Merge remote branch 'origin/master' into lp-binning

Conflicts:
	src/gallium/drivers/llvmpipe/lp_quad.h
	src/gallium/drivers/llvmpipe/lp_setup.c

282 files changed, 11877 insertions, 1890 deletions

diff --git a/src/egl/drivers/glx/egl_glx.c b/src/egl/drivers/glx/egl_glx.c
index 96292b0e9e8..7c6e8637a1c 100644
--- a/src/egl/drivers/glx/egl_glx.c
+++ b/src/egl/drivers/glx/egl_glx.c
@@ -899,7 +899,7 @@ GLX_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *draw)
  * Called from eglGetProcAddress() via drv->API.GetProcAddress().
  */
 static _EGLProc
-GLX_eglGetProcAddress(const char *procname)
+GLX_eglGetProcAddress(_EGLDriver *drv, const char *procname)
 {
    return (_EGLProc) glXGetProcAddress((const GLubyte *) procname);
 }
diff --git a/src/egl/drivers/xdri/egl_xdri.c b/src/egl/drivers/xdri/egl_xdri.c
index d2affc66dd4..8425b3d11ef 100644
--- a/src/egl/drivers/xdri/egl_xdri.c
+++ b/src/egl/drivers/xdri/egl_xdri.c
@@ -355,7 +355,7 @@ xdri_eglTerminate(_EGLDriver *drv, _EGLDisplay *dpy)
  * Called from eglGetProcAddress() via drv->API.GetProcAddress().
  */
 static _EGLProc
-xdri_eglGetProcAddress(const char *procname)
+xdri_eglGetProcAddress(_EGLDriver *drv, const char *procname)
 {
    /* the symbol is defined in libGL.so */
    return (_EGLProc) _glapi_get_proc_address(procname);
@@ -562,7 +562,7 @@ xdri_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw)
    struct xdri_egl_display *xdri_dpy = lookup_display(dpy);
    struct xdri_egl_surface *xdri_surf = lookup_surface(draw);
 
-   xdri_dpy->psc->driScreen->swapBuffers(xdri_surf->driDrawable);
+   xdri_dpy->psc->driScreen->swapBuffers(xdri_surf->driDrawable, 0, 0, 0);
 
    return EGL_TRUE;
 }
diff --git a/src/egl/main/Makefile b/src/egl/main/Makefile
index c951b070f1f..ec326a845df 100644
--- a/src/egl/main/Makefile
+++ b/src/egl/main/Makefile
@@ -4,7 +4,10 @@ TOP = ../../..
 include $(TOP)/configs/current
 
 
-INCLUDE_DIRS = -I$(TOP)/include -I$(TOP)/src/mesa/glapi $(X11_INCLUDES)
+EGL_MAJOR = 1
+EGL_MINOR = 0
+
+INCLUDE_DIRS = -I$(TOP)/include
 
 HEADERS = \
 	eglcompiler.h \
@@ -43,7 +46,7 @@ SOURCES = \
 OBJECTS = $(SOURCES:.c=.o)
 
 
-# Undefined for now
+# use dl*() to load drivers
 LOCAL_CFLAGS = -D_EGL_PLATFORM_X=1
 
 
@@ -56,21 +59,21 @@ default: depend library
 
 
 # EGL Library
-library: $(TOP)/$(LIB_DIR)/libEGL.so
+library: $(TOP)/$(LIB_DIR)/$(EGL_LIB_NAME)
 
-$(TOP)/$(LIB_DIR)/libEGL.so: $(OBJECTS)
-	$(MKLIB) -o EGL -linker '$(CC)' -ldflags '$(LDFLAGS)' \
-		-major 1 -minor 0 \
-		-install $(TOP)/$(LIB_DIR) \
+$(TOP)/$(LIB_DIR)/$(EGL_LIB_NAME): $(OBJECTS)
+	$(MKLIB) -o $(EGL_LIB) -linker '$(CC)' -ldflags '$(LDFLAGS)' \
+		-major $(EGL_MAJOR) -minor $(EGL_MINOR) \
+		-install $(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) \
 		$(EGL_LIB_DEPS) $(OBJECTS)
 
 install: default
 	$(INSTALL) -d $(DESTDIR)$(INSTALL_LIB_DIR)
-	$(MINSTALL) $(TOP)/$(LIB_DIR)/libEGL.so* $(DESTDIR)$(INSTALL_LIB_DIR)
+	$(MINSTALL) $(TOP)/$(LIB_DIR)/$(EGL_LIB_GLOB) \
+		$(DESTDIR)$(INSTALL_LIB_DIR)
 
 clean:
-	-rm -f *.o *.so*
-	-rm -f core.*
+	-rm -f *.o
 	-rm -f depend depend.bak
 
 
@@ -82,5 +85,5 @@ depend: $(SOURCES) $(HEADERS)
 		$(SOURCES) $(HEADERS) > /dev/null 2>/dev/null
 
 
-include depend
+-include depend
 # DO NOT DELETE
diff --git a/src/egl/main/eglapi.c b/src/egl/main/eglapi.c
index 14cc5fa6137..26e0602453a 100644
--- a/src/egl/main/eglapi.c
+++ b/src/egl/main/eglapi.c
@@ -716,7 +716,8 @@ void (* EGLAPIENTRY eglGetProcAddress(const char *procname))()
 
    /* now loop over drivers to query their procs */
    for (i = 0; i < _eglGlobal.NumDrivers; i++) {
-      _EGLProc p = _eglGlobal.Drivers[i]->API.GetProcAddress(procname);
+      _EGLDriver *drv = _eglGlobal.Drivers[i];
+      _EGLProc p = drv->API.GetProcAddress(drv, procname);
       if (p)
          return p;
    }
diff --git a/src/egl/main/eglapi.h b/src/egl/main/eglapi.h
index aa0abe3eb6b..080f2155e3a 100644
--- a/src/egl/main/eglapi.h
+++ b/src/egl/main/eglapi.h
@@ -44,7 +44,7 @@ typedef const char *(*QueryString_t)(_EGLDriver *drv, _EGLDisplay *dpy, EGLint n
 typedef EGLBoolean (*WaitClient_t)(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx);
 typedef EGLBoolean (*WaitNative_t)(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine);
 
-typedef _EGLProc (*GetProcAddress_t)(const char *procname);
+typedef _EGLProc (*GetProcAddress_t)(_EGLDriver *drv, const char *procname);
 
 
 
diff --git a/src/egl/main/eglconfig.c b/src/egl/main/eglconfig.c
index 31d69a77080..4d149603ae5 100644
--- a/src/egl/main/eglconfig.c
+++ b/src/egl/main/eglconfig.c
@@ -324,6 +324,7 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
             mask = EGL_PBUFFER_BIT |
                    EGL_PIXMAP_BIT |
                    EGL_WINDOW_BIT |
+                   EGL_SCREEN_BIT_MESA | /* XXX should check the extension */
                    EGL_VG_COLORSPACE_LINEAR_BIT |
                    EGL_VG_ALPHA_FORMAT_PRE_BIT |
                    EGL_MULTISAMPLE_RESOLVE_BOX_BIT |
diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index e3af41c6e04..8f937e3b4e9 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -111,6 +111,7 @@ C_SOURCES = \
 	util/u_math.c \
 	util/u_mm.c \
 	util/u_rect.c \
+	util/u_ringbuffer.c \
 	util/u_simple_shaders.c \
 	util/u_snprintf.c \
 	util/u_stream_stdc.c \
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index 782eb533863..f957090b5fb 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -147,6 +147,7 @@ source = [
     'util/u_math.c',
     'util/u_mm.c',
     'util/u_rect.c',
+    'util/u_ringbuffer.c',
     'util/u_simple_shaders.c',
     'util/u_snprintf.c',
     'util/u_stream_stdc.c',
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 667aa46b208..e90dfc5aec4 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -95,6 +95,7 @@ void draw_destroy( struct draw_context *draw )
    draw_pipeline_destroy( draw );
    draw_pt_destroy( draw );
    draw_vs_destroy( draw );
+   draw_gs_destroy( draw );
 
    FREE( draw );
 }
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index b716209df29..8a64c06efcd 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -164,6 +164,14 @@ void draw_set_mapped_constant_buffer(struct draw_context *draw,
 void draw_arrays(struct draw_context *draw, unsigned prim,
 		 unsigned start, unsigned count);
 
+void
+draw_arrays_instanced(struct draw_context *draw,
+                      unsigned mode,
+                      unsigned start,
+                      unsigned count,
+                      unsigned startInstance,
+                      unsigned instanceCount);
+
 void draw_flush(struct draw_context *draw);
 
 
diff --git a/src/gallium/auxiliary/draw/draw_gs.c b/src/gallium/auxiliary/draw/draw_gs.c
index 5db2e755423..daf8d071f12 100644
--- a/src/gallium/auxiliary/draw/draw_gs.c
+++ b/src/gallium/auxiliary/draw/draw_gs.c
@@ -59,6 +59,15 @@ draw_gs_init( struct draw_context *draw )
    return TRUE;
 }
 
+void draw_gs_destroy( struct draw_context *draw )
+{
+   if (!draw->gs.machine)
+      return;
+
+   align_free(draw->gs.machine->Primitives);
+
+   tgsi_exec_machine_destroy(draw->gs.machine);
+}
 
 void draw_gs_set_constants( struct draw_context *draw,
                             const float (*constants)[4],
diff --git a/src/gallium/auxiliary/draw/draw_pipe_clip.c b/src/gallium/auxiliary/draw/draw_pipe_clip.c
index 205cda5eabe..51a6115ebf5 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_clip.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_clip.c
@@ -55,7 +55,7 @@
 
 
 
-struct clipper {
+struct clip_stage {
    struct draw_stage stage;      /**< base class */
 
    /* Basically duplicate some of the flatshading logic here:
@@ -70,9 +70,9 @@ struct clipper {
 
 /* This is a bit confusing:
  */
-static INLINE struct clipper *clipper_stage( struct draw_stage *stage )
+static INLINE struct clip_stage *clip_stage( struct draw_stage *stage )
 {
-   return (struct clipper *)stage;
+   return (struct clip_stage *)stage;
 }
 
 
@@ -92,11 +92,12 @@ static void interp_attr( float *fdst,
    fdst[3] = LINTERP( t, fout[3], fin[3] );
 }
 
+
 static void copy_colors( struct draw_stage *stage,
 			 struct vertex_header *dst,
 			 const struct vertex_header *src )
 {
-   const struct clipper *clipper = clipper_stage(stage);
+   const struct clip_stage *clipper = clip_stage(stage);
    uint i;
    for (i = 0; i < clipper->num_color_attribs; i++) {
       const uint attr = clipper->color_attribs[i];
@@ -108,7 +109,7 @@ static void copy_colors( struct draw_stage *stage,
 
 /* Interpolate between two vertices to produce a third.  
  */
-static void interp( const struct clipper *clip,
+static void interp( const struct clip_stage *clip,
 		    struct vertex_header *dst,
 		    float t,
 		    const struct vertex_header *out, 
@@ -179,7 +180,7 @@ static void emit_poly( struct draw_stage *stage,
       header.v[2] = inlist[0];	/* keep in v[2] for flatshading */
 
       if (i == n-1)
-        header.flags |= edge_last;
+         header.flags |= edge_last;
 
       if (0) {
          const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader;
@@ -200,13 +201,14 @@ static void emit_poly( struct draw_stage *stage,
    }
 }
 
+
 static INLINE float
 dot4(const float *a, const float *b)
 {
-   return (a[0]*b[0] +
-           a[1]*b[1] +
-           a[2]*b[2] +
-           a[3]*b[3]);
+   return (a[0] * b[0] +
+           a[1] * b[1] +
+           a[2] * b[2] +
+           a[3] * b[3]);
 }
 
 
@@ -217,7 +219,7 @@ do_clip_tri( struct draw_stage *stage,
 	     struct prim_header *header,
 	     unsigned clipmask )
 {
-   struct clipper *clipper = clipper_stage( stage );
+   struct clip_stage *clipper = clip_stage( stage );
    struct vertex_header *a[MAX_CLIPPED_VERTICES];
    struct vertex_header *b[MAX_CLIPPED_VERTICES];
    struct vertex_header **inlist = a;
@@ -280,6 +282,7 @@ do_clip_tri( struct draw_stage *stage,
 	 dp_prev = dp;
       }
 
+      /* swap in/out lists */
       {
 	 struct vertex_header **tmp = inlist;
 	 inlist = outlist;
@@ -291,15 +294,11 @@ do_clip_tri( struct draw_stage *stage,
    /* If flat-shading, copy color to new provoking vertex.
     */
    if (clipper->flat && inlist[0] != header->v[2]) {
-      if (1) {
-	 inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
-      }
+      inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
 
       copy_colors(stage, inlist[0], header->v[2]);
    }
 
-
-
    /* Emit the polygon as triangles to the setup stage:
     */
    if (n >= 3)
@@ -314,7 +313,7 @@ do_clip_line( struct draw_stage *stage,
 	      struct prim_header *header,
 	      unsigned clipmask )
 {
-   const struct clipper *clipper = clipper_stage( stage );
+   const struct clip_stage *clipper = clip_stage( stage );
    struct vertex_header *v0 = header->v[0];
    struct vertex_header *v1 = header->v[1];
    const float *pos0 = v0->clip;
@@ -416,13 +415,14 @@ clip_tri( struct draw_stage *stage,
    }
 }
 
+
 /* Update state.  Could further delay this until we hit the first
  * primitive that really requires clipping.
  */
 static void 
 clip_init_state( struct draw_stage *stage )
 {
-   struct clipper *clipper = clipper_stage( stage );
+   struct clip_stage *clipper = clip_stage( stage );
 
    clipper->flat = stage->draw->rasterizer->flatshade ? TRUE : FALSE;
 
@@ -488,7 +488,7 @@ static void clip_destroy( struct draw_stage *stage )
  */
 struct draw_stage *draw_clip_stage( struct draw_context *draw )
 {
-   struct clipper *clipper = CALLOC_STRUCT(clipper);
+   struct clip_stage *clipper = CALLOC_STRUCT(clip_stage);
    if (clipper == NULL)
       goto fail;
 
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index 1a5269c0de9..d40c0352401 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -138,7 +138,7 @@ emit_vertex( struct vbuf_stage *vbuf,
       /* Note: we really do want data[0] here, not data[pos]: 
        */
       vbuf->translate->set_buffer(vbuf->translate, 0, vertex->data[0], 0);
-      vbuf->translate->run(vbuf->translate, 0, 1, vbuf->vertex_ptr);
+      vbuf->translate->run(vbuf->translate, 0, 1, 0, vbuf->vertex_ptr);
 
       if (0) draw_dump_emitted_vertex(vbuf->vinfo, (uint8_t *)vbuf->vertex_ptr);
       
@@ -271,10 +271,12 @@ vbuf_start_prim( struct vbuf_stage *vbuf, uint prim )
 	 emit_sz = 0;
 	 break;
       }
-      
+
+      hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       hw_key.element[i].input_buffer = src_buffer;
       hw_key.element[i].input_offset = src_offset;
+      hw_key.element[i].instance_divisor = 0;
       hw_key.element[i].output_format = output_format;
       hw_key.element[i].output_offset = dst_offset;
 
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index e49041556bd..ef49e575366 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -172,6 +172,8 @@ struct draw_context
 
    boolean force_passthrough; /**< never clip or shade */
 
+   boolean dump_vs;
+
    double mrd;  /**< minimum resolvable depth value, for polygon offset */
 
    /* pipe state that we need: */
@@ -239,6 +241,8 @@ struct draw_context
 
    unsigned reduced_prim;
 
+   unsigned instance_id;
+
    void *driver_private;
 };
 
@@ -265,6 +269,7 @@ boolean draw_gs_init( struct draw_context *draw );
 void draw_gs_set_constants( struct draw_context *,
                             const float (*constants)[4],
                             unsigned size );
+void draw_gs_destroy( struct draw_context *draw );
 
 /*******************************************************************************
  * Common shading code:
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 2801dbafe47..a5ddec52863 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -280,20 +280,33 @@ void
 draw_arrays(struct draw_context *draw, unsigned prim,
             unsigned start, unsigned count)
 {
-   unsigned reduced_prim = u_reduced_prim(prim);
+   draw_arrays_instanced(draw, prim, start, count, 0, 1);
+}
+
+void
+draw_arrays_instanced(struct draw_context *draw,
+                      unsigned mode,
+                      unsigned start,
+                      unsigned count,
+                      unsigned startInstance,
+                      unsigned instanceCount)
+{
+   unsigned reduced_prim = u_reduced_prim(mode);
+   unsigned instance;
+
    if (reduced_prim != draw->reduced_prim) {
-      draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+      draw_do_flush(draw, DRAW_FLUSH_STATE_CHANGE);
       draw->reduced_prim = reduced_prim;
    }
 
    if (0)
-      draw_print_arrays(draw, prim, start, MIN2(count, 20));
+      draw_print_arrays(draw, mode, start, MIN2(count, 20));
 
 #if 0
    {
       int i;
-      debug_printf("draw_arrays(prim=%u start=%u count=%u):\n",
-                   prim, start, count);
+      debug_printf("draw_arrays(mode=%u start=%u count=%u):\n",
+                   mode, start, count);
       tgsi_dump(draw->vs.vertex_shader->state.tokens, 0);
       debug_printf("Elements:\n");
       for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
@@ -311,6 +324,8 @@ draw_arrays(struct draw_context *draw, unsigned prim,
    }
 #endif
 
-   /* drawing done here: */
-   draw_pt_arrays(draw, prim, start, count);
+   for (instance = 0; instance < instanceCount; instance++) {
+      draw->instance_id = instance + startInstance;
+      draw_pt_arrays(draw, mode, start, count);
+   }
 }
diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h
index 20edf7a227e..d5e0d92a605 100644
--- a/src/gallium/auxiliary/draw/draw_pt.h
+++ b/src/gallium/auxiliary/draw/draw_pt.h
@@ -183,7 +183,8 @@ struct pt_emit *draw_pt_emit_create( struct draw_context *draw );
 struct pt_fetch;
 void draw_pt_fetch_prepare( struct pt_fetch *fetch,
                             unsigned vertex_input_count,
-			    unsigned vertex_size );
+                            unsigned vertex_size,
+                            unsigned instance_id_index );
 
 void draw_pt_fetch_run( struct pt_fetch *fetch,
 			const unsigned *elts,
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index 064e16c295c..4fb53276bbe 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -121,10 +121,12 @@ void draw_pt_emit_prepare( struct pt_emit *emit,
 	 emit_sz = 0;
 	 break;
       }
-      
+
+      hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       hw_key.element[i].input_buffer = src_buffer;
       hw_key.element[i].input_offset = src_offset;
+      hw_key.element[i].instance_divisor = 0;
       hw_key.element[i].output_format = output_format;
       hw_key.element[i].output_offset = dst_offset;
 
@@ -204,6 +206,7 @@ void draw_pt_emit( struct pt_emit *emit,
    translate->run( translate,
 		   0, 
 		   vertex_count,
+                   draw->instance_id,
 		   hw_verts );
 
    render->unmap_vertices( render, 
@@ -263,6 +266,7 @@ void draw_pt_emit_linear(struct pt_emit *emit,
    translate->run(translate,
                   0,
                   count,
+                  draw->instance_id,
                   hw_verts);
 
    if (0) {
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index 305bfef4352..55e7a7b81ad 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -58,12 +58,14 @@ struct pt_fetch {
  */
 void draw_pt_fetch_prepare( struct pt_fetch *fetch,
                             unsigned vs_input_count,
-			    unsigned vertex_size )
+                            unsigned vertex_size,
+                            unsigned instance_id_index )
 {
    struct draw_context *draw = fetch->draw;
    unsigned nr_inputs;
-   unsigned i, nr = 0;
+   unsigned i, nr = 0, ei = 0;
    unsigned dst_offset = 0;
+   unsigned num_extra_inputs = 0;
    struct translate_key key;
 
    fetch->vertex_size = vertex_size;
@@ -78,9 +80,11 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
    {
       /* Need to set header->vertex_id = 0xffff somehow.
        */
+      key.element[nr].type = TRANSLATE_ELEMENT_NORMAL;
       key.element[nr].input_format = PIPE_FORMAT_R32_FLOAT;
       key.element[nr].input_buffer = draw->pt.nr_vertex_buffers;
       key.element[nr].input_offset = 0;
+      key.element[nr].instance_divisor = 0;
       key.element[nr].output_format = PIPE_FORMAT_R32_FLOAT;
       key.element[nr].output_offset = dst_offset;
       dst_offset += 1 * sizeof(float);
@@ -91,19 +95,36 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
        */
       dst_offset += 4 * sizeof(float);
    }
-      
-   assert( draw->pt.nr_vertex_elements >= vs_input_count );
 
