radeon.state.stencil.clear = 0xffff0000;
       break;
    default:
-      fprintf( stderr, "Error: Unsupported depth %d... exiting\n",
-	       ctx->Visual.depthBits );
-      exit( -1 );
+      break;
    }
 
-   /* Only have hw stencil when depth buffer is 24 bits deep */
-   rmesa->state.stencil.hwBuffer = ( ctx->Visual.stencilBits > 0 &&
-				     ctx->Visual.depthBits == 24 );
+   rmesa->radeon.Fallback = 0;
 
-   rmesa->Fallback = 0;
-
-   if ( ctx->Visual.doubleBufferMode && rmesa->sarea->pfCurrentPage == 0 ) {
-      drawOffset = rmesa->radeonScreen->backOffset;
-      drawPitch  = rmesa->radeonScreen->backPitch;
-   } else {
-      drawOffset = rmesa->radeonScreen->frontOffset;
-      drawPitch  = rmesa->radeonScreen->frontPitch;
-   }
 
-   rmesa->hw.max_state_size = 0;
+   rmesa->radeon.hw.max_state_size = 0;
 
-#define ALLOC_STATE( ATOM, CHK, SZ, NM, FLAG )				\
+#define ALLOC_STATE_IDX( ATOM, CHK, SZ, NM, FLAG, IDX )		\
    do {								\
       rmesa->hw.ATOM.cmd_size = SZ;				\
-      rmesa->hw.ATOM.cmd = (int *)CALLOC(SZ * sizeof(int));	\
-      rmesa->hw.ATOM.lastcmd = (int *)CALLOC(SZ * sizeof(int));	\
-      rmesa->hw.ATOM.name = NM;					\
+      rmesa->hw.ATOM.cmd = (GLuint *)CALLOC(SZ * sizeof(int));	\
+      rmesa->hw.ATOM.lastcmd = (GLuint *)CALLOC(SZ * sizeof(int)); \
+      rmesa->hw.ATOM.name = NM;						\
       rmesa->hw.ATOM.is_tcl = FLAG;					\
       rmesa->hw.ATOM.check = check_##CHK;				\
-      rmesa->hw.ATOM.dirty = GL_TRUE;				\
-      rmesa->hw.max_state_size += SZ * sizeof(int);		\
+      rmesa->hw.ATOM.dirty = GL_TRUE;					\
+      rmesa->hw.ATOM.idx = IDX;					\
+      rmesa->radeon.hw.max_state_size += SZ * sizeof(int);		\
    } while (0)
-      
-      
+
+#define ALLOC_STATE( ATOM, CHK, SZ, NM, FLAG )		\
+   ALLOC_STATE_IDX(ATOM, CHK, SZ, NM, FLAG, 0)
+
    /* Allocate state buffers:
     */
    ALLOC_STATE( ctx, always, CTX_STATE_SIZE, "CTX/context", 0 );
+   if (rmesa->radeon.radeonScreen->kernel_mm)
+     rmesa->hw.ctx.emit = ctx_emit_cs;
+   else
+     rmesa->hw.ctx.emit = ctx_emit;
    ALLOC_STATE( lin, always, LIN_STATE_SIZE, "LIN/line", 0 );
    ALLOC_STATE( msk, always, MSK_STATE_SIZE, "MSK/mask", 0 );
    ALLOC_STATE( vpt, always, VPT_STATE_SIZE, "VPT/viewport", 0 );
@@ -233,20 +645,29 @@ void radeonInitState( radeonContextPtr rmesa )
    ALLOC_STATE( fog, fog, FOG_STATE_SIZE, "FOG/fog", 1 );
    ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 1 );
    ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 1 );
-   ALLOC_STATE( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0 );
-   ALLOC_STATE( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0 );
-   ALLOC_STATE( tex[2], tex2, TEX_STATE_SIZE, "TEX/tex-2", 0 );
-   if (rmesa->radeonScreen->drmSupportsCubeMapsR100)
+   ALLOC_STATE_IDX( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0, 0);
+   ALLOC_STATE_IDX( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0, 1);
+   ALLOC_STATE_IDX( tex[2], tex2, TEX_STATE_SIZE, "TEX/tex-2", 0, 2);
+
+   for (i = 0; i < 3; i++) {
+      if (rmesa->radeon.radeonScreen->kernel_mm)
+          rmesa->hw.tex[i].emit = tex_emit_cs;
+      else
+          rmesa->hw.tex[i].emit = tex_emit;
+   }
+   if (rmesa->radeon.radeonScreen->drmSupportsCubeMapsR100)
    {
-      ALLOC_STATE( cube[0], cube0, CUBE_STATE_SIZE, "CUBE/cube-0", 0 );
-      ALLOC_STATE( cube[1], cube1, CUBE_STATE_SIZE, "CUBE/cube-1", 0 );
-      ALLOC_STATE( cube[2], cube2, CUBE_STATE_SIZE, "CUBE/cube-2", 0 );
+      ALLOC_STATE_IDX( cube[0], cube0, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 );
+      ALLOC_STATE_IDX( cube[1], cube1, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 );
+      ALLOC_STATE_IDX( cube[2], cube2, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 );
+      for (i = 0; i < 3; i++)
+         rmesa->hw.cube[i].emit = cube_emit;
    }
    else
    {
-      ALLOC_STATE( cube[0], never, CUBE_STATE_SIZE, "CUBE/cube-0", 0 );
-      ALLOC_STATE( cube[1], never, CUBE_STATE_SIZE, "CUBE/cube-1", 0 );
-      ALLOC_STATE( cube[2], never, CUBE_STATE_SIZE, "CUBE/cube-2", 0 );
+      ALLOC_STATE_IDX( cube[0], never, CUBE_STATE_SIZE, "CUBE/cube-0", 0, 0 );
+      ALLOC_STATE_IDX( cube[1], never, CUBE_STATE_SIZE, "CUBE/cube-1", 0, 1 );
+      ALLOC_STATE_IDX( cube[2], never, CUBE_STATE_SIZE, "CUBE/cube-2", 0, 2 );
    }
    ALLOC_STATE( mat[0], tcl, MAT_STATE_SIZE, "MAT/modelproject", 1 );
    ALLOC_STATE( mat[1], tcl_eyespace_or_fog, MAT_STATE_SIZE, "MAT/modelview", 1 );
@@ -268,43 +689,43 @@ void radeonInitState( radeonContextPtr rmesa )
    ALLOC_STATE( lit[5], tcl_lit5, LIT_STATE_SIZE, "LIT/light-5", 1 );
    ALLOC_STATE( lit[6], tcl_lit6, LIT_STATE_SIZE, "LIT/light-6", 1 );
    ALLOC_STATE( lit[7], tcl_lit7, LIT_STATE_SIZE, "LIT/light-7", 1 );
-   ALLOC_STATE( txr[0], txr0, TXR_STATE_SIZE, "TXR/txr-0", 0 );
-   ALLOC_STATE( txr[1], txr1, TXR_STATE_SIZE, "TXR/txr-1", 0 );
-   ALLOC_STATE( txr[2], txr2, TXR_STATE_SIZE, "TXR/txr-2", 0 );
+   ALLOC_STATE_IDX( txr[0], txr0, TXR_STATE_SIZE, "TXR/txr-0", 0, 0 );
+   ALLOC_STATE_IDX( txr[1], txr1, TXR_STATE_SIZE, "TXR/txr-1", 0, 1 );
+   ALLOC_STATE_IDX( txr[2], txr2, TXR_STATE_SIZE, "TXR/txr-2", 0, 2 );
 
    radeonSetUpAtomList( rmesa );
 
    /* Fill in the packet headers:
     */
-   rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(RADEON_EMIT_PP_MISC);
-   rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(RADEON_EMIT_PP_CNTL);
-   rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(RADEON_EMIT_RB3D_COLORPITCH);
-   rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(RADEON_EMIT_RE_LINE_PATTERN);
-   rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(RADEON_EMIT_SE_LINE_WIDTH);
-   rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(RADEON_EMIT_RB3D_STENCILREFMASK);
-   rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(RADEON_EMIT_SE_VPORT_XSCALE);
-   rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(RADEON_EMIT_SE_CNTL);
-   rmesa->hw.set.cmd[SET_CMD_1] = cmdpkt(RADEON_EMIT_SE_CNTL_STATUS);
-   rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(RADEON_EMIT_RE_MISC);
-   rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_0);
-   rmesa->hw.tex[0].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_0);
-   rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_1);
-   rmesa->hw.tex[1].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_1);
-   rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_2);
-   rmesa->hw.tex[2].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_2);
-   rmesa->hw.cube[0].cmd[CUBE_CMD_0] = cmdpkt(RADEON_EMIT_PP_CUBIC_FACES_0);
-   rmesa->hw.cube[0].cmd[CUBE_CMD_1] = cmdpkt(RADEON_EMIT_PP_CUBIC_OFFSETS_T0);
-   rmesa->hw.cube[1].cmd[CUBE_CMD_0] = cmdpkt(RADEON_EMIT_PP_CUBIC_FACES_1);
-   rmesa->hw.cube[1].cmd[CUBE_CMD_1] = cmdpkt(RADEON_EMIT_PP_CUBIC_OFFSETS_T1);
-   rmesa->hw.cube[2].cmd[CUBE_CMD_0] = cmdpkt(RADEON_EMIT_PP_CUBIC_FACES_2);
-   rmesa->hw.cube[2].cmd[CUBE_CMD_1] = cmdpkt(RADEON_EMIT_PP_CUBIC_OFFSETS_T2);
-   rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(RADEON_EMIT_SE_ZBIAS_FACTOR);
-   rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT);
+   rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_MISC);
+   rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CNTL);
+   rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(rmesa, RADEON_EMIT_RB3D_COLORPITCH);
+   rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RE_LINE_PATTERN);
+   rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_SE_LINE_WIDTH);
+   rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RB3D_STENCILREFMASK);
+   rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_VPORT_XSCALE);
+   rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_CNTL);
+   rmesa->hw.set.cmd[SET_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_SE_CNTL_STATUS);
+   rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_RE_MISC);
+   rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TXFILTER_0);
+   rmesa->hw.tex[0].cmd[TEX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_BORDER_COLOR_0);
+   rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TXFILTER_1);
+   rmesa->hw.tex[1].cmd[TEX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_BORDER_COLOR_1);
+   rmesa->hw.tex[2].cmd[TEX_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TXFILTER_2);
+   rmesa->hw.tex[2].cmd[TEX_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_BORDER_COLOR_2);
+   rmesa->hw.cube[0].cmd[CUBE_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_FACES_0);
+   rmesa->hw.cube[0].cmd[CUBE_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_OFFSETS_T0);
+   rmesa->hw.cube[1].cmd[CUBE_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_FACES_1);
+   rmesa->hw.cube[1].cmd[CUBE_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_OFFSETS_T1);
+   rmesa->hw.cube[2].cmd[CUBE_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_FACES_2);
+   rmesa->hw.cube[2].cmd[CUBE_CMD_1] = cmdpkt(rmesa, RADEON_EMIT_PP_CUBIC_OFFSETS_T2);
+   rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_ZBIAS_FACTOR);
+   rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT);
    rmesa->hw.mtl.cmd[MTL_CMD_0] = 
-      cmdpkt(RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED);
-   rmesa->hw.txr[0].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_0);
-   rmesa->hw.txr[1].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_1);
-   rmesa->hw.txr[2].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_2);
+      cmdpkt(rmesa, RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED);
+   rmesa->hw.txr[0].cmd[TXR_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TEX_SIZE_0);
+   rmesa->hw.txr[1].cmd[TXR_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TEX_SIZE_1);
+   rmesa->hw.txr[2].cmd[TXR_CMD_0] = cmdpkt(rmesa, RADEON_EMIT_PP_TEX_SIZE_2);
    rmesa->hw.grd.cmd[GRD_CMD_0] = 
       cmdscl( RADEON_SS_VERT_GUARD_CLIP_ADJ_ADDR, 1, 4 );
    rmesa->hw.fog.cmd[FOG_CMD_0] = 
@@ -331,6 +752,22 @@ void radeonInitState( radeonContextPtr rmesa )
 	 cmdvec( RADEON_VS_UCP_ADDR + i, 1, 4 );
    }
 
+   if (rmesa->radeon.radeonScreen->kernel_mm) {
+      rmesa->hw.grd.emit = scl_emit;
+      rmesa->hw.fog.emit = vec_emit;
+      rmesa->hw.glt.emit = vec_emit;
+      rmesa->hw.eye.emit = vec_emit;
+      
+      for (i = 0; i <= 6; i++)
+	 rmesa->hw.mat[i].emit = vec_emit;
+
+      for (i = 0; i < 8; i++)
+	 rmesa->hw.lit[i].emit = lit_emit;
+
+      for (i = 0; i < 6; i++)
+	 rmesa->hw.ucp[i].emit = vec_emit;
+   }
+
    rmesa->last_ReallyEnabled = -1;
 
    /* Initial Harware state:
@@ -352,19 +789,7 @@ void radeonInitState( radeonContextPtr rmesa )
 					    RADEON_SRC_BLEND_GL_ONE |
 					    RADEON_DST_BLEND_GL_ZERO );
 
-   rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHOFFSET] =
-      rmesa->radeonScreen->depthOffset + rmesa->radeonScreen->fbLocation;
-
-   rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] = 
-      ((rmesa->radeonScreen->depthPitch &
-	RADEON_DEPTHPITCH_MASK) |
-       RADEON_DEPTH_ENDIAN_NO_SWAP);
-       
-   if (rmesa->using_hyperz)
-       rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] |= RADEON_DEPTH_HYPERZ;
-
-   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (depth_fmt |
-					       RADEON_Z_TEST_LESS |
+   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (RADEON_Z_TEST_LESS |
 					       RADEON_STENCIL_TEST_ALWAYS |
 					       RADEON_STENCIL_FAIL_KEEP |
 					       RADEON_STENCIL_ZPASS_KEEP |
@@ -374,7 +799,7 @@ void radeonInitState( radeonContextPtr rmesa )
    if (rmesa->using_hyperz) {
        rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_COMPRESSION_ENABLE |
 						   RADEON_Z_DECOMPRESSION_ENABLE;
-      if (rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
+      if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
 	 /* works for q3, but slight rendering errors with glxgears ? */
 /*	 rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_HIERARCHY_ENABLE;*/
 	 /* need this otherwise get lots of lockups with q3 ??? */
@@ -386,10 +811,9 @@ void radeonInitState( radeonContextPtr rmesa )
 				     RADEON_ANTI_ALIAS_NONE);
 
    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = (RADEON_PLANE_MASK_ENABLE |
-				       color_fmt |
 				       RADEON_ZBLOCK16);
 
-   switch ( driQueryOptioni( &rmesa->optionCache, "dither_mode" ) ) {
+   switch ( driQueryOptioni( &rmesa->radeon.optionCache, "dither_mode" ) ) {
    case DRI_CONF_DITHER_XERRORDIFFRESET:
       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_DITHER_INIT;
       break;
@@ -397,31 +821,18 @@ void radeonInitState( radeonContextPtr rmesa )
       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_SCALE_DITHER_ENABLE;
       break;
    }
-   if ( driQueryOptioni( &rmesa->optionCache, "round_mode" ) ==
+   if ( driQueryOptioni( &rmesa->radeon.optionCache, "round_mode" ) ==
 	DRI_CONF_ROUND_ROUND )
-      rmesa->state.color.roundEnable = RADEON_ROUND_ENABLE;
+      rmesa->radeon.state.color.roundEnable = RADEON_ROUND_ENABLE;
    else
-      rmesa->state.color.roundEnable = 0;
-   if ( driQueryOptioni (&rmesa->optionCache, "color_reduction" ) ==
+      rmesa->radeon.state.color.roundEnable = 0;
+   if ( driQueryOptioni (&rmesa->radeon.optionCache, "color_reduction" ) ==
 	DRI_CONF_COLOR_REDUCTION_DITHER )
       rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= RADEON_DITHER_ENABLE;
    else
-      rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->state.color.roundEnable;
-
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = ((drawOffset +
-					       rmesa->radeonScreen->fbLocation)
-					      & RADEON_COLOROFFSET_MASK);
-
-   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = ((drawPitch &
-					      RADEON_COLORPITCH_MASK) |
-					     RADEON_COLOR_ENDIAN_NO_SWAP);
+      rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |= rmesa->radeon.state.color.roundEnable;
 
 
-   /* (fixed size) sarea is initialized to zero afaics so can omit version check. Phew! */
-   if (rmesa->sarea->tiling_enabled) {
-      rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= RADEON_COLOR_TILE_ENABLE;
-   }
-
    rmesa->hw.set.cmd[SET_SE_CNTL] = (RADEON_FFACE_CULL_CCW |
 				     RADEON_BFACE_SOLID |
 				     RADEON_FFACE_SOLID |
@@ -444,7 +855,7 @@ void radeonInitState( radeonContextPtr rmesa )
   					    RADEON_VC_NO_SWAP;
 #endif
 
-   if (!(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
+   if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) {
      rmesa->hw.set.cmd[SET_SE_CNTL_STATUS] |= RADEON_TCL_BYPASS;
    }
 
@@ -491,8 +902,8 @@ void radeonInitState( radeonContextPtr rmesa )
 	   (2 << RADEON_TXFORMAT_HEIGHT_SHIFT));
 
       /* Initialize the texture offset to the start of the card texture heap */
-      rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET] =
-	  rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
+      //      rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET] =
+      //	  rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
 
       rmesa->hw.tex[i].cmd[TEX_PP_BORDER_COLOR] = 0;
       rmesa->hw.tex[i].cmd[TEX_PP_TXCBLEND] =  
@@ -513,15 +924,15 @@ void radeonInitState( radeonContextPtr rmesa )
 
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_FACES] = 0;
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_0] =
-	  rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	  rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_1] =
-	  rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	  rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_2] =
-	  rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	  rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_3] =
-	  rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	  rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
       rmesa->hw.cube[i].cmd[CUBE_PP_CUBIC_OFFSET_4] =
-	  rmesa->radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
+	  rmesa->radeon.radeonScreen->texOffset[RADEON_LOCAL_TEX_HEAP];
    }
 
    /* Can only add ST1 at the time of doing some multitex but can keep
@@ -613,5 +1024,7 @@ void radeonInitState( radeonContextPtr rmesa )
    rmesa->hw.eye.cmd[EYE_Z] = IEEE_ONE;
    rmesa->hw.eye.cmd[EYE_RESCALE_FACTOR] = IEEE_ONE;
    
-   rmesa->hw.all_dirty = GL_TRUE;
+   rmesa->radeon.hw.all_dirty = GL_TRUE;
+
+   rcommonInitCmdBuf(&rmesa->radeon);
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.c b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
index ebea1fecdca..e31f045991c 100644
--- a/src/mesa/drivers/dri/radeon/radeon_swtcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
@@ -52,8 +52,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_tcl.h"
 
 
-static void flush_last_swtcl_prim( radeonContextPtr rmesa  );
-
 /* R100: xyzw, c0, c1/fog, stq[0..2]  = 4+1+1+3*3 = 15  right? */
 /* R200: xyzw, c0, c1/fog, strq[0..5] = 4+1+1+4*6 = 30 */
 #define RADEON_MAX_TNL_VERTEX_SIZE (15 * sizeof(GLfloat))	/* for mesa _tnl stage */
@@ -64,18 +62,18 @@ static void flush_last_swtcl_prim( radeonContextPtr rmesa  );
 