-   nr_inputs = MIN2( vs_input_count, draw->pt.nr_vertex_elements );
+   if (instance_id_index != ~0) {
+      num_extra_inputs++;
+   }
+
+   assert(draw->pt.nr_vertex_elements + num_extra_inputs >= vs_input_count);
+
+   nr_inputs = MIN2(vs_input_count, draw->pt.nr_vertex_elements + num_extra_inputs);
 
    for (i = 0; i < nr_inputs; i++) {
-      key.element[nr].input_format = draw->pt.vertex_element[i].src_format;
-      key.element[nr].input_buffer = draw->pt.vertex_element[i].vertex_buffer_index;
-      key.element[nr].input_offset = draw->pt.vertex_element[i].src_offset;
-      key.element[nr].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-      key.element[nr].output_offset = dst_offset;
+      if (i == instance_id_index) {
+         key.element[nr].type = TRANSLATE_ELEMENT_INSTANCE_ID;
+         key.element[nr].input_format = PIPE_FORMAT_R32_USCALED;
+         key.element[nr].output_format = PIPE_FORMAT_R32_USCALED;
+         key.element[nr].output_offset = dst_offset;
+
+         dst_offset += sizeof(uint);
+      } else {
+         key.element[nr].type = TRANSLATE_ELEMENT_NORMAL;
+         key.element[nr].input_format = draw->pt.vertex_element[ei].src_format;
+         key.element[nr].input_buffer = draw->pt.vertex_element[ei].vertex_buffer_index;
+         key.element[nr].input_offset = draw->pt.vertex_element[ei].src_offset;
+         key.element[nr].instance_divisor = draw->pt.vertex_element[ei].instance_divisor;
+         key.element[nr].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+         key.element[nr].output_offset = dst_offset;
+
+         ei++;
+         dst_offset += 4 * sizeof(float);
+      }
 
-      dst_offset += 4 * sizeof(float);
       nr++;
    }
 
@@ -158,6 +179,7 @@ void draw_pt_fetch_run( struct pt_fetch *fetch,
    translate->run_elts( translate,
 			elts, 
 			count,
+                        draw->instance_id,
 			verts );
 
 }
@@ -183,6 +205,7 @@ void draw_pt_fetch_run_linear( struct pt_fetch *fetch,
    translate->run( translate,
                    start,
                    count,
+                   draw->instance_id,
                    verts );
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index e7fe6b3b768..2a604470e9a 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -166,9 +166,11 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
 	 continue;
       }
 
+      key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       key.element[i].input_format = input_format;
       key.element[i].input_buffer = input_buffer;
       key.element[i].input_offset = input_offset;
+      key.element[i].instance_divisor = src->instance_divisor;
       key.element[i].output_format = output_format;
       key.element[i].output_offset = dst_offset;
       
@@ -256,6 +258,7 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle,
    feme->translate->run_elts( feme->translate, 
 			      fetch_elts,
 			      fetch_count,
+                              draw->instance_id,
 			      hw_verts );
 
    if (0) {
@@ -314,6 +317,7 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
    feme->translate->run( feme->translate,
                          start,
                          count,
+                         draw->instance_id,
                          hw_verts );
 
    if (0) {
@@ -374,6 +378,7 @@ static boolean fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
    feme->translate->run( feme->translate,
                          start,
                          count,
+                         draw->instance_id,
                          hw_verts );
 
    draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
index 1a9df4cac5d..23da556f797 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
@@ -59,6 +59,8 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
    struct draw_context *draw = fpme->draw;
    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+   unsigned i;
+   unsigned instance_id_index = ~0;
 
    /* Add one to num_outputs because the pipeline occasionally tags on
     * an additional texcoord, eg for AA lines.
@@ -66,6 +68,15 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
    unsigned nr = MAX2( vs->info.num_inputs,
 		       vs->info.num_outputs + 1 );
 
+   /* Scan for instanceID system value.
+    */
+   for (i = 0; i < vs->info.num_inputs; i++) {
+      if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
+         instance_id_index = i;
+         break;
+      }
+   }
+
    fpme->prim = prim;
    fpme->opt = opt;
 
@@ -79,7 +90,8 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
 
    draw_pt_fetch_prepare( fpme->fetch, 
                           vs->info.num_inputs,
-			  fpme->vertex_size );
+                          fpme->vertex_size,
+                          instance_id_index );
    /* XXX: it's not really gl rasterization rules we care about here,
     * but gl vs dx9 clip spaces.
     */
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index 35536895326..e03ac8c2291 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -43,11 +43,11 @@
 #include "translate/translate.h"
 #include "translate/translate_cache.h"
 
+#include "tgsi/tgsi_dump.h"
 #include "tgsi/tgsi_exec.h"
 
 
 
-
 void draw_vs_set_constants( struct draw_context *draw,
                             const float (*constants)[4],
                             unsigned size )
@@ -83,6 +83,10 @@ draw_create_vertex_shader(struct draw_context *draw,
 {
    struct draw_vertex_shader *vs;
 
+   if (draw->dump_vs) {
+      tgsi_dump(shader->tokens, 0);
+   }
+
    vs = draw_create_vs_llvm( draw, shader );
    if (!vs) {
       vs = draw_create_vs_sse( draw, shader );
@@ -152,6 +156,8 @@ draw_delete_vertex_shader(struct draw_context *draw,
 boolean 
 draw_vs_init( struct draw_context *draw )
 {
+   draw->dump_vs = debug_get_bool_option("GALLIUM_DUMP_VS", FALSE);
+
    draw->vs.machine = tgsi_exec_machine_create();
    if (!draw->vs.machine)
       return FALSE;
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index e3b807ebd0e..00036cfe68b 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -43,6 +43,7 @@ struct draw_varient_input
    enum pipe_format format;
    unsigned buffer;
    unsigned offset; 
+   unsigned instance_divisor;
 };
 
 struct draw_varient_output
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
index ad184bd696d..da9f3e3d35c 100644
--- a/src/gallium/auxiliary/draw/draw_vs_ppc.c
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -98,9 +98,9 @@ vs_ppc_run_linear( struct draw_vertex_shader *base,
    /* loop over verts */
    for (i = 0; i < count; i += MAX_VERTICES) {
       const uint max_vertices = MIN2(MAX_VERTICES, count - i);
-      float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4] ALIGN16_ATTRIB;
-      float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4] ALIGN16_ATTRIB;
-      float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4] ALIGN16_ATTRIB;
+      PIPE_ALIGN_VAR(16) float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4];
+      PIPE_ALIGN_VAR(16) float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4];
+      PIPE_ALIGN_VAR(16) float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4];
       uint attr;
 
       /* convert (up to) four input verts to SoA format */
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index d16692584e5..9f40030f39f 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -142,6 +142,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
    vsvg->fetch->run_elts( vsvg->fetch, 
                           elts,
                           count,
+                          vsvg->draw->instance_id,
                           temp_buffer );
 
    vsvg->base.vs->run_linear( vsvg->base.vs, 
@@ -181,6 +182,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
 
    vsvg->emit->run( vsvg->emit,
                     0, count,
+                    vsvg->draw->instance_id,
                     output_buffer );
 
    FREE(temp_buffer);
@@ -203,6 +205,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
    vsvg->fetch->run( vsvg->fetch, 
                      start,
                      count,
+                     vsvg->draw->instance_id,
                      temp_buffer );
 
    vsvg->base.vs->run_linear( vsvg->base.vs, 
@@ -239,6 +242,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
    
    vsvg->emit->run( vsvg->emit,
                     0, count,
+                    vsvg->draw->instance_id,
                     output_buffer );
 
    FREE(temp_buffer);
@@ -281,9 +285,11 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    fetch.nr_elements = key->nr_inputs;
    fetch.output_stride = vsvg->temp_vertex_stride;
    for (i = 0; i < key->nr_inputs; i++) {
+      fetch.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       fetch.element[i].input_format = key->element[i].in.format;
       fetch.element[i].input_buffer = key->element[i].in.buffer;
       fetch.element[i].input_offset = key->element[i].in.offset;
+      fetch.element[i].instance_divisor = 0;
       fetch.element[i].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       fetch.element[i].output_offset = i * 4 * sizeof(float);
       assert(fetch.element[i].output_offset < fetch.output_stride);
@@ -295,17 +301,21 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    for (i = 0; i < key->nr_outputs; i++) {
       if (key->element[i].out.format != EMIT_1F_PSIZE)
       {      
+         emit.element[i].type = TRANSLATE_ELEMENT_NORMAL;
          emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
          emit.element[i].input_buffer = 0;
          emit.element[i].input_offset = key->element[i].out.vs_output * 4 * sizeof(float);
+         emit.element[i].instance_divisor = 0;
          emit.element[i].output_format = draw_translate_vinfo_format(key->element[i].out.format);
          emit.element[i].output_offset = key->element[i].out.offset;
          assert(emit.element[i].input_offset <= fetch.output_stride);
       }
       else {
+         emit.element[i].type = TRANSLATE_ELEMENT_NORMAL;
          emit.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
          emit.element[i].input_buffer = 1;
          emit.element[i].input_offset = 0;
+         emit.element[i].instance_divisor = 0;
          emit.element[i].output_format = PIPE_FORMAT_R32_FLOAT;
          emit.element[i].output_offset = key->element[i].out.offset;
       }
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
index ba6f7b15f9e..a4b78f14943 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -80,27 +80,11 @@ struct fenced_buffer_list
  */
 struct fenced_buffer
 {
-   /*
-    * Immutable members.
-    */
-
    struct pb_buffer base;
+   
    struct pb_buffer *buffer;
-   struct fenced_buffer_list *list;
-
-   /**
-    * Protected by fenced_buffer_list::mutex
-    */
-   struct list_head head;
 
-   /**
-    * Following members are mutable and protected by this mutex.
-    * 
-    * You may lock this mutex alone, or lock it with fenced_buffer_list::mutex
-    * held, but in order to prevent deadlocks you must never lock 
-    * fenced_buffer_list::mutex with this mutex held.
-    */
-   pipe_mutex mutex;
+   /* FIXME: protect access with mutex */
 
    /**
     * A bitmask of PIPE_BUFFER_USAGE_CPU/GPU_READ/WRITE describing the current
@@ -112,6 +96,9 @@ struct fenced_buffer
    struct pb_validate *vl;
    unsigned validation_flags;
    struct pipe_fence_handle *fence;
+
+   struct list_head head;
+   struct fenced_buffer_list *list;
 };
 
 
@@ -123,24 +110,15 @@ fenced_buffer(struct pb_buffer *buf)
 }
 
 
-/**
- * Add the buffer to the fenced list.
- * 
- * fenced_buffer_list::mutex and fenced_buffer::mutex must be held, in this
- * order, before calling this function.
- * 
- * Reference count should be incremented before calling this function.
- */
 static INLINE void
-fenced_buffer_add_locked(struct fenced_buffer_list *fenced_list, 
-                         struct fenced_buffer *fenced_buf)
+_fenced_buffer_add(struct fenced_buffer *fenced_buf)
 {
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+
    assert(pipe_is_referenced(&fenced_buf->base.base.reference));
    assert(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
    assert(fenced_buf->fence);
 
-   /* TODO: Move the reference count increment here */
-   
 #ifdef DEBUG
    LIST_DEL(&fenced_buf->head);
    assert(fenced_list->numUnfenced);
@@ -152,16 +130,32 @@ fenced_buffer_add_locked(struct fenced_buffer_list *fenced_list,
 
 
 /**
- * Remove the buffer from the fenced list.
- * 
- * fenced_buffer_list::mutex and fenced_buffer::mutex must be held, in this 
- * order before calling this function.
- * 
- * Reference count should be decremented after calling this function.
+ * Actually destroy the buffer.
  */
 static INLINE void
-fenced_buffer_remove_locked(struct fenced_buffer_list *fenced_list,
-                            struct fenced_buffer *fenced_buf)
+_fenced_buffer_destroy(struct fenced_buffer *fenced_buf)
+{
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+   
+   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));
+   assert(!fenced_buf->fence);
+#ifdef DEBUG
+   assert(fenced_buf->head.prev);
+   assert(fenced_buf->head.next);
+   LIST_DEL(&fenced_buf->head);
+   assert(fenced_list->numUnfenced);
+   --fenced_list->numUnfenced;
+#else
+   (void)fenced_list;
+#endif
+   pb_reference(&fenced_buf->buffer, NULL);
+   FREE(fenced_buf);
+}
+
+
+static INLINE void
+_fenced_buffer_remove(struct fenced_buffer_list *fenced_list,
+                      struct fenced_buffer *fenced_buf)
 {
    struct pb_fence_ops *ops = fenced_list->ops;
 
@@ -183,56 +177,37 @@ fenced_buffer_remove_locked(struct fenced_buffer_list *fenced_list,
    ++fenced_list->numUnfenced;
 #endif
    
-   /* TODO: Move the reference count decrement and destruction here */
+   /**
+    * FIXME!!!
+    */
+
+   if(!pipe_is_referenced(&fenced_buf->base.base.reference))
+      _fenced_buffer_destroy(fenced_buf);
 }
 
 
-/**
- * Wait for the fence to expire, and remove it from the fenced list.
- * 
- * fenced_buffer::mutex must be held. fenced_buffer_list::mutex must not be 
- * held -- it will be acquired internally.
- */
 static INLINE enum pipe_error
-fenced_buffer_finish_locked(struct fenced_buffer_list *fenced_list,
-                              struct fenced_buffer *fenced_buf)
+_fenced_buffer_finish(struct fenced_buffer *fenced_buf)
 {
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
    struct pb_fence_ops *ops = fenced_list->ops;
-   enum pipe_error ret = PIPE_ERROR;
 
 #if 0
    debug_warning("waiting for GPU");
 #endif
 
-   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
    assert(fenced_buf->fence);
-
-   /*
-    * Acquire the global lock. Must release buffer mutex first to preserve
-    * lock order.
-    */
-   pipe_mutex_unlock(fenced_buf->mutex);
-   pipe_mutex_lock(fenced_list->mutex);
-   pipe_mutex_lock(fenced_buf->mutex);
-
    if(fenced_buf->fence) {
-      if(ops->fence_finish(ops, fenced_buf->fence, 0) == 0) {
-         /* Remove from the fenced list */
-         /* TODO: remove consequents */
-         fenced_buffer_remove_locked(fenced_list, fenced_buf);
-
-         p_atomic_dec(&fenced_buf->base.base.reference.count);
-         assert(pipe_is_referenced(&fenced_buf->base.base.reference));
-
-         fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
-
-         ret = PIPE_OK;
+      if(ops->fence_finish(ops, fenced_buf->fence, 0) != 0) {
+	 return PIPE_ERROR;
       }
+      /* Remove from the fenced list */
+      /* TODO: remove consequents */
+      _fenced_buffer_remove(fenced_list, fenced_buf);
    }
 
-   pipe_mutex_unlock(fenced_list->mutex);
-
-   return ret;
+   fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+   return PIPE_OK;
 }
 
 
@@ -240,8 +215,8 @@ fenced_buffer_finish_locked(struct fenced_buffer_list *fenced_list,
  * Free as many fenced buffers from the list head as possible. 
  */
 static void
-fenced_buffer_list_check_free_locked(struct fenced_buffer_list *fenced_list, 
-                                     int wait)
+_fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                               int wait)
 {
    struct pb_fence_ops *ops = fenced_list->ops;
    struct list_head *curr, *next;
@@ -254,29 +229,21 @@ fenced_buffer_list_check_free_locked(struct fenced_buffer_list *fenced_list,
    while(curr != &fenced_list->delayed) {
       fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
 
-      pipe_mutex_lock(fenced_buf->mutex);
-
       if(fenced_buf->fence != prev_fence) {
 	 int signaled;
 	 if (wait)
 	    signaled = ops->fence_finish(ops, fenced_buf->fence, 0);
 	 else
 	    signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
-	 if (signaled != 0) {
-            pipe_mutex_unlock(fenced_buf->mutex);
+	 if (signaled != 0)
 	    break;
-         }
 	 prev_fence = fenced_buf->fence;
       }
       else {
 	 assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
       }
 
-      fenced_buffer_remove_locked(fenced_list, fenced_buf);
-      pipe_mutex_unlock(fenced_buf->mutex);
-
-      pb_buf = &fenced_buf->base;
-      pb_reference(&pb_buf, NULL);
+      _fenced_buffer_remove(fenced_list, fenced_buf);
 
       curr = next; 
       next = curr->next;
@@ -290,25 +257,30 @@ fenced_buffer_destroy(struct pb_buffer *buf)
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);   
    struct fenced_buffer_list *fenced_list = fenced_buf->list;
 
-   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));
-   assert(!fenced_buf->fence);
-
-#ifdef DEBUG
    pipe_mutex_lock(fenced_list->mutex);
-   assert(fenced_buf->head.prev);
-   assert(fenced_buf->head.next);
-   LIST_DEL(&fenced_buf->head);
-   assert(fenced_list->numUnfenced);
-   --fenced_list->numUnfenced;
+   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));
+   if (fenced_buf->fence) {
+      struct pb_fence_ops *ops = fenced_list->ops;
+      if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) {
+	 struct list_head *curr, *prev;
+	 curr = &fenced_buf->head;
+	 prev = curr->prev;
+	 do {
+	    fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
+	    assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
+	    _fenced_buffer_remove(fenced_list, fenced_buf);
+	    curr = prev;
+	    prev = curr->prev;
+	 } while (curr != &fenced_list->delayed);
+      }	  
+      else {
+	 /* delay destruction */
+      }
+   }
+   else {
+      _fenced_buffer_destroy(fenced_buf);
+   }
    pipe_mutex_unlock(fenced_list->mutex);
-#else
-   (void)fenced_list;
-#endif
-
-   pb_reference(&fenced_buf->buffer, NULL);
-
-   pipe_mutex_destroy(fenced_buf->mutex);
-   FREE(fenced_buf);
 }
 
 
@@ -319,23 +291,24 @@ fenced_buffer_map(struct pb_buffer *buf,
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
    struct fenced_buffer_list *fenced_list = fenced_buf->list;
    struct pb_fence_ops *ops = fenced_list->ops;
-   void *map = NULL;
-
-   pipe_mutex_lock(fenced_buf->mutex);
+   void *map;
 
    assert(!(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE));
    
    /* Serialize writes */
    if((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) ||
       ((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ) && (flags & PIPE_BUFFER_USAGE_CPU_WRITE))) {
-      if((flags & PIPE_BUFFER_USAGE_DONTBLOCK) &&
-          ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) {
+      if(flags & PIPE_BUFFER_USAGE_DONTBLOCK) {
          /* Don't wait for the GPU to finish writing */
-         goto done;
+         if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0)
+            _fenced_buffer_remove(fenced_list, fenced_buf);
+         else
+            return NULL;
+      }
+      else {
+         /* Wait for the GPU to finish writing */
+         _fenced_buffer_finish(fenced_buf);
       }
-
-      /* Wait for the GPU to finish writing */
-      fenced_buffer_finish_locked(fenced_list, fenced_buf);
    }
 
 #if 0
@@ -352,9 +325,6 @@ fenced_buffer_map(struct pb_buffer *buf,
       fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE;
    }
 
-done:
-   pipe_mutex_unlock(fenced_buf->mutex);
-   
    return map;
 }
 
@@ -363,9 +333,6 @@ static void
 fenced_buffer_unmap(struct pb_buffer *buf)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
-   
-   pipe_mutex_lock(fenced_buf->mutex);
-   
    assert(fenced_buf->mapcount);
    if(fenced_buf->mapcount) {
       pb_unmap(fenced_buf->buffer);
@@ -373,8 +340,6 @@ fenced_buffer_unmap(struct pb_buffer *buf)
       if(!fenced_buf->mapcount)
 	 fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE;
    }
-   
-   pipe_mutex_unlock(fenced_buf->mutex);
 }
 
 
@@ -386,14 +351,11 @@ fenced_buffer_validate(struct pb_buffer *buf,
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
    enum pipe_error ret;
    