 #define EMIT_ATTR( ATTR, STYLE, F0 )					\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = (ATTR);	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = (STYLE);	\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR);	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE);	\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
    fmt_0 |= F0;								\
 } while (0)
 
 #define EMIT_PAD( N )							\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = 0;		\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = EMIT_PAD;	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].offset = (N);		\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0;		\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD;	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N);		\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
 } while (0)
 
 static GLuint radeon_cp_vc_frmts[3][2] =
@@ -87,7 +85,7 @@ static GLuint radeon_cp_vc_frmts[3][2] =
 
 static void radeonSetVertexFormat( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   r100ContextPtr rmesa = R100_CONTEXT( ctx );
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    DECLARE_RENDERINPUTS(index_bitset);
@@ -106,7 +104,7 @@ static void radeonSetVertexFormat( GLcontext *ctx )
    }
 
    assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
-   rmesa->swtcl.vertex_attr_count = 0;
+   rmesa->radeon.swtcl.vertex_attr_count = 0;
 
    /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
     * build up a hardware vertex.
@@ -204,33 +202,33 @@ static void radeonSetVertexFormat( GLcontext *ctx )
       }
    }
 
-   if (!RENDERINPUTS_EQUAL( rmesa->tnl_index_bitset, index_bitset ) ||
+   if (!RENDERINPUTS_EQUAL( rmesa->radeon.tnl_index_bitset, index_bitset ) ||
 	fmt_0 != rmesa->swtcl.vertex_format) {
       RADEON_NEWPRIM(rmesa);
       rmesa->swtcl.vertex_format = fmt_0;
-      rmesa->swtcl.vertex_size =
+      rmesa->radeon.swtcl.vertex_size =
 	  _tnl_install_attrs( ctx,
-			      rmesa->swtcl.vertex_attrs, 
-			      rmesa->swtcl.vertex_attr_count,
+			      rmesa->radeon.swtcl.vertex_attrs, 
+			      rmesa->radeon.swtcl.vertex_attr_count,
 			      NULL, 0 );
-      rmesa->swtcl.vertex_size /= 4;
-      RENDERINPUTS_COPY( rmesa->tnl_index_bitset, index_bitset );
+      rmesa->radeon.swtcl.vertex_size /= 4;
+      RENDERINPUTS_COPY( rmesa->radeon.tnl_index_bitset, index_bitset );
       if (RADEON_DEBUG & DEBUG_VERTS)
 	 fprintf( stderr, "%s: vertex_size= %d floats\n",
-		  __FUNCTION__, rmesa->swtcl.vertex_size);
+		  __FUNCTION__, rmesa->radeon.swtcl.vertex_size);
    }
 }
 
 
 static void radeonRenderStart( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   r100ContextPtr rmesa = R100_CONTEXT( ctx );
 
    radeonSetVertexFormat( ctx );
    
-   if (rmesa->dma.flush != 0 && 
-       rmesa->dma.flush != flush_last_swtcl_prim)
-      rmesa->dma.flush( rmesa );
+   if (rmesa->radeon.dma.flush != 0 && 
+       rmesa->radeon.dma.flush != rcommon_flush_last_swtcl_prim)
+      rmesa->radeon.dma.flush( ctx );
 }
 
 
@@ -241,7 +239,7 @@ static void radeonRenderStart( GLcontext *ctx )
  */
 void radeonChooseVertexState( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   r100ContextPtr rmesa = R100_CONTEXT( ctx );
    TNLcontext *tnl = TNL_CONTEXT(ctx);
 
    GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
@@ -254,7 +252,7 @@ void radeonChooseVertexState( GLcontext *ctx )
     * rasterization fallback.  As this function will be called again when we
     * leave a rasterization fallback, we can just skip it for now.
     */
-   if (rmesa->Fallback != 0)
+   if (rmesa->radeon.Fallback != 0)
       return;
 
    /* HW perspective divide is a win, but tiny vertex formats are a
@@ -281,80 +279,29 @@ void radeonChooseVertexState( GLcontext *ctx )
    }
 }
 
-
-/* Flush vertices in the current dma region.
- */
-static void flush_last_swtcl_prim( radeonContextPtr rmesa  )
+void r100_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
 {
-   if (RADEON_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   rmesa->dma.flush = NULL;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
-   if (rmesa->dma.current.buf) {
-      struct radeon_dma_region *current = &rmesa->dma.current;
-      GLuint current_offset = (rmesa->radeonScreen->gart_buffer_offset +
-			       current->buf->buf->idx * RADEON_BUFFER_SIZE + 
-			       current->start);
+   rcommonEnsureCmdBufSpace(&rmesa->radeon,
+			    rmesa->radeon.hw.max_state_size + (12*sizeof(int)),
+			    __FUNCTION__);
 
-      assert (!(rmesa->swtcl.hw_primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
 
-      assert (current->start + 
-	      rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-	      current->ptr);
+   radeonEmitState(&rmesa->radeon);
+   radeonEmitVertexAOS( rmesa,
+			rmesa->radeon.swtcl.vertex_size,
+			rmesa->radeon.dma.current,
+			current_offset);
 
-      if (rmesa->dma.current.start != rmesa->dma.current.ptr) {
-	 radeonEnsureCmdBufSpace( rmesa, VERT_AOS_BUFSZ +
-			          rmesa->hw.max_state_size + VBUF_BUFSZ );
-
-	 radeonEmitVertexAOS( rmesa,
-			      rmesa->swtcl.vertex_size,
-			      current_offset);
-
-	 radeonEmitVbufPrim( rmesa,
-			     rmesa->swtcl.vertex_format,
-			     rmesa->swtcl.hw_primitive,
-			     rmesa->swtcl.numverts);
-      }
+		      
+   radeonEmitVbufPrim( rmesa,
+		       rmesa->swtcl.vertex_format,
+		       rmesa->radeon.swtcl.hw_primitive,
+		       rmesa->radeon.swtcl.numverts);
 
-      rmesa->swtcl.numverts = 0;
-      current->start = current->ptr;
-   }
 }
 
-
-/* Alloc space in the current dma region.
- */
-static INLINE void *
-radeonAllocDmaLowVerts( radeonContextPtr rmesa, int nverts, int vsize )
-{
-   GLuint bytes = vsize * nverts;
-
-   if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) 
-      radeonRefillCurrentDmaRegion( rmesa );
-
-   if (!rmesa->dma.flush) {
-      rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
-      rmesa->dma.flush = flush_last_swtcl_prim;
-   }
-
-   assert( vsize == rmesa->swtcl.vertex_size * 4 );
-   assert( rmesa->dma.flush == flush_last_swtcl_prim );
-   assert (rmesa->dma.current.start + 
-	   rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-	   rmesa->dma.current.ptr);
-
-
-   {
-      GLubyte *head = (GLubyte *)(rmesa->dma.current.address + rmesa->dma.current.ptr);
-      rmesa->dma.current.ptr += bytes;
-      rmesa->swtcl.numverts += nverts;
-      return head;
-   }
-
-}
-
-
 /*
  * Render unclipped vertex buffers by emitting vertices directly to
  * dma buffers.  Use strip/fan hardware primitives where possible.
@@ -387,22 +334,22 @@ static const GLuint hw_prim[GL_POLYGON+1] = {
 };
 
 static INLINE void
-radeonDmaPrimitive( radeonContextPtr rmesa, GLenum prim )
+radeonDmaPrimitive( r100ContextPtr rmesa, GLenum prim )
 {
    RADEON_NEWPRIM( rmesa );
-   rmesa->swtcl.hw_primitive = hw_prim[prim];
-   assert(rmesa->dma.current.ptr == rmesa->dma.current.start);
+   rmesa->radeon.swtcl.hw_primitive = hw_prim[prim];
+   //   assert(rmesa->radeon.dma.current.ptr == rmesa->radeon.dma.current.start);
 }
 
-#define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx)
+#define LOCAL_VARS r100ContextPtr rmesa = R100_CONTEXT(ctx)
 #define INIT( prim ) radeonDmaPrimitive( rmesa, prim )
 #define FLUSH()  RADEON_NEWPRIM( rmesa )
-#define GET_CURRENT_VB_MAX_VERTS() \
-  (((int)rmesa->dma.current.end - (int)rmesa->dma.current.ptr) / (rmesa->swtcl.vertex_size*4))
+#define GET_CURRENT_VB_MAX_VERTS()					10\
+//  (((int)rmesa->radeon.dma.current.end - (int)rmesa->radeon.dma.current.ptr) / (rmesa->radeon.swtcl.vertex_size*4))
 #define GET_SUBSEQUENT_VB_MAX_VERTS() \
-  ((RADEON_BUFFER_SIZE) / (rmesa->swtcl.vertex_size*4))
+  ((RADEON_BUFFER_SIZE) / (rmesa->radeon.swtcl.vertex_size*4))
 #define ALLOC_VERTS( nr ) \
-  radeonAllocDmaLowVerts( rmesa, nr, rmesa->swtcl.vertex_size * 4 )
+  rcommonAllocDmaLowVerts( &rmesa->radeon, nr, rmesa->radeon.swtcl.vertex_size * 4 )
 #define EMIT_VERTS( ctx, j, nr, buf ) \
   _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf)
 
@@ -418,16 +365,13 @@ radeonDmaPrimitive( radeonContextPtr rmesa, GLenum prim )
 static GLboolean radeon_run_render( GLcontext *ctx,
 				    struct tnl_pipeline_stage *stage )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    tnl_render_func *tab = TAG(render_tab_verts);
    GLuint i;
 
-   if (rmesa->swtcl.indexed_verts.buf) 
-      RELEASE_ELT_VERTS();
-   	
-   if (rmesa->swtcl.RenderIndex != 0 ||   
+   if (rmesa->radeon.swtcl.RenderIndex != 0 ||   
        !radeon_dma_validate_render( ctx, VB ))
       return GL_TRUE;		
 
@@ -496,13 +440,13 @@ static void radeonResetLineStipple( GLcontext *ctx );
 
 #undef LOCAL_VARS
 #undef ALLOC_VERTS
-#define CTX_ARG radeonContextPtr rmesa
-#define GET_VERTEX_DWORDS() rmesa->swtcl.vertex_size
-#define ALLOC_VERTS( n, size ) radeonAllocDmaLowVerts( rmesa, n, (size) * 4 )
+#define CTX_ARG r100ContextPtr rmesa
+#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
+#define ALLOC_VERTS( n, size ) rcommonAllocDmaLowVerts( &rmesa->radeon, n, (size) * 4 )
 #undef LOCAL_VARS
 #define LOCAL_VARS						\
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);		\
-   const char *radeonverts = (char *)rmesa->swtcl.verts;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);		\
+   const char *radeonverts = (char *)rmesa->radeon.swtcl.verts;
 #define VERT(x) (radeonVertex *)(radeonverts + ((x) * (vertsize) * sizeof(int)))
 #define VERTEX radeonVertex 
 #undef TAG
@@ -560,7 +504,7 @@ static struct {
 #define VERT_Y(_v) _v->v.y
 #define VERT_Z(_v) _v->v.z
 #define AREA_IS_CCW( a ) (a < 0)
-#define GET_VERTEX(e) (rmesa->swtcl.verts + ((e) * rmesa->swtcl.vertex_size * sizeof(int)))
+#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + ((e) * rmesa->radeon.swtcl.vertex_size * sizeof(int)))
 
 #define VERT_SET_RGBA( v, c )  					\
 do {								\
@@ -606,7 +550,7 @@ do {							\
 #undef INIT
 
 #define LOCAL_VARS(n)							\
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);			\
    GLuint color[n], spec[n];						\
    GLuint coloroffset = rmesa->swtcl.coloroffset;	\
    GLuint specoffset = rmesa->swtcl.specoffset;			\
@@ -617,7 +561,7 @@ do {							\
  ***********************************************************************/
 
 #define RASTERIZE(x) radeonRasterPrimitive( ctx, reduced_hw_prim[x] )
-#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive
+#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive
 #undef TAG
 #define TAG(x) x
 #include "tnl_dd/t_dd_unfilled.h"
@@ -673,9 +617,9 @@ static void init_rast_tab( void )
 } while (0)
 #undef LOCAL_VARS
 #define LOCAL_VARS						\
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);		\
-   const GLuint vertsize = rmesa->swtcl.vertex_size;		\
-   const char *radeonverts = (char *)rmesa->swtcl.verts;		\
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);		\
+   const GLuint vertsize = rmesa->radeon.swtcl.vertex_size;		\
+   const char *radeonverts = (char *)rmesa->radeon.swtcl.verts;		\
    const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
    const GLboolean stipple = ctx->Line.StippleFlag;		\
    (void) elt; (void) stipple;
@@ -700,17 +644,17 @@ static void init_rast_tab( void )
 void radeonChooseRenderState( GLcontext *ctx )
 {
    TNLcontext *tnl = TNL_CONTEXT(ctx);
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint index = 0;
    GLuint flags = ctx->_TriangleCaps;
 
-   if (!rmesa->TclFallback || rmesa->Fallback) 
+   if (!rmesa->radeon.TclFallback || rmesa->radeon.Fallback) 
       return;
 
    if (flags & DD_TRI_LIGHT_TWOSIDE) index |= RADEON_TWOSIDE_BIT;
    if (flags & DD_TRI_UNFILLED)      index |= RADEON_UNFILLED_BIT;
 
-   if (index != rmesa->swtcl.RenderIndex) {
+   if (index != rmesa->radeon.swtcl.RenderIndex) {
       tnl->Driver.Render.Points = rast_tab[index].points;
       tnl->Driver.Render.Line = rast_tab[index].line;
       tnl->Driver.Render.ClippedLine = rast_tab[index].line;
@@ -727,7 +671,7 @@ void radeonChooseRenderState( GLcontext *ctx )
 	 tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
       }
 
-      rmesa->swtcl.RenderIndex = index;
+      rmesa->radeon.swtcl.RenderIndex = index;
    }
 }
 
@@ -739,18 +683,18 @@ void radeonChooseRenderState( GLcontext *ctx )
 
 static void radeonRasterPrimitive( GLcontext *ctx, GLuint hwprim )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
-   if (rmesa->swtcl.hw_primitive != hwprim) {
+   if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
       RADEON_NEWPRIM( rmesa );
-      rmesa->swtcl.hw_primitive = hwprim;
+      rmesa->radeon.swtcl.hw_primitive = hwprim;
    }
 }
 
 static void radeonRenderPrimitive( GLcontext *ctx, GLenum prim )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   rmesa->swtcl.render_primitive = prim;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   rmesa->radeon.swtcl.render_primitive = prim;
    if (prim < GL_TRIANGLES || !(ctx->_TriangleCaps & DD_TRI_UNFILLED)) 
       radeonRasterPrimitive( ctx, reduced_hw_prim[prim] );
 }
@@ -761,7 +705,7 @@ static void radeonRenderFinish( GLcontext *ctx )
 
 static void radeonResetLineStipple( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    RADEON_STATECHANGE( rmesa, lin );
 }
 
@@ -795,17 +739,17 @@ static const char *getFallbackString(GLuint bit)
 
 void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
-   GLuint oldfallback = rmesa->Fallback;
+   GLuint oldfallback = rmesa->radeon.Fallback;
 
    if (mode) {
-      rmesa->Fallback |= bit;
+      rmesa->radeon.Fallback |= bit;
       if (oldfallback == 0) {
-	 RADEON_FIREVERTICES( rmesa );
+	 radeon_firevertices(&rmesa->radeon);
 	 TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_TRUE );
 	 _swsetup_Wakeup( ctx );
-	 rmesa->swtcl.RenderIndex = ~0;
+	 rmesa->radeon.swtcl.RenderIndex = ~0;
          if (RADEON_DEBUG & DEBUG_FALLBACKS) {
             fprintf(stderr, "Radeon begin rasterization fallback: 0x%x %s\n",
                     bit, getFallbackString(bit));
@@ -813,7 +757,7 @@ void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
       }
    }
    else {
-      rmesa->Fallback &= ~bit;
+      rmesa->radeon.Fallback &= ~bit;
       if (oldfallback == bit) {
 	 _swrast_flush( ctx );
 	 tnl->Driver.Render.Start = radeonRenderStart;
@@ -826,14 +770,14 @@ void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 
 	 tnl->Driver.Render.ResetLineStipple = radeonResetLineStipple;
 	 TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_FALSE );
-	 if (rmesa->TclFallback) {
-	    /* These are already done if rmesa->TclFallback goes to
+	 if (rmesa->radeon.TclFallback) {
+	    /* These are already done if rmesa->radeon.TclFallback goes to
 	     * zero above. But not if it doesn't (RADEON_NO_TCL for
 	     * example?)
 	     */
 	    _tnl_invalidate_vertex_state( ctx, ~0 );
 	    _tnl_invalidate_vertices( ctx, ~0 );
-	    RENDERINPUTS_ZERO( rmesa->tnl_index_bitset );
+	    RENDERINPUTS_ZERO( rmesa->radeon.tnl_index_bitset );
 	    radeonChooseVertexState( ctx );
 	    radeonChooseRenderState( ctx );
 	 }
@@ -853,7 +797,7 @@ void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 void radeonInitSwtcl( GLcontext *ctx )
 {
    TNLcontext *tnl = TNL_CONTEXT(ctx);
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    static int firsttime = 1;
 
    if (firsttime) {
@@ -872,18 +816,9 @@ void radeonInitSwtcl( GLcontext *ctx )
    _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 
 		       RADEON_MAX_TNL_VERTEX_SIZE);
    
-   rmesa->swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
-   rmesa->swtcl.RenderIndex = ~0;
-   rmesa->swtcl.render_primitive = GL_TRIANGLES;
-   rmesa->swtcl.hw_primitive = 0;
+   rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
+   rmesa->radeon.swtcl.RenderIndex = ~0;
+   rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES;
+   rmesa->radeon.swtcl.hw_primitive = 0;
 }
 
-
-void radeonDestroySwtcl( GLcontext *ctx )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if (rmesa->swtcl.indexed_verts.buf) 
-      radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, 
-			      __FUNCTION__ );
-}
diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.h b/src/mesa/drivers/dri/radeon/radeon_swtcl.h
index e485052ad77..da89158eeb9 100644
--- a/src/mesa/drivers/dri/radeon/radeon_swtcl.h
+++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.h
@@ -40,7 +40,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_context.h"
 
 extern void radeonInitSwtcl( GLcontext *ctx );
-extern void radeonDestroySwtcl( GLcontext *ctx );
 
 extern void radeonChooseRenderState( GLcontext *ctx );
 extern void radeonChooseVertexState( GLcontext *ctx );
@@ -63,5 +62,5 @@ extern void radeon_translate_vertex( GLcontext *ctx,
 
 extern void radeon_print_vertex( GLcontext *ctx, const radeonVertex *v );
 
-
+extern void r100_swtcl_flush(GLcontext *ctx, uint32_t current_offset);
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c
index 779e9ae5df0..df6708f05e3 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c
@@ -42,6 +42,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/tnl.h"
 #include "tnl/t_pipeline.h"
 
+#include "radeon_common.h"
 #include "radeon_context.h"
 #include "radeon_state.h"
 #include "radeon_ioctl.h"
@@ -104,7 +105,7 @@ static GLboolean discrete_prim[0x10] = {
 };
    
 
-#define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx)
+#define LOCAL_VARS r100ContextPtr rmesa = R100_CONTEXT(ctx)
 #define ELT_TYPE  GLushort
 
 #define ELT_INIT(prim, hw_prim) \
@@ -125,7 +126,7 @@ static GLboolean discrete_prim[0x10] = {
 
 #define RESET_STIPPLE() do {			\
    RADEON_STATECHANGE( rmesa, lin );		\
-   radeonEmitState( rmesa );			\
+   radeonEmitState(&rmesa->radeon);			\
 } while (0)
 
 #define AUTO_STIPPLE( mode )  do {		\
@@ -136,31 +137,29 @@ static GLboolean discrete_prim[0x10] = {
    else						\
       rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &=	\
 	 ~RADEON_LINE_PATTERN_AUTO_RESET;	\
-   radeonEmitState( rmesa );			\
+   radeonEmitState(&rmesa->radeon);		\
 } while (0)
 
 
 
 #define ALLOC_ELTS(nr)	radeonAllocElts( rmesa, nr )
 
-static GLushort *radeonAllocElts( radeonContextPtr rmesa, GLuint nr ) 
+static GLushort *radeonAllocElts( r100ContextPtr rmesa, GLuint nr ) 
 {
-   if (rmesa->dma.flush)
-      rmesa->dma.flush( rmesa );
+      if (rmesa->radeon.dma.flush)
+	 rmesa->radeon.dma.flush( rmesa->radeon.glCtx );
 
-   radeonEnsureCmdBufSpace(rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
-			   rmesa->hw.max_state_size + ELTS_BUFSZ(nr));
+      rcommonEnsureCmdBufSpace(&rmesa->radeon, rmesa->radeon.hw.max_state_size + ELTS_BUFSZ(nr) + 
+			       AOS_BUFSZ(rmesa->radeon.tcl.aos_count), __FUNCTION__);
 
-   radeonEmitAOS( rmesa,
-		rmesa->tcl.aos_components,
-		rmesa->tcl.nr_aos_components, 0 );
+      radeonEmitAOS( rmesa,
+		     rmesa->radeon.tcl.aos_count, 0 );
 
-   return radeonAllocEltsOpenEnded( rmesa,
-				    rmesa->tcl.vertex_format, 
-				    rmesa->tcl.hw_primitive, nr );
+      return radeonAllocEltsOpenEnded( rmesa, rmesa->tcl.vertex_format,
+				       rmesa->tcl.hw_primitive, nr );
 }
 
-#define CLOSE_ELTS()  RADEON_NEWPRIM( rmesa )
+#define CLOSE_ELTS() if (0)  RADEON_NEWPRIM( rmesa )
 
 
 
@@ -174,15 +173,15 @@ static void radeonEmitPrim( GLcontext *ctx,
 		       GLuint start, 
 		       GLuint count)	
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   r100ContextPtr rmesa = R100_CONTEXT( ctx );
    radeonTclPrimitive( ctx, prim, hwprim );
    
-   radeonEnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
-			    rmesa->hw.max_state_size + VBUF_BUFSZ );
+   rcommonEnsureCmdBufSpace( &rmesa->radeon,
+			     AOS_BUFSZ(rmesa->radeon.tcl.aos_count) +
+			     rmesa->radeon.hw.max_state_size + VBUF_BUFSZ, __FUNCTION__ );
 
    radeonEmitAOS( rmesa,
-		  rmesa->tcl.aos_components,
-		  rmesa->tcl.nr_aos_components,
+		  rmesa->radeon.tcl.aos_count,
 		  start );
    
    /* Why couldn't this packet have taken an offset param?
@@ -254,7 +253,7 @@ void radeonTclPrimitive( GLcontext *ctx,
 			 GLenum prim,
 			 int hw_prim )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint se_cntl;
    GLuint newprim = hw_prim | RADEON_CP_VC_CNTL_TCL_ENABLE;
 
@@ -371,7 +370,7 @@ radeonComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord )
 static GLboolean radeon_run_tcl_render( GLcontext *ctx,
 					struct tnl_pipeline_stage *stage )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    GLuint inputs = VERT_BIT_POS | VERT_BIT_COLOR0;
@@ -379,7 +378,7 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx,
 
    /* TODO: separate this from the swtnl pipeline 
     */
-   if (rmesa->TclFallback)
+   if (rmesa->radeon.TclFallback)
       return GL_TRUE;	/* fallback to software t&l */
 
    if (VB->Count == 0)
@@ -461,7 +460,7 @@ const struct tnl_pipeline_stage _radeon_tcl_stage =
 
 static void transition_to_swtnl( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    GLuint se_cntl;
 
@@ -490,7 +489,7 @@ static void transition_to_swtnl( GLcontext *ctx )
 
 static void transition_to_hwtnl( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
 
@@ -509,15 +508,15 @@ static void transition_to_hwtnl( GLcontext *ctx )
 
    tnl->Driver.NotifyMaterialChange = radeonUpdateMaterial;
 
-   if ( rmesa->dma.flush )			
-      rmesa->dma.flush( rmesa );	
+   if ( rmesa->radeon.dma.flush )			
+      rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	
 
-   rmesa->dma.flush = NULL;
+   rmesa->radeon.dma.flush = NULL;
    rmesa->swtcl.vertex_format = 0;
    
-   if (rmesa->swtcl.indexed_verts.buf) 
-      radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, 
-			      __FUNCTION__ );
+   //   if (rmesa->swtcl.indexed_verts.buf) 
+   //      radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, 
+   //			      __FUNCTION__ );
 
    if (RADEON_DEBUG & DEBUG_FALLBACKS) 
       fprintf(stderr, "Radeon end tcl fallback\n");
@@ -550,11 +549,11 @@ static char *getFallbackString(GLuint bit)
 
 void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   GLuint oldfallback = rmesa->TclFallback;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   GLuint oldfallback = rmesa->radeon.TclFallback;
 
    if (mode) {
-      rmesa->TclFallback |= bit;
+      rmesa->radeon.TclFallback |= bit;
       if (oldfallback == 0) {
 	 if (RADEON_DEBUG & DEBUG_FALLBACKS) 
 	    fprintf(stderr, "Radeon begin tcl fallback %s\n",
@@ -563,7 +562,7 @@ void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
       }
    }
    else {
-      rmesa->TclFallback &= ~bit;
+      rmesa->radeon.TclFallback &= ~bit;
       if (oldfallback == bit) {
 	 if (RADEON_DEBUG & DEBUG_FALLBACKS) 
 	    fprintf(stderr, "Radeon end tcl fallback %s\n",
diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.c b/src/mesa/drivers/dri/radeon/radeon_tex.c
index f2b6deb9c04..2549d5cb5cb 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tex.c
+++ b/src/mesa/drivers/dri/radeon/radeon_tex.c
@@ -44,6 +44,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/texobj.h"
 
 #include "radeon_context.h"
+#include "radeon_mipmap_tree.h"
 #include "radeon_state.h"
 #include "radeon_ioctl.h"
 #include "radeon_swtcl.h"
@@ -170,10 +171,13 @@ static void radeonSetTexFilter( radeonTexObjPtr t, GLenum minf, GLenum magf )
 {
    GLuint anisotropy = (t->pp_txfilter & RADEON_MAX_ANISO_MASK);
 
+   /* Force revalidation to account for switches from/to mipmapping. */
+   t->validated = GL_FALSE;
+
    t->pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK);
 
    /* r100 chips can't handle mipmaps/aniso for cubemap/volume textures */
-   if ( t->base.tObj->Target == GL_TEXTURE_CUBE_MAP ) {
+   if ( t->base.Target == GL_TEXTURE_CUBE_MAP ) {
       switch ( minf ) {
       case GL_NEAREST:
       case GL_NEAREST_MIPMAP_NEAREST:
@@ -249,433 +253,13 @@ static void radeonSetTexBorderColor( radeonTexObjPtr t, const GLfloat color[4] )
    t->pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
 }
 
-
-/**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-static radeonTexObjPtr radeonAllocTexObj( struct gl_texture_object *texObj )
-{
-   radeonTexObjPtr t;
-
-   t = CALLOC_STRUCT( radeon_tex_obj );
-   texObj->DriverData = t;
-   if ( t != NULL ) {
-      if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
-	 fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, (void *)texObj, (void *)t );
-      }
-
-      /* Initialize non-image-dependent parts of the state:
-       */
-      t->base.tObj = texObj;
-      t->border_fallback = GL_FALSE;
-
-      t->pp_txfilter = RADEON_BORDER_MODE_OGL;
-      t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP |
-			RADEON_TXFORMAT_PERSPECTIVE_ENABLE);
-
-      make_empty_list( & t->base );
-
-      radeonSetTexWrap( t, texObj->WrapS, texObj->WrapT );
-      radeonSetTexMaxAnisotropy( t, texObj->MaxAnisotropy );
-      radeonSetTexFilter( t, texObj->MinFilter, texObj->MagFilter );
-      radeonSetTexBorderColor( t, texObj->BorderColor );
-   }
-
-   return t;
-}
-
-
-static const struct gl_texture_format *
-radeonChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
-                           GLenum format, GLenum type )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   const GLboolean do32bpt =
-       ( rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32 );
-   const GLboolean force16bpt =
-       ( rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16 );
-   (void) format;
-
-   switch ( internalFormat ) {
-   case 4:
-   case GL_RGBA:
-   case GL_COMPRESSED_RGBA:
-      switch ( type ) {
-      case GL_UNSIGNED_INT_10_10_10_2:
-      case GL_UNSIGNED_INT_2_10_10_10_REV:
-	 return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_argb1555;
-      case GL_UNSIGNED_SHORT_4_4_4_4:
-      case GL_UNSIGNED_SHORT_4_4_4_4_REV:
-	 return _dri_texformat_argb4444;
-      case GL_UNSIGNED_SHORT_5_5_5_1:
-      case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-	 return _dri_texformat_argb1555;
-      default:
-         return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_argb4444;
-      }
-
-   case 3:
-   case GL_RGB:
-   case GL_COMPRESSED_RGB:
-      switch ( type ) {
-      case GL_UNSIGNED_SHORT_4_4_4_4:
-      case GL_UNSIGNED_SHORT_4_4_4_4_REV:
-	 return _dri_texformat_argb4444;
-      case GL_UNSIGNED_SHORT_5_5_5_1:
-      case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-	 return _dri_texformat_argb1555;
-      case GL_UNSIGNED_SHORT_5_6_5:
-      case GL_UNSIGNED_SHORT_5_6_5_REV:
-	 return _dri_texformat_rgb565;
-      default:
-         return do32bpt ? _dri_texformat_argb8888 : _dri_texformat_rgb565;
-      }
-
-   case GL_RGBA8:
-   case GL_RGB10_A2:
-   case GL_RGBA12:
-   case GL_RGBA16:
-      return !force16bpt ?
-	  _dri_texformat_argb8888 : _dri_texformat_argb4444;
-
-   case GL_RGBA4:
-   case GL_RGBA2:
-      return _dri_texformat_argb4444;
-
-   case GL_RGB5_A1:
-      return _dri_texformat_argb1555;
-
-   case GL_RGB8:
-   case GL_RGB10:
-   case GL_RGB12:
-   case GL_RGB16:
-      return !force16bpt ? _dri_texformat_argb8888 : _dri_texformat_rgb565;
-
-   case GL_RGB5:
-   case GL_RGB4:
-   case GL_R3_G3_B2:
-      return _dri_texformat_rgb565;
-
-   case GL_ALPHA:
-   case GL_ALPHA4:
-   case GL_ALPHA8:
-   case GL_ALPHA12:
-   case GL_ALPHA16:
-   case GL_COMPRESSED_ALPHA:
-      return _dri_texformat_a8;
-
-   case 1:
-   case GL_LUMINANCE:
-   case GL_LUMINANCE4:
-   case GL_LUMINANCE8:
-   case GL_LUMINANCE12:
-   case GL_LUMINANCE16:
-   case GL_COMPRESSED_LUMINANCE:
-      return _dri_texformat_l8;
-
-   case 2:
-   case GL_LUMINANCE_ALPHA:
-   case GL_LUMINANCE4_ALPHA4:
-   case GL_LUMINANCE6_ALPHA2:
-   case GL_LUMINANCE8_ALPHA8:
-   case GL_LUMINANCE12_ALPHA4:
-   case GL_LUMINANCE12_ALPHA12:
-   case GL_LUMINANCE16_ALPHA16:
-   case GL_COMPRESSED_LUMINANCE_ALPHA:
-      return _dri_texformat_al88;
-
-   case GL_INTENSITY:
-   case GL_INTENSITY4:
-   case GL_INTENSITY8:
-   case GL_INTENSITY12:
-   case GL_INTENSITY16:
-   case GL_COMPRESSED_INTENSITY:
-      return _dri_texformat_i8;
-
-   case GL_YCBCR_MESA:
-      if (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
-          type == GL_UNSIGNED_BYTE)
-         return &_mesa_texformat_ycbcr;
-      else
-         return &_mesa_texformat_ycbcr_rev;
-
-   case GL_RGB_S3TC:
-   case GL_RGB4_S3TC:
-   case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
-      return &_mesa_texformat_rgb_dxt1;
-
-   case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
-      return &_mesa_texformat_rgba_dxt1;
-
-   case GL_RGBA_S3TC:
-   case GL_RGBA4_S3TC:
-   case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
-      return &_mesa_texformat_rgba_dxt3;
-
-   case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
-      return &_mesa_texformat_rgba_dxt5;
-
-   default:
-      _mesa_problem(ctx, "unexpected texture format in %s", __FUNCTION__);
-      return NULL;
-   }
-
-   return NULL; /* never get here */
-}
-
-
-static void radeonTexImage1D( GLcontext *ctx, GLenum target, GLint level,
-                              GLint internalFormat,
-                              GLint width, GLint border,
-                              GLenum format, GLenum type, const GLvoid *pixels,
-                              const struct gl_pixelstore_attrib *packing,
-                              struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) radeonAllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
-         return;
-      }
-   }
-
-   /* Note, this will call ChooseTextureFormat */
-   _mesa_store_teximage1d(ctx, target, level, internalFormat,
-                          width, border, format, type, pixels,
-                          &ctx->Unpack, texObj, texImage);
-
-   t->dirty_images[0] |= (1 << level);
-}
-
-
-static void radeonTexSubImage1D( GLcontext *ctx, GLenum target, GLint level,
-                                 GLint xoffset,
-                                 GLsizei width,
-                                 GLenum format, GLenum type,
-                                 const GLvoid *pixels,
-                                 const struct gl_pixelstore_attrib *packing,
-                                 struct gl_texture_object *texObj,
-                                 struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) radeonAllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D");
-         return;
-      }
-   }
-
-   _mesa_store_texsubimage1d(ctx, target, level, xoffset, width,
-			     format, type, pixels, packing, texObj,
-			     texImage);
-
-   t->dirty_images[0] |= (1 << level);
-}
-
-
-static void radeonTexImage2D( GLcontext *ctx, GLenum target, GLint level,
-                              GLint internalFormat,
-                              GLint width, GLint height, GLint border,
-                              GLenum format, GLenum type, const GLvoid *pixels,
-                              const struct gl_pixelstore_attrib *packing,
-                              struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   if ( t != NULL ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) radeonAllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
-         return;
-      }
-   }
-
-   /* Note, this will call ChooseTextureFormat */
-   _mesa_store_teximage2d(ctx, target, level, internalFormat,
-                          width, height, border, format, type, pixels,
-                          &ctx->Unpack, texObj, texImage);
-
-   t->dirty_images[face] |= (1 << level);
-}
-
-
-static void radeonTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
-                                 GLint xoffset, GLint yoffset,
-                                 GLsizei width, GLsizei height,
-                                 GLenum format, GLenum type,
-                                 const GLvoid *pixels,
-                                 const struct gl_pixelstore_attrib *packing,
-                                 struct gl_texture_object *texObj,
-                                 struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) radeonAllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D");
-         return;
-      }
-   }
-
-   _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
-			     height, format, type, pixels, packing, texObj,
-			     texImage);
-
-   t->dirty_images[face] |= (1 << level);
-}
-
-static void radeonCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level,
-                              GLint internalFormat,
-                              GLint width, GLint height, GLint border,
-                              GLsizei imageSize, const GLvoid *data,
-                              struct gl_texture_object *texObj,
-                              struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   if ( t != NULL ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) radeonAllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage2D");
-         return;
-      }
-   }
-
-   /* Note, this will call ChooseTextureFormat */
-   _mesa_store_compressed_teximage2d(ctx, target, level, internalFormat, width,
-                                 height, border, imageSize, data, texObj, texImage);
-
-   t->dirty_images[face] |= (1 << level);
-}
-
-
-static void radeonCompressedTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
-                                 GLint xoffset, GLint yoffset,
-                                 GLsizei width, GLsizei height,
-                                 GLenum format,
-                                 GLsizei imageSize, const GLvoid *data,
-                                 struct gl_texture_object *texObj,
-                                 struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
-
-
-   /* which cube face or ordinary 2D image */
-   switch (target) {
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
-   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
-   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
-      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
-      ASSERT(face < 6);
-      break;
-   default:
-      face = 0;
-   }
-
-   assert( t ); /* this _should_ be true */
-   if ( t ) {
-      driSwapOutTextureObject( t );
-   }
-   else {
-      t = (driTextureObject *) radeonAllocTexObj( texObj );
-      if (!t) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexSubImage2D");
-         return;
-      }
-   }
-
-   _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
-                                 height, format, imageSize, data, texObj, texImage);
-
-   t->dirty_images[face] |= (1 << level);
-}
-
 #define SCALED_FLOAT_TO_BYTE( x, scale ) \
 		(((GLuint)((255.0F / scale) * (x))) / 2)
 
 static void radeonTexEnv( GLcontext *ctx, GLenum target,
 			  GLenum pname, const GLfloat *param )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint unit = ctx->Texture.CurrentUnit;
    struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
 
@@ -706,7 +290,7 @@ static void radeonTexEnv( GLcontext *ctx, GLenum target,
        * functions, one mapping [-1.0,0.0] to [-128,0] and one mapping
        * [0.0,4.0] to [0,127].
        */
-      min = driQueryOptionb (&rmesa->optionCache, "no_neg_lod_bias") ?
+      min = driQueryOptionb (&rmesa->radeon.optionCache, "no_neg_lod_bias") ?
 	  0.0 : -1.0;
       bias = CLAMP( *param, min, 4.0 );
       if ( bias == 0 ) {
@@ -739,7 +323,7 @@ static void radeonTexParameter( GLcontext *ctx, GLenum target,
 				struct gl_texture_object *texObj,
 				GLenum pname, const GLfloat *params )
 {
-   radeonTexObjPtr t = (radeonTexObjPtr) texObj->DriverData;
+   radeonTexObj* t = radeon_tex_obj(texObj);
 
    if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
       fprintf( stderr, "%s( %s )\n", __FUNCTION__,
@@ -767,57 +351,51 @@ static void radeonTexParameter( GLcontext *ctx, GLenum target,
    case GL_TEXTURE_MAX_LEVEL:
    case GL_TEXTURE_MIN_LOD:
    case GL_TEXTURE_MAX_LOD:
+
       /* This isn't the most efficient solution but there doesn't appear to
        * be a nice alternative.  Since there's no LOD clamping,
        * we just have to rely on loading the right subset of mipmap levels
        * to simulate a clamped LOD.
        */
-      driSwapOutTextureObject( (driTextureObject *) t );
+      if (t->mt) {
+         radeon_miptree_unreference(t->mt);
+	 t->mt = 0;
+	 t->validated = GL_FALSE;
+      }
       break;
 
    default:
       return;
    }
-
-   /* Mark this texobj as dirty (one bit per tex unit)
-    */
-   t->dirty_state = TEX_ALL;
-}
-
-
-static void radeonBindTexture( GLcontext *ctx, GLenum target,
-			       struct gl_texture_object *texObj )
-{
-   if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
-      fprintf( stderr, "%s( %p ) unit=%d\n", __FUNCTION__, (void *)texObj,
-	       ctx->Texture.CurrentUnit );
-   }
-
-   assert( (target != GL_TEXTURE_1D && target != GL_TEXTURE_2D &&
-            target != GL_TEXTURE_RECTANGLE_NV && target != GL_TEXTURE_CUBE_MAP) ||
-           (texObj->DriverData != NULL) );
 }
 