-   pipe_mutex_lock(fenced_buf->mutex);
-
    if(!vl) {
       /* invalidate */
       fenced_buf->vl = NULL;
       fenced_buf->validation_flags = 0;
-      ret = PIPE_OK;
-      goto done;
+      return PIPE_OK;
    }
    
    assert(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
@@ -401,17 +363,14 @@ fenced_buffer_validate(struct pb_buffer *buf,
    flags &= PIPE_BUFFER_USAGE_GPU_READ_WRITE;
 
    /* Buffer cannot be validated in two different lists */ 
-   if(fenced_buf->vl && fenced_buf->vl != vl) {
-      ret = PIPE_ERROR_RETRY;
-      goto done;
-   }
+   if(fenced_buf->vl && fenced_buf->vl != vl)
+      return PIPE_ERROR_RETRY;
    
 #if 0
    /* Do not validate if buffer is still mapped */
    if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) {
       /* TODO: wait for the thread that mapped the buffer to unmap it */
-      ret = PIPE_ERROR_RETRY;
-      goto done;
+      return PIPE_ERROR_RETRY;
    }
    /* Final sanity checking */
    assert(!(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE));
@@ -421,21 +380,17 @@ fenced_buffer_validate(struct pb_buffer *buf,
    if(fenced_buf->vl == vl &&
       (fenced_buf->validation_flags & flags) == flags) {
       /* Nothing to do -- buffer already validated */
-      ret = PIPE_OK;
-      goto done;
+      return PIPE_OK;
    }
    
    ret = pb_validate(fenced_buf->buffer, vl, flags);
    if (ret != PIPE_OK)
-      goto done;
+      return ret;
    
    fenced_buf->vl = vl;
    fenced_buf->validation_flags |= flags;
    
-done:
-   pipe_mutex_unlock(fenced_buf->mutex);
-
-   return ret;
+   return PIPE_OK;
 }
 
 
@@ -450,36 +405,29 @@ fenced_buffer_fence(struct pb_buffer *buf,
    fenced_buf = fenced_buffer(buf);
    fenced_list = fenced_buf->list;
    ops = fenced_list->ops;
-
-   pipe_mutex_lock(fenced_list->mutex);
-   pipe_mutex_lock(fenced_buf->mutex);
-
-   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
-
-   if(fence != fenced_buf->fence) {
-      assert(fenced_buf->vl);
-      assert(fenced_buf->validation_flags);
-      
-      if (fenced_buf->fence) {
-         fenced_buffer_remove_locked(fenced_list, fenced_buf);
-         p_atomic_dec(&fenced_buf->base.base.reference.count);
-         assert(pipe_is_referenced(&fenced_buf->base.base.reference));
-      }
-      if (fence) {
-         ops->fence_reference(ops, &fenced_buf->fence, fence);
-         fenced_buf->flags |= fenced_buf->validation_flags;
-         p_atomic_inc(&fenced_buf->base.base.reference.count);
-         fenced_buffer_add_locked(fenced_list, fenced_buf);
-      }
-
-      pb_fence(fenced_buf->buffer, fence);
    
-      fenced_buf->vl = NULL;
-      fenced_buf->validation_flags = 0;
+   if(fence == fenced_buf->fence) {
+      /* Nothing to do */
+      return;
    }
 
-   pipe_mutex_unlock(fenced_buf->mutex);
+   assert(fenced_buf->vl);
+   assert(fenced_buf->validation_flags);
+   
+   pipe_mutex_lock(fenced_list->mutex);
+   if (fenced_buf->fence)
+      _fenced_buffer_remove(fenced_list, fenced_buf);
+   if (fence) {
+      ops->fence_reference(ops, &fenced_buf->fence, fence);
+      fenced_buf->flags |= fenced_buf->validation_flags;
+      _fenced_buffer_add(fenced_buf);
+   }
    pipe_mutex_unlock(fenced_list->mutex);
+   
+   pb_fence(fenced_buf->buffer, fence);
+
+   fenced_buf->vl = NULL;
+   fenced_buf->validation_flags = 0;
 }
 
 
@@ -489,7 +437,6 @@ fenced_buffer_get_base_buffer(struct pb_buffer *buf,
                               pb_size *offset)
 {
    struct fenced_buffer *fenced_buf = fenced_buffer(buf);
-   /* NOTE: accesses immutable members only -- mutex not necessary */
    pb_get_base_buffer(fenced_buf->buffer, base_buf, offset);
 }
 
@@ -529,8 +476,6 @@ fenced_buffer_create(struct fenced_buffer_list *fenced_list,
    buf->buffer = buffer;
    buf->list = fenced_list;
    
-   pipe_mutex_init(buf->mutex);
-
 #ifdef DEBUG
    pipe_mutex_lock(fenced_list->mutex);
    LIST_ADDTAIL(&buf->head, &fenced_list->unfenced);
@@ -572,7 +517,7 @@ fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list,
                               int wait)
 {
    pipe_mutex_lock(fenced_list->mutex);
-   fenced_buffer_list_check_free_locked(fenced_list, wait);
+   _fenced_buffer_list_check_free(fenced_list, wait);
    pipe_mutex_unlock(fenced_list->mutex);
 }
 
@@ -594,13 +539,11 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list)
    next = curr->next;
    while(curr != &fenced_list->unfenced) {
       fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
-      pipe_mutex_lock(fenced_buf->mutex);
       assert(!fenced_buf->fence);
       debug_printf("%10p %7u %7u\n",
                    (void *) fenced_buf,
                    fenced_buf->base.base.size,
                    p_atomic_read(&fenced_buf->base.base.reference.count));
-      pipe_mutex_unlock(fenced_buf->mutex);
       curr = next; 
       next = curr->next;
    }
@@ -610,7 +553,6 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list)
    while(curr != &fenced_list->delayed) {
       int signaled;
       fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
-      pipe_mutex_lock(fenced_buf->mutex);
       signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
       debug_printf("%10p %7u %7u %10p %s\n",
                    (void *) fenced_buf,
@@ -618,7 +560,6 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list)
                    p_atomic_read(&fenced_buf->base.base.reference.count),
                    (void *) fenced_buf->fence,
                    signaled == 0 ? "y" : "n");
-      pipe_mutex_unlock(fenced_buf->mutex);
       curr = next; 
       next = curr->next;
    }
@@ -639,8 +580,8 @@ fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list)
 #if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
       sched_yield();
 #endif
+      _fenced_buffer_list_check_free(fenced_list, 1);
       pipe_mutex_lock(fenced_list->mutex);
-      fenced_buffer_list_check_free_locked(fenced_list, 1);
    }
 
 #ifdef DEBUG
@@ -648,7 +589,6 @@ fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list)
 #endif
       
    pipe_mutex_unlock(fenced_list->mutex);
-   pipe_mutex_destroy(fenced_list->mutex);
    
    fenced_list->ops->destroy(fenced_list->ops);
    
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index 1acf3c373eb..f675427d987 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -673,6 +673,13 @@ void x86_and( struct x86_function *p,
    emit_op_modrm( p, 0x23, 0x21, dst, src );
 }
 
+void x86_div( struct x86_function *p,
+              struct x86_reg src )
+{
+   assert(src.file == file_REG32 && src.mod == mod_REG);
+   emit_op_modrm(p, 0xf7, 0, x86_make_reg(file_REG32, 6), src);
+}
+
 
 
 /***********************************************************************
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
index 731a6517968..f7612d416a0 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
@@ -244,6 +244,7 @@ void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_sahf( struct x86_function *p );
+void x86_div( struct x86_function *p, struct x86_reg src );
 
 
 void x86_cdecl_caller_push_regs( struct x86_function *p );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index e2e5394f86f..d7ff262f30a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -123,7 +123,8 @@ static const char *semantic_names[] =
    "NORMAL",
    "FACE",
    "EDGEFLAG",
-   "PRIM_ID"
+   "PRIM_ID",
+   "INSTANCEID"
 };
 
 static const char *immediate_type_names[] =
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 2bcb33392a8..118a638ab48 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -1041,11 +1041,19 @@ fetch_src_file_channel(
 
       default:
          assert( 0 );
+         chan->u[0] = 0;
+         chan->u[1] = 0;
+         chan->u[2] = 0;
+         chan->u[3] = 0;
       }
       break;
 
    default:
       assert( 0 );
+      chan->u[0] = 0;
+      chan->u[1] = 0;
+      chan->u[2] = 0;
+      chan->u[3] = 0;
    }
 }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
index 138d2d095bb..ad553c71a57 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ppc.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
@@ -51,7 +51,8 @@
  * Since it's pretty much impossible to form PPC vector immediates, load
  * them from memory here:
  */
-const float ppc_builtin_constants[] ALIGN16_ATTRIB = {
+PIPE_ALIGN_VAR(16) const float 
+ppc_builtin_constants[] = {
    1.0f, -128.0f, 128.0, 0.0
 };
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index 7f1c8e5dd68..e1e4f97967d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -335,13 +335,13 @@ iter_instruction(
          fill_scan_register1d(ind_reg,
                               inst->Src[i].Indirect.File,
                               inst->Src[i].Indirect.Index);
-         if (!(reg->file == TGSI_FILE_ADDRESS || reg->file == TGSI_FILE_LOOP) ||
-             reg->indices[0] != 0) {
+         if (!(ind_reg->file == TGSI_FILE_ADDRESS || ind_reg->file == TGSI_FILE_LOOP) ||
+             ind_reg->indices[0] != 0) {
             report_warning(ctx, "Indirect register neither ADDR[0] nor LOOP[0]");
          }
          check_register_usage(
             ctx,
-            reg,
+            ind_reg,
             "indirect",
             FALSE );
       }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 2e13a7aaf9d..a85cc4659e0 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -2146,40 +2146,50 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_XPD:
+      /* Note: we do all stores after all operands have been fetched
+       * to avoid src/dst register aliasing issues for an instruction
+       * such as:  XPD TEMP[2].xyz, TEMP[0], TEMP[2];
+       */
       if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
           IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
-         FETCH( func, *inst, 1, 1, CHAN_Z );
-         FETCH( func, *inst, 3, 0, CHAN_Z );
+         FETCH( func, *inst, 1, 1, CHAN_Z ); /* xmm[1] = src[1].z */
+         FETCH( func, *inst, 3, 0, CHAN_Z ); /* xmm[3] = src[0].z */
       }
       if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
           IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
-         FETCH( func, *inst, 0, 0, CHAN_Y );
-         FETCH( func, *inst, 4, 1, CHAN_Y );
+         FETCH( func, *inst, 0, 0, CHAN_Y ); /* xmm[0] = src[0].y */
+         FETCH( func, *inst, 4, 1, CHAN_Y ); /* xmm[4] = src[1].y */
       }
       IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
-         emit_MOV( func, 2, 0 );
-         emit_mul( func, 2, 1 );
-         emit_MOV( func, 5, 3 );
-         emit_mul( func, 5, 4 );
-         emit_sub( func, 2, 5 );
-         STORE( func, *inst, 2, 0, CHAN_X );
+         emit_MOV( func, 7, 0 );  /* xmm[7] = xmm[0] */
+         emit_mul( func, 7, 1 );  /* xmm[7] = xmm[2] * xmm[1] */
+         emit_MOV( func, 5, 3 );  /* xmm[5] = xmm[3] */
+         emit_mul( func, 5, 4 );  /* xmm[5] = xmm[5] * xmm[4] */
+         emit_sub( func, 7, 5 );  /* xmm[7] = xmm[2] - xmm[5] */
+         /* store xmm[7] in dst.x below */
       }
       if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
           IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
-         FETCH( func, *inst, 2, 1, CHAN_X );
-         FETCH( func, *inst, 5, 0, CHAN_X );
+         FETCH( func, *inst, 2, 1, CHAN_X ); /* xmm[2] = src[1].x */
+         FETCH( func, *inst, 5, 0, CHAN_X ); /* xmm[5] = src[0].x */
       }
       IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
-         emit_mul( func, 3, 2 );
-         emit_mul( func, 1, 5 );
-         emit_sub( func, 3, 1 );
-         STORE( func, *inst, 3, 0, CHAN_Y );
+         emit_mul( func, 3, 2 );  /* xmm[3] = xmm[3] * xmm[2] */
+         emit_mul( func, 1, 5 );  /* xmm[1] = xmm[1] * xmm[5] */
+         emit_sub( func, 3, 1 );  /* xmm[3] = xmm[3] - xmm[1] */
+         /* store xmm[3] in dst.y below */
       }
       IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
-         emit_mul( func, 5, 4 );
-         emit_mul( func, 0, 2 );
-         emit_sub( func, 5, 0 );
-         STORE( func, *inst, 5, 0, CHAN_Z );
+         emit_mul( func, 5, 4 );  /* xmm[5] = xmm[5] * xmm[4] */
+         emit_mul( func, 0, 2 );  /* xmm[0] = xmm[0] * xmm[2] */
+         emit_sub( func, 5, 0 );  /* xmm[5] = xmm[5] - xmm[0] */
+         STORE( func, *inst, 5, 0, CHAN_Z ); /* dst.z = xmm[5] */
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         STORE( func, *inst, 7, 0, CHAN_X ); /* dst.x = xmm[7] */
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         STORE( func, *inst, 3, 0, CHAN_Y ); /* dst.y = xmm[3] */
       }
       IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
 	 emit_tempf(
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index e64e2b731df..8bd6f68dcc0 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -40,6 +40,8 @@ union tgsi_any_token {
    struct tgsi_header header;
    struct tgsi_processor processor;
    struct tgsi_token token;
+   struct tgsi_property prop;
+   struct tgsi_property_data prop_data;
    struct tgsi_declaration decl;
    struct tgsi_declaration_range decl_range;
    struct tgsi_declaration_semantic decl_semantic;
@@ -64,6 +66,7 @@ struct ureg_tokens {
 };
 
 #define UREG_MAX_INPUT PIPE_MAX_ATTRIBS
+#define UREG_MAX_SYSTEM_VALUE PIPE_MAX_ATTRIBS
 #define UREG_MAX_OUTPUT PIPE_MAX_ATTRIBS
 #define UREG_MAX_CONSTANT_RANGE 32
 #define UREG_MAX_IMMEDIATE 32
@@ -95,6 +98,13 @@ struct ureg_program
    unsigned nr_gs_inputs;
 
    struct {
+      unsigned index;
+      unsigned semantic_name;
+      unsigned semantic_index;
+   } system_value[UREG_MAX_SYSTEM_VALUE];
+   unsigned nr_system_values;
+
+   struct {
       unsigned semantic_name;
       unsigned semantic_index;
    } output[UREG_MAX_OUTPUT];
@@ -123,6 +133,8 @@ struct ureg_program
    } constant_range[UREG_MAX_CONSTANT_RANGE];
    unsigned nr_constant_ranges;
 
+   unsigned property_gs_input_prim;
+
    unsigned nr_addrs;
    unsigned nr_preds;
    unsigned nr_loops;
@@ -234,19 +246,28 @@ ureg_src_register( unsigned file,
    src.SwizzleY = TGSI_SWIZZLE_Y;
    src.SwizzleZ = TGSI_SWIZZLE_Z;
    src.SwizzleW = TGSI_SWIZZLE_W;
-   src.Pad      = 0;
    src.Indirect = 0;
    src.IndirectIndex = 0;
    src.IndirectSwizzle = 0;
    src.Absolute = 0;
    src.Index    = index;
    src.Negate   = 0;
+   src.Dimension = 0;
+   src.DimensionIndex = 0;
 
    return src;
 }
 
 
 
+void
+ureg_property_gs_input_prim(struct ureg_program *ureg,
+                            unsigned gs_input_prim)
+{
+   ureg->property_gs_input_prim = gs_input_prim;
+}
+
+
 
 struct ureg_src 
 ureg_DECL_fs_input( struct ureg_program *ureg,
@@ -304,6 +325,25 @@ ureg_DECL_gs_input(struct ureg_program *ureg,
 }
 
 
+struct ureg_src
+ureg_DECL_system_value(struct ureg_program *ureg,
+                       unsigned index,
+                       unsigned semantic_name,
+                       unsigned semantic_index)
+{
+   if (ureg->nr_system_values < UREG_MAX_SYSTEM_VALUE) {
+      ureg->system_value[ureg->nr_system_values].index = index;
+      ureg->system_value[ureg->nr_system_values].semantic_name = semantic_name;
+      ureg->system_value[ureg->nr_system_values].semantic_index = semantic_index;
+      ureg->nr_system_values++;
+   } else {
+      set_bad(ureg);
+   }
+
+   return ureg_src_register(TGSI_FILE_SYSTEM_VALUE, index);
+}
+
+
 struct ureg_dst 
 ureg_DECL_output( struct ureg_program *ureg,
                   unsigned name,
@@ -628,7 +668,7 @@ void
 ureg_emit_src( struct ureg_program *ureg,
                struct ureg_src src )
 {
-   unsigned size = 1 + (src.Indirect ? 1 : 0);
+   unsigned size = 1 + (src.Indirect ? 1 : 0) + (src.Dimension ? 1 : 0);
 
    union tgsi_any_token *out = get_tokens( ureg, DOMAIN_INSN, size );
    unsigned n = 0;
@@ -660,6 +700,15 @@ ureg_emit_src( struct ureg_program *ureg,
       n++;
    }
 
+   if (src.Dimension) {
+      out[0].src.Dimension = 1;
+      out[n].dim.Indirect = 0;
+      out[n].dim.Dimension = 0;
+      out[n].dim.Padding = 0;
+      out[n].dim.Index = src.DimensionIndex;
+      n++;
+   }
+
    assert(n == size);
 }
 
@@ -1027,13 +1076,34 @@ emit_immediate( struct ureg_program *ureg,
    out[4].imm_data.Uint = v[3];
 }
 
+static void
+emit_property(struct ureg_program *ureg,
+              unsigned name,
+              unsigned data)
+{
+   union tgsi_any_token *out = get_tokens(ureg, DOMAIN_DECL, 2);
 
+   out[0].value = 0;
+   out[0].prop.Type = TGSI_TOKEN_TYPE_PROPERTY;
+   out[0].prop.NrTokens = 2;
+   out[0].prop.PropertyName = name;
+
+   out[1].prop_data.Data = data;
+}
 
 
 static void emit_decls( struct ureg_program *ureg )
 {
    unsigned i;
 
+   if (ureg->property_gs_input_prim != ~0) {
+      assert(ureg->processor == TGSI_PROCESSOR_GEOMETRY);
+
+      emit_property(ureg,
+                    TGSI_PROPERTY_GS_INPUT_PRIM,
+                    ureg->property_gs_input_prim);
+   }
+
    if (ureg->processor == TGSI_PROCESSOR_VERTEX) {
       for (i = 0; i < UREG_MAX_INPUT; i++) {
          if (ureg->vs_inputs[i/32] & (1 << (i%32))) {
@@ -1058,6 +1128,15 @@ static void emit_decls( struct ureg_program *ureg )
       }
    }
 
+   for (i = 0; i < ureg->nr_system_values; i++) {
+      emit_decl(ureg,
+                TGSI_FILE_SYSTEM_VALUE,
+                ureg->system_value[i].index,
+                ureg->system_value[i].semantic_name,
+                ureg->system_value[i].semantic_index,
+                TGSI_INTERPOLATE_CONSTANT);
+   }
+
    for (i = 0; i < ureg->nr_outputs; i++) {
       emit_decl( ureg, 
                  TGSI_FILE_OUTPUT, 
@@ -1234,6 +1313,7 @@ struct ureg_program *ureg_create( unsigned processor )
       return NULL;
 
    ureg->processor = processor;
+   ureg->property_gs_input_prim = ~0;
    return ureg;
 }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index 6f11273320a..03eaf24854c 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -47,13 +47,14 @@ struct ureg_src
    unsigned SwizzleY    : 2;  /* TGSI_SWIZZLE_ */
    unsigned SwizzleZ    : 2;  /* TGSI_SWIZZLE_ */
    unsigned SwizzleW    : 2;  /* TGSI_SWIZZLE_ */
-   unsigned Pad         : 1;  /* BOOL */
    unsigned Indirect    : 1;  /* BOOL */
+   unsigned Dimension   : 1;  /* BOOL */
    unsigned Absolute    : 1;  /* BOOL */
-   int      Index       : 16; /* SINT */
    unsigned Negate      : 1;  /* BOOL */
+   int      Index       : 16; /* SINT */
    int      IndirectIndex   : 16; /* SINT */
    int      IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
+   int      DimensionIndex : 16;  /* SINT */
 };
 
 /* Very similar to a tgsi_dst_register, removing unsupported fields
@@ -118,6 +119,14 @@ ureg_create_shader_and_destroy( struct ureg_program *p,
 }
 