-
 static void radeonDeleteTexture( GLcontext *ctx,
 				 struct gl_texture_object *texObj )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   radeonTexObj* t = radeon_tex_obj(texObj);
+   int i;
 
    if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
       fprintf( stderr, "%s( %p (target = %s) )\n", __FUNCTION__, (void *)texObj,
 	       _mesa_lookup_enum_by_nr( texObj->Target ) );
    }
 
-   if ( t != NULL ) {
-      if ( rmesa ) {
-         RADEON_FIREVERTICES( rmesa );
-      }
-
-      driDestroyTextureObject( t );
+   if ( rmesa ) {
+     radeon_firevertices(&rmesa->radeon);
+     for ( i = 0 ; i < rmesa->radeon.glCtx->Const.MaxTextureUnits ; i++ ) {
+       if ( t == rmesa->state.texture.unit[i].texobj ) {
+	 rmesa->state.texture.unit[i].texobj = NULL;
+	 rmesa->hw.tex[i].dirty = GL_FALSE;
+	 rmesa->hw.cube[i].dirty = GL_FALSE;
+       }
+     }
    }
 
+   if (t->mt) {
+      radeon_miptree_unreference(t->mt);
+      t->mt = 0;
+   }
    /* Free mipmap images and the texture object itself */
    _mesa_delete_texture_object(ctx, texObj);
 }
@@ -837,7 +415,7 @@ static void radeonTexGen( GLcontext *ctx,
 			  GLenum pname,
 			  const GLfloat *params )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLuint unit = ctx->Texture.CurrentUnit;
    rmesa->recheck_texgen[unit] = GL_TRUE;
 }
@@ -851,29 +429,40 @@ static void radeonTexGen( GLcontext *ctx,
 static struct gl_texture_object *
 radeonNewTextureObject( GLcontext *ctx, GLuint name, GLenum target )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   struct gl_texture_object *obj;
-   obj = _mesa_new_texture_object(ctx, name, target);
-   if (!obj)
-      return NULL;
-   obj->MaxAnisotropy = rmesa->initialMaxAnisotropy;
-   radeonAllocTexObj( obj );
-   return obj;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   radeonTexObj* t = CALLOC_STRUCT(radeon_tex_obj);
+
+   _mesa_initialize_texture_object(&t->base, name, target);
+   t->base.MaxAnisotropy = rmesa->radeon.initialMaxAnisotropy;
+
+   t->border_fallback = GL_FALSE;
+
+   t->pp_txfilter = RADEON_BORDER_MODE_OGL;
+   t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP |
+		     RADEON_TXFORMAT_PERSPECTIVE_ENABLE);
+   
+   radeonSetTexWrap( t, t->base.WrapS, t->base.WrapT );
+   radeonSetTexMaxAnisotropy( t, t->base.MaxAnisotropy );
+   radeonSetTexFilter( t, t->base.MinFilter, t->base.MagFilter );
+   radeonSetTexBorderColor( t, t->base.BorderColor );
+   return &t->base;
 }
 
 
+
 void radeonInitTextureFuncs( struct dd_function_table *functions )
 {
-   functions->ChooseTextureFormat	= radeonChooseTextureFormat;
+   functions->ChooseTextureFormat	= radeonChooseTextureFormat_mesa;
    functions->TexImage1D		= radeonTexImage1D;
    functions->TexImage2D		= radeonTexImage2D;
    functions->TexSubImage1D		= radeonTexSubImage1D;
    functions->TexSubImage2D		= radeonTexSubImage2D;
+   functions->GetTexImage               = radeonGetTexImage;
+   functions->GetCompressedTexImage     = radeonGetCompressedTexImage;
 
    functions->NewTextureObject		= radeonNewTextureObject;
-   functions->BindTexture		= radeonBindTexture;
+   //   functions->BindTexture		= radeonBindTexture;
    functions->DeleteTexture		= radeonDeleteTexture;
-   functions->IsTextureResident		= driIsTextureResident;
 
    functions->TexEnv			= radeonTexEnv;
    functions->TexParameter		= radeonTexParameter;
@@ -882,5 +471,12 @@ void radeonInitTextureFuncs( struct dd_function_table *functions )
    functions->CompressedTexImage2D	= radeonCompressedTexImage2D;
    functions->CompressedTexSubImage2D	= radeonCompressedTexSubImage2D;
 
+   functions->GenerateMipmap = radeonGenerateMipmap;
+
+   functions->NewTextureImage = radeonNewTextureImage;
+   functions->FreeTexImageData = radeonFreeTexImageData;
+   functions->MapTexture = radeonMapTexture;
+   functions->UnmapTexture = radeonUnmapTexture;
+
    driInitTextureFormats();
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.h b/src/mesa/drivers/dri/radeon/radeon_tex.h
index 80008808289..a4aaddc74fa 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tex.h
+++ b/src/mesa/drivers/dri/radeon/radeon_tex.h
@@ -41,12 +41,16 @@ extern void radeonSetTexOffset(__DRIcontext *pDRICtx, GLint texname,
                                unsigned long long offset, GLint depth,
                                GLuint pitch);
 
+extern void radeonSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv);
+extern void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format,
+			       __DRIdrawable *dPriv);
+
 extern void radeonUpdateTextureState( GLcontext *ctx );
 
-extern int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t,
+extern int radeonUploadTexImages( r100ContextPtr rmesa, radeonTexObjPtr t,
 				  GLuint face );
 
-extern void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
+extern void radeonDestroyTexObj( r100ContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonInitTextureFuncs( struct dd_function_table *functions );
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_texmem.c b/src/mesa/drivers/dri/radeon/radeon_texmem.c
deleted file mode 100644
index 5f7bbe6a4cb..00000000000
--- a/src/mesa/drivers/dri/radeon/radeon_texmem.c
+++ /dev/null
@@ -1,404 +0,0 @@
-/**************************************************************************
-
-Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
-                     VA Linux Systems Inc., Fremont, California.
-
-All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation on the rights to use, copy, modify, merge, publish,
-distribute, sub license, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice (including the
-next paragraph) shall be included in all copies or substantial
-portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-NON-INFRINGEMENT. IN NO EVENT SHALL ATI, VA LINUX SYSTEMS AND/OR THEIR
-SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
-IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR
-IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Kevin E. Martin <[email protected]>
- *   Gareth Hughes <[email protected]>
- *
- */
-#include <errno.h> 
-
-#include "main/glheader.h"
-#include "main/imports.h"
-#include "main/context.h"
-#include "main/macros.h"
-
-#include "radeon_context.h"
-#include "radeon_ioctl.h"
-#include "radeon_tex.h"
-
-#include <unistd.h>  /* for usleep() */
-
-
-/**
- * Destroy any device-dependent state associated with the texture.  This may
- * include NULLing out hardware state that points to the texture.
- */
-void
-radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-   if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
-      fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, (void *)t, (void *)t->base.tObj );
-   }
-
-   if ( rmesa != NULL ) {
-      unsigned   i;
-
-
-      for ( i = 0 ; i < rmesa->glCtx->Const.MaxTextureUnits ; i++ ) {
-	 if ( t == rmesa->state.texture.unit[i].texobj ) {
-	    rmesa->state.texture.unit[i].texobj = NULL;
-	 }
-      }
-   }
-}
-
-
-/* ------------------------------------------------------------
- * Texture image conversions
- */
-
-
-static void radeonUploadRectSubImage( radeonContextPtr rmesa,
-				      radeonTexObjPtr t, 
-				      struct gl_texture_image *texImage,
-				      GLint x, GLint y, 
-				      GLint width, GLint height )
-{
-   const struct gl_texture_format *texFormat = texImage->TexFormat;
-   int blit_format, dstPitch, done;
-
-   switch ( texFormat->TexelBytes ) {
-   case 1:
-      blit_format = RADEON_GMC_DST_8BPP_CI;
-      break;
-   case 2:
-      blit_format = RADEON_GMC_DST_16BPP;
-      break;
-   case 4:
-      blit_format = RADEON_GMC_DST_32BPP;
-      break;
-   default:
-      fprintf( stderr, "radeonUploadRectSubImage: unknown blit_format (texelbytes=%d)\n", 
-      	       texFormat->TexelBytes);
-      return;
-   }
-
-   t->image[0][0].data = texImage->Data;
-
-   /* Currently don't need to cope with small pitches.
-    */
-   width = texImage->Width;
-   height = texImage->Height;
-   dstPitch = t->pp_txpitch + 32;
-
-   {	/* FIXME: prefer GART-texturing if possible */
-      /* Data not in GART memory, or bad pitch.
-       */
-      for (done = 0; done < height ; ) {
-	 struct radeon_dma_region region;
-	 int lines = MIN2( height - done, RADEON_BUFFER_SIZE / dstPitch );
-	 int src_pitch;
-	 char *tex;
-
-         src_pitch = texImage->RowStride * texFormat->TexelBytes;
-
-	 tex = (char *)texImage->Data + done * src_pitch;
-
-	 memset(&region, 0, sizeof(region));
-	 radeonAllocDmaRegion( rmesa, &region, lines * dstPitch, 1024 );
-
-	 /* Copy texdata to dma:
-	  */
-	 if (0)
-	    fprintf(stderr, "%s: src_pitch %d dst_pitch %d\n",
-		    __FUNCTION__, src_pitch, dstPitch);
-
-	 if (src_pitch == dstPitch) {
-	    memcpy( region.address + region.start, tex, lines * src_pitch );
-	 } 
-	 else {
-	    char *buf = region.address + region.start;
-	    int i;
-	    for (i = 0 ; i < lines ; i++) {
-	       memcpy( buf, tex, src_pitch );
-	       buf += dstPitch;
-	       tex += src_pitch;
-	    }
-	 }
-
-	 radeonEmitWait( rmesa, RADEON_WAIT_3D );
-
-	 
-
-	 /* Blit to framebuffer
-	  */
-	 radeonEmitBlit( rmesa,
-		       blit_format,
-		       dstPitch, GET_START( &region ),
-		       dstPitch, t->bufAddr,
-		       0, 0,
-		       0, done,
-		       width, lines );
-	 
-	 radeonEmitWait( rmesa, RADEON_WAIT_2D );
-
-	 radeonReleaseDmaRegion( rmesa, &region, __FUNCTION__ );
-	 done += lines;
-      }
-   }
-}
-
-
-/**
- * Upload the texture image associated with texture \a t at the specified
- * level at the address relative to \a start.
- */
-static void uploadSubImage( radeonContextPtr rmesa, radeonTexObjPtr t, 
-			    GLint hwlevel,
-			    GLint x, GLint y, GLint width, GLint height,
-			    GLuint face )
-{
-   struct gl_texture_image *texImage = NULL;
-   GLuint offset;
-   GLint imageWidth, imageHeight;
-   GLint ret;
-   drm_radeon_texture_t tex;
-   drm_radeon_tex_image_t tmp;
-   const int level = hwlevel + t->base.firstLevel;
-
-   if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
-      fprintf( stderr, "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n", 
-	       __FUNCTION__, (void *)t, (void *)t->base.tObj, level, width, height, face );
-   }
-
-   ASSERT(face < 6);
-
-   /* Ensure we have a valid texture to upload */
-   if ( ( hwlevel < 0 ) || ( hwlevel >= RADEON_MAX_TEXTURE_LEVELS ) ) {
-      _mesa_problem(NULL, "bad texture level in %s", __FUNCTION__);
-      return;
-   }
-
-   texImage = t->base.tObj->Image[face][level];
-
-   if ( !texImage ) {
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: texImage %d is NULL!\n", __FUNCTION__, level );
-      return;
-   }
-   if ( !texImage->Data ) {
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is NULL!\n", __FUNCTION__ );
-      return;
-   }
-
-
-   if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-      assert(level == 0);
-      assert(hwlevel == 0);
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is rectangular\n", __FUNCTION__);
-      radeonUploadRectSubImage( rmesa, t, texImage, x, y, width, height );
-      return;
-   }
-
-   imageWidth = texImage->Width;
-   imageHeight = texImage->Height;
-
-   offset = t->bufAddr + t->base.totalSize * face / 6;
-
-   if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
-      GLint imageX = 0;
-      GLint imageY = 0;
-      GLint blitX = t->image[face][hwlevel].x;
-      GLint blitY = t->image[face][hwlevel].y;
-      GLint blitWidth = t->image[face][hwlevel].width;
-      GLint blitHeight = t->image[face][hwlevel].height;
-      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
-	       imageWidth, imageHeight, imageX, imageY );
-      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
-	       blitWidth, blitHeight, blitX, blitY );
-      fprintf( stderr, "       blit ofs: 0x%07x level: %d/%d\n",
-	       (GLuint)offset, hwlevel, level );
-   }
-
-   t->image[face][hwlevel].data = texImage->Data;
-
-   /* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct.
-    * NOTE: we're always use a 1KB-wide blit and I8 texture format.
-    * We used to use 1, 2 and 4-byte texels and used to use the texture
-    * width to dictate the blit width - but that won't work for compressed
-    * textures. (Brian)
-    * NOTE: can't do that with texture tiling. (sroland)
-    */
-   tex.offset = offset;
-   tex.image = &tmp;
-   /* copy (x,y,width,height,data) */
-   memcpy( &tmp, &t->image[face][hwlevel], sizeof(drm_radeon_tex_image_t) );
-
-   if (texImage->TexFormat->TexelBytes) {
-      /* use multi-byte upload scheme */
-      tex.height = imageHeight;
-      tex.width = imageWidth;
-      tex.format = t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK;
-      tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1);
-      tex.offset += tmp.x & ~1023;
-      tmp.x = tmp.x % 1024;
-      if (t->tile_bits & RADEON_TXO_MICRO_TILE_X2) {
-	 /* need something like "tiled coordinates" ? */
-	 tmp.y = tmp.x / (tex.pitch * 128) * 2;
-	 tmp.x = tmp.x % (tex.pitch * 128) / 2 / texImage->TexFormat->TexelBytes;
-	 tex.pitch |= RADEON_DST_TILE_MICRO >> 22;
-      }
-      else {
-	 tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1);
-      }
-      if ((t->tile_bits & RADEON_TXO_MACRO_TILE) &&
-	 (texImage->Width * texImage->TexFormat->TexelBytes >= 256)) {
-	 /* radeon switches off macro tiling for small textures/mipmaps it seems */
-	 tex.pitch |= RADEON_DST_TILE_MACRO >> 22;
-      }
-   }
-   else {
-      /* In case of for instance 8x8 texture (2x2 dxt blocks), padding after the first two blocks is
-         needed (only with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */
-      /* set tex.height to 1/4 since 1 "macropixel" (dxt-block) has 4 real pixels. Needed
-         so the kernel module reads the right amount of data. */
-      tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */
-      tex.pitch = (BLIT_WIDTH_BYTES / 64);
-      tex.height = (imageHeight + 3) / 4;
-      tex.width = (imageWidth + 3) / 4;
-      switch (t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK) {
-      case RADEON_TXFORMAT_DXT1:
-         tex.width *= 8;
-         break;
-      case RADEON_TXFORMAT_DXT23:
-      case RADEON_TXFORMAT_DXT45:
-         tex.width *= 16;
-         break;
-      }
-   }
-
-   LOCK_HARDWARE( rmesa );
-   do {
-      ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_TEXTURE,
-                                 &tex, sizeof(drm_radeon_texture_t) );
-   } while ( ret == -EAGAIN );
-
-   UNLOCK_HARDWARE( rmesa );
-
-   if ( ret ) {
-      fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret );
-      fprintf( stderr, "   offset=0x%08x\n",
-	       offset );
-      fprintf( stderr, "   image width=%d height=%d\n",
-	       imageWidth, imageHeight );
-      fprintf( stderr, "    blit width=%d height=%d data=%p\n",
-	       t->image[face][hwlevel].width, t->image[face][hwlevel].height,
-	       t->image[face][hwlevel].data );
-      exit( 1 );
-   }
-}
-
-
-/**
- * Upload the texture images associated with texture \a t.  This might
- * require the allocation of texture memory.
- * 
- * \param rmesa Context pointer
- * \param t Texture to be uploaded
- * \param face Cube map face to be uploaded.  Zero for non-cube maps.
- */
-
-int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t, GLuint face )
-{
-   int numLevels;
-
-   if ( !t || t->base.totalSize == 0 || t->image_override )
-      return 0;
-
-   if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
-      fprintf( stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__,
-	       (void *)rmesa->glCtx, (void *)t->base.tObj, t->base.totalSize,
-	       t->base.firstLevel, t->base.lastLevel );
-   }
-
-   numLevels = t->base.lastLevel - t->base.firstLevel + 1;
-
-   if (RADEON_DEBUG & DEBUG_SYNC) {
-      fprintf(stderr, "%s: Syncing\n", __FUNCTION__ );
-      radeonFinish( rmesa->glCtx );
-   }
-
-   LOCK_HARDWARE( rmesa );
-
-   if ( t->base.memBlock == NULL ) {
-      int heap;
-
-      heap = driAllocateTexture( rmesa->texture_heaps, rmesa->nr_heaps,
-				 (driTextureObject *) t );
-      if ( heap == -1 ) {
-	 UNLOCK_HARDWARE( rmesa );
-	 return -1;
-      }
-
-      /* Set the base offset of the texture image */
-      t->bufAddr = rmesa->radeonScreen->texOffset[heap] 
-	   + t->base.memBlock->ofs;
-      t->pp_txoffset = t->bufAddr;
-
-      if (!(t->base.tObj->Image[0][0]->IsClientData)) {
-	 /* hope it's safe to add that here... */
-	 t->pp_txoffset |= t->tile_bits;
-      }
-
-      /* Mark this texobj as dirty on all units:
-       */
-      t->dirty_state = TEX_ALL;
-   }
-
-
-   /* Let the world know we've used this memory recently.
-    */
-   driUpdateTextureLRU( (driTextureObject *) t );
-   UNLOCK_HARDWARE( rmesa );
-
-
-   /* Upload any images that are new */
-   if (t->base.dirty_images[face]) {
-      int i;
-      for ( i = 0 ; i < numLevels ; i++ ) {
-         if ( (t->base.dirty_images[face] & (1 << (i+t->base.firstLevel))) != 0 ) {
-            uploadSubImage( rmesa, t, i, 0, 0, t->image[face][i].width,
-			    t->image[face][i].height, face );
-         }
-      }
-      t->base.dirty_images[face] = 0;
-   }
-
-   if (RADEON_DEBUG & DEBUG_SYNC) {
-      fprintf(stderr, "%s: Syncing\n", __FUNCTION__ );
-      radeonFinish( rmesa->glCtx );
-   }
-
-   return 0;
-}
diff --git a/src/mesa/drivers/dri/radeon/radeon_texstate.c b/src/mesa/drivers/dri/radeon/radeon_texstate.c
index b165205c093..279bcd4df6a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texstate.c
+++ b/src/mesa/drivers/dri/radeon/radeon_texstate.c
@@ -39,10 +39,12 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/context.h"
 #include "main/macros.h"
 #include "main/texformat.h"
+#include "main/teximage.h"
 #include "main/texobj.h"
 #include "main/enums.h"
 