 
+/***********************************************************************
+ * Build shader properties:
+ */
+
+void
+ureg_property_gs_input_prim(struct ureg_program *ureg,
+                            unsigned gs_input_prim);
+
 
 /***********************************************************************
  * Build shader declarations:
@@ -137,6 +146,12 @@ struct ureg_src
 ureg_DECL_gs_input(struct ureg_program *,
                    unsigned index);
 
+struct ureg_src
+ureg_DECL_system_value(struct ureg_program *,
+                       unsigned index,
+                       unsigned semantic_name,
+                       unsigned semantic_index);
+
 struct ureg_dst
 ureg_DECL_output( struct ureg_program *,
                   unsigned semantic_name,
@@ -760,6 +775,15 @@ ureg_src_indirect( struct ureg_src reg, struct ureg_src addr )
    return reg;
 }
 
+static INLINE struct ureg_src 
+ureg_src_dimension( struct ureg_src reg, int index )
+{
+   assert(reg.File != TGSI_FILE_NULL);
+   reg.Dimension = 1;
+   reg.DimensionIndex = index;
+   return reg;
+}
+
 static INLINE struct ureg_dst
 ureg_dst( struct ureg_src src )
 {
@@ -792,13 +816,14 @@ ureg_src( struct ureg_dst dst )
    src.SwizzleY  = TGSI_SWIZZLE_Y;
    src.SwizzleZ  = TGSI_SWIZZLE_Z;
    src.SwizzleW  = TGSI_SWIZZLE_W;
-   src.Pad       = 0;
    src.Indirect  = dst.Indirect;
    src.IndirectIndex = dst.IndirectIndex;
    src.IndirectSwizzle = dst.IndirectSwizzle;
    src.Absolute  = 0;
    src.Index     = dst.Index;
    src.Negate    = 0;
+   src.Dimension = 0;
+   src.DimensionIndex = 0;
 
    return src;
 }
@@ -837,13 +862,14 @@ ureg_src_undef( void )
    src.SwizzleY  = 0;
    src.SwizzleZ  = 0;
    src.SwizzleW  = 0;
-   src.Pad       = 0;
    src.Indirect  = 0;
    src.IndirectIndex = 0;
    src.IndirectSwizzle = 0;
    src.Absolute  = 0;
    src.Index     = 0;
    src.Negate    = 0;
+   src.Dimension = 0;
+   src.DimensionIndex = 0;
    
    return src;
 }
diff --git a/src/gallium/auxiliary/translate/translate.h b/src/gallium/auxiliary/translate/translate.h
index 34526eb0617..54ed2c1a4be 100644
--- a/src/gallium/auxiliary/translate/translate.h
+++ b/src/gallium/auxiliary/translate/translate.h
@@ -44,12 +44,19 @@
 #include "pipe/p_format.h"
 #include "pipe/p_state.h"
 
+enum translate_element_type {
+   TRANSLATE_ELEMENT_NORMAL,
+   TRANSLATE_ELEMENT_INSTANCE_ID
+};
+
 struct translate_element 
 {
+   enum translate_element_type type;
    enum pipe_format input_format;
    enum pipe_format output_format;
    unsigned input_buffer:8;
    unsigned input_offset:24;
+   unsigned instance_divisor;
    unsigned output_offset;
 };
 
@@ -74,11 +81,13 @@ struct translate {
    void (PIPE_CDECL *run_elts)( struct translate *,
                                 const unsigned *elts,
                                 unsigned count,
+                                unsigned instance_id,
                                 void *output_buffer);
 
    void (PIPE_CDECL *run)( struct translate *,
                            unsigned start,
                            unsigned count,
+                           unsigned instance_id,
                            void *output_buffer);
 };
 
@@ -103,8 +112,13 @@ static INLINE int translate_keysize( const struct translate_key *key )
 static INLINE int translate_key_compare( const struct translate_key *a,
                                          const struct translate_key *b )
 {
-   int keysize = translate_keysize(a);
-   return memcmp(a, b, keysize);
+   int keysize_a = translate_keysize(a);
+   int keysize_b = translate_keysize(b);
+
+   if (keysize_a != keysize_b) {
+      return keysize_a - keysize_b;
+   }
+   return memcmp(a, b, keysize_a);
 }
 
 
diff --git a/src/gallium/auxiliary/translate/translate_generic.c b/src/gallium/auxiliary/translate/translate_generic.c
index 266e7ee81e6..24727d49888 100644
--- a/src/gallium/auxiliary/translate/translate_generic.c
+++ b/src/gallium/auxiliary/translate/translate_generic.c
@@ -46,9 +46,12 @@ struct translate_generic {
    struct translate translate;
 
    struct {
+      enum translate_element_type type;
+
       fetch_func fetch;
       unsigned buffer;
       unsigned input_offset;
+      unsigned instance_divisor;
 
       emit_func emit;
       unsigned output_offset;
@@ -568,6 +571,7 @@ static emit_func get_emit_func( enum pipe_format format )
 static void PIPE_CDECL generic_run_elts( struct translate *translate,
                                          const unsigned *elts,
                                          unsigned count,
+                                         unsigned instance_id,
                                          void *output_buffer )
 {
    struct translate_generic *tg = translate_generic(translate);
@@ -583,13 +587,20 @@ static void PIPE_CDECL generic_run_elts( struct translate *translate,
 
       for (attr = 0; attr < nr_attrs; attr++) {
 	 float data[4];
-
-	 const char *src = (tg->attrib[attr].input_ptr + 
-			    tg->attrib[attr].input_stride * elt);
+         const char *src;
 
 	 char *dst = (vert + 
 		      tg->attrib[attr].output_offset);
 
+         if (tg->attrib[attr].instance_divisor) {
+            src = tg->attrib[attr].input_ptr +
+                  tg->attrib[attr].input_stride *
+                  (instance_id / tg->attrib[attr].instance_divisor);
+         } else {
+            src = tg->attrib[attr].input_ptr +
+                  tg->attrib[attr].input_stride * elt;
+         }
+
 	 tg->attrib[attr].fetch( src, data );
 
          if (0) debug_printf("vert %d/%d attr %d: %f %f %f %f\n",
@@ -607,6 +618,7 @@ static void PIPE_CDECL generic_run_elts( struct translate *translate,
 static void PIPE_CDECL generic_run( struct translate *translate,
                                     unsigned start,
                                     unsigned count,
+                                    unsigned instance_id,
                                     void *output_buffer )
 {
    struct translate_generic *tg = translate_generic(translate);
@@ -623,13 +635,25 @@ static void PIPE_CDECL generic_run( struct translate *translate,
       for (attr = 0; attr < nr_attrs; attr++) {
 	 float data[4];
 
-	 const char *src = (tg->attrib[attr].input_ptr + 
-			    tg->attrib[attr].input_stride * elt);
-
 	 char *dst = (vert + 
 		      tg->attrib[attr].output_offset);
 
-	 tg->attrib[attr].fetch( src, data );
+         if (tg->attrib[attr].type == TRANSLATE_ELEMENT_NORMAL) {
+            const char *src;
+
+            if (tg->attrib[attr].instance_divisor) {
+               src = tg->attrib[attr].input_ptr +
+                     tg->attrib[attr].input_stride *
+                     (instance_id / tg->attrib[attr].instance_divisor);
+            } else {
+               src = tg->attrib[attr].input_ptr +
+                     tg->attrib[attr].input_stride * elt;
+            }
+
+            tg->attrib[attr].fetch( src, data );
+         } else {
+            data[0] = (float)instance_id;
+         }
 
          if (0) debug_printf("vert %d attr %d: %f %f %f %f\n",
                              i, attr, data[0], data[1], data[2], data[3]);
@@ -683,10 +707,12 @@ struct translate *translate_generic_create( const struct translate_key *key )
    tg->translate.run = generic_run;
 
    for (i = 0; i < key->nr_elements; i++) {
+      tg->attrib[i].type = key->element[i].type;
 
       tg->attrib[i].fetch = get_fetch_func(key->element[i].input_format);
       tg->attrib[i].buffer = key->element[i].input_buffer;
       tg->attrib[i].input_offset = key->element[i].input_offset;
+      tg->attrib[i].instance_divisor = key->element[i].instance_divisor;
 
       tg->attrib[i].emit = get_emit_func(key->element[i].output_format);
       tg->attrib[i].output_offset = key->element[i].output_offset;
diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c
index b62db8d8f33..c13e7427387 100644
--- a/src/gallium/auxiliary/translate/translate_sse.c
+++ b/src/gallium/auxiliary/translate/translate_sse.c
@@ -49,19 +49,29 @@
 typedef void (PIPE_CDECL *run_func)( struct translate *translate,
                                      unsigned start,
                                      unsigned count,
-                                     void *output_buffer );
+                                     unsigned instance_id,
+                                     void *output_buffer);
 
 typedef void (PIPE_CDECL *run_elts_func)( struct translate *translate,
                                           const unsigned *elts,
                                           unsigned count,
-                                          void *output_buffer );
+                                          unsigned instance_id,
+                                          void *output_buffer);
 
 struct translate_buffer {
    const void *base_ptr;
    unsigned stride;
-   void *ptr;                   /* updated per vertex */
 };
 
+struct translate_buffer_varient {
+   unsigned buffer_index;
+   unsigned instance_divisor;
+   void *ptr;                    /* updated either per vertex or per instance */
+};
+
+
+#define ELEMENT_BUFFER_INSTANCE_ID  1001
+
 
 struct translate_sse {
    struct translate translate;
@@ -81,6 +91,16 @@ struct translate_sse {
    struct translate_buffer buffer[PIPE_MAX_ATTRIBS];
    unsigned nr_buffers;
 
+   /* Multiple buffer varients can map to a single buffer. */
+   struct translate_buffer_varient buffer_varient[PIPE_MAX_ATTRIBS];
+   unsigned nr_buffer_varients;
+
+   /* Multiple elements can map to a single buffer varient. */
+   unsigned element_to_buffer_varient[PIPE_MAX_ATTRIBS];
+
+   boolean use_instancing;
+   unsigned instance_id;
+
    run_func      gen_run;
    run_elts_func gen_run_elts;
 
@@ -359,32 +379,61 @@ static boolean init_inputs( struct translate_sse *p,
                             boolean linear )
 {
    unsigned i;
-   if (linear) {
-      for (i = 0; i < p->nr_buffers; i++) {
+   struct x86_reg instance_id = x86_make_disp(p->machine_EDX,
+                                              get_offset(p, &p->instance_id));
+
+   for (i = 0; i < p->nr_buffer_varients; i++) {
+      struct translate_buffer_varient *varient = &p->buffer_varient[i];
+      struct translate_buffer *buffer = &p->buffer[varient->buffer_index];
+
+      if (linear || varient->instance_divisor) {
          struct x86_reg buf_stride   = x86_make_disp(p->machine_EDX,
-                                                     get_offset(p, &p->buffer[i].stride));
+                                                     get_offset(p, &buffer->stride));
          struct x86_reg buf_ptr      = x86_make_disp(p->machine_EDX,
-                                                     get_offset(p, &p->buffer[i].ptr));
+                                                     get_offset(p, &varient->ptr));
          struct x86_reg buf_base_ptr = x86_make_disp(p->machine_EDX,
-                                                     get_offset(p, &p->buffer[i].base_ptr));
+                                                     get_offset(p, &buffer->base_ptr));
          struct x86_reg elt = p->idx_EBX;
-         struct x86_reg tmp = p->tmp_EAX;
-
+         struct x86_reg tmp_EAX = p->tmp_EAX;
 
          /* Calculate pointer to first attrib:
+          *   base_ptr + stride * index, where index depends on instance divisor
           */
-         x86_mov(p->func, tmp, buf_stride);
-         x86_imul(p->func, tmp, elt);
-         x86_add(p->func, tmp, buf_base_ptr);
+         if (varient->instance_divisor) {
+            /* Our index is instance ID divided by instance divisor.
+             */
+            x86_mov(p->func, tmp_EAX, instance_id);
+
+            if (varient->instance_divisor != 1) {
+               struct x86_reg tmp_EDX = p->machine_EDX;
+               struct x86_reg tmp_ECX = p->outbuf_ECX;
+
+               /* TODO: Add x86_shr() to rtasm and use it whenever
+                *       instance divisor is power of two.
+                */
+
+               x86_push(p->func, tmp_EDX);
+               x86_push(p->func, tmp_ECX);
+               x86_xor(p->func, tmp_EDX, tmp_EDX);
+               x86_mov_reg_imm(p->func, tmp_ECX, varient->instance_divisor);
+               x86_div(p->func, tmp_ECX);    /* EAX = EDX:EAX / ECX */
+               x86_pop(p->func, tmp_ECX);
+               x86_pop(p->func, tmp_EDX);
+            }
+         } else {
+            x86_mov(p->func, tmp_EAX, elt);
+         }
+         x86_imul(p->func, tmp_EAX, buf_stride);
+         x86_add(p->func, tmp_EAX, buf_base_ptr);
 
 
          /* In the linear case, keep the buffer pointer instead of the
           * index number.
           */
-         if (p->nr_buffers == 1) 
-            x86_mov( p->func, elt, tmp );
+         if (linear && p->nr_buffer_varients == 1)
+            x86_mov(p->func, elt, tmp_EAX);
          else
-            x86_mov( p->func, buf_ptr, tmp );
+            x86_mov(p->func, buf_ptr, tmp_EAX);
       }
    }
 
@@ -394,31 +443,36 @@ static boolean init_inputs( struct translate_sse *p,
 
 static struct x86_reg get_buffer_ptr( struct translate_sse *p,
                                       boolean linear,
-                                      unsigned buf_idx,
+                                      unsigned var_idx,
                                       struct x86_reg elt )
 {
-   if (linear && p->nr_buffers == 1) {
+   if (var_idx == ELEMENT_BUFFER_INSTANCE_ID) {
+      return x86_make_disp(p->machine_EDX,
+                           get_offset(p, &p->instance_id));
+   }
+   if (linear && p->nr_buffer_varients == 1) {
       return p->idx_EBX;
    }
-   else if (linear) {
+   else if (linear || p->buffer_varient[var_idx].instance_divisor) {
       struct x86_reg ptr = p->tmp_EAX;
       struct x86_reg buf_ptr = 
          x86_make_disp(p->machine_EDX, 
-                       get_offset(p, &p->buffer[buf_idx].ptr));
+                       get_offset(p, &p->buffer_varient[var_idx].ptr));
       
       x86_mov(p->func, ptr, buf_ptr);
       return ptr;
    }
    else {
       struct x86_reg ptr = p->tmp_EAX;
+      const struct translate_buffer_varient *varient = &p->buffer_varient[var_idx];
 
       struct x86_reg buf_stride = 
          x86_make_disp(p->machine_EDX, 
-                       get_offset(p, &p->buffer[buf_idx].stride));
+                       get_offset(p, &p->buffer[varient->buffer_index].stride));
 
       struct x86_reg buf_base_ptr = 
          x86_make_disp(p->machine_EDX, 
-                       get_offset(p, &p->buffer[buf_idx].base_ptr));
+                       get_offset(p, &p->buffer[varient->buffer_index].base_ptr));
 
 
 
@@ -436,28 +490,33 @@ static struct x86_reg get_buffer_ptr( struct translate_sse *p,
 static boolean incr_inputs( struct translate_sse *p, 
                             boolean linear )
 {
-   if (linear && p->nr_buffers == 1) {
+   if (linear && p->nr_buffer_varients == 1) {
       struct x86_reg stride = x86_make_disp(p->machine_EDX,
                                             get_offset(p, &p->buffer[0].stride));
 
-      x86_add(p->func, p->idx_EBX, stride);
-      sse_prefetchnta(p->func, x86_make_disp(p->idx_EBX, 192));
+      if (p->buffer_varient[0].instance_divisor == 0) {
+         x86_add(p->func, p->idx_EBX, stride);
+         sse_prefetchnta(p->func, x86_make_disp(p->idx_EBX, 192));
+      }
    }
    else if (linear) {
       unsigned i;
 
       /* Is this worthwhile??
        */
-      for (i = 0; i < p->nr_buffers; i++) {
+      for (i = 0; i < p->nr_buffer_varients; i++) {
+         struct translate_buffer_varient *varient = &p->buffer_varient[i];
          struct x86_reg buf_ptr = x86_make_disp(p->machine_EDX,
-                                                get_offset(p, &p->buffer[i].ptr));
+                                                get_offset(p, &varient->ptr));
          struct x86_reg buf_stride = x86_make_disp(p->machine_EDX,
-                                                   get_offset(p, &p->buffer[i].stride));
+                                                   get_offset(p, &p->buffer[varient->buffer_index].stride));
 
-         x86_mov(p->func, p->tmp_EAX, buf_ptr);
-         x86_add(p->func, p->tmp_EAX, buf_stride);
-         if (i == 0) sse_prefetchnta(p->func, x86_make_disp(p->tmp_EAX, 192));
-         x86_mov(p->func, buf_ptr, p->tmp_EAX);
+         if (varient->instance_divisor == 0) {
+            x86_mov(p->func, p->tmp_EAX, buf_ptr);
+            x86_add(p->func, p->tmp_EAX, buf_stride);
+            if (i == 0) sse_prefetchnta(p->func, x86_make_disp(p->tmp_EAX, 192));
+            x86_mov(p->func, buf_ptr, p->tmp_EAX);
+         }
       }
    } 
    else {
@@ -514,7 +573,18 @@ static boolean build_vertex_emit( struct translate_sse *p,
    x86_mov(p->func, p->machine_EDX, x86_fn_arg(p->func, 1));
    x86_mov(p->func, p->idx_EBX, x86_fn_arg(p->func, 2));
    x86_mov(p->func, p->count_ESI, x86_fn_arg(p->func, 3));
-   x86_mov(p->func, p->outbuf_ECX, x86_fn_arg(p->func, 4));
+   x86_mov(p->func, p->outbuf_ECX, x86_fn_arg(p->func, 5));
+
+   /* Load instance ID.
+    */
+   if (p->use_instancing) {
+      x86_mov(p->func,
+              p->tmp_EAX,
+              x86_fn_arg(p->func, 4));
+      x86_mov(p->func,
+              x86_make_disp(p->machine_EDX, get_offset(p, &p->instance_id)),
+              p->tmp_EAX);
+   }
 
    /* Get vertex count, compare to zero
     */
@@ -531,17 +601,18 @@ static boolean build_vertex_emit( struct translate_sse *p,
    label = x86_get_label(p->func);
    {
       struct x86_reg elt = linear ? p->idx_EBX : x86_deref(p->idx_EBX);
-      int last_vb = -1;
+      int last_varient = -1;
       struct x86_reg vb;
 
       for (j = 0; j < p->translate.key.nr_elements; j++) {
          const struct translate_element *a = &p->translate.key.element[j];
+         unsigned varient = p->element_to_buffer_varient[j];
 
          /* Figure out source pointer address:
           */
-         if (a->input_buffer != last_vb) {
-            last_vb = a->input_buffer;
-            vb = get_buffer_ptr(p, linear, a->input_buffer, elt);
+         if (varient != last_varient) {
+            last_varient = varient;
+            vb = get_buffer_ptr(p, linear, varient, elt);
          }
          
          if (!translate_attr( p, a, 
@@ -624,6 +695,7 @@ static void translate_sse_release( struct translate *translate )
 static void PIPE_CDECL translate_sse_run_elts( struct translate *translate,
 			      const unsigned *elts,
 			      unsigned count,
+                              unsigned instance_id,
 			      void *output_buffer )
 {
    struct translate_sse *p = (struct translate_sse *)translate;
@@ -631,12 +703,14 @@ static void PIPE_CDECL translate_sse_run_elts( struct translate *translate,
    p->gen_run_elts( translate,
 		    elts,
 		    count,
-		    output_buffer );
+                    instance_id,
+                    output_buffer);
 }
 
 static void PIPE_CDECL translate_sse_run( struct translate *translate,
 			 unsigned start,
 			 unsigned count,
+                         unsigned instance_id,
 			 void *output_buffer )
 {
    struct translate_sse *p = (struct translate_sse *)translate;
@@ -644,7 +718,8 @@ static void PIPE_CDECL translate_sse_run( struct translate *translate,
    p->gen_run( translate,
 	       start,
 	       count,
-	       output_buffer );
+               instance_id,
+               output_buffer);
 }
 
 
@@ -666,8 +741,37 @@ struct translate *translate_sse2_create( const struct translate_key *key )
    p->translate.run_elts = translate_sse_run_elts;
    p->translate.run = translate_sse_run;
 