 #include "radeon_context.h"
+#include "radeon_mipmap_tree.h"
 #include "radeon_state.h"
 #include "radeon_ioctl.h"
 #include "radeon_swtcl.h"
@@ -75,10 +77,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_RGBA_DXT5) \
 			     && (tx_table[f].format != 0xffffffff) )
 
-static const struct {
+struct tx_table {
    GLuint format, filter;
-}
-tx_table[] =
+};
+
+static const struct tx_table tx_table[] =
 {
    _ALPHA(RGBA8888),
    _ALPHA_REV(RGBA8888),
@@ -111,252 +114,6 @@ tx_table[] =
 #undef _ALPHA
 #undef _INVALID
 
-/**
- * This function computes the number of bytes of storage needed for
- * the given texture object (all mipmap levels, all cube faces).
- * The \c image[face][level].x/y/width/height parameters for upload/blitting
- * are computed here.  \c pp_txfilter, \c pp_txformat, etc. will be set here
- * too.
- * 
- * \param rmesa Context pointer
- * \param tObj GL texture object whose images are to be posted to
- *                 hardware state.
- */
-static void radeonSetTexImages( radeonContextPtr rmesa,
-				struct gl_texture_object *tObj )
-{
-   radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   GLint curOffset, blitWidth;
-   GLint i, texelBytes;
-   GLint numLevels;
-   GLint log2Width, log2Height, log2Depth;
-
-   /* Set the hardware texture format
-    */
-   if ( !t->image_override ) {
-      t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
-                          RADEON_TXFORMAT_ALPHA_IN_MAP);
-      t->pp_txfilter &= ~RADEON_YUV_TO_RGB;
-
-      if ( VALID_FORMAT( baseImage->TexFormat->MesaFormat ) ) {
-         t->pp_txformat |= tx_table[ baseImage->TexFormat->MesaFormat ].format;
-         t->pp_txfilter |= tx_table[ baseImage->TexFormat->MesaFormat ].filter;
-      }
-      else {
-         _mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__);
-         return;
-      }
-   }
-
-   texelBytes = baseImage->TexFormat->TexelBytes;
-
-   /* Compute which mipmap levels we really want to send to the hardware.
-    */
-
-   if (tObj->Target != GL_TEXTURE_CUBE_MAP)
-      driCalculateTextureFirstLastLevel( (driTextureObject *) t );
-   else {
-      /* r100 can't handle mipmaps for cube/3d textures, so don't waste
-         memory for them */
-      t->base.firstLevel = t->base.lastLevel = tObj->BaseLevel;
-   }
-   log2Width  = tObj->Image[0][t->base.firstLevel]->WidthLog2;
-   log2Height = tObj->Image[0][t->base.firstLevel]->HeightLog2;
-   log2Depth  = tObj->Image[0][t->base.firstLevel]->DepthLog2;
-
-   numLevels = t->base.lastLevel - t->base.firstLevel + 1;
-
-   assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS);
-
-   /* Calculate mipmap offsets and dimensions for blitting (uploading)
-    * The idea is that we lay out the mipmap levels within a block of
-    * memory organized as a rectangle of width BLIT_WIDTH_BYTES.
-    */
-   curOffset = 0;
-   blitWidth = BLIT_WIDTH_BYTES;
-   t->tile_bits = 0;
-
-   /* figure out if this texture is suitable for tiling. */
-   if (texelBytes && (tObj->Target != GL_TEXTURE_RECTANGLE_NV)) {
-      if (rmesa->texmicrotile && (baseImage->Height > 1)) {
-	 /* allow 32 (bytes) x 1 mip (which will use two times the space
-	    the non-tiled version would use) max if base texture is large enough */
-	 if ((numLevels == 1) ||
-	   (((baseImage->Width * texelBytes / baseImage->Height) <= 32) &&
-	       (baseImage->Width * texelBytes > 64)) ||
-	    ((baseImage->Width * texelBytes / baseImage->Height) <= 16)) {
-	    /* R100 has two microtile bits (only the txoffset reg, not the blitter)
-	       weird: X2 + OPT: 32bit correct, 16bit completely hosed
-		      X2: 32bit correct, 16bit correct
-		      OPT: 32bit large mips correct, small mips hosed, 16bit completely hosed */
-	    t->tile_bits |= RADEON_TXO_MICRO_TILE_X2 /*| RADEON_TXO_MICRO_TILE_OPT*/;
-	 }
-      }
-      if ((baseImage->Width * texelBytes >= 256) && (baseImage->Height >= 16)) {
-	 /* R100 disables macro tiling only if mip width is smaller than 256 bytes, and not
-	    in the case if height is smaller than 16 (not 100% sure), as does the r200,
-	    so need to disable macro tiling in that case */
-	 if ((numLevels == 1) || ((baseImage->Width * texelBytes / baseImage->Height) <= 4)) {
-	    t->tile_bits |= RADEON_TXO_MACRO_TILE;
-	 }
-      }
-   }
-
-   for (i = 0; i < numLevels; i++) {
-      const struct gl_texture_image *texImage;
-      GLuint size;
-
-      texImage = tObj->Image[0][i + t->base.firstLevel];
-      if ( !texImage )
-	 break;
-
-      /* find image size in bytes */
-      if (texImage->IsCompressed) {
-      /* need to calculate the size AFTER padding even though the texture is
-         submitted without padding.
-         Only handle pot textures currently - don't know if npot is even possible,
-         size calculation would certainly need (trivial) adjustments.
-         Align (and later pad) to 32byte, not sure what that 64byte blit width is
-         good for? */
-         if ((t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK) == RADEON_TXFORMAT_DXT1) {
-            /* RGB_DXT1/RGBA_DXT1, 8 bytes per block */
-            if ((texImage->Width + 3) < 8) /* width one block */
-               size = texImage->CompressedSize * 4;
-            else if ((texImage->Width + 3) < 16)
-               size = texImage->CompressedSize * 2;
-            else size = texImage->CompressedSize;
-         }
-         else /* DXT3/5, 16 bytes per block */
-            if ((texImage->Width + 3) < 8)
-               size = texImage->CompressedSize * 2;
-            else size = texImage->CompressedSize;
-      }
-      else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-	 size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height;
-      }
-      else if (t->tile_bits & RADEON_TXO_MICRO_TILE_X2) {
-	 /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
-	    though the actual offset may be different (if texture is less than
-	    32 bytes width) to the untiled case */
-	 int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
-	 size = (w * ((texImage->Height + 1) / 2)) * texImage->Depth;
-	 blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-      }
-      else {
-	 int w = (texImage->Width * texelBytes + 31) & ~31;
-	 size = w * texImage->Height * texImage->Depth;
-	 blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-      }
-      assert(size > 0);
-
-      /* Align to 32-byte offset.  It is faster to do this unconditionally
-       * (no branch penalty).
-       */
-
-      curOffset = (curOffset + 0x1f) & ~0x1f;
-
-      if (texelBytes) {
-	 t->image[0][i].x = curOffset; /* fix x and y coords up later together with offset */
-	 t->image[0][i].y = 0;
-	 t->image[0][i].width = MIN2(size / texelBytes, blitWidth);
-	 t->image[0][i].height = (size / texelBytes) / t->image[0][i].width;
-      }
-      else {
-         t->image[0][i].x = curOffset % BLIT_WIDTH_BYTES;
-         t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES;
-         t->image[0][i].width  = MIN2(size, BLIT_WIDTH_BYTES);
-         t->image[0][i].height = size / t->image[0][i].width;     
-      }
-
-#if 0
-      /* for debugging only and only  applicable to non-rectangle targets */
-      assert(size % t->image[0][i].width == 0);
-      assert(t->image[0][i].x == 0
-             || (size < BLIT_WIDTH_BYTES && t->image[0][i].height == 1));
-#endif
-
-      if (0)
-         fprintf(stderr,
-                 "level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
-                 i, texImage->Width, texImage->Height,
-                 t->image[0][i].x, t->image[0][i].y,
-                 t->image[0][i].width, t->image[0][i].height, size, curOffset);
-
-      curOffset += size;
-
-   }
-
-   /* Align the total size of texture memory block.
-    */
-   t->base.totalSize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
-
-   /* Setup remaining cube face blits, if needed */
-   if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-      const GLuint faceSize = t->base.totalSize;
-      GLuint face;
-      /* reuse face 0 x/y/width/height - just update the offset when uploading */
-      for (face = 1; face < 6; face++) {
-         for (i = 0; i < numLevels; i++) {
-            t->image[face][i].x =  t->image[0][i].x;
-            t->image[face][i].y =  t->image[0][i].y;
-            t->image[face][i].width  = t->image[0][i].width;
-            t->image[face][i].height = t->image[0][i].height;
-         }
-      }
-      t->base.totalSize = 6 * faceSize; /* total texmem needed */
-   }
-
-   /* Hardware state:
-    */
-   t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
-   t->pp_txfilter |= (numLevels - 1) << RADEON_MAX_MIP_LEVEL_SHIFT;
-
-   t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
-		       RADEON_TXFORMAT_HEIGHT_MASK |
-                       RADEON_TXFORMAT_CUBIC_MAP_ENABLE |
-                       RADEON_TXFORMAT_F5_WIDTH_MASK |
-                       RADEON_TXFORMAT_F5_HEIGHT_MASK);
-   t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
-		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
-
-   if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-      assert(log2Width == log2Height);
-      t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_F5_WIDTH_SHIFT) |
-                         (log2Height << RADEON_TXFORMAT_F5_HEIGHT_SHIFT) |
-                         (RADEON_TXFORMAT_CUBIC_MAP_ENABLE));
-      t->pp_cubic_faces = ((log2Width << RADEON_FACE_WIDTH_1_SHIFT) |
-                           (log2Height << RADEON_FACE_HEIGHT_1_SHIFT) |
-                           (log2Width << RADEON_FACE_WIDTH_2_SHIFT) |
-                           (log2Height << RADEON_FACE_HEIGHT_2_SHIFT) |
-                           (log2Width << RADEON_FACE_WIDTH_3_SHIFT) |
-                           (log2Height << RADEON_FACE_HEIGHT_3_SHIFT) |
-                           (log2Width << RADEON_FACE_WIDTH_4_SHIFT) |
-                           (log2Height << RADEON_FACE_HEIGHT_4_SHIFT));
-   }
-
-   t->pp_txsize = (((tObj->Image[0][t->base.firstLevel]->Width - 1) << 0) |
-                   ((tObj->Image[0][t->base.firstLevel]->Height - 1) << 16));
-
-   /* Only need to round to nearest 32 for textures, but the blitter
-    * requires 64-byte aligned pitches, and we may/may not need the
-    * blitter.   NPOT only!
-    */
-   if ( !t->image_override ) {
-      if (baseImage->IsCompressed)
-         t->pp_txpitch = (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63);
-      else
-         t->pp_txpitch = ((tObj->Image[0][t->base.firstLevel]->Width * texelBytes) + 63) & ~(63);
-      t->pp_txpitch -= 32;
-   }
-
-   t->dirty_state = TEX_ALL;
-
-   /* FYI: radeonUploadTexImages( rmesa, t ); used to be called here */
-}
-
-
-
 /* ================================================================
  * Texture combine functions
  */
@@ -503,7 +260,7 @@ do {							\
 
 static GLboolean radeonUpdateTextureEnv( GLcontext *ctx, int unit )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
    GLuint color_combine, alpha_combine;
    const GLuint color_combine0 = RADEON_COLOR_ARG_A_ZERO | RADEON_COLOR_ARG_B_ZERO
@@ -846,22 +603,21 @@ static GLboolean radeonUpdateTextureEnv( GLcontext *ctx, int unit )
 void radeonSetTexOffset(__DRIcontext * pDRICtx, GLint texname,
                         unsigned long long offset, GLint depth, GLuint pitch)
 {
-	radeonContextPtr rmesa = pDRICtx->driverPrivate;
+	r100ContextPtr rmesa = pDRICtx->driverPrivate;
 	struct gl_texture_object *tObj =
-	    _mesa_lookup_texture(rmesa->glCtx, texname);
-	radeonTexObjPtr t;
+	    _mesa_lookup_texture(rmesa->radeon.glCtx, texname);
+	radeonTexObjPtr t = radeon_tex_obj(tObj);
 
 	if (tObj == NULL)
 		return;
 
-	t = (radeonTexObjPtr) tObj->DriverData;
-
 	t->image_override = GL_TRUE;
 
 	if (!offset)
 		return;
-
-	t->pp_txoffset = offset;
+	
+	t->bo = NULL;
+	t->override_offset = offset;
 	t->pp_txpitch = pitch - 32;
 
 	switch (depth) {
@@ -881,6 +637,125 @@ void radeonSetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 	}
 }
 
+void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format,
+			 __DRIdrawable *dPriv)
+{
+	struct gl_texture_unit *texUnit;
+	struct gl_texture_object *texObj;
+	struct gl_texture_image *texImage;
+	struct radeon_renderbuffer *rb;
+	radeon_texture_image *rImage;
+	radeonContextPtr radeon;
+	r100ContextPtr rmesa;
+	struct radeon_framebuffer *rfb;
+	radeonTexObjPtr t;
+	uint32_t pitch_val;
+	uint32_t internalFormat, type, format;
+
+	type = GL_BGRA;
+	format = GL_UNSIGNED_BYTE;
+	internalFormat = (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT ? 3 : 4);
+
+	radeon = pDRICtx->driverPrivate;
+	rmesa = pDRICtx->driverPrivate;
+
+	rfb = dPriv->driverPrivate;
+        texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit];
+	texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target);
+        texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0);
+
+	rImage = get_radeon_texture_image(texImage);
+	t = radeon_tex_obj(texObj);
+        if (t == NULL) {
+    	    return;
+    	}
+
+	radeon_update_renderbuffers(pDRICtx, dPriv);
+	/* back & depth buffer are useless free them right away */
+	rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer;
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+        rb->bo = NULL;
+	}
+	rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+	if (rb && rb->bo) {
+		radeon_bo_unref(rb->bo);
+		rb->bo = NULL;
+	}
+	rb = rfb->color_rb[0];
+	if (rb->bo == NULL) {
+		/* Failed to BO for the buffer */
+		return;
+	}
+	
+	_mesa_lock_texture(radeon->glCtx, texObj);
+	if (t->bo) {
+		radeon_bo_unref(t->bo);
+		t->bo = NULL;
+	}
+	if (rImage->bo) {
+		radeon_bo_unref(rImage->bo);
+		rImage->bo = NULL;
+	}
+	if (t->mt) {
+		radeon_miptree_unreference(t->mt);
+		t->mt = NULL;
+	}
+	if (rImage->mt) {
+		radeon_miptree_unreference(rImage->mt);
+		rImage->mt = NULL;
+	}
+	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
+				   rb->width, rb->height, 1, 0, rb->cpp);
+	texImage->RowStride = rb->pitch / rb->cpp;
+	texImage->TexFormat = radeonChooseTextureFormat(radeon->glCtx,
+							internalFormat,
+							type, format, 0);
+	rImage->bo = rb->bo;
+	radeon_bo_ref(rImage->bo);
+	t->bo = rb->bo;
+	radeon_bo_ref(t->bo);
+	t->tile_bits = 0;
+	t->image_override = GL_TRUE;
+	t->override_offset = 0;
+	t->pp_txpitch &= (1 << 13) -1;
+	pitch_val = rb->pitch;
+	switch (rb->cpp) {
+	case 4:
+		if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT)
+			t->pp_txformat = tx_table[MESA_FORMAT_RGB888].format;
+		else
+			t->pp_txformat = tx_table[MESA_FORMAT_ARGB8888].format;
+		t->pp_txfilter |= tx_table[MESA_FORMAT_ARGB8888].filter;
+		break;
+	case 3:
+	default:
+		t->pp_txformat = tx_table[MESA_FORMAT_RGB888].format;
+		t->pp_txfilter |= tx_table[MESA_FORMAT_RGB888].filter;
+		break;
+	case 2:
+		t->pp_txformat = tx_table[MESA_FORMAT_RGB565].format;
+		t->pp_txfilter |= tx_table[MESA_FORMAT_RGB565].filter;
+		break;
+	}
+        t->pp_txsize = ((rb->width - 1) << RADEON_TEX_USIZE_SHIFT)
+		   | ((rb->height - 1) << RADEON_TEX_VSIZE_SHIFT);
+        t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2;
+	t->pp_txpitch = pitch_val;
+        t->pp_txpitch -= 32;
+
+	t->validated = GL_TRUE;
+	_mesa_unlock_texture(radeon->glCtx, texObj);
+	return;
+}
+
+
+void radeonSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
+{
+        radeonSetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv);
+}
+
+
 #define TEXOBJ_TXFILTER_MASK (RADEON_MAX_MIP_LEVEL_MASK |	\
 			      RADEON_MIN_FILTER_MASK | 		\
 			      RADEON_MAG_FILTER_MASK |		\
@@ -901,12 +776,53 @@ void radeonSetTexOffset(__DRIcontext * pDRICtx, GLint texname,
                               RADEON_TXFORMAT_NON_POWER2)
 
 
-static void import_tex_obj_state( radeonContextPtr rmesa,
+static void disable_tex_obj_state( r100ContextPtr rmesa, 
+				   int unit )
+{
+   RADEON_STATECHANGE( rmesa, tex[unit] );
+
+   RADEON_STATECHANGE( rmesa, tcl );
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_ST_BIT(unit) |
+					     RADEON_Q_BIT(unit));
+   
+   if (rmesa->radeon.TclFallback & (RADEON_TCL_FALLBACK_TEXGEN_0<<unit)) {
+     TCL_FALLBACK( rmesa->radeon.glCtx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE);
+     rmesa->recheck_texgen[unit] = GL_TRUE;
+   }
+
+   if (rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] & RADEON_TXFORMAT_CUBIC_MAP_ENABLE) {
+     /* this seems to be a genuine (r100 only?) hw bug. Need to remove the
+	cubic_map bit on unit 2 when the unit is disabled, otherwise every
+	2nd (2d) mipmap on unit 0 will be broken (may not be needed for other
+	units, better be safe than sorry though).*/
+     RADEON_STATECHANGE( rmesa, tex[unit] );
+     rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] &= ~RADEON_TXFORMAT_CUBIC_MAP_ENABLE;
+   }
+
+   {
+      GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4;
+      GLuint tmp = rmesa->TexGenEnabled;
+
+      rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_TEXMAT_0_ENABLE<<unit);
+      rmesa->TexGenEnabled &= ~(RADEON_TEXMAT_0_ENABLE<<unit);
+      rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_INPUT_MASK<<inputshift);
+      rmesa->TexGenNeedNormals[unit] = 0;
+      rmesa->TexGenEnabled |= 
+	(RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift;
+
+      if (tmp != rmesa->TexGenEnabled) {
+	rmesa->recheck_texgen[unit] = GL_TRUE;
+	rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
+      }
+   }
+}
+
+static void import_tex_obj_state( r100ContextPtr rmesa,
 				  int unit,
 				  radeonTexObjPtr texobj )
 {
 /* do not use RADEON_DB_STATE to avoid stale texture caches */
-   int *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0];
+   uint32_t *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0];
    GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
 