-   for (i = 0; i < key->nr_elements; i++) 
-      p->nr_buffers = MAX2( p->nr_buffers, key->element[i].input_buffer + 1 );
+   for (i = 0; i < key->nr_elements; i++) {
+      if (key->element[i].type == TRANSLATE_ELEMENT_NORMAL) {
+         unsigned j;
+
+         p->nr_buffers = MAX2(p->nr_buffers, key->element[i].input_buffer + 1);
+
+         if (key->element[i].instance_divisor) {
+            p->use_instancing = TRUE;
+         }
+
+         /*
+          * Map vertex element to vertex buffer varient.
+          */
+         for (j = 0; j < p->nr_buffer_varients; j++) {
+            if (p->buffer_varient[j].buffer_index == key->element[i].input_buffer &&
+                p->buffer_varient[j].instance_divisor == key->element[i].instance_divisor) {
+               break;
+            }
+         }
+         if (j == p->nr_buffer_varients) {
+            p->buffer_varient[j].buffer_index = key->element[i].input_buffer;
+            p->buffer_varient[j].instance_divisor = key->element[i].instance_divisor;
+            p->nr_buffer_varients++;
+         }
+         p->element_to_buffer_varient[i] = j;
+      } else {
+         assert(key->element[i].type == TRANSLATE_ELEMENT_INSTANCE_ID);
+
+         p->element_to_buffer_varient[i] = ELEMENT_BUFFER_INSTANCE_ID;
+      }
+   }
 
    if (0) debug_printf("nr_buffers: %d\n", p->nr_buffers);
 
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index cef3b69e46d..249a0375fc5 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -163,6 +163,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    rs_state.cull_mode = PIPE_WINDING_NONE;
    rs_state.bypass_vs_clip_and_viewport = 1;
    rs_state.gl_rasterization_rules = 1;
+   rs_state.flatshade = 1;
    ctx->rs_state = pipe->create_rasterizer_state(pipe, &rs_state);
 
    /* fragment shaders are created on-demand */
@@ -378,9 +379,16 @@ static void blitter_set_texcoords_cube(struct blitter_context_priv *ctx,
    float t1 = y1 / (float)surf->height;
    float s2 = x2 / (float)surf->width;
    float t2 = y2 / (float)surf->height;
-   const float st[4][2] = {
-      {s1, t1}, {s2, t1}, {s2, t2}, {s1, t2}
-   };
+   float st[4][2];
+
+   st[0][0] = s1;
+   st[0][1] = t1;
+   st[1][0] = s2;
+   st[1][1] = t1;
+   st[2][0] = s2;
+   st[2][1] = t2;
+   st[3][0] = s1;
+   st[3][1] = t2;
 
    util_map_texcoords2d_onto_cubemap(surf->face,
                                      /* pointer, stride in floats */
@@ -567,45 +575,29 @@ void util_blitter_clear(struct blitter_context *blitter,
    blitter_restore_CSOs(ctx);
 }
 
-void util_blitter_copy(struct blitter_context *blitter,
-                       struct pipe_surface *dst,
-                       unsigned dstx, unsigned dsty,
-                       struct pipe_surface *src,
-                       unsigned srcx, unsigned srcy,
-                       unsigned width, unsigned height,
-                       boolean ignore_stencil)
+static boolean
+is_overlap(unsigned sx1, unsigned sx2, unsigned sy1, unsigned sy2,
+           unsigned dx1, unsigned dx2, unsigned dy1, unsigned dy2)
+{
+    if (sx1 >= dx2 || sx2 <= dx1 || sy1 >= dy2 || sy2 <= dy1) {
+        return FALSE;
+    } else {
+        return TRUE;
+    }
+}
+
+static void util_blitter_do_copy(struct blitter_context *blitter,
+				 struct pipe_surface *dst,
+				 unsigned dstx, unsigned dsty,
+				 struct pipe_surface *src,
+				 unsigned srcx, unsigned srcy,
+				 unsigned width, unsigned height,
+				 boolean is_depth)
 {
    struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
    struct pipe_context *pipe = ctx->pipe;
-   struct pipe_screen *screen = pipe->screen;
    struct pipe_framebuffer_state fb_state;
-   boolean is_stencil, is_depth;
-   unsigned dst_tex_usage;
-
-   /* give up if textures are not set */
-   assert(dst->texture && src->texture);
-   if (!dst->texture || !src->texture)
-      return;
-
-   is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
-   is_stencil = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 1) != 0;
-   dst_tex_usage = is_depth || is_stencil ? PIPE_TEXTURE_USAGE_DEPTH_STENCIL :
-                                            PIPE_TEXTURE_USAGE_RENDER_TARGET;
 
-   /* check if we can sample from and render to the surfaces */
-   /* (assuming copying a stencil buffer is not possible) */
-   if ((!ignore_stencil && is_stencil) ||
-       !screen->is_format_supported(screen, dst->format, dst->texture->target,
-                                    dst_tex_usage, 0) ||
-       !screen->is_format_supported(screen, src->format, src->texture->target,
-                                    PIPE_TEXTURE_USAGE_SAMPLER, 0)) {
-      util_surface_copy(pipe, FALSE, dst, dstx, dsty, src, srcx, srcy,
-                        width, height);
-      return;
-   }
-
-   /* check whether the states are properly saved */
-   blitter_check_saved_CSOs(ctx);
    assert(blitter->saved_fb_state.nr_cbufs != ~0);
    assert(blitter->saved_num_textures != ~0);
    assert(blitter->saved_num_sampler_states != ~0);
@@ -663,6 +655,108 @@ void util_blitter_copy(struct blitter_context *blitter,
 
    blitter_set_rectangle(ctx, dstx, dsty, dstx+width, dsty+height, 0);
    blitter_draw_quad(ctx);
+
+}
+
+static void util_blitter_overlap_copy(struct blitter_context *blitter,
+				      struct pipe_surface *dst,
+				      unsigned dstx, unsigned dsty,
+				      struct pipe_surface *src,
+				      unsigned srcx, unsigned srcy,
+				      unsigned width, unsigned height)
+{
+   struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+
+   struct pipe_texture texTemp;
+   struct pipe_texture *texture;
+   struct pipe_surface *tex_surf;
+
+   /* check whether the states are properly saved */
+   blitter_check_saved_CSOs(ctx);
+
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = dst->texture->format; /* XXX verify supported by driver! */
+   texTemp.last_level = 0;
+   texTemp.width0 = width;
+   texTemp.height0 = height;
+   texTemp.depth0 = 1;
+
+   texture = screen->texture_create(screen, &texTemp);
+   if (!texture)
+      return;
+
+   tex_surf = screen->get_tex_surface(screen, texture, 0, 0, 0,
+				      PIPE_BUFFER_USAGE_GPU_READ | 
+				      PIPE_BUFFER_USAGE_GPU_WRITE);
+
+   /* blit from the src to the temp */
+   util_blitter_do_copy(blitter, tex_surf, 0, 0,
+			src, srcx, srcy,
+			width, height,
+			FALSE);
+   util_blitter_do_copy(blitter, dst, dstx, dsty,
+			tex_surf, 0, 0,
+			width, height,
+			FALSE);
+   pipe_surface_reference(&tex_surf, NULL);
+   pipe_texture_reference(&texture, NULL);
+   blitter_restore_CSOs(ctx);
+}
+
+void util_blitter_copy(struct blitter_context *blitter,
+                       struct pipe_surface *dst,
+                       unsigned dstx, unsigned dsty,
+                       struct pipe_surface *src,
+                       unsigned srcx, unsigned srcy,
+                       unsigned width, unsigned height,
+                       boolean ignore_stencil)
+{
+   struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   boolean is_stencil, is_depth;
+   unsigned dst_tex_usage;
+
+   /* give up if textures are not set */
+   assert(dst->texture && src->texture);
+   if (!dst->texture || !src->texture)
+      return;
+
+   if (dst->texture == src->texture) {
+      if (is_overlap(srcx, srcx + width, srcy, srcy + height,
+		             dstx, dstx + width, dsty, dsty + height)) {
+         util_blitter_overlap_copy(blitter, dst, dstx, dsty, src, srcx, srcy,
+                                   width, height);
+         return;
+      }
+   }
+		   
+   is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
+   is_stencil = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 1) != 0;
+   dst_tex_usage = is_depth || is_stencil ? PIPE_TEXTURE_USAGE_DEPTH_STENCIL :
+                                            PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+   /* check if we can sample from and render to the surfaces */
+   /* (assuming copying a stencil buffer is not possible) */
+   if ((!ignore_stencil && is_stencil) ||
+       !screen->is_format_supported(screen, dst->format, dst->texture->target,
+                                    dst_tex_usage, 0) ||
+       !screen->is_format_supported(screen, src->format, src->texture->target,
+                                    PIPE_TEXTURE_USAGE_SAMPLER, 0)) {
+      util_surface_copy(pipe, FALSE, dst, dstx, dsty, src, srcx, srcy,
+                        width, height);
+      return;
+   }
+
+   /* check whether the states are properly saved */
+   blitter_check_saved_CSOs(ctx);
+   util_blitter_do_copy(blitter,
+			dst, dstx, dsty,
+			src, srcx, srcy,
+			width, height, is_depth);
    blitter_restore_CSOs(ctx);
 }
 
diff --git a/src/gallium/auxiliary/util/u_draw_quad.c b/src/gallium/auxiliary/util/u_draw_quad.c
index 4110485fb19..e2e23c3cdd8 100644
--- a/src/gallium/auxiliary/util/u_draw_quad.c
+++ b/src/gallium/auxiliary/util/u_draw_quad.c
@@ -61,6 +61,7 @@ util_draw_vertex_buffer(struct pipe_context *pipe,
    /* tell pipe about the vertex attributes */
    for (i = 0; i < num_attribs; i++) {
       velements[i].src_offset = i * 4 * sizeof(float);
+      velements[i].instance_divisor = 0;
       velements[i].vertex_buffer_index = 0;
       velements[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       velements[i].nr_components = 4;
diff --git a/src/gallium/auxiliary/util/u_format.csv b/src/gallium/auxiliary/util/u_format.csv
index 9f16b42944e..01f7931aed1 100644
--- a/src/gallium/auxiliary/util/u_format.csv
+++ b/src/gallium/auxiliary/util/u_format.csv
@@ -62,10 +62,10 @@ PIPE_FORMAT_R16G16_SSCALED        , array , 1, 1, s16 , s16 ,     ,     , xy01,
 PIPE_FORMAT_R16G16B16_SSCALED     , array , 1, 1, s16 , s16 , s16 ,     , xyz1, rgb
 PIPE_FORMAT_R16G16B16A16_SSCALED  , array , 1, 1, s16 , s16 , s16 , s16 , xyzw, rgb
 PIPE_FORMAT_R8_UNORM              , array , 1, 1, un8 ,     ,     ,     , x001, rgb
-PIPE_FORMAT_R8G8_UNORM            , array , 1, 1, un8 , un8 ,     ,     , xy01, rgb
-PIPE_FORMAT_R8G8B8_UNORM          , array , 1, 1, un8 , un8 , un8 ,     , xyz1, rgb
-PIPE_FORMAT_R8G8B8A8_UNORM        , array , 1, 1, un8 , un8 , un8 , un8 , xyzw, rgb
-PIPE_FORMAT_R8G8B8X8_UNORM        , array , 1, 1, un8 , un8 , un8 , un8 , xyz1, rgb
+PIPE_FORMAT_R8G8_UNORM            , array , 1, 1, un8 , un8 ,     ,     , yx01, rgb
+PIPE_FORMAT_R8G8B8_UNORM          , array , 1, 1, un8 , un8 , un8 ,     , zyx1, rgb
+PIPE_FORMAT_R8G8B8A8_UNORM        , array , 1, 1, un8 , un8 , un8 , un8 , wzyx, rgb
+PIPE_FORMAT_R8G8B8X8_UNORM        , array , 1, 1, un8 , un8 , un8 , un8 , wzy1, rgb
 PIPE_FORMAT_R8_USCALED            , array , 1, 1, u8  ,     ,     ,     , x001, rgb
 PIPE_FORMAT_R8G8_USCALED          , array , 1, 1, u8  , u8  ,     ,     , xy01, rgb
 PIPE_FORMAT_R8G8B8_USCALED        , array , 1, 1, u8  , u8  , u8  ,     , xyz1, rgb
diff --git a/src/gallium/auxiliary/util/u_pack_color.h b/src/gallium/auxiliary/util/u_pack_color.h
index 43eb0153ee7..0ab53c75dd6 100644
--- a/src/gallium/auxiliary/util/u_pack_color.h
+++ b/src/gallium/auxiliary/util/u_pack_color.h
@@ -425,6 +425,8 @@ util_pack_z(enum pipe_format format, double z)
       if (z == 1.0)
          return 0xffffffff;
       return (uint) (z * 0xffffffff);
+   case PIPE_FORMAT_Z32_FLOAT:
+      return (uint)z;
    case PIPE_FORMAT_S8Z24_UNORM:
    case PIPE_FORMAT_X8Z24_UNORM:
       if (z == 1.0)
diff --git a/src/gallium/auxiliary/util/u_ringbuffer.c b/src/gallium/auxiliary/util/u_ringbuffer.c
new file mode 100644
index 00000000000..3f43a19e018
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_ringbuffer.c
@@ -0,0 +1,145 @@
+
+#include "pipe/p_thread.h"
+#include "pipe/p_defines.h"
+#include "util/u_ringbuffer.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+/* Generic ringbuffer: 
+ */
+struct util_ringbuffer 
+{
+   struct util_packet *buf;
+   unsigned mask;
+
+   /* Can this be done with atomic variables??
+    */
+   unsigned head;
+   unsigned tail;
+   pipe_condvar change;
+   pipe_mutex mutex;
+};
+
+
+struct util_ringbuffer *util_ringbuffer_create( unsigned dwords )
+{
+   struct util_ringbuffer *ring = CALLOC_STRUCT(util_ringbuffer);
+   if (ring == NULL)
+      return NULL;
+
+   assert(util_is_power_of_two(dwords));
+   
+   ring->buf = MALLOC( dwords * sizeof(unsigned) );
+   if (ring->buf == NULL)
+      goto fail;
+
+   ring->mask = dwords - 1;
+
+   pipe_condvar_init(ring->change);
+   pipe_mutex_init(ring->mutex);
+   return ring;
+
+fail:
+   FREE(ring->buf);
+   FREE(ring);
+   return NULL;
+}
+
+void util_ringbuffer_destroy( struct util_ringbuffer *ring )
+{
+   pipe_condvar_destroy(ring->change);
+   pipe_mutex_destroy(ring->mutex);
+   FREE(ring->buf);
+   FREE(ring);
+}
+
+static INLINE unsigned util_ringbuffer_space( const struct util_ringbuffer *ring )
+{
+   return (ring->tail - (ring->head + 1)) & ring->mask;
+}
+
+void util_ringbuffer_enqueue( struct util_ringbuffer *ring,
+                              const struct util_packet *packet )
+{
+   unsigned i;
+
+   /* XXX: over-reliance on mutexes, etc:
+    */
+   pipe_mutex_lock(ring->mutex);
+
+   /* Wait for free space:
+    */
+   while (util_ringbuffer_space(ring) < packet->dwords)
+      pipe_condvar_wait(ring->change, ring->mutex);
+
+   /* Copy data to ring:
+    */
+   for (i = 0; i < packet->dwords; i++) {
+
+      /* Copy all dwords of the packet.  Note we're abusing the
+       * typesystem a little - we're being passed a pointer to
+       * something, but probably not an array of packet structs:
+       */
+      ring->buf[ring->head] = packet[i];
+      ring->head++;
+      ring->head &= ring->mask;
+   }
+
+   /* Signal change:
+    */
+   pipe_condvar_signal(ring->change);
+   pipe_mutex_unlock(ring->mutex);
+}
+
+enum pipe_error util_ringbuffer_dequeue( struct util_ringbuffer *ring,
+                                         struct util_packet *packet,
+                                         unsigned max_dwords,
+                                         boolean wait )
+{
+   const struct util_packet *ring_packet;
+   unsigned i;
+   int ret = PIPE_OK;
+
+   /* XXX: over-reliance on mutexes, etc:
+    */
+   pipe_mutex_lock(ring->mutex);
+
+   /* Wait for free space:
+    */
+   if (wait) {
+      while (util_ringbuffer_space(ring) == 0)
+         pipe_condvar_wait(ring->change, ring->mutex);
+   }
+   else {
+      if (util_ringbuffer_space(ring) == 0) {
+         ret = PIPE_ERROR_OUT_OF_MEMORY;
+         goto out;
+      }
+   }
+
+   ring_packet = &ring->buf[ring->tail];
+
+   /* Both of these are considered bugs.  Raise an assert on debug builds.
+    */
+   if (ring_packet->dwords > ring->mask + 1 - util_ringbuffer_space(ring) ||
+       ring_packet->dwords > max_dwords) {
+      assert(0);
+      ret = PIPE_ERROR_BAD_INPUT;
+      goto out;
+   }
+
+   /* Copy data from ring:
+    */
+   for (i = 0; i < ring_packet->dwords; i++) {
+      packet[i] = ring->buf[ring->tail];
+      ring->tail++;
+      ring->tail &= ring->mask;
+   }
+
+out:
+   /* Signal change:
+    */
+   pipe_condvar_signal(ring->change);
+   pipe_mutex_unlock(ring->mutex);
+   return ret;
+}
diff --git a/src/gallium/auxiliary/util/u_ringbuffer.h b/src/gallium/auxiliary/util/u_ringbuffer.h
new file mode 100644
index 00000000000..85f0ad6c1f6
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_ringbuffer.h
@@ -0,0 +1,29 @@
+
+#ifndef UTIL_RINGBUFFER_H
+#define UTIL_RINGBUFFER_H
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_defines.h"       /* only for pipe_error! */
+
+/* Generic header
+ */
+struct util_packet {
+   unsigned dwords:8;
+   unsigned data24:24;
+};
+
+struct util_ringbuffer;
+
+struct util_ringbuffer *util_ringbuffer_create( unsigned dwords );
+
+void util_ringbuffer_destroy( struct util_ringbuffer *ring );
+
+void util_ringbuffer_enqueue( struct util_ringbuffer *ring,
+                              const struct util_packet *packet );
+
+enum pipe_error util_ringbuffer_dequeue( struct util_ringbuffer *ring,
+                                         struct util_packet *packet,
+                                         unsigned max_dwords,
+                                         boolean wait );
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c
index 1ba82bb21f0..f9936eb1cb2 100644
--- a/src/gallium/auxiliary/util/u_tile.c
+++ b/src/gallium/auxiliary/util/u_tile.c
@@ -1357,7 +1357,10 @@ pipe_put_tile_rgba(struct pipe_transfer *pt,
       /*z24s8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
       break;
    default:
-      debug_printf("%s: unsupported format %s\n", __FUNCTION__, pf_name(format));
+      util_format_write_4f(format,
+                           p, src_stride * sizeof(float),
+                           packed, util_format_get_stride(format, w),
+                           0, 0, w, h);
    }
 
    pipe_put_tile_raw(pt, x, y, w, h, packed, 0);
diff --git a/src/gallium/auxiliary/vl/vl_compositor.c b/src/gallium/auxiliary/vl/vl_compositor.c
index fc2a1c59a6b..a524e2fdfb3 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.c
+++ b/src/gallium/auxiliary/vl/vl_compositor.c
@@ -316,6 +316,7 @@ init_buffers(struct vl_compositor *c)
    pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[0].buffer);
 
    c->vertex_elems[0].src_offset = 0;
+   c->vertex_elems[0].instance_divisor = 0;
    c->vertex_elems[0].vertex_buffer_index = 0;
    c->vertex_elems[0].nr_components = 2;
    c->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
@@ -345,6 +346,7 @@ init_buffers(struct vl_compositor *c)
    pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[1].buffer);
 
    c->vertex_elems[1].src_offset = 0;
+   c->vertex_elems[1].instance_divisor = 0;
    c->vertex_elems[1].vertex_buffer_index = 1;
    c->vertex_elems[1].nr_components = 2;
    c->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
@@ -353,7 +355,7 @@ init_buffers(struct vl_compositor *c)
     * Create our vertex shader's constant buffer
     * Const buffer contains scaling and translation vectors
     */
-   c->vs_const_buf.buffer = pipe_buffer_create
+   c->vs_const_buf = pipe_buffer_create
    (
       c->pipe->screen,
       1,
@@ -365,7 +367,7 @@ init_buffers(struct vl_compositor *c)
     * Create our fragment shader's constant buffer
     * Const buffer contains the color conversion matrix and bias vectors
     */
-   c->fs_const_buf.buffer = pipe_buffer_create
+   c->fs_const_buf = pipe_buffer_create
    (
       c->pipe->screen,
       1,
@@ -390,8 +392,8 @@ cleanup_buffers(struct vl_compositor *c)
    for (i = 0; i < 2; ++i)
       pipe_buffer_reference(&c->vertex_bufs[i].buffer, NULL);
 
-   pipe_buffer_reference(&c->vs_const_buf.buffer, NULL);
-   pipe_buffer_reference(&c->fs_const_buf.buffer, NULL);
+   pipe_buffer_reference(&c->vs_const_buf, NULL);
+   pipe_buffer_reference(&c->fs_const_buf, NULL);
 }
 
 bool vl_compositor_init(struct vl_compositor *compositor, struct pipe_context *pipe)
@@ -483,13 +485,13 @@ void vl_compositor_render(struct vl_compositor          *compositor,
    compositor->pipe->bind_fs_state(compositor->pipe, compositor->fragment_shader);
    compositor->pipe->set_vertex_buffers(compositor->pipe, 2, compositor->vertex_bufs);
    compositor->pipe->set_vertex_elements(compositor->pipe, 2, compositor->vertex_elems);
-   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_VERTEX, 0, &compositor->vs_const_buf);
-   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_FRAGMENT, 0, &compositor->fs_const_buf);
+   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_VERTEX, 0, compositor->vs_const_buf);
+   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_FRAGMENT, 0, compositor->fs_const_buf);
 
    vs_consts = pipe_buffer_map
    (
       compositor->pipe->screen,
-      compositor->vs_const_buf.buffer,
+      compositor->vs_const_buf,
       PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
    );
 
@@ -511,7 +513,7 @@ void vl_compositor_render(struct vl_compositor          *compositor,
    vs_consts->src_trans.z = 0;
    vs_consts->src_trans.w = 0;
 
-   pipe_buffer_unmap(compositor->pipe->screen, compositor->vs_const_buf.buffer);
+   pipe_buffer_unmap(compositor->pipe->screen, compositor->vs_const_buf);
 
    compositor->pipe->draw_arrays(compositor->pipe, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
    compositor->pipe->flush(compositor->pipe, PIPE_FLUSH_RENDER_CACHE, fence);
@@ -525,10 +527,10 @@ void vl_compositor_set_csc_matrix(struct vl_compositor *compositor, const float
 
    memcpy
    (
-      pipe_buffer_map(compositor->pipe->screen, compositor->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      pipe_buffer_map(compositor->pipe->screen, compositor->fs_const_buf, PIPE_BUFFER_USAGE_CPU_WRITE),
       mat,
       sizeof(struct fragment_shader_consts)
    );
 
-   pipe_buffer_unmap(compositor->pipe->screen, compositor->fs_const_buf.buffer);
+   pipe_buffer_unmap(compositor->pipe->screen, compositor->fs_const_buf);
 }
diff --git a/src/gallium/auxiliary/vl/vl_compositor.h b/src/gallium/auxiliary/vl/vl_compositor.h
index f441901a751..6a9a3fd7af1 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.h
+++ b/src/gallium/auxiliary/vl/vl_compositor.h
@@ -47,7 +47,7 @@ struct vl_compositor
    struct pipe_scissor_state scissor;
    struct pipe_vertex_buffer vertex_bufs[2];
    struct pipe_vertex_element vertex_elems[2];
-   struct pipe_constant_buffer vs_const_buf, fs_const_buf;
+   struct pipe_buffer *vs_const_buf, *fs_const_buf;
 };
 
 bool vl_compositor_init(struct vl_compositor *compositor, struct pipe_context *pipe);
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
index caf581aca60..e43187545c5 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -891,53 +891,61 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
 
    /* Position element */
    r->vertex_elems[0].src_offset = 0;
+   r->vertex_elems[0].instance_divisor = 0;
    r->vertex_elems[0].vertex_buffer_index = 0;
    r->vertex_elems[0].nr_components = 2;
    r->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* Luma, texcoord element */
    r->vertex_elems[1].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[1].instance_divisor = 0;
    r->vertex_elems[1].vertex_buffer_index = 0;
    r->vertex_elems[1].nr_components = 2;
    r->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* Chroma Cr texcoord element */
    r->vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
+   r->vertex_elems[2].instance_divisor = 0;
    r->vertex_elems[2].vertex_buffer_index = 0;
    r->vertex_elems[2].nr_components = 2;
    r->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* Chroma Cb texcoord element */
    r->vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
+   r->vertex_elems[3].instance_divisor = 0;
    r->vertex_elems[3].vertex_buffer_index = 0;
    r->vertex_elems[3].nr_components = 2;
    r->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* First ref surface top field texcoord element */
    r->vertex_elems[4].src_offset = 0;
+   r->vertex_elems[4].instance_divisor = 0;
    r->vertex_elems[4].vertex_buffer_index = 1;
    r->vertex_elems[4].nr_components = 2;
    r->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* First ref surface bottom field texcoord element */
    r->vertex_elems[5].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[5].instance_divisor = 0;
    r->vertex_elems[5].vertex_buffer_index = 1;
    r->vertex_elems[5].nr_components = 2;
    r->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* Second ref surface top field texcoord element */
    r->vertex_elems[6].src_offset = 0;
+   r->vertex_elems[6].instance_divisor = 0;
    r->vertex_elems[6].vertex_buffer_index = 2;
    r->vertex_elems[6].nr_components = 2;
    r->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* Second ref surface bottom field texcoord element */
    r->vertex_elems[7].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[7].instance_divisor = 0;
    r->vertex_elems[7].vertex_buffer_index = 2;
    r->vertex_elems[7].nr_components = 2;
    r->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
 
-   r->vs_const_buf.buffer = pipe_buffer_create
+   r->vs_const_buf = pipe_buffer_create
    (
       r->pipe->screen,
       DEFAULT_BUF_ALIGNMENT,
@@ -945,7 +953,7 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
       sizeof(struct vertex_shader_consts)
    );
 
-   r->fs_const_buf.buffer = pipe_buffer_create
+   r->fs_const_buf = pipe_buffer_create
    (
       r->pipe->screen,
       DEFAULT_BUF_ALIGNMENT,
@@ -954,11 +962,11 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
 
    memcpy
    (
-      pipe_buffer_map(r->pipe->screen, r->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      pipe_buffer_map(r->pipe->screen, r->fs_const_buf, PIPE_BUFFER_USAGE_CPU_WRITE),
       &fs_consts, sizeof(struct fragment_shader_consts)
    );
 
-   pipe_buffer_unmap(r->pipe->screen, r->fs_const_buf.buffer);
+   pipe_buffer_unmap(r->pipe->screen, r->fs_const_buf);
 
    return true;
 }
@@ -970,8 +978,8 @@ cleanup_buffers(struct vl_mpeg12_mc_renderer *r)
 
    assert(r);
 
-   pipe_buffer_reference(&r->vs_const_buf.buffer, NULL);
-   pipe_buffer_reference(&r->fs_const_buf.buffer, NULL);
+   pipe_buffer_reference(&r->vs_const_buf, NULL);
+   pipe_buffer_reference(&r->fs_const_buf, NULL);
 
    for (i = 0; i < 3; ++i)
       pipe_buffer_reference(&r->vertex_bufs.all[i].buffer, NULL);
@@ -1284,19 +1292,19 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    vs_consts = pipe_buffer_map
    (
-      r->pipe->screen, r->vs_const_buf.buffer,
+      r->pipe->screen, r->vs_const_buf,
       PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
    );
 
    vs_consts->denorm.x = r->surface->width0;
    vs_consts->denorm.y = r->surface->height0;
 
-   pipe_buffer_unmap(r->pipe->screen, r->vs_const_buf.buffer);
+   pipe_buffer_unmap(r->pipe->screen, r->vs_const_buf);
 
    r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_VERTEX, 0,
-                                &r->vs_const_buf);
+                                r->vs_const_buf);
    r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_FRAGMENT, 0,
-                                &r->fs_const_buf);
+                                r->fs_const_buf);
 
    if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
index 64184337a06..f00b8c7b8b1 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -63,8 +63,8 @@ struct vl_mpeg12_mc_renderer
 
    struct pipe_viewport_state viewport;
    struct pipe_scissor_state scissor;
-   struct pipe_constant_buffer vs_const_buf;
-   struct pipe_constant_buffer fs_const_buf;
+   struct pipe_buffer *vs_const_buf;
+   struct pipe_buffer *fs_const_buf;
    struct pipe_framebuffer_state fb_state;
    struct pipe_vertex_element vertex_elems[8];
 	
diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
index 21f5f9111a0..d394f5b4f1b 100644
--- a/src/gallium/docs/source/context.rst
+++ b/src/gallium/docs/source/context.rst
@@ -33,7 +33,11 @@ This state describes how resources in various flavours (textures,
 buffers, surfaces) are bound to the driver.
 
 
-* ``set_constant_buffer``
+* ``set_constant_buffer`` sets a constant buffer to be used for a given shader
+  type. index is used to indicate which buffer to set (some apis may allow
+  multiple ones to be set, and binding a specific one later, though drivers
+  are mostly restricted to the first one right now).
+
 * ``set_framebuffer_state``
 * ``set_fragment_sampler_textures``
 * ``set_vertex_sampler_textures``
@@ -47,7 +51,6 @@ These pieces of state are too small, variable, and/or trivial to have CSO
 objects. They all follow simple, one-method binding calls, e.g.
 ``set_edgeflags``.
 
-* ``set_edgeflags``
 * ``set_blend_color``
 * ``set_clip_state``
 * ``set_polygon_stipple``
@@ -72,12 +75,67 @@ stencil-only clears of packed depth-stencil buffers.
 Drawing
 ^^^^^^^
 
-``draw_arrays``
+``draw_arrays`` draws a specified primitive.
+
+This command is equivalent to calling ``draw_arrays_instanced``
+with ``startInstance`` set to 0 and ``instanceCount`` set to 1.
+
+``draw_elements`` draws a specified primitive using an optional
+index buffer.
 
-``draw_elements``
+This command is equivalent to calling ``draw_elements_instanced``
+with ``startInstance`` set to 0 and ``instanceCount`` set to 1.
 
 ``draw_range_elements``
 
+XXX: this is (probably) a temporary entrypoint, as the range
+information should be available from the vertex_buffer state.
+Using this to quickly evaluate a specialized path in the draw
+module.
+
+``draw_arrays_instanced`` draws multiple instances of the same primitive.
+
+This command is equivalent to calling ``draw_elements_instanced``
+with ``indexBuffer`` set to NULL and ``indexSize`` set to 0.
+
+``draw_elements_instanced`` draws multiple instances of the same primitive
+using an optional index buffer.
+
+For instanceID in the range between ``startInstance``
+and ``startInstance``+``instanceCount``-1, inclusive, draw a primitive
+specified by ``mode`` and sequential numbers in the range between ``start``
+and ``start``+``count``-1, inclusive.
+
+If ``indexBuffer`` is not NULL, it specifies an index buffer with index
+byte size of ``indexSize``. The sequential numbers are used to lookup
+the index buffer and the resulting indices in turn are used to fetch
+vertex attributes.
+
+If ``indexBuffer`` is NULL, the sequential numbers are used directly
+as indices to fetch vertex attributes.
+
+If a given vertex element has ``instance_divisor`` set to 0, it is said
+it contains per-vertex data and effective vertex attribute address needs
+to be recalculated for every index.
+
+  attribAddr = ``stride`` * index + ``src_offset``
+
+If a given vertex element has ``instance_divisor`` set to non-zero,
+it is said it contains per-instance data and effective vertex attribute
+address needs to recalculated for every ``instance_divisor``-th instance.
+
+  attribAddr = ``stride`` * instanceID / ``instance_divisor`` + ``src_offset``
+
+In the above formulas, ``src_offset`` is taken from the given vertex element
+and ``stride`` is taken from a vertex buffer associated with the given
+vertex element.
+
+The calculated attribAddr is used as an offset into the vertex buffer to
+fetch the attribute data.
+
+The value of ``instanceID`` can be read in a vertex shader through a system
+value register declared with INSTANCEID semantic name.
+
 
 Queries
 ^^^^^^^
diff --git a/src/gallium/docs/source/distro.rst b/src/gallium/docs/source/distro.rst
index 33e846e33d2..a448203c6fa 100644
--- a/src/gallium/docs/source/distro.rst
+++ b/src/gallium/docs/source/distro.rst
@@ -118,6 +118,9 @@ Gallivm
 Indices
 ^^^^^^^
 
+Tool for translating or generating element indices for element-based
+rendering.
+
 Pipe Buffer Manager
 ^^^^^^^^^^^^^^^^^^^
 
@@ -127,12 +130,11 @@ Remote Debugger
 Runtime Assembly Emission
 ^^^^^^^^^^^^^^^^^^^^^^^^^
 
-Surface Context Tracker
-^^^^^^^^^^^^^^^^^^^^^^^
-
 TGSI
 ^^^^
 
+Basic utilities for manipulating TGSI streams.
+
 Translate
 ^^^^^^^^^
 
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index d5f5c7bbba8..aa29dcb3947 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -358,6 +358,7 @@ struct cell_spu_function_info
 
 
 /** This is the object passed to spe_create_thread() */
+PIPE_ALIGN_TYPE(16,
 struct cell_init_info
 {
    unsigned id;
@@ -370,7 +371,7 @@ struct cell_init_info
    uint *buffer_status;  /**< points at cell_context->buffer_status */
 
    struct cell_spu_function_info *spu_functions;
-} ALIGN16_ATTRIB;
+});
 
 
 #endif /* CELL_COMMON_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index 5c3188e7f9d..e402ed29220 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -89,7 +89,7 @@ struct cell_buffer_node;
  */
 struct cell_buffer_list
 {
-   struct cell_fence fence ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) struct cell_fence fence;
    struct cell_buffer_node *head;
 };
 
@@ -115,7 +115,7 @@ struct cell_context
 
    struct pipe_blend_color blend_color;
    struct pipe_clip_state clip;
-   struct pipe_constant_buffer constants[2];
+   struct pipe_buffer *constants[2];
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
@@ -150,18 +150,18 @@ struct cell_context
    /** Mapped constant buffers */
    void *mapped_constants[PIPE_SHADER_TYPES];
 
-   struct cell_spu_function_info spu_functions ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) struct cell_spu_function_info spu_functions;
 
    uint num_cells, num_spus;
 
    /** Buffers for command batches, vertex/index data */
    uint buffer_size[CELL_NUM_BUFFERS];
-   ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE] ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE];
 
    int cur_batch;  /**< which buffer is being filled w/ commands */
 
    /** [4] to ensure 16-byte alignment for each status word */
-   uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4] ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4];
 
 
    /** Associated with each command/batch buffer is a list of pipe_buffers
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index 3fa8b975d39..0a4da8ecc85 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -51,17 +51,17 @@ cell_map_constant_buffers(struct cell_context *sp)
    struct pipe_winsys *ws = sp->pipe.winsys;
    uint i;
    for (i = 0; i < 2; i++) {
-      if (sp->constants[i].buffer && sp->constants[i].buffer->size) {
-         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer,
+      if (sp->constants[i] && sp->constants[i]->size) {
+         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i],
                                                    PIPE_BUFFER_USAGE_CPU_READ);
          cell_flush_buffer_range(sp, sp->mapped_constants[i], 
-                                 sp->constants[i].buffer->size);
+                                 sp->constants[i]->size);
       }
    }
 
    draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX,
                                    sp->mapped_constants[PIPE_SHADER_VERTEX],
-                                   sp->constants[PIPE_SHADER_VERTEX].buffer->size);
+                                   sp->constants[PIPE_SHADER_VERTEX]->size);
 }
 
 static void
@@ -70,8 +70,8 @@ cell_unmap_constant_buffers(struct cell_context *sp)
    struct pipe_winsys *ws = sp->pipe.winsys;
    uint i;
    for (i = 0; i < 2; i++) {
-      if (sp->constants[i].buffer && sp->constants[i].buffer->size)
-         ws->buffer_unmap(ws, sp->constants[i].buffer);
+      if (sp->constants[i] && sp->constants[i]->size)
+         ws->buffer_unmap(ws, sp->constants[i]);
       sp->mapped_constants[i] = NULL;
    }
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 5b87286d4c5..f1e1dcb9eb0 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -240,12 +240,12 @@ cell_emit_state(struct cell_context *cell)
 
    if (cell->dirty & (CELL_NEW_FS_CONSTANTS)) {
       const uint shader = PIPE_SHADER_FRAGMENT;
-      const uint num_const = cell->constants[shader].buffer->size / sizeof(float);
+      const uint num_const = cell->constants[shader]->size / sizeof(float);
       uint i, j;
       float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float)));
       uint32_t *ibuf = (uint32_t *) buf;
       const float *constants = pipe_buffer_map(cell->pipe.screen,
-                                               cell->constants[shader].buffer,
+                                               cell->constants[shader],
                                                PIPE_BUFFER_USAGE_CPU_READ);
       ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS;
       ibuf[4] = num_const;
@@ -253,7 +253,7 @@ cell_emit_state(struct cell_context *cell)
       for (i = 0; i < num_const; i++) {
          buf[j++] = constants[i];
       }
-      pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader].buffer);
+      pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader]);
    }
 
    if (cell->dirty & (CELL_NEW_FRAMEBUFFER |
diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index 6568c784fec..1b09cf7f7d7 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -183,7 +183,7 @@ cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 static void
 cell_set_constant_buffer(struct pipe_context *pipe,
                          uint shader, uint index,
-                         const struct pipe_constant_buffer *buf)
+                         struct pipe_buffer *buf)
 {
    struct cell_context *cell = cell_context(pipe);
 
@@ -193,7 +193,7 @@ cell_set_constant_buffer(struct pipe_context *pipe,
    draw_flush(cell->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&cell->constants[shader].buffer, buf->buffer);
+   pipe_buffer_reference(&cell->constants[shader], buf);
 
    if (shader == PIPE_SHADER_VERTEX)
       cell->dirty |= CELL_NEW_VS_CONSTANTS;
diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c
index 12b855a3db2..55bd85bde2b 100644
--- a/src/gallium/drivers/cell/spu/spu_command.c
+++ b/src/gallium/drivers/cell/spu/spu_command.c
@@ -53,8 +53,7 @@ struct spu_vs_context draw;
 /**
  * Buffers containing dynamically generated SPU code:
  */
-static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS]
-    ALIGN16_ATTRIB;
+PIPE_ALIGN_VAR(16) static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS];
 
 
 
@@ -543,7 +542,7 @@ cmd_batch(uint opcode)
 {
    const uint buf = (opcode >> 8) & 0xff;
    uint size = (opcode >> 16);
-   qword buffer[CELL_BUFFER_SIZE / 16] ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) qword buffer[CELL_BUFFER_SIZE / 16];
    const unsigned usize = ROUNDUP16(size) / sizeof(buffer[0]);
    uint pos;
 
diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c
index d86d8e09a51..d2166a49016 100644
--- a/src/gallium/drivers/cell/spu/spu_exec.c
+++ b/src/gallium/drivers/cell/spu/spu_exec.c
@@ -1839,10 +1839,11 @@ spu_exec_machine_run( struct spu_exec_machine *mach )
    /* execute declarations (interpolants) */
    if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
       for (i = 0; i < mach->NumDeclarations; i++) {
+         PIPE_ALIGN_VAR(16)
          union {
             struct tgsi_full_declaration decl;
             qword buffer[ROUNDUP16(sizeof(struct tgsi_full_declaration)) / 16];
-         } d ALIGN16_ATTRIB;
+         } d;
          unsigned ea = (unsigned) (mach->Declarations + pc);
 
          spu_dcache_fetch_unaligned(d.buffer, ea, sizeof(d.decl));
@@ -1853,10 +1854,11 @@ spu_exec_machine_run( struct spu_exec_machine *mach )
 