    RADEON_STATECHANGE( rmesa, tex[unit] );
@@ -915,10 +831,9 @@ static void import_tex_obj_state( radeonContextPtr rmesa,
    cmd[TEX_PP_TXFILTER] |= texobj->pp_txfilter & TEXOBJ_TXFILTER_MASK;
    cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
    cmd[TEX_PP_TXFORMAT] |= texobj->pp_txformat & TEXOBJ_TXFORMAT_MASK;
-   cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset;
    cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color;
 
-   if (texobj->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+   if (texobj->base.Target == GL_TEXTURE_RECTANGLE_NV) {
       GLuint *txr_cmd = RADEON_DB_STATE( txr[unit] );
       txr_cmd[TXR_PP_TEX_SIZE] = texobj->pp_txsize; /* NPOT only! */
       txr_cmd[TXR_PP_TEX_PITCH] = texobj->pp_txpitch; /* NPOT only! */
@@ -928,22 +843,12 @@ static void import_tex_obj_state( radeonContextPtr rmesa,
    else {
       se_coord_fmt &= ~(RADEON_VTX_ST0_NONPARAMETRIC << unit);
 
-      if (texobj->base.tObj->Target == GL_TEXTURE_CUBE_MAP) {
-	 int *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0];
-	 GLuint bytesPerFace = texobj->base.totalSize / 6;
-	 ASSERT(texobj->base.totalSize % 6 == 0);
+      if (texobj->base.Target == GL_TEXTURE_CUBE_MAP) {
+	 uint32_t *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0];
 
 	 RADEON_STATECHANGE( rmesa, cube[unit] );
 	 cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
-	 /* dont know if this setup conforms to OpenGL.. 
-	  * at least it matches the behavior of mesa software renderer
-	  */
-	 cube_cmd[CUBE_PP_CUBIC_OFFSET_0] = texobj->pp_txoffset; /* right */
-	 cube_cmd[CUBE_PP_CUBIC_OFFSET_1] = texobj->pp_txoffset + 1 * bytesPerFace; /* left */
-	 cube_cmd[CUBE_PP_CUBIC_OFFSET_2] = texobj->pp_txoffset + 2 * bytesPerFace; /* top */
-	 cube_cmd[CUBE_PP_CUBIC_OFFSET_3] = texobj->pp_txoffset + 3 * bytesPerFace; /* bottom */
-	 cube_cmd[CUBE_PP_CUBIC_OFFSET_4] = texobj->pp_txoffset + 4 * bytesPerFace; /* front */
-	 cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset + 5 * bytesPerFace; /* back */
+	 /* state filled out in the cube_emit */
       }
    }
 
@@ -952,13 +857,11 @@ static void import_tex_obj_state( radeonContextPtr rmesa,
       rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
    }
 
-   texobj->dirty_state &= ~(1<<unit);
+   rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
 }
 
 
-
-
-static void set_texgen_matrix( radeonContextPtr rmesa, 
+static void set_texgen_matrix( r100ContextPtr rmesa, 
 			       GLuint unit,
 			       const GLfloat *s_plane,
 			       const GLfloat *t_plane,
@@ -986,14 +889,14 @@ static void set_texgen_matrix( radeonContextPtr rmesa,
    rmesa->TexGenMatrix[unit].m[15] = q_plane[3];
 
    rmesa->TexGenEnabled |= RADEON_TEXMAT_0_ENABLE << unit;
-   rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+   rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
 }
 
 /* Returns GL_FALSE if fallback required.
  */
 static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
    GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4;
    GLuint tmp = rmesa->TexGenEnabled;
@@ -1094,283 +997,187 @@ static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit )
    }
 
    if (tmp != rmesa->TexGenEnabled) {
-      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+      rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
    }
 
    return GL_TRUE;
 }
 
-
-static void disable_tex( GLcontext *ctx, int unit )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-
-   if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<unit)) {
-      /* Texture unit disabled */
-      if ( rmesa->state.texture.unit[unit].texobj != NULL ) {
-	 /* The old texture is no longer bound to this texture unit.
-	  * Mark it as such.
-	  */
-
-	 rmesa->state.texture.unit[unit].texobj->base.bound &= ~(1UL << unit);
-	 rmesa->state.texture.unit[unit].texobj = NULL;
-      }
-
-      RADEON_STATECHANGE( rmesa, ctx );
-      rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= 
-	  ~((RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit);
-
-      RADEON_STATECHANGE( rmesa, tcl );
-      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_ST_BIT(unit) |
-						RADEON_Q_BIT(unit));
-
-      if (rmesa->TclFallback & (RADEON_TCL_FALLBACK_TEXGEN_0<<unit)) {
-	 TCL_FALLBACK( ctx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE);
-	 rmesa->recheck_texgen[unit] = GL_TRUE;
-      }
-
-      if (rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] & RADEON_TXFORMAT_CUBIC_MAP_ENABLE) {
-      /* this seems to be a genuine (r100 only?) hw bug. Need to remove the
-         cubic_map bit on unit 2 when the unit is disabled, otherwise every
-	 2nd (2d) mipmap on unit 0 will be broken (may not be needed for other
-	 units, better be safe than sorry though).*/
-	 RADEON_STATECHANGE( rmesa, tex[unit] );
-	 rmesa->hw.tex[unit].cmd[TEX_PP_TXFORMAT] &= ~RADEON_TXFORMAT_CUBIC_MAP_ENABLE;
-      }
-
-      {
-	 GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4;
-	 GLuint tmp = rmesa->TexGenEnabled;
-
-	 rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_TEXMAT_0_ENABLE<<unit);
-	 rmesa->TexGenEnabled &= ~(RADEON_TEXMAT_0_ENABLE<<unit);
-	 rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_INPUT_MASK<<inputshift);
-	 rmesa->TexGenNeedNormals[unit] = 0;
-	 rmesa->TexGenEnabled |= 
-	     (RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift;
-
-	 if (tmp != rmesa->TexGenEnabled) {
-	    rmesa->recheck_texgen[unit] = GL_TRUE;
-	    rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
-	 }
-      }
-   }
-}
-
-static GLboolean enable_tex_2d( GLcontext *ctx, int unit )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
-
-   /* Need to load the 2d images associated with this unit.
-    */
-   if (t->pp_txformat & RADEON_TXFORMAT_NON_POWER2) {
-      t->pp_txformat &= ~RADEON_TXFORMAT_NON_POWER2;
-      t->base.dirty_images[0] = ~0;
-   }
-
-   ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D);
-
-   if ( t->base.dirty_images[0] ) {
-      RADEON_FIREVERTICES( rmesa );
-      radeonSetTexImages( rmesa, tObj );
-      radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, 0 );
-      if ( !t->base.memBlock && !t->image_override ) 
-	return GL_FALSE;
-   }
-
-   return GL_TRUE;
-}
-
-static GLboolean enable_tex_cube( GLcontext *ctx, int unit )
+/**
+ * Compute the cached hardware register values for the given texture object.
+ *
+ * \param rmesa Context pointer
+ * \param t the r300 texture object
+ */
+static GLboolean setup_hardware_state(r100ContextPtr rmesa, radeonTexObj *t, int unit)
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
-   GLuint face;
+   const struct gl_texture_image *firstImage;
+   GLint log2Width, log2Height, log2Depth, texelBytes;
 
-   /* Need to load the 2d images associated with this unit.
-    */
-   if (t->pp_txformat & RADEON_TXFORMAT_NON_POWER2) {
-      t->pp_txformat &= ~RADEON_TXFORMAT_NON_POWER2;
-      for (face = 0; face < 6; face++)
-         t->base.dirty_images[face] = ~0;
+   if ( t->bo ) {
+	return GL_TRUE;
    }
 
-   ASSERT(tObj->Target == GL_TEXTURE_CUBE_MAP);
+   firstImage = t->base.Image[0][t->mt->firstLevel];   
 
-   if ( t->base.dirty_images[0] || t->base.dirty_images[1] ||
-        t->base.dirty_images[2] || t->base.dirty_images[3] ||
-        t->base.dirty_images[4] || t->base.dirty_images[5] ) {
-      /* flush */
-      RADEON_FIREVERTICES( rmesa );
-      /* layout memory space, once for all faces */
-      radeonSetTexImages( rmesa, tObj );
+   if (firstImage->Border > 0) {
+      fprintf(stderr, "%s: border\n", __FUNCTION__);
+      return GL_FALSE;
    }
 
-   /* upload (per face) */
-   for (face = 0; face < 6; face++) {
-      if (t->base.dirty_images[face]) {
-         radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, face );
+   log2Width  = firstImage->WidthLog2;
+   log2Height = firstImage->HeightLog2;
+   log2Depth  = firstImage->DepthLog2;
+   texelBytes = firstImage->TexFormat->TexelBytes;
+
+   if (!t->image_override) {
+      if (VALID_FORMAT(firstImage->TexFormat->MesaFormat)) {
+	const struct tx_table *table = tx_table;
+
+	 t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+			     RADEON_TXFORMAT_ALPHA_IN_MAP);
+	 t->pp_txfilter &= ~RADEON_YUV_TO_RGB;	 
+	 
+	 t->pp_txformat |= table[ firstImage->TexFormat->MesaFormat ].format;
+	 t->pp_txfilter |= table[ firstImage->TexFormat->MesaFormat ].filter;
+      } else {
+	 _mesa_problem(NULL, "unexpected texture format in %s",
+		       __FUNCTION__);
+	 return GL_FALSE;
       }
    }
-      
-   if ( !t->base.memBlock ) {
-      /* texmem alloc failed, use s/w fallback */
-      return GL_FALSE;
+   
+   t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
+   t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << RADEON_MAX_MIP_LEVEL_SHIFT;
+	
+   t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
+		       RADEON_TXFORMAT_HEIGHT_MASK |
+		       RADEON_TXFORMAT_CUBIC_MAP_ENABLE |
+		       RADEON_TXFORMAT_F5_WIDTH_MASK |
+		       RADEON_TXFORMAT_F5_HEIGHT_MASK);
+   t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
+		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
+   
+   t->tile_bits = 0;
+   
+   if (t->base.Target == GL_TEXTURE_CUBE_MAP) {
+      ASSERT(log2Width == log2Height);
+      t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_F5_WIDTH_SHIFT) |
+			 (log2Height << RADEON_TXFORMAT_F5_HEIGHT_SHIFT) |
+			 /* don't think we need this bit, if it exists at all - fglrx does not set it */
+			 (RADEON_TXFORMAT_CUBIC_MAP_ENABLE));
+      t->pp_cubic_faces = ((log2Width << RADEON_FACE_WIDTH_1_SHIFT) |
+                           (log2Height << RADEON_FACE_HEIGHT_1_SHIFT) |
+                           (log2Width << RADEON_FACE_WIDTH_2_SHIFT) |
+                           (log2Height << RADEON_FACE_HEIGHT_2_SHIFT) |
+                           (log2Width << RADEON_FACE_WIDTH_3_SHIFT) |
+                           (log2Height << RADEON_FACE_HEIGHT_3_SHIFT) |
+                           (log2Width << RADEON_FACE_WIDTH_4_SHIFT) |
+                           (log2Height << RADEON_FACE_HEIGHT_4_SHIFT));
    }
 
-   return GL_TRUE;
-}
-
-static GLboolean enable_tex_rect( GLcontext *ctx, int unit )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+   t->pp_txsize = (((firstImage->Width - 1) << RADEON_TEX_USIZE_SHIFT)
+		   | ((firstImage->Height - 1) << RADEON_TEX_VSIZE_SHIFT));
 
-   if (!(t->pp_txformat & RADEON_TXFORMAT_NON_POWER2)) {
-      t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2;
-      t->base.dirty_images[0] = ~0;
+   if ( !t->image_override ) {
+      if (firstImage->IsCompressed)
+         t->pp_txpitch = (firstImage->Width + 63) & ~(63);
+      else
+         t->pp_txpitch = ((firstImage->Width * texelBytes) + 63) & ~(63);
+      t->pp_txpitch -= 32;
    }
 
-   ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV);
-
-   if ( t->base.dirty_images[0] ) {
-      RADEON_FIREVERTICES( rmesa );
-      radeonSetTexImages( rmesa, tObj );
-      radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, 0 );
-      if ( !t->base.memBlock &&
-           !t->image_override /* && !rmesa->prefer_gart_client_texturing  FIXME */ ) {
-	 fprintf(stderr, "%s: upload failed\n", __FUNCTION__);
-	 return GL_FALSE;
-      }
+   if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) {
+      t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2;
    }
 
    return GL_TRUE;
 }
 
-
-static GLboolean update_tex_common( GLcontext *ctx, int unit )
+static GLboolean radeon_validate_texture(GLcontext *ctx, struct gl_texture_object *texObj, int unit)
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
-   GLenum format;
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
+   radeonTexObj *t = radeon_tex_obj(texObj);
+   int ret;
 
-   /* Fallback if there's a texture border */
-   if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 ) {
-      fprintf(stderr, "%s: border\n", __FUNCTION__);
+   if (!radeon_validate_texture_miptree(ctx, texObj))
       return GL_FALSE;
-   }
+
+   ret = setup_hardware_state(rmesa, t, unit);
+   if (ret == GL_FALSE)
+     return GL_FALSE;
+
    /* yuv conversion only works in first unit */
    if (unit != 0 && (t->pp_txfilter & RADEON_YUV_TO_RGB))
       return GL_FALSE;
 
-   /* Update state if this is a different texture object to last
-    * time.
-    */
-   if ( rmesa->state.texture.unit[unit].texobj != t ) {
-      if ( rmesa->state.texture.unit[unit].texobj != NULL ) {
-	 /* The old texture is no longer bound to this texture unit.
-	  * Mark it as such.
-	  */
-
-	 rmesa->state.texture.unit[unit].texobj->base.bound &= 
-	     ~(1UL << unit);
-      }
-
-      rmesa->state.texture.unit[unit].texobj = t;
-      t->base.bound |= (1UL << unit);
-      t->dirty_state |= 1<<unit;
-      driUpdateTextureLRU( (driTextureObject *) t ); /* XXX: should be locked! */
-   }
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= 
+     (RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit;
 
+   RADEON_STATECHANGE( rmesa, tcl );
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_ST_BIT(unit);
 
-   /* Newly enabled?
-    */
-   if ( !(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<unit))) {
-      RADEON_STATECHANGE( rmesa, ctx );
-      rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= 
-	  (RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit;
+   rmesa->recheck_texgen[unit] = GL_TRUE;
 
-      RADEON_STATECHANGE( rmesa, tcl );
-
-      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_ST_BIT(unit);
-
-      rmesa->recheck_texgen[unit] = GL_TRUE;
-   }
-
-   if (t->dirty_state & (1<<unit)) {
-      import_tex_obj_state( rmesa, unit, t );
-      /* may need to update texture matrix (for texrect adjustments) */
-      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
-   }
+   import_tex_obj_state( rmesa, unit, t );
 
    if (rmesa->recheck_texgen[unit]) {
       GLboolean fallback = !radeon_validate_texgen( ctx, unit );
       TCL_FALLBACK( ctx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), fallback);
       rmesa->recheck_texgen[unit] = 0;
-      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+      rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
    }
 
-   format = tObj->Image[0][tObj->BaseLevel]->_BaseFormat;
-   if ( rmesa->state.texture.unit[unit].format != format ||
-	rmesa->state.texture.unit[unit].envMode != texUnit->EnvMode ) {
-      rmesa->state.texture.unit[unit].format = format;
-      rmesa->state.texture.unit[unit].envMode = texUnit->EnvMode;
-      if ( ! radeonUpdateTextureEnv( ctx, unit ) ) {
-	 return GL_FALSE;
-      }
+   if ( ! radeonUpdateTextureEnv( ctx, unit ) ) {
+     return GL_FALSE;
    }
-
    FALLBACK( rmesa, RADEON_FALLBACK_BORDER_MODE, t->border_fallback );
+
+   t->validated = GL_TRUE;
    return !t->border_fallback;
 }
 
-
-
 static GLboolean radeonUpdateTextureUnit( GLcontext *ctx, int unit )
 {
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
 
-   if ( texUnit->_ReallyEnabled & (TEXTURE_RECT_BIT) ) {
-      return (enable_tex_rect( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
-   }
-   else if ( texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) {
-      return (enable_tex_2d( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
-   }
-   else if ( texUnit->_ReallyEnabled & (TEXTURE_CUBE_BIT) ) {
-      return (enable_tex_cube( ctx, unit ) &&
-	      update_tex_common( ctx, unit ));
+   if (ctx->Texture.Unit[unit]._ReallyEnabled & TEXTURE_3D_BIT) {
+     return GL_FALSE;
    }
-   else if ( texUnit->_ReallyEnabled ) {
-      return GL_FALSE;
+
+   if (!ctx->Texture.Unit[unit]._ReallyEnabled) {
+     /* disable the unit */
+     disable_tex_obj_state(rmesa, unit);
+     return GL_TRUE;
    }
-   else {
-      disable_tex( ctx, unit );
-      return GL_TRUE;
+
+   if (!radeon_validate_texture(ctx, ctx->Texture.Unit[unit]._Current, unit)) {
+    _mesa_warning(ctx,
+		  "failed to validate texture for unit %d.\n",
+		  unit);
+    rmesa->state.texture.unit[unit].texobj = NULL;
+    return GL_FALSE;
    }
+   rmesa->state.texture.unit[unit].texobj = radeon_tex_obj(ctx->Texture.Unit[unit]._Current);
+   return GL_TRUE;
 }
 
 void radeonUpdateTextureState( GLcontext *ctx )
 {
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   r100ContextPtr rmesa = R100_CONTEXT(ctx);
    GLboolean ok;
 
+   /* set the ctx all textures off */
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~((RADEON_TEX_ENABLE_MASK) | (RADEON_TEX_BLEND_ENABLE_MASK));
+
    ok = (radeonUpdateTextureUnit( ctx, 0 ) &&
 	 radeonUpdateTextureUnit( ctx, 1 ) &&
 	 radeonUpdateTextureUnit( ctx, 2 ));
 
    FALLBACK( rmesa, RADEON_FALLBACK_TEXTURE, !ok );
 