    /* execute instructions, until pc is set to -1 */
    while (pc != -1) {
+      PIPE_ALIGN_VAR(16)
       union {
          struct tgsi_full_instruction inst;
          qword buffer[ROUNDUP16(sizeof(struct tgsi_full_instruction)) / 16];
-      } i ALIGN16_ATTRIB;
+      } i;
       unsigned ea = (unsigned) (mach->Instructions + pc);
 
       spu_dcache_fetch_unaligned(i.buffer, ea, sizeof(i.inst));
diff --git a/src/gallium/drivers/cell/spu/spu_exec.h b/src/gallium/drivers/cell/spu/spu_exec.h
index 86056799405..0ca92af248d 100644
--- a/src/gallium/drivers/cell/spu/spu_exec.h
+++ b/src/gallium/drivers/cell/spu/spu_exec.h
@@ -98,9 +98,9 @@ struct spu_exec_machine
     * 4  internal temporaries
     * 1  address
     */
+   PIPE_ALIGN_VAR(16)
    struct spu_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS 
-                                      + TGSI_EXEC_NUM_TEMP_EXTRAS + 1]
-       ALIGN16_ATTRIB;
+                                      + TGSI_EXEC_NUM_TEMP_EXTRAS + 1];
 
    struct spu_exec_vector       *Addrs;
 
diff --git a/src/gallium/drivers/cell/spu/spu_funcs.c b/src/gallium/drivers/cell/spu/spu_funcs.c
index ff3d609d258..98919c43ffc 100644
--- a/src/gallium/drivers/cell/spu/spu_funcs.c
+++ b/src/gallium/drivers/cell/spu/spu_funcs.c
@@ -144,7 +144,7 @@ export_func(struct cell_spu_function_info *spu_functions,
 void
 return_function_info(void)
 {
-   struct cell_spu_function_info funcs ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) struct cell_spu_function_info funcs;
    int tag = TAG_MISC;
 
    ASSERT(sizeof(funcs) == 256); /* must be multiple of 16 bytes */
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 33767e7c51d..b18f4c22ef1 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -93,6 +93,7 @@ typedef vector unsigned int (*spu_fragment_program_func)(vector float *inputs,
                                                          vector float *constants);
 
 
+PIPE_ALIGN_TYPE(16,
 struct spu_framebuffer
 {
    void *color_start;              /**< addr of color surface in main memory */
@@ -107,10 +108,11 @@ struct spu_framebuffer
 
    uint zsize;                     /**< 0, 2 or 4 bytes per Z */
    float zscale;                   /**< 65535.0, 2^24-1 or 2^32-1 */
-} ALIGN16_ATTRIB;
+});
 
 
 /** per-texture level info */
+PIPE_ALIGN_TYPE(16,
 struct spu_texture_level
 {
    void *start;
@@ -123,20 +125,22 @@ struct spu_texture_level
    vector signed int mask_s, mask_t, mask_r;
    /** texcoord clamp limits */
    vector signed int max_s, max_t, max_r;
-} ALIGN16_ATTRIB;
+});
 
 
+PIPE_ALIGN_TYPE(16,
 struct spu_texture
 {
    struct spu_texture_level level[CELL_MAX_TEXTURE_LEVELS];
    uint max_level;
    uint target;  /**< PIPE_TEXTURE_x */
-} ALIGN16_ATTRIB;
+});
 
 
 /**
  * All SPU global/context state will be in a singleton object of this type:
  */
+PIPE_ALIGN_TYPE(16,
 struct spu_global
 {
    /** One-time init/constant info */
@@ -155,8 +159,8 @@ struct spu_global
    struct vertex_info vertex_info;
 
    /** Current color and Z tiles */
-   tile_t ctile ALIGN16_ATTRIB;
-   tile_t ztile ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) tile_t ctile;
+   PIPE_ALIGN_VAR(16) tile_t ztile;
 
    /** Read depth/stencil tiles? */
    boolean read_depth_stencil;
@@ -165,8 +169,8 @@ struct spu_global
    ubyte cur_ctile_status, cur_ztile_status;
 
    /** Status of all tiles in framebuffer */
-   ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
-   ubyte ztile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE];
+   PIPE_ALIGN_VAR(16) ubyte ztile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE];
 
    /** Current fragment ops machine code, at 8-byte boundary */
    uint *fragment_ops_code;
@@ -175,7 +179,7 @@ struct spu_global
    spu_fragment_ops_func fragment_ops[2];
 
    /** Current fragment program machine code, at 8-byte boundary */
-   uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS] ALIGN8_ATTRIB;
+   PIPE_ALIGN_VAR(8) uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS];
    /** Current fragment ops function */
    spu_fragment_program_func fragment_program;
 
@@ -187,7 +191,7 @@ struct spu_global
    /** Fragment program constants */
    vector float constants[4 * CELL_MAX_CONSTANTS];
 
-} ALIGN16_ATTRIB;
+});
 
 
 extern struct spu_global spu;
diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c
index 5ffb7073abf..14987e3c3a2 100644
--- a/src/gallium/drivers/cell/spu/spu_render.c
+++ b/src/gallium/drivers/cell/spu/spu_render.c
@@ -169,7 +169,7 @@ void
 cmd_render(const struct cell_command_render *render, uint *pos_incr)
 {
    /* we'll DMA into these buffers */
-   ubyte vertex_data[CELL_BUFFER_SIZE] ALIGN16_ATTRIB;
+   PIPE_ALIGN_VAR(16) ubyte vertex_data[CELL_BUFFER_SIZE];
    const uint vertex_size = render->vertex_size; /* in bytes */
    /*const*/ uint total_vertex_bytes = render->num_verts * vertex_size;
    uint index_bytes;
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
index 03375d84a57..087963960df 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
@@ -43,7 +43,8 @@ typedef void (*spu_fetch_func)(qword *out, const qword *in,
 			       const qword *shuffle_data);
 
 
-static const qword fetch_shuffle_data[5] ALIGN16_ATTRIB = {
+PIPE_ALIGN_VAR(16) static const qword
+fetch_shuffle_data[5] = {
    /* Shuffle used by CVT_64_FLOAT
     */
    {
@@ -110,7 +111,7 @@ static void generic_vertex_fetch(struct spu_vs_context *draw,
       unsigned idx;
       const unsigned bytes_per_entry = draw->vertex_fetch.size[attr];
       const unsigned quads_per_entry = (bytes_per_entry + 15) / 16;
-      qword in[2 * 4] ALIGN16_ATTRIB;
+      PIPE_ALIGN_VAR(16) qword in[2 * 4];
 
 
       /* Fetch four attributes for four vertices.  
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.c b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
index fbe5b34d397..3e9804bf8ee 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_shader.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
@@ -107,8 +107,8 @@ run_vertex_program(struct spu_vs_context *draw,
    struct spu_exec_machine *machine = &draw->machine;
    unsigned int j;
 
-   ALIGN16_DECL(struct spu_exec_vector, inputs, PIPE_MAX_ATTRIBS);
-   ALIGN16_DECL(struct spu_exec_vector, outputs, PIPE_MAX_ATTRIBS);
+   PIPE_ALIGN_VAR(16) struct spu_exec_vector inputs[PIPE_MAX_ATTRIBS];
+   PIPE_ALIGN_VAR(16) struct spu_exec_vector outputs[PIPE_MAX_ATTRIBS];
    const float *scale = draw->viewport.scale;
    const float *trans = draw->viewport.translate;
 
@@ -119,8 +119,8 @@ run_vertex_program(struct spu_vs_context *draw,
    ASSERT_ALIGN16(draw->constants);
    machine->Consts = (float (*)[4]) draw->constants;
 
-   machine->Inputs = ALIGN16_ASSIGN(inputs);
-   machine->Outputs = ALIGN16_ASSIGN(outputs);
+   machine->Inputs = inputs;
+   machine->Outputs = outputs;
 
    spu_vertex_fetch( draw, machine, elts, count );
 
@@ -132,8 +132,9 @@ run_vertex_program(struct spu_vs_context *draw,
    for (j = 0; j < count; j++) {
       unsigned slot;
       float x, y, z, w;
+      PIPE_ALIGN_VAR(16)
       unsigned char buffer[sizeof(struct vertex_header)
-          + MAX_VERTEX_SIZE] ALIGN16_ATTRIB;
+          + MAX_VERTEX_SIZE];
       struct vertex_header *const tmpOut =
           (struct vertex_header *) buffer;
       const unsigned vert_size = ROUNDUP16(sizeof(struct vertex_header)
@@ -186,8 +187,8 @@ run_vertex_program(struct spu_vs_context *draw,
 }
 
 
-unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]
-    ALIGN16_ATTRIB;
+PIPE_ALIGN_VAR(16) unsigned char
+immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]);
 
 
 void
diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h
index 149393712a3..191a44c3dfc 100644
--- a/src/gallium/drivers/failover/fo_context.h
+++ b/src/gallium/drivers/failover/fo_context.h
@@ -125,7 +125,7 @@ failover_context( struct pipe_context *pipe )
 void
 failover_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
-                             const struct pipe_constant_buffer *buf);
+                             struct pipe_buffer *buf);
 
 
 #endif /* FO_CONTEXT_H */
diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c
index 3f5f5560323..d6ec4d13136 100644
--- a/src/gallium/drivers/failover/fo_state.c
+++ b/src/gallium/drivers/failover/fo_state.c
@@ -495,7 +495,7 @@ failover_set_vertex_elements(struct pipe_context *pipe,
 void
 failover_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
-                             const struct pipe_constant_buffer *buf)
+                             struct pipe_buffer *buf)
 {
    struct failover_context *failover = failover_context(pipe);
 
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index 234b441ce6e..37cbd56036b 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -233,7 +233,8 @@ struct i915_context
 
    struct pipe_blend_color blend_color;
    struct pipe_clip_state clip;
-   struct pipe_constant_buffer constants[PIPE_SHADER_TYPES];
+   /* XXX unneded */
+   struct pipe_buffer *constants[PIPE_SHADER_TYPES];
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index 5f5b6f8e185..0fab6e1bc36 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -518,7 +518,7 @@ static void i915_delete_vs_state(struct pipe_context *pipe, void *shader)
 
 static void i915_set_constant_buffer(struct pipe_context *pipe,
                                      uint shader, uint index,
-                                     const struct pipe_constant_buffer *buf)
+                                     struct pipe_buffer *buf)
 {
    struct i915_context *i915 = i915_context(pipe);
    struct pipe_screen *screen = pipe->screen;
@@ -538,13 +538,13 @@ static void i915_set_constant_buffer(struct pipe_context *pipe,
     */
    if (buf) {
       void *mapped;
-      if (buf->buffer && buf->buffer->size &&
-          (mapped = pipe_buffer_map(screen, buf->buffer,
+      if (buf->size &&
+          (mapped = pipe_buffer_map(screen, buf,
                                     PIPE_BUFFER_USAGE_CPU_READ))) {
-         memcpy(i915->current.constants[shader], mapped, buf->buffer->size);
-         pipe_buffer_unmap(screen, buf->buffer);
+         memcpy(i915->current.constants[shader], mapped, buf->size);
+         pipe_buffer_unmap(screen, buf);
          i915->current.num_user_constants[shader]
-            = buf->buffer->size / (4 * sizeof(float));
+            = buf->size / (4 * sizeof(float));
       }
       else {
          i915->current.num_user_constants[shader] = 0;
diff --git a/src/gallium/drivers/i965/brw_pipe_shader.c b/src/gallium/drivers/i965/brw_pipe_shader.c
index bb32d90e331..e389587f3e1 100644
--- a/src/gallium/drivers/i965/brw_pipe_shader.c
+++ b/src/gallium/drivers/i965/brw_pipe_shader.c
@@ -262,7 +262,7 @@ static void brw_delete_vs_state( struct pipe_context *pipe, void *prog )
 
 static void brw_set_constant_buffer(struct pipe_context *pipe,
                                      uint shader, uint index,
-                                     const struct pipe_constant_buffer *buf)
+                                     struct pipe_buffer *buf)
 {
    struct brw_context *brw = brw_context(pipe);
 
@@ -270,13 +270,13 @@ static void brw_set_constant_buffer(struct pipe_context *pipe,
 
    if (shader == PIPE_SHADER_FRAGMENT) {
       pipe_buffer_reference( &brw->curr.fragment_constants,
-                             buf->buffer );
+                             buf );
 
       brw->state.dirty.mesa |= PIPE_NEW_FRAGMENT_CONSTANTS;
    }
    else {
       pipe_buffer_reference( &brw->curr.vertex_constants,
-                             buf->buffer );
+                             buf );
 
       brw->state.dirty.mesa |= PIPE_NEW_VERTEX_CONSTANTS;
    }
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index 9f5b4e63236..f9063d90fb1 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -404,17 +404,17 @@ static void
 identity_set_constant_buffer(struct pipe_context *_pipe,
                              uint shader,
                              uint index,
-                             const struct pipe_constant_buffer *_buffer)
+                             struct pipe_buffer *_buffer)
 {
    struct identity_context *id_pipe = identity_context(_pipe);
    struct pipe_context *pipe = id_pipe->pipe;
-   struct pipe_constant_buffer unwrapped_buffer;
-   struct pipe_constant_buffer *buffer = NULL;
+   struct pipe_buffer *unwrapped_buffer;
+   struct pipe_buffer *buffer = NULL;
 
-   /* unwrap the input state */
+   /* XXX hmm? unwrap the input state */
    if (_buffer) {
-      unwrapped_buffer.buffer = identity_buffer_unwrap(_buffer->buffer);
-      buffer = &unwrapped_buffer;
+      unwrapped_buffer = identity_buffer_unwrap(_buffer);
+      buffer = unwrapped_buffer;
    }
 
    pipe->set_constant_buffer(pipe,
diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README
index 0c3f00fd58f..72d9f39658f 100644
--- a/src/gallium/drivers/llvmpipe/README
+++ b/src/gallium/drivers/llvmpipe/README
@@ -59,27 +59,16 @@ Requirements
    
    See /proc/cpuinfo to know what your CPU supports.
  
- - LLVM 2.5 or greater. LLVM 2.6 is preferred.
+ - LLVM 2.6.
  
-   On Debian based distributions do:
+   For Linux, on a recent Debian based distribution do:
  
      aptitude install llvm-dev
 
-   There is a typo in one of the llvm 2.5 headers, that may cause compilation
-   errors. To fix it apply the change:
-
-     --- /usr/include/llvm-c/Core.h.orig	2009-08-10 15:38:54.000000000 +0100
-     +++ /usr/include/llvm-c/Core.h	2009-08-10 15:38:25.000000000 +0100
-     @@ -831,7 +831,7 @@
-        template<typename T>
-        inline T **unwrap(LLVMValueRef *Vals, unsigned Length) {
-          #if DEBUG
-     -    for (LLVMValueRef *I = Vals, E = Vals + Length; I != E; ++I)
-     +    for (LLVMValueRef *I = Vals, *E = Vals + Length; I != E; ++I)
-            cast<T>(*I);
-          #endif
-          return reinterpret_cast<T**>(Vals);
- 
+   For Windows download pre-built MSVC 9.0 or MinGW binaries from
+   http://people.freedesktop.org/~jrfonseca/llvm/ and set the LLVM environment
+   variable to the extracted path.
+
  - scons (optional)
 
  - udis86, http://udis86.sourceforge.net/ (optional):
@@ -95,9 +84,9 @@ Requirements
 Building
 ========
 
-To build everything invoke scons as:
+To build everything on Linux invoke scons as:
 
-  scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false -k
+  scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=xlib dri=false
 
 Alternatively, you can build it with GNU make, if you prefer, by invoking it as
 
@@ -105,12 +94,15 @@ Alternatively, you can build it with GNU make, if you prefer, by invoking it as
 
 but the rest of these instructions assume that scons is used.
 
+For windows is everything the except except the winsys:
+
+  scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=gdi dri=false
 
 Using
 =====
 
-Building will create a drop-in alternative for libGL.so. To use it set the
-environment variables:
+On Linux, building will create a drop-in alternative for libGL.so. To use it
+set the environment variables:
 
   export LD_LIBRARY_PATH=$PWD/build/linux-x86_64-debug/lib:$LD_LIBRARY_PATH
 
@@ -121,6 +113,11 @@ or
 For performance evaluation pass debug=no to scons, and use the corresponding
 lib directory without the "-debug" suffix.
 
+On Windows, building will create a drop-in alternative for opengl32.dll. To use
+it put it in the same directory as the application. It can also be used by
+replacing the native ICD driver, but it's quite an advanced usage, so if you
+need to ask, don't even try it.
+
 
 Unit testing
 ============
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp b/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp
index d3f78c06d92..6e79438ead0 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp
+++ b/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp
@@ -59,3 +59,17 @@ LLVMInitializeNativeTarget(void)
 
 
 #endif
+
+
+/* 
+ * Hack to allow the linking of release LLVM static libraries on a debug build.
+ *
+ * See also:
+ * - http://social.msdn.microsoft.com/Forums/en-US/vclanguage/thread/7234ea2b-0042-42ed-b4e2-5d8644dfb57d
+ */
+#if defined(_MSC_VER) && defined(_DEBUG)
+#include <crtdefs.h>
+extern "C" {
+   _CRTIMP void __cdecl _invalid_parameter_noinfo(void) {}
+}
+#endif
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index bd549d4028b..c5b00f8e23f 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -74,8 +74,8 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
    }
 
    for (i = 0; i < Elements(llvmpipe->constants); i++) {
-      if (llvmpipe->constants[i].buffer) {
-         pipe_buffer_reference(&llvmpipe->constants[i].buffer, NULL);
+      if (llvmpipe->constants[i]) {
+         pipe_buffer_reference(&llvmpipe->constants[i], NULL);
       }
    }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 5390a713699..3af7b62a53b 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -62,7 +62,7 @@ struct llvmpipe_context {
    /** Other rendering state */
    struct pipe_blend_color blend_color;
    struct pipe_clip_state clip;
-   struct pipe_constant_buffer constants[PIPE_SHADER_TYPES];
+   struct pipe_buffer *constants[PIPE_SHADER_TYPES];
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h
index 21ebfa7ca91..e4c56f153fc 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast.h
@@ -86,7 +86,7 @@ struct lp_rast_shader_inputs {
    float (*dady)[4];
 
    /* edge/step info for 3 edges and 4x4 block of pixels */
-   int ALIGN16_ATTRIB step[3][16];
+   PIPE_ALIGN_VAR(16) int step[3][16];
 };
 
 
@@ -121,7 +121,7 @@ struct lp_rast_triangle {
    int c1, c2, c3;
 
    /* inputs for the shader */
-   struct lp_rast_shader_inputs ALIGN16_ATTRIB inputs;
+   PIPE_ALIGN_VAR(16) struct lp_rast_shader_inputs inputs;
 };
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index 224b6e523c3..4242653c909 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -162,7 +162,7 @@ void llvmpipe_set_clip_state( struct pipe_context *,
 
 void llvmpipe_set_constant_buffer(struct pipe_context *,
                                   uint shader, uint index,
-                                  const struct pipe_constant_buffer *buf);
+                                  struct pipe_buffer *buf);
 
 void *llvmpipe_create_fs_state(struct pipe_context *,
                                const struct pipe_shader_state *);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_derived.c b/src/gallium/drivers/llvmpipe/lp_state_derived.c
index 28af477914c..bdd906e1a73 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_derived.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_derived.c
@@ -180,7 +180,7 @@ void llvmpipe_update_derived( struct llvmpipe_context *llvmpipe )
 
    if (llvmpipe->dirty & LP_NEW_CONSTANTS)
       lp_setup_set_fs_constants(llvmpipe->setup, 
-                                llvmpipe->constants[PIPE_SHADER_FRAGMENT].buffer);
+                                llvmpipe->constants[PIPE_SHADER_FRAGMENT]);
 
    if (llvmpipe->dirty & LP_NEW_TEXTURE)
       lp_setup_set_sampler_textures(llvmpipe->setup, 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index a8f4a4ed463..4b51d6b964f 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -998,23 +998,22 @@ llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
 void
 llvmpipe_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
-                             const struct pipe_constant_buffer *constants)
+                             struct pipe_buffer *constants)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
-   struct pipe_buffer *buffer = constants ? constants->buffer : NULL;
-   unsigned size = buffer ? buffer->size : 0;
-   const void *data = buffer ? llvmpipe_buffer(buffer)->data : NULL;
+   unsigned size = constants ? constants->size : 0;
+   const void *data = constants ? llvmpipe_buffer(constants)->data : NULL;
 
    assert(shader < PIPE_SHADER_TYPES);
    assert(index == 0);
 