-   if (rmesa->TclFallback)
+   if (rmesa->radeon.TclFallback)
       radeonChooseVertexState( ctx );
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.c b/src/mesa/drivers/dri/radeon/radeon_texture.c
new file mode 100644
index 00000000000..0d87f152e99
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_texture.c
@@ -0,0 +1,996 @@
+/*
+ * Copyright (C) 2008 Nicolai Haehnle.
+ * Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+ *
+ * The Weather Channel (TM) funded Tungsten Graphics to develop the
+ * initial release of the Radeon 8500 driver under the XFree86 license.
+ * This notice must be preserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "main/glheader.h"
+#include "main/imports.h"
+#include "main/context.h"
+#include "main/convolve.h"
+#include "main/mipmap.h"
+#include "main/texcompress.h"
+#include "main/texformat.h"
+#include "main/texstore.h"
+#include "main/teximage.h"
+#include "main/texobj.h"
+#include "main/texgetimage.h"
+
+#include "xmlpool.h"		/* for symbolic values of enum-type options */
+
+#include "radeon_common.h"
+
+#include "radeon_mipmap_tree.h"
+
+
+static void copy_rows(void* dst, GLuint dststride, const void* src, GLuint srcstride,
+	GLuint numrows, GLuint rowsize)
+{
+	assert(rowsize <= dststride);
+	assert(rowsize <= srcstride);
+
+	if (rowsize == srcstride && rowsize == dststride) {
+		memcpy(dst, src, numrows*rowsize);
+	} else {
+		GLuint i;
+		for(i = 0; i < numrows; ++i) {
+			memcpy(dst, src, rowsize);
+			dst += dststride;
+			src += srcstride;
+		}
+	}
+}
+
+/* textures */
+/**
+ * Allocate an empty texture image object.
+ */
+struct gl_texture_image *radeonNewTextureImage(GLcontext *ctx)
+{
+	return CALLOC(sizeof(radeon_texture_image));
+}
+
+/**
+ * Free memory associated with this texture image.
+ */
+void radeonFreeTexImageData(GLcontext *ctx, struct gl_texture_image *timage)
+{
+	radeon_texture_image* image = get_radeon_texture_image(timage);
+
+	if (image->mt) {
+		radeon_miptree_unreference(image->mt);
+		image->mt = 0;
+		assert(!image->base.Data);
+	} else {
+		_mesa_free_texture_image_data(ctx, timage);
+	}
+	if (image->bo) {
+		radeon_bo_unref(image->bo);
+		image->bo = NULL;
+	}
+	if (timage->Data) {
+		_mesa_free_texmemory(timage->Data);
+		timage->Data = NULL;
+	}
+}
+
+/* Set Data pointer and additional data for mapped texture image */
+static void teximage_set_map_data(radeon_texture_image *image)
+{
+	radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel];
+
+	image->base.Data = image->mt->bo->ptr + lvl->faces[image->mtface].offset;
+	image->base.RowStride = lvl->rowstride / image->mt->bpp;
+}
+
+
+/**
+ * Map a single texture image for glTexImage and friends.
+ */
+void radeon_teximage_map(radeon_texture_image *image, GLboolean write_enable)
+{
+	if (image->mt) {
+		assert(!image->base.Data);
+
+		radeon_bo_map(image->mt->bo, write_enable);
+		teximage_set_map_data(image);
+	}
+}
+
+
+void radeon_teximage_unmap(radeon_texture_image *image)
+{
+	if (image->mt) {
+		assert(image->base.Data);
+
+		image->base.Data = 0;
+		radeon_bo_unmap(image->mt->bo);
+	}
+}
+
+static void map_override(GLcontext *ctx, radeonTexObj *t)
+{
+	radeon_texture_image *img = get_radeon_texture_image(t->base.Image[0][0]);
+
+	radeon_bo_map(t->bo, GL_FALSE);
+
+	img->base.Data = t->bo->ptr;
+	_mesa_set_fetch_functions(&img->base, 2);
+}
+
+static void unmap_override(GLcontext *ctx, radeonTexObj *t)
+{
+	radeon_texture_image *img = get_radeon_texture_image(t->base.Image[0][0]);
+
+	radeon_bo_unmap(t->bo);
+
+	img->base.Data = NULL;
+}
+
+/**
+ * Map a validated texture for reading during software rendering.
+ */
+void radeonMapTexture(GLcontext *ctx, struct gl_texture_object *texObj)
+{
+	radeonTexObj* t = radeon_tex_obj(texObj);
+	int face, level;
+
+	if (!radeon_validate_texture_miptree(ctx, texObj))
+	  return;
+
+	/* for r100 3D sw fallbacks don't have mt */
+	if (t->image_override && t->bo)
+		map_override(ctx, t);
+
+	if (!t->mt)
+		return;
+
+	radeon_bo_map(t->mt->bo, GL_FALSE);
+	for(face = 0; face < t->mt->faces; ++face) {
+		for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level)
+			teximage_set_map_data(get_radeon_texture_image(texObj->Image[face][level]));
+	}
+}
+
+void radeonUnmapTexture(GLcontext *ctx, struct gl_texture_object *texObj)
+{
+	radeonTexObj* t = radeon_tex_obj(texObj);
+	int face, level;
+
+	if (t->image_override && t->bo)
+		unmap_override(ctx, t);
+	/* for r100 3D sw fallbacks don't have mt */
+	if (!t->mt)
+	  return;
+
+	for(face = 0; face < t->mt->faces; ++face) {
+		for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level)
+			texObj->Image[face][level]->Data = 0;
+	}
+	radeon_bo_unmap(t->mt->bo);
+}
+
+GLuint radeon_face_for_target(GLenum target)
+{
+	switch (target) {
+	case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
+	case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
+	case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
+	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
+	case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
+	case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
+		return (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
+	default:
+		return 0;
+	}
+}
+
+/**
+ * Wraps Mesa's implementation to ensure that the base level image is mapped.
+ *
+ * This relies on internal details of _mesa_generate_mipmap, in particular
+ * the fact that the memory for recreated texture images is always freed.
+ */
+static void radeon_generate_mipmap(GLcontext *ctx, GLenum target,
+				   struct gl_texture_object *texObj)
+{
+	radeonTexObj* t = radeon_tex_obj(texObj);
+	GLuint nr_faces = (t->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+	int i, face;
+
+
+	_mesa_generate_mipmap(ctx, target, texObj);
+
+	for (face = 0; face < nr_faces; face++) {
+		for (i = texObj->BaseLevel + 1; i < texObj->MaxLevel; i++) {
+			radeon_texture_image *image;
+
+			image = get_radeon_texture_image(texObj->Image[face][i]);
+
+			if (image == NULL)
+				break;
+
+			image->mtlevel = i;
+			image->mtface = face;
+
+			radeon_miptree_unreference(image->mt);
+			image->mt = NULL;
+		}
+	}
+	
+}
+
+void radeonGenerateMipmap(GLcontext* ctx, GLenum target, struct gl_texture_object *texObj)
+{
+	GLuint face = radeon_face_for_target(target);
+	radeon_texture_image *baseimage = get_radeon_texture_image(texObj->Image[face][texObj->BaseLevel]);
+
+	radeon_teximage_map(baseimage, GL_FALSE);
+	radeon_generate_mipmap(ctx, target, texObj);
+	radeon_teximage_unmap(baseimage);
+}
+
+
+/* try to find a format which will only need a memcopy */
+static const struct gl_texture_format *radeonChoose8888TexFormat(radeonContextPtr rmesa,
+								 GLenum srcFormat,
+								 GLenum srcType, GLboolean fbo)
+{
+	const GLuint ui = 1;
+	const GLubyte littleEndian = *((const GLubyte *)&ui);
+
+	/* r100 can only do this */
+	if (IS_R100_CLASS(rmesa->radeonScreen) || fbo)
+	  return _dri_texformat_argb8888;
+
+	if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
+	    (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
+	    (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
+	    (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && littleEndian)) {
+		return &_mesa_texformat_rgba8888;
+	} else if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
+		   (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && littleEndian) ||
+		   (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
+		   (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && !littleEndian)) {
+		return &_mesa_texformat_rgba8888_rev;
+	} else if (IS_R200_CLASS(rmesa->radeonScreen)) {
+		return _dri_texformat_argb8888;
+	} else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
+					    srcType == GL_UNSIGNED_INT_8_8_8_8)) {
+		return &_mesa_texformat_argb8888_rev;
+	} else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && littleEndian) ||
+					    srcType == GL_UNSIGNED_INT_8_8_8_8_REV)) {
+		return &_mesa_texformat_argb8888;
+	} else
+		return _dri_texformat_argb8888;
+}
+
+const struct gl_texture_format *radeonChooseTextureFormat_mesa(GLcontext * ctx,
+							  GLint internalFormat,
+							  GLenum format,
+							  GLenum type)
+{
+	return radeonChooseTextureFormat(ctx, internalFormat, format,
+					 type, 0);
+}
+
+const struct gl_texture_format *radeonChooseTextureFormat(GLcontext * ctx,
+							  GLint internalFormat,
+							  GLenum format,
+							  GLenum type, GLboolean fbo)
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	const GLboolean do32bpt =
+	    (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32);
+	const GLboolean force16bpt =
+	    (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16);
+	(void)format;
+
+#if 0
+	fprintf(stderr, "InternalFormat=%s(%d) type=%s format=%s\n",
+		_mesa_lookup_enum_by_nr(internalFormat), internalFormat,
+		_mesa_lookup_enum_by_nr(type), _mesa_lookup_enum_by_nr(format));
+	fprintf(stderr, "do32bpt=%d force16bpt=%d\n", do32bpt, force16bpt);
+#endif
+
+	switch (internalFormat) {
+	case 4:
+	case GL_RGBA:
+	case GL_COMPRESSED_RGBA:
+		switch (type) {
+		case GL_UNSIGNED_INT_10_10_10_2:
+		case GL_UNSIGNED_INT_2_10_10_10_REV:
+			return do32bpt ? _dri_texformat_argb8888 :
+			    _dri_texformat_argb1555;
+		case GL_UNSIGNED_SHORT_4_4_4_4:
+		case GL_UNSIGNED_SHORT_4_4_4_4_REV:
+			return _dri_texformat_argb4444;
+		case GL_UNSIGNED_SHORT_5_5_5_1:
+		case GL_UNSIGNED_SHORT_1_5_5_5_REV:
+			return _dri_texformat_argb1555;
+		default:
+			return do32bpt ? radeonChoose8888TexFormat(rmesa, format, type, fbo) :
+			    _dri_texformat_argb4444;
+		}
+
+	case 3:
+	case GL_RGB:
+	case GL_COMPRESSED_RGB:
+		switch (type) {
+		case GL_UNSIGNED_SHORT_4_4_4_4:
+		case GL_UNSIGNED_SHORT_4_4_4_4_REV:
+			return _dri_texformat_argb4444;
+		case GL_UNSIGNED_SHORT_5_5_5_1:
+		case GL_UNSIGNED_SHORT_1_5_5_5_REV:
+			return _dri_texformat_argb1555;
+		case GL_UNSIGNED_SHORT_5_6_5:
+		case GL_UNSIGNED_SHORT_5_6_5_REV:
+			return _dri_texformat_rgb565;
+		default:
+			return do32bpt ? _dri_texformat_argb8888 :
+			    _dri_texformat_rgb565;
+		}
+
+	case GL_RGBA8:
+	case GL_RGB10_A2:
+	case GL_RGBA12:
+	case GL_RGBA16:
+		return !force16bpt ?
+			radeonChoose8888TexFormat(rmesa, format, type, fbo) :
+			_dri_texformat_argb4444;
+
+	case GL_RGBA4:
+	case GL_RGBA2:
+		return _dri_texformat_argb4444;
+
+	case GL_RGB5_A1:
+		return _dri_texformat_argb1555;
+
+	case GL_RGB8:
+	case GL_RGB10:
+	case GL_RGB12:
+	case GL_RGB16:
+		return !force16bpt ? _dri_texformat_argb8888 :
+		    _dri_texformat_rgb565;
+
+	case GL_RGB5:
+	case GL_RGB4:
+	case GL_R3_G3_B2:
+		return _dri_texformat_rgb565;
+
+	case GL_ALPHA:
+	case GL_ALPHA4:
+	case GL_ALPHA8:
+	case GL_ALPHA12:
+	case GL_ALPHA16:
+	case GL_COMPRESSED_ALPHA:
+		/* r200: can't use a8 format since interpreting hw I8 as a8 would result
+		   in wrong rgb values (same as alpha value instead of 0). */
+		if (IS_R200_CLASS(rmesa->radeonScreen))
+			return _dri_texformat_al88;
+		else
+			return _dri_texformat_a8;
+	case 1:
+	case GL_LUMINANCE:
+	case GL_LUMINANCE4:
+	case GL_LUMINANCE8:
+	case GL_LUMINANCE12:
+	case GL_LUMINANCE16:
+	case GL_COMPRESSED_LUMINANCE:
+		return _dri_texformat_l8;
+
+	case 2:
+	case GL_LUMINANCE_ALPHA:
+	case GL_LUMINANCE4_ALPHA4:
+	case GL_LUMINANCE6_ALPHA2:
+	case GL_LUMINANCE8_ALPHA8:
+	case GL_LUMINANCE12_ALPHA4:
+	case GL_LUMINANCE12_ALPHA12:
+	case GL_LUMINANCE16_ALPHA16:
+	case GL_COMPRESSED_LUMINANCE_ALPHA:
+		return _dri_texformat_al88;
+
+	case GL_INTENSITY:
+	case GL_INTENSITY4:
+	case GL_INTENSITY8:
+	case GL_INTENSITY12:
+	case GL_INTENSITY16:
+	case GL_COMPRESSED_INTENSITY:
+		return _dri_texformat_i8;
+
+	case GL_YCBCR_MESA:
+		if (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
+		    type == GL_UNSIGNED_BYTE)
+			return &_mesa_texformat_ycbcr;
+		else
+			return &_mesa_texformat_ycbcr_rev;
+
+	case GL_RGB_S3TC:
+	case GL_RGB4_S3TC:
+	case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
+		return &_mesa_texformat_rgb_dxt1;
+
+	case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
+		return &_mesa_texformat_rgba_dxt1;
+
+	case GL_RGBA_S3TC:
+	case GL_RGBA4_S3TC:
+	case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
+		return &_mesa_texformat_rgba_dxt3;
+
+	case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
+		return &_mesa_texformat_rgba_dxt5;
+
+	case GL_ALPHA16F_ARB:
+		return &_mesa_texformat_alpha_float16;
+	case GL_ALPHA32F_ARB:
+		return &_mesa_texformat_alpha_float32;
+	case GL_LUMINANCE16F_ARB:
+		return &_mesa_texformat_luminance_float16;
+	case GL_LUMINANCE32F_ARB:
+		return &_mesa_texformat_luminance_float32;
+	case GL_LUMINANCE_ALPHA16F_ARB:
+		return &_mesa_texformat_luminance_alpha_float16;
+	case GL_LUMINANCE_ALPHA32F_ARB:
+		return &_mesa_texformat_luminance_alpha_float32;
+	case GL_INTENSITY16F_ARB:
+		return &_mesa_texformat_intensity_float16;
+	case GL_INTENSITY32F_ARB:
+		return &_mesa_texformat_intensity_float32;
+	case GL_RGB16F_ARB:
+		return &_mesa_texformat_rgba_float16;
+	case GL_RGB32F_ARB:
+		return &_mesa_texformat_rgba_float32;
+	case GL_RGBA16F_ARB:
+		return &_mesa_texformat_rgba_float16;
+	case GL_RGBA32F_ARB:
+		return &_mesa_texformat_rgba_float32;
+
+	case GL_DEPTH_COMPONENT:
+	case GL_DEPTH_COMPONENT16:
+	case GL_DEPTH_COMPONENT24:
+	case GL_DEPTH_COMPONENT32:
+	case GL_DEPTH_STENCIL_EXT:
+	case GL_DEPTH24_STENCIL8_EXT:
+		return &_mesa_texformat_s8_z24;
+	default:
+		_mesa_problem(ctx,
+			      "unexpected internalFormat 0x%x in %s",
+			      (int)internalFormat, __func__);
+		return NULL;
+	}
+
+	return NULL;		/* never get here */
+}
+
+/**
+ * All glTexImage calls go through this function.
+ */
+static void radeon_teximage(
+	GLcontext *ctx, int dims,
+	GLint face, GLint level,
+	GLint internalFormat,
+	GLint width, GLint height, GLint depth,
+	GLsizei imageSize,
+	GLenum format, GLenum type, const GLvoid * pixels,
+	const struct gl_pixelstore_attrib *packing,
+	struct gl_texture_object *texObj,
+	struct gl_texture_image *texImage,
+	int compressed)
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	radeonTexObj* t = radeon_tex_obj(texObj);
+	radeon_texture_image* image = get_radeon_texture_image(texImage);
+	GLuint dstRowStride;
+	GLint postConvWidth = width;
+	GLint postConvHeight = height;
+	GLuint texelBytes;
+
+	radeon_firevertices(rmesa);
+
+	t->validated = GL_FALSE;
+
+	if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) {
+	       _mesa_adjust_image_for_convolution(ctx, dims, &postConvWidth,
+						  &postConvHeight);
+	}
+
+	/* Choose and fill in the texture format for this image */
+	texImage->TexFormat = radeonChooseTextureFormat(ctx, internalFormat, format, type, 0);
+	_mesa_set_fetch_functions(texImage, dims);
+
+	if (texImage->TexFormat->TexelBytes == 0) {
+		texelBytes = 0;
+		texImage->IsCompressed = GL_TRUE;
+		texImage->CompressedSize =
+			ctx->Driver.CompressedTextureSize(ctx, texImage->Width,
+					   texImage->Height, texImage->Depth,
+					   texImage->TexFormat->MesaFormat);
+	} else {
+		texImage->IsCompressed = GL_FALSE;
+		texImage->CompressedSize = 0;
+
+		texelBytes = texImage->TexFormat->TexelBytes;
+		/* Minimum pitch of 32 bytes */
+		if (postConvWidth * texelBytes < 32) {
+		  postConvWidth = 32 / texelBytes;
+		  texImage->RowStride = postConvWidth;
+		}
+		if (!image->mt) {      
+			assert(texImage->RowStride == postConvWidth);
+		}
+	}
+
+	/* Allocate memory for image */
+	radeonFreeTexImageData(ctx, texImage); /* Mesa core only clears texImage->Data but not image->mt */
+
+	if (t->mt &&
+	    t->mt->firstLevel == level &&
+	    t->mt->lastLevel == level &&
+	    t->mt->target != GL_TEXTURE_CUBE_MAP_ARB &&
+	    !radeon_miptree_matches_image(t->mt, texImage, face, level)) {
+	  radeon_miptree_unreference(t->mt);
+	  t->mt = NULL;
+	}
+
+	if (!t->mt)
+		radeon_try_alloc_miptree(rmesa, t, texImage, face, level);
+	if (t->mt && radeon_miptree_matches_image(t->mt, texImage, face, level)) {
+		radeon_mipmap_level *lvl;
+		image->mt = t->mt;
+		image->mtlevel = level - t->mt->firstLevel;
+		image->mtface = face;
+		radeon_miptree_reference(t->mt);
+		lvl = &image->mt->levels[image->mtlevel];
+		dstRowStride = lvl->rowstride;
+	} else {
+		int size;
+		if (texImage->IsCompressed) {
+			size = texImage->CompressedSize;
+		} else {
+			size = texImage->Width * texImage->Height * texImage->Depth * texImage->TexFormat->TexelBytes;
+		}
+		texImage->Data = _mesa_alloc_texmemory(size);
+	}
+
+	/* Upload texture image; note that the spec allows pixels to be NULL */
+	if (compressed) {
+		pixels = _mesa_validate_pbo_compressed_teximage(
+			ctx, imageSize, pixels, packing, "glCompressedTexImage");
+	} else {
+		pixels = _mesa_validate_pbo_teximage(
+			ctx, dims, width, height, depth,
+			format, type, pixels, packing, "glTexImage");
+	}
+
+	if (pixels) {
+		radeon_teximage_map(image, GL_TRUE);
+
+		if (compressed) {
+			memcpy(texImage->Data, pixels, imageSize);
+		} else {
+			GLuint dstRowStride;
+			if (image->mt) {
+				radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel];
+				dstRowStride = lvl->rowstride;
+			} else {
+				dstRowStride = texImage->Width * texImage->TexFormat->TexelBytes;
+			}
+
+			if (!texImage->TexFormat->StoreImage(ctx, dims,
+						texImage->_BaseFormat,
+						texImage->TexFormat,
+						texImage->Data, 0, 0, 0, /* dstX/Y/Zoffset */
+						dstRowStride,
+						texImage->ImageOffsets,
+						width, height, depth,
+						format, type, pixels, packing))
+				_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
+		}
+
+		/* SGIS_generate_mipmap */
+		if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+			radeon_generate_mipmap(ctx, texObj->Target, texObj);
+		}
+	}
+
+	_mesa_unmap_teximage_pbo(ctx, packing);
+
+	if (pixels)
+	  radeon_teximage_unmap(image);
+
+
+}
+
+void radeonTexImage1D(GLcontext * ctx, GLenum target, GLint level,
+		      GLint internalFormat,
+		      GLint width, GLint border,
+		      GLenum format, GLenum type, const GLvoid * pixels,
+		      const struct gl_pixelstore_attrib *packing,
+		      struct gl_texture_object *texObj,
+		      struct gl_texture_image *texImage)
+{
+	radeon_teximage(ctx, 1, 0, level, internalFormat, width, 1, 1,
+		0, format, type, pixels, packing, texObj, texImage, 0);
+}
+
+void radeonTexImage2D(GLcontext * ctx, GLenum target, GLint level,
+			   GLint internalFormat,
+			   GLint width, GLint height, GLint border,
+			   GLenum format, GLenum type, const GLvoid * pixels,
+			   const struct gl_pixelstore_attrib *packing,
+			   struct gl_texture_object *texObj,
+			   struct gl_texture_image *texImage)
+
+{
+	GLuint face = radeon_face_for_target(target);
+
+	radeon_teximage(ctx, 2, face, level, internalFormat, width, height, 1,
+		0, format, type, pixels, packing, texObj, texImage, 0);
+}
+
+void radeonCompressedTexImage2D(GLcontext * ctx, GLenum target,
+				     GLint level, GLint internalFormat,
+				     GLint width, GLint height, GLint border,
+				     GLsizei imageSize, const GLvoid * data,
+				     struct gl_texture_object *texObj,
+				     struct gl_texture_image *texImage)
+{
+	GLuint face = radeon_face_for_target(target);
+
+	radeon_teximage(ctx, 2, face, level, internalFormat, width, height, 1,
+		imageSize, 0, 0, data, &ctx->Unpack, texObj, texImage, 1);
+}
+
+void radeonTexImage3D(GLcontext * ctx, GLenum target, GLint level,
+		      GLint internalFormat,
+		      GLint width, GLint height, GLint depth,
+		      GLint border,
+		      GLenum format, GLenum type, const GLvoid * pixels,
+		      const struct gl_pixelstore_attrib *packing,
+		      struct gl_texture_object *texObj,
+		      struct gl_texture_image *texImage)
+{
+	radeon_teximage(ctx, 3, 0, level, internalFormat, width, height, depth,
+		0, format, type, pixels, packing, texObj, texImage, 0);
+}
+
+/**
+ * Update a subregion of the given texture image.
+ */
+static void radeon_texsubimage(GLcontext* ctx, int dims, int level,
+		GLint xoffset, GLint yoffset, GLint zoffset,
+		GLsizei width, GLsizei height, GLsizei depth,
+		GLsizei imageSize,
+		GLenum format, GLenum type,
+		const GLvoid * pixels,
+		const struct gl_pixelstore_attrib *packing,
+		struct gl_texture_object *texObj,
+		struct gl_texture_image *texImage,
+		int compressed)
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	radeonTexObj* t = radeon_tex_obj(texObj);
+	radeon_texture_image* image = get_radeon_texture_image(texImage);
+
+	radeon_firevertices(rmesa);
+
+	t->validated = GL_FALSE;
+	if (compressed) {
+		pixels = _mesa_validate_pbo_compressed_teximage(
+			ctx, imageSize, pixels, packing, "glCompressedTexImage");
+	} else {
+		pixels = _mesa_validate_pbo_teximage(ctx, dims,
+			width, height, depth, format, type, pixels, packing, "glTexSubImage1D");
+	}
+
+	if (pixels) {
+		GLint dstRowStride;
+		radeon_teximage_map(image, GL_TRUE);
+
+		if (image->mt) {
+			radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel];
+			dstRowStride = lvl->rowstride;
+		} else {
+			dstRowStride = texImage->RowStride * texImage->TexFormat->TexelBytes;
+		}
+
+		if (compressed) {
+			uint32_t srcRowStride, bytesPerRow, rows; 
+			dstRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, texImage->Width);
+			srcRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width);
+			bytesPerRow = srcRowStride;
+			rows = height / 4;
+
+			copy_rows(texImage->Data, dstRowStride,  image->base.Data, srcRowStride, rows,
+				  bytesPerRow);
+			
+		} else {
+			if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat,
+							     texImage->TexFormat, texImage->Data,
+							     xoffset, yoffset, zoffset,
+							     dstRowStride,
+							     texImage->ImageOffsets,
+							     width, height, depth,
+							     format, type, pixels, packing))
+				_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage");
+		}
+
+		/* GL_SGIS_generate_mipmap */
+		if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+			radeon_generate_mipmap(ctx, texObj->Target, texObj);
+		}
+	}
+
+	radeon_teximage_unmap(image);
+
+	_mesa_unmap_teximage_pbo(ctx, packing);
+
+
+}
+
+void radeonTexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
+			 GLint xoffset,
+			 GLsizei width,
+			 GLenum format, GLenum type,
+			 const GLvoid * pixels,
+			 const struct gl_pixelstore_attrib *packing,
+			 struct gl_texture_object *texObj,
+			 struct gl_texture_image *texImage)
+{
+	radeon_texsubimage(ctx, 1, level, xoffset, 0, 0, width, 1, 1, 0,
+		format, type, pixels, packing, texObj, texImage, 0);
+}
+
+void radeonTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
+			 GLint xoffset, GLint yoffset,
+			 GLsizei width, GLsizei height,
+			 GLenum format, GLenum type,
+			 const GLvoid * pixels,
+			 const struct gl_pixelstore_attrib *packing,
+			 struct gl_texture_object *texObj,
+			 struct gl_texture_image *texImage)
+{
+	radeon_texsubimage(ctx, 2, level, xoffset, yoffset, 0, width, height, 1,
+			   0, format, type, pixels, packing, texObj, texImage,
+			   0);
+}
+
+void radeonCompressedTexSubImage2D(GLcontext * ctx, GLenum target,
+				   GLint level, GLint xoffset,
+				   GLint yoffset, GLsizei width,
+				   GLsizei height, GLenum format,
+				   GLsizei imageSize, const GLvoid * data,
+				   struct gl_texture_object *texObj,
+				   struct gl_texture_image *texImage)
+{
+	radeon_texsubimage(ctx, 2, level, xoffset, yoffset, 0, width, height, 1,
+		imageSize, format, 0, data, &ctx->Unpack, texObj, texImage, 1);
+}
+
+
+void radeonTexSubImage3D(GLcontext * ctx, GLenum target, GLint level,
+			 GLint xoffset, GLint yoffset, GLint zoffset,
+			 GLsizei width, GLsizei height, GLsizei depth,
+			 GLenum format, GLenum type,
+			 const GLvoid * pixels,
+			 const struct gl_pixelstore_attrib *packing,
+			 struct gl_texture_object *texObj,
+			 struct gl_texture_image *texImage)
+{
+	radeon_texsubimage(ctx, 3, level, xoffset, yoffset, zoffset, width, height, depth, 0,
+		format, type, pixels, packing, texObj, texImage, 0);
+}
+
+
+
+/**
+ * Ensure that the given image is stored in the given miptree from now on.
+ */
+static void migrate_image_to_miptree(radeon_mipmap_tree *mt, radeon_texture_image *image, int face, int level)
+{
+	radeon_mipmap_level *dstlvl = &mt->levels[level - mt->firstLevel];
+	unsigned char *dest;
+
+	assert(image->mt != mt);
+	assert(dstlvl->width == image->base.Width);
+	assert(dstlvl->height == image->base.Height);
+	assert(dstlvl->depth == image->base.Depth);
+
+
+	radeon_bo_map(mt->bo, GL_TRUE);
+	dest = mt->bo->ptr + dstlvl->faces[face].offset;
+
+	if (image->mt) {
+		/* Format etc. should match, so we really just need a memcpy().
+		 * In fact, that memcpy() could be done by the hardware in many
+		 * cases, provided that we have a proper memory manager.
+		 */
+		radeon_mipmap_level *srclvl = &image->mt->levels[image->mtlevel-image->mt->firstLevel];
+
+		assert(srclvl->size == dstlvl->size);
+		assert(srclvl->rowstride == dstlvl->rowstride);
+
+		radeon_bo_map(image->mt->bo, GL_FALSE);
+
+		memcpy(dest,
+			image->mt->bo->ptr + srclvl->faces[face].offset,
+			dstlvl->size);
+		radeon_bo_unmap(image->mt->bo);
+
+		radeon_miptree_unreference(image->mt);
+	} else {
+		uint32_t srcrowstride;
+		uint32_t height;
+		/* need to confirm this value is correct */
+		if (mt->compressed) {
+			height = image->base.Height / 4;
+			srcrowstride = image->base.RowStride * mt->bpp;
+		} else {
+			height = image->base.Height * image->base.Depth;
+			srcrowstride = image->base.Width * image->base.TexFormat->TexelBytes;
+		}
+
+//		if (mt->tilebits)
+//			WARN_ONCE("%s: tiling not supported yet", __FUNCTION__);
+
+		copy_rows(dest, dstlvl->rowstride, image->base.Data, srcrowstride,
+			  height, srcrowstride);
+
+		_mesa_free_texmemory(image->base.Data);
+		image->base.Data = 0;
+	}
+
+	radeon_bo_unmap(mt->bo);
+
+	image->mt = mt;
+	image->mtface = face;
+	image->mtlevel = level;
+	radeon_miptree_reference(image->mt);
+}
+
+int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj)
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	radeonTexObj *t = radeon_tex_obj(texObj);
+	radeon_texture_image *baseimage = get_radeon_texture_image(texObj->Image[0][texObj->BaseLevel]);
+	int face, level;
+
+	if (t->validated || t->image_override)
+		return GL_TRUE;
+
+	if (RADEON_DEBUG & DEBUG_TEXTURE)
+		fprintf(stderr, "%s: Validating texture %p now\n", __FUNCTION__, texObj);
+
+	if (baseimage->base.Border > 0)
+		return GL_FALSE;
+
+	/* Ensure a matching miptree exists.
+	 *
+	 * Differing mipmap trees can result when the app uses TexImage to
+	 * change texture dimensions.
+	 *
+	 * Prefer to use base image's miptree if it
+	 * exists, since that most likely contains more valid data (remember
+	 * that the base level is usually significantly larger than the rest
+	 * of the miptree, so cubemaps are the only possible exception).
+	 */
+	if (baseimage->mt &&
+	    baseimage->mt != t->mt &&
+	    radeon_miptree_matches_texture(baseimage->mt, &t->base)) {
+		radeon_miptree_unreference(t->mt);
+		t->mt = baseimage->mt;
+		radeon_miptree_reference(t->mt);
+	} else if (t->mt && !radeon_miptree_matches_texture(t->mt, &t->base)) {
+		radeon_miptree_unreference(t->mt);
+		t->mt = 0;
+	}
+
+	if (!t->mt) {
+		if (RADEON_DEBUG & DEBUG_TEXTURE)
+			fprintf(stderr, " Allocate new miptree\n");
+		radeon_try_alloc_miptree(rmesa, t, &baseimage->base, 0, texObj->BaseLevel);
+		if (!t->mt) {
+			_mesa_problem(ctx, "r300_validate_texture failed to alloc miptree");
+			return GL_FALSE;
+		}
+	}
+
+	/* Ensure all images are stored in the single main miptree */
+	for(face = 0; face < t->mt->faces; ++face) {
+		for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level) {
+			radeon_texture_image *image = get_radeon_texture_image(texObj->Image[face][level]);
+			if (RADEON_DEBUG & DEBUG_TEXTURE)
+				fprintf(stderr, " face %i, level %i... %p vs %p ", face, level, t->mt, image->mt);
+			if (t->mt == image->mt) {
+				if (RADEON_DEBUG & DEBUG_TEXTURE)
+					fprintf(stderr, "OK\n");
+				continue;
+			}
+
+			if (RADEON_DEBUG & DEBUG_TEXTURE)
+				fprintf(stderr, "migrating\n");
+			migrate_image_to_miptree(t->mt, image, face, level);
+		}
+	}
+
+	return GL_TRUE;
+}
+
+
+/**
+ * Need to map texture image into memory before copying image data,
+ * then unmap it.
+ */
+static void
+radeon_get_tex_image(GLcontext * ctx, GLenum target, GLint level,
+		     GLenum format, GLenum type, GLvoid * pixels,
+		     struct gl_texture_object *texObj,
+		     struct gl_texture_image *texImage, int compressed)
+{
+	radeon_texture_image *image = get_radeon_texture_image(texImage);
+
+	if (image->mt) {
+		/* Map the texture image read-only */
+		radeon_teximage_map(image, GL_FALSE);
+	} else {
+		/* Image hasn't been uploaded to a miptree yet */
+		assert(image->base.Data);
+	}
+
+	if (compressed) {
+		_mesa_get_compressed_teximage(ctx, target, level, pixels,
+					      texObj, texImage);
+	} else {
+		_mesa_get_teximage(ctx, target, level, format, type, pixels,
+				   texObj, texImage);
+	}
+     
+	if (image->mt) {
+		radeon_teximage_unmap(image);
+	}
+}
+
+void
+radeonGetTexImage(GLcontext * ctx, GLenum target, GLint level,
+		  GLenum format, GLenum type, GLvoid * pixels,
+		  struct gl_texture_object *texObj,
+		  struct gl_texture_image *texImage)
+{
+	radeon_get_tex_image(ctx, target, level, format, type, pixels,
+			     texObj, texImage, 0);
+}
+
+void
+radeonGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level,
+			    GLvoid *pixels,
+			    struct gl_texture_object *texObj,
+			    struct gl_texture_image *texImage)
+{
+	radeon_get_tex_image(ctx, target, level, 0, 0, pixels,
+			     texObj, texImage, 1);
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.h b/src/mesa/drivers/dri/radeon/radeon_texture.h
new file mode 100644
index 00000000000..888a55ba911
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_texture.h
@@ -0,0 +1,122 @@
+/*
+ * Copyright (C) 2008 Nicolai Haehnle.
+ * Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+ *
+ * The Weather Channel (TM) funded Tungsten Graphics to develop the
+ * initial release of the Radeon 8500 driver under the XFree86 license.
+ * This notice must be preserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef RADEON_TEXTURE_H
+#define RADEON_TEXTURE_H
+struct gl_texture_image *radeonNewTextureImage(GLcontext *ctx);
+void radeonFreeTexImageData(GLcontext *ctx, struct gl_texture_image *timage);
+
+void radeon_teximage_map(radeon_texture_image *image, GLboolean write_enable);
+void radeon_teximage_unmap(radeon_texture_image *image);
+void radeonMapTexture(GLcontext *ctx, struct gl_texture_object *texObj);
+void radeonUnmapTexture(GLcontext *ctx, struct gl_texture_object *texObj);
+void radeonGenerateMipmap(GLcontext* ctx, GLenum target, struct gl_texture_object *texObj);
+int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj);
+GLuint radeon_face_for_target(GLenum target);
+const struct gl_texture_format *radeonChooseTextureFormat_mesa(GLcontext * ctx,
+							  GLint internalFormat,
+							  GLenum format,
+							  GLenum type);
+const struct gl_texture_format *radeonChooseTextureFormat(GLcontext * ctx,
+							  GLint internalFormat,
+							  GLenum format,
+							  GLenum type, GLboolean fbo);
+
+void radeonTexImage1D(GLcontext * ctx, GLenum target, GLint level,
+		      GLint internalFormat,
+		      GLint width, GLint border,
+		      GLenum format, GLenum type, const GLvoid * pixels,
+		      const struct gl_pixelstore_attrib *packing,
+		      struct gl_texture_object *texObj,
+		      struct gl_texture_image *texImage);
+void radeonTexImage2D(GLcontext * ctx, GLenum target, GLint level,
+		      GLint internalFormat,
+		      GLint width, GLint height, GLint border,
+		      GLenum format, GLenum type, const GLvoid * pixels,
+		      const struct gl_pixelstore_attrib *packing,
+		      struct gl_texture_object *texObj,
+		      struct gl_texture_image *texImage);
+void radeonCompressedTexImage2D(GLcontext * ctx, GLenum target,
+				GLint level, GLint internalFormat,
+				GLint width, GLint height, GLint border,
+				GLsizei imageSize, const GLvoid * data,
+				struct gl_texture_object *texObj,
+				struct gl_texture_image *texImage);
+void radeonTexImage3D(GLcontext * ctx, GLenum target, GLint level,
+		      GLint internalFormat,
+		      GLint width, GLint height, GLint depth,
+		      GLint border,
+		      GLenum format, GLenum type, const GLvoid * pixels,
+		      const struct gl_pixelstore_attrib *packing,
+		      struct gl_texture_object *texObj,
+		      struct gl_texture_image *texImage);
+void radeonTexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
+			 GLint xoffset,
+			 GLsizei width,
+			 GLenum format, GLenum type,
+			 const GLvoid * pixels,
+			 const struct gl_pixelstore_attrib *packing,
+			 struct gl_texture_object *texObj,
+			 struct gl_texture_image *texImage);
+void radeonTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
+				GLint xoffset, GLint yoffset,
+				GLsizei width, GLsizei height,
+				GLenum format, GLenum type,
+				const GLvoid * pixels,
+				const struct gl_pixelstore_attrib *packing,
+				struct gl_texture_object *texObj,
+				struct gl_texture_image *texImage);
+void radeonCompressedTexSubImage2D(GLcontext * ctx, GLenum target,
+				   GLint level, GLint xoffset,
+				   GLint yoffset, GLsizei width,
+				   GLsizei height, GLenum format,
+				   GLsizei imageSize, const GLvoid * data,
+				   struct gl_texture_object *texObj,
+				   struct gl_texture_image *texImage);
+
+void radeonTexSubImage3D(GLcontext * ctx, GLenum target, GLint level,
+			 GLint xoffset, GLint yoffset, GLint zoffset,
+			 GLsizei width, GLsizei height, GLsizei depth,
+			 GLenum format, GLenum type,
+			 const GLvoid * pixels,
+			 const struct gl_pixelstore_attrib *packing,
+			 struct gl_texture_object *texObj,
+			 struct gl_texture_image *texImage);
+
+void radeonGetTexImage(GLcontext * ctx, GLenum target, GLint level,
+		       GLenum format, GLenum type, GLvoid * pixels,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage);
+void radeonGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level,
+				 GLvoid *pixels,
+				 struct gl_texture_object *texObj,
+				 struct gl_texture_image *texImage);
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_reg.h b/src/mesa/drivers/dri/radeon/server/radeon_reg.h
index ae2ccdf292a..866807462a4 100644
--- a/src/mesa/drivers/dri/radeon/server/radeon_reg.h
+++ b/src/mesa/drivers/dri/radeon/server/radeon_reg.h
@@ -2031,6 +2031,9 @@
 #define RADEON_CP_PACKET3_3D_DRAW_INDX              0xC0002A00
 #define RADEON_CP_PACKET3_LOAD_PALETTE              0xC0002C00
 #define RADEON_CP_PACKET3_3D_LOAD_VBPNTR            0xC0002F00
+#define R200_CP_CMD_3D_DRAW_VBUF_2      0xC0003400
+#define R200_CP_CMD_3D_DRAW_IMMD_2      0xC0003500
+#define R200_CP_CMD_3D_DRAW_INDX_2      0xC0003600
 #define RADEON_CP_PACKET3_CNTL_PAINT                0xC0009100
 #define RADEON_CP_PACKET3_CNTL_BITBLT               0xC0009200
 #define RADEON_CP_PACKET3_CNTL_SMALLTEXT            0xC0009300