-   if(llvmpipe->constants[shader].buffer == buffer)
+   if(llvmpipe->constants[shader] == constants)
       return;
 
    draw_flush(llvmpipe->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&llvmpipe->constants[shader].buffer, buffer);
+   pipe_buffer_reference(&llvmpipe->constants[shader], constants);
 
    if(shader == PIPE_SHADER_VERTEX) {
       draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX,
diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c
index 29fff91981a..6c29e8d8ace 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c
@@ -462,7 +462,7 @@ compute_blend_ref(const struct pipe_blend_state *blend,
 }
 
 
-ALIGN_STACK
+PIPE_ALIGN_STACK
 static boolean
 test_one(unsigned verbose,
          FILE *fp,
@@ -531,11 +531,11 @@ test_one(unsigned verbose,
    success = TRUE;
    for(i = 0; i < n && success; ++i) {
       if(mode == AoS) {
-         ALIGN16_ATTRIB uint8_t src[LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t dst[LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t con[LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t res[LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t ref[LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t src[LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t dst[LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t con[LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t res[LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t ref[LP_NATIVE_VECTOR_WIDTH/8];
          int64_t start_counter = 0;
          int64_t end_counter = 0;
 
@@ -596,11 +596,11 @@ test_one(unsigned verbose,
 
       if(mode == SoA) {
          const unsigned stride = type.length*type.width/8;
-         ALIGN16_ATTRIB uint8_t src[4*LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t dst[4*LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t con[4*LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t res[4*LP_NATIVE_VECTOR_WIDTH/8];
-         ALIGN16_ATTRIB uint8_t ref[4*LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t src[4*LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t dst[4*LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t con[4*LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t res[4*LP_NATIVE_VECTOR_WIDTH/8];
+         PIPE_ALIGN_VAR(16) uint8_t ref[4*LP_NATIVE_VECTOR_WIDTH/8];
          int64_t start_counter = 0;
          int64_t end_counter = 0;
          boolean mismatch;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c
index faddfb96779..c1abee424c9 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c
@@ -142,7 +142,7 @@ add_conv_test(LLVMModuleRef module,
 }
 
 
-ALIGN_STACK
+PIPE_ALIGN_STACK
 static boolean
 test_one(unsigned verbose,
          FILE *fp,
@@ -230,8 +230,8 @@ test_one(unsigned verbose,
    for(i = 0; i < n && success; ++i) {
       unsigned src_stride = src_type.length*src_type.width/8;
       unsigned dst_stride = dst_type.length*dst_type.width/8;
-      ALIGN16_ATTRIB uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
-      ALIGN16_ATTRIB uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
+      PIPE_ALIGN_VAR(16) uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
+      PIPE_ALIGN_VAR(16) uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
       double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
       uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
       int64_t start_counter = 0;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 23ea9ebbe7d..2b258f1052e 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -199,7 +199,7 @@ add_store_rgba_test(LLVMModuleRef module,
 }
 
 
-ALIGN_STACK
+PIPE_ALIGN_STACK
 static boolean
 test_format(unsigned verbose, FILE *fp, const struct pixel_test_case *test)
 {
diff --git a/src/gallium/drivers/nv04/nv04_state.c b/src/gallium/drivers/nv04/nv04_state.c
index e3dc4c5bf44..b67f1e16b1c 100644
--- a/src/gallium/drivers/nv04/nv04_state.c
+++ b/src/gallium/drivers/nv04/nv04_state.c
@@ -332,7 +332,7 @@ nv04_set_clip_state(struct pipe_context *pipe,
 
 static void
 nv04_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
-			 const struct pipe_constant_buffer *buf )
+			 struct pipe_buffer *buf )
 {
 	struct nv04_context *nv04 = nv04_context(pipe);
 	struct pipe_screen *pscreen = pipe->screen;
@@ -342,13 +342,13 @@ nv04_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
 
 	if (buf) {
 		void *mapped;
-		if (buf->buffer && buf->buffer->size &&
-                    (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)))
+		if (buf && buf->size &&
+                    (mapped = pipe_buffer_map(pscreen, buf, PIPE_BUFFER_USAGE_CPU_READ)))
 		{
-			memcpy(nv04->constbuf[shader], mapped, buf->buffer->size);
+			memcpy(nv04->constbuf[shader], mapped, buf->size);
 			nv04->constbuf_nr[shader] =
-				buf->buffer->size / (4 * sizeof(float));
-			pipe_buffer_unmap(pscreen, buf->buffer);
+				buf->size / (4 * sizeof(float));
+			pipe_buffer_unmap(pscreen, buf);
 		}
 	}
 }
diff --git a/src/gallium/drivers/nv10/nv10_state.c b/src/gallium/drivers/nv10/nv10_state.c
index ffc6be3c401..ad7def53b12 100644
--- a/src/gallium/drivers/nv10/nv10_state.c
+++ b/src/gallium/drivers/nv10/nv10_state.c
@@ -458,7 +458,7 @@ nv10_set_clip_state(struct pipe_context *pipe,
 
 static void
 nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
-			 const struct pipe_constant_buffer *buf )
+			 struct pipe_buffer *buf )
 {
 	struct nv10_context *nv10 = nv10_context(pipe);
 	struct pipe_screen *pscreen = pipe->screen;
@@ -468,13 +468,13 @@ nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
 
 	if (buf) {
 		void *mapped;
-		if (buf->buffer && buf->buffer->size &&
-                    (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)))
+		if (buf->size &&
+                    (mapped = pipe_buffer_map(pscreen, buf, PIPE_BUFFER_USAGE_CPU_READ)))
 		{
-			memcpy(nv10->constbuf[shader], mapped, buf->buffer->size);
+			memcpy(nv10->constbuf[shader], mapped, buf->size);
 			nv10->constbuf_nr[shader] =
-				buf->buffer->size / (4 * sizeof(float));
-			pipe_buffer_unmap(pscreen, buf->buffer);
+				buf->size / (4 * sizeof(float));
+			pipe_buffer_unmap(pscreen, buf);
 		}
 	}
 }
diff --git a/src/gallium/drivers/nv20/nv20_state.c b/src/gallium/drivers/nv20/nv20_state.c
index 3a82e63423d..45697a60efd 100644
--- a/src/gallium/drivers/nv20/nv20_state.c
+++ b/src/gallium/drivers/nv20/nv20_state.c
@@ -451,7 +451,7 @@ nv20_set_clip_state(struct pipe_context *pipe,
 
 static void
 nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
-			 const struct pipe_constant_buffer *buf )
+			 struct pipe_buffer *buf )
 {
 	struct nv20_context *nv20 = nv20_context(pipe);
 	struct pipe_screen *pscreen = pipe->screen;
@@ -461,13 +461,13 @@ nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
 
 	if (buf) {
 		void *mapped;
-		if (buf->buffer && buf->buffer->size &&
-                    (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)))
+		if (buf->size &&
+                    (mapped = pipe_buffer_map(pscreen, buf, PIPE_BUFFER_USAGE_CPU_READ)))
 		{
-			memcpy(nv20->constbuf[shader], mapped, buf->buffer->size);
+			memcpy(nv20->constbuf[shader], mapped, buf->size);
 			nv20->constbuf_nr[shader] =
-				buf->buffer->size / (4 * sizeof(float));
-			pipe_buffer_unmap(pscreen, buf->buffer);
+				buf->size / (4 * sizeof(float));
+			pipe_buffer_unmap(pscreen, buf);
 		}
 	}
 }
diff --git a/src/gallium/drivers/nv30/nv30_fragtex.c b/src/gallium/drivers/nv30/nv30_fragtex.c
index 98935678911..0cc3172dcd5 100644
--- a/src/gallium/drivers/nv30/nv30_fragtex.c
+++ b/src/gallium/drivers/nv30/nv30_fragtex.c
@@ -43,7 +43,6 @@ static struct nv30_texture_format *
 nv30_fragtex_format(uint pipe_format)
 {
 	struct nv30_texture_format *tf = nv30_texture_formats;
-	char fs[128];
 
 	while (tf->defined) {
 		if (tf->pipe == pipe_format)
@@ -65,7 +64,7 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit)
 	struct nouveau_bo *bo = nouveau_bo(nv30mt->buffer);
 	struct nv30_texture_format *tf;
 	struct nouveau_stateobj *so;
-	uint32_t txf, txs , txp;
+	uint32_t txf, txs;
 	unsigned tex_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
 
 	tf = nv30_fragtex_format(pt->format);
@@ -97,13 +96,6 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit)
 		return NULL;
 	}
 
-	if (!(pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		txp = 0;
-	} else {
-		txp  = nv30mt->level[0].pitch;
-		txf |= (1<<13) /*FIXME: NV34TCL_TX_FORMAT_LINEAR ? */;
-	}
-
 	txs = tf->swizzle;
 
 	so = so_new(1, 8, 2);
diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c
index a80dfb04880..065c927a10d 100644
--- a/src/gallium/drivers/nv30/nv30_state.c
+++ b/src/gallium/drivers/nv30/nv30_state.c
@@ -590,12 +590,12 @@ nv30_set_clip_state(struct pipe_context *pipe,
 
 static void
 nv30_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
-			 const struct pipe_constant_buffer *buf )
+			 struct pipe_buffer *buf )
 {
 	struct nv30_context *nv30 = nv30_context(pipe);
 
-	nv30->constbuf[shader] = buf->buffer;
-	nv30->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float));
+	nv30->constbuf[shader] = buf;
+	nv30->constbuf_nr[shader] = buf->size / (4 * sizeof(float));
 
 	if (shader == PIPE_SHADER_VERTEX) {
 		nv30->dirty |= NV30_NEW_VERTPROG;
diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c
index ed0ca9e02c3..7d990f7d567 100644
--- a/src/gallium/drivers/nv40/nv40_state.c
+++ b/src/gallium/drivers/nv40/nv40_state.c
@@ -605,12 +605,12 @@ nv40_set_clip_state(struct pipe_context *pipe,
 
 static void
 nv40_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
-			 const struct pipe_constant_buffer *buf )
+			 struct pipe_buffer *buf )
 {
 	struct nv40_context *nv40 = nv40_context(pipe);
 
-	nv40->constbuf[shader] = buf->buffer;
-	nv40->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float));
+	nv40->constbuf[shader] = buf;
+	nv40->constbuf_nr[shader] = buf->size / (4 * sizeof(float));
 
 	if (shader == PIPE_SHADER_VERTEX) {
 		nv40->dirty |= NV40_NEW_VERTPROG;
diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index 5997456e4c9..22a8498c69b 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -67,8 +67,12 @@ nv50_destroy(struct pipe_context *pipe)
 		so_ref(NULL, &nv50->state.vertprog);
 	if (nv50->state.fragprog)
 		so_ref(NULL, &nv50->state.fragprog);
-	if (nv50->state.programs)
-		so_ref(NULL, &nv50->state.programs);
+	if (nv50->state.geomprog)
+		so_ref(NULL, &nv50->state.geomprog);
+	if (nv50->state.fp_linkage)
+		so_ref(NULL, &nv50->state.fp_linkage);
+	if (nv50->state.gp_linkage)
+		so_ref(NULL, &nv50->state.gp_linkage);
 	if (nv50->state.vtxfmt)
 		so_ref(NULL, &nv50->state.vtxfmt);
 	if (nv50->state.vtxbuf)
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index cbd4c3ff86d..d024be6ea95 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -45,9 +45,11 @@
 #define NV50_NEW_VERTPROG_CB	(1 << 9)
 #define NV50_NEW_FRAGPROG	(1 << 10)
 #define NV50_NEW_FRAGPROG_CB	(1 << 11)
-#define NV50_NEW_ARRAYS		(1 << 12)
-#define NV50_NEW_SAMPLER	(1 << 13)
-#define NV50_NEW_TEXTURE	(1 << 14)
+#define NV50_NEW_GEOMPROG	(1 << 12)
+#define NV50_NEW_GEOMPROG_CB	(1 << 13)
+#define NV50_NEW_ARRAYS		(1 << 14)
+#define NV50_NEW_SAMPLER	(1 << 15)
+#define NV50_NEW_TEXTURE	(1 << 16)
 
 struct nv50_blend_stateobj {
 	struct pipe_blend_state pipe;
@@ -129,7 +131,9 @@ struct nv50_state {
 	unsigned miptree_nr[PIPE_SHADER_TYPES];
 	struct nouveau_stateobj *vertprog;
 	struct nouveau_stateobj *fragprog;
-	struct nouveau_stateobj *programs;
+	struct nouveau_stateobj *geomprog;
+	struct nouveau_stateobj *fp_linkage;
+	struct nouveau_stateobj *gp_linkage;
 	struct nouveau_stateobj *vtxfmt;
 	struct nouveau_stateobj *vtxbuf;
 	struct nouveau_stateobj *vtxattr;
@@ -157,6 +161,7 @@ struct nv50_context {
 	struct pipe_framebuffer_state framebuffer;
 	struct nv50_program *vertprog;
 	struct nv50_program *fragprog;
+	struct nv50_program *geomprog;
 	struct pipe_buffer *constbuf[PIPE_SHADER_TYPES];
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
@@ -207,7 +212,9 @@ extern void nv50_clear(struct pipe_context *pipe, unsigned buffers,
 /* nv50_program.c */
 extern void nv50_vertprog_validate(struct nv50_context *nv50);
 extern void nv50_fragprog_validate(struct nv50_context *nv50);
-extern void nv50_linkage_validate(struct nv50_context *nv50);
+extern void nv50_geomprog_validate(struct nv50_context *nv50);
+extern void nv50_fp_linkage_validate(struct nv50_context *nv50);
+extern void nv50_gp_linkage_validate(struct nv50_context *nv50);
 extern void nv50_program_destroy(struct nv50_context *nv50,
 				 struct nv50_program *p);
 
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 3f1edf0a139..dc8364ced7e 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -92,12 +92,23 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 	case PIPE_FORMAT_Z24S8_UNORM:
 		tile_flags = 0x1800;
 		break;
+	case PIPE_FORMAT_Z16_UNORM:
+		tile_flags = 0x6c00;
+		break;
 	case PIPE_FORMAT_X8Z24_UNORM:
 	case PIPE_FORMAT_S8Z24_UNORM:
 		tile_flags = 0x2800;
 		break;
+	case PIPE_FORMAT_R32G32B32A32_FLOAT:
+	case PIPE_FORMAT_R32G32B32_FLOAT:
+		tile_flags = 0x7400;
+		break;
 	default:
-		tile_flags = 0x7000;
+		if ((pt->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) &&
+		    util_format_get_blocksizebits(pt->format) == 32)
+			tile_flags = 0x7a00;
+		else
+			tile_flags = 0x7000;
 		break;
 	}
 
@@ -145,7 +156,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 				  mt->level[0].tile_mode, tile_flags,
 				  &mt->base.bo);
 	if (ret) {
-		for (l = 0; l < pt->last_level; ++l)
+		for (l = 0; l <= pt->last_level; ++l)
 			FREE(mt->level[l].image_offset);
 		FREE(mt);
 		return NULL;
@@ -188,7 +199,7 @@ nv50_miptree_destroy(struct pipe_texture *pt)
 	struct nv50_miptree *mt = nv50_miptree(pt);
 	unsigned l;
 
-	for (l = 0; l < pt->last_level; ++l)
+	for (l = 0; l <= pt->last_level; ++l)
 		FREE(mt->level[l].image_offset);
 
 	nouveau_bo_ref(NULL, &mt->base.bo);
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 53f9f0adf31..acd695b39bc 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -92,6 +92,9 @@ struct nv50_reg {
 
 	int rhw; /* result hw for FP outputs, or interpolant index */
 	int acc; /* instruction where this reg is last read (first insn == 1) */
+
+	int vtx; /* vertex index, for GP inputs (TGSI Dimension.Index) */
+	int indirect[2]; /* index into pc->addr, or -1 */
 };
 
 #define NV50_MOD_NEG 1
@@ -135,9 +138,11 @@ struct nv50_pc {
 	int immd_nr;
 	struct nv50_reg **addr;
 	int addr_nr;
-	uint8_t addr_alloc; /* set bit indicates used for TGSI_FILE_ADDRESS */
+	struct nv50_reg *sysval;
+	int sysval_nr;
 
 	struct nv50_reg *temp_temp[16];
+	struct nv50_program_exec *temp_temp_exec[16];
 	unsigned temp_temp_nr;
 
 	/* broadcast and destination replacement regs */
@@ -170,6 +175,8 @@ struct nv50_pc {
 	uint8_t edgeflag_out;
 };
 
+static struct nv50_reg *get_address_reg(struct nv50_pc *, struct nv50_reg *);
+
 static INLINE void
 ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw)
 {
@@ -178,7 +185,9 @@ ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw)
 	reg->hw = hw;
 	reg->mod = 0;
 	reg->rhw = -1;
+	reg->vtx = -1;
 	reg->acc = 0;
+	reg->indirect[0] = reg->indirect[1] = -1;
 }
 
 static INLINE unsigned
@@ -196,7 +205,8 @@ terminate_mbb(struct nv50_pc *pc)
 
 	/* remove records of temporary address register values */
 	for (i = 0; i < NV50_SU_MAX_ADDR; ++i)
-		pc->r_addr[i].rhw = -1;
+		if (pc->r_addr[i].index < 0)
+			pc->r_addr[i].acc = 0;
 }
 
 static void
@@ -245,7 +255,8 @@ alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg)
 		}
 	}
 
-	assert(0);
+	NOUVEAU_ERR("out of registers\n");
+	abort();
 }
 
 static INLINE struct nv50_reg *
@@ -258,6 +269,7 @@ reg_instance(struct nv50_pc *pc, struct nv50_reg *reg)
 	if (reg) {
 		alloc_reg(pc, reg);
 		*ri = *reg;
+		reg->indirect[0] = reg->indirect[1] = -1;
 		reg->mod = 0;
 	}
 	return ri;
@@ -285,7 +297,8 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst)
 		}
 	}
 
-	assert(0);
+	NOUVEAU_ERR("out of registers\n");
+	abort();
 	return NULL;
 }
 
@@ -347,23 +360,29 @@ free_temp4(struct nv50_pc *pc, struct nv50_reg *reg[4])
 }
 
 static struct nv50_reg *
-temp_temp(struct nv50_pc *pc)
+temp_temp(struct nv50_pc *pc, struct nv50_program_exec *e)
 {
 	if (pc->temp_temp_nr >= 16)
 		assert(0);
 
 	pc->temp_temp[pc->temp_temp_nr] = alloc_temp(pc, NULL);
+	pc->temp_temp_exec[pc->temp_temp_nr] = e;
 	return pc->temp_temp[pc->temp_temp_nr++];
 }
 
+/* This *must* be called for all nv50_program_exec that have been
+ * given as argument to temp_temp, or the temps will be leaked !
+ */
 static void
-kill_temp_temp(struct nv50_pc *pc)
+kill_temp_temp(struct nv50_pc *pc, struct nv50_program_exec *e)
 {
 	int i;
 
 	for (i = 0; i < pc->temp_temp_nr; i++)
-		free_temp(pc, pc->temp_temp[i]);
-	pc->temp_temp_nr = 0;
+		if (pc->temp_temp_exec[i] == e)
+			free_temp(pc, pc->temp_temp[i]);
+	if (!e)
+		pc->temp_temp_nr = 0;
 }
 
 static int
@@ -425,6 +444,8 @@ emit(struct nv50_pc *pc, struct nv50_program_exec *e)
 		p->exec_head = e;
 	p->exec_tail = e;
 	p->exec_size += (e->inst[0] & 1) ? 2 : 1;
+
+	kill_temp_temp(pc, e);
 }
 
 static INLINE void set_long(struct nv50_pc *, struct nv50_program_exec *);
@@ -514,11 +535,33 @@ set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e)
 static INLINE void
 set_addr(struct nv50_program_exec *e, struct nv50_reg *a)
 {
+	assert(a->type == P_ADDR);
+
 	assert(!(e->inst[0] & 0x0c000000));
 	assert(!(e->inst[1] & 0x00000004));
 
 	e->inst[0] |= (a->hw & 3) << 26;
-	e->inst[1] |= (a->hw >> 2) << 2;
+	e->inst[1] |= a->hw & 4;
+}
+
+static void
+emit_arl(struct nv50_pc *, struct nv50_reg *, struct nv50_reg *, uint8_t);
+
+static void
+emit_shl_imm(struct nv50_pc *, struct nv50_reg *, struct nv50_reg *, int);
+
+static void
+emit_mov_from_addr(struct nv50_pc *pc, struct nv50_reg *dst,
+		   struct nv50_reg *src)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[1] = 0x40000000;
+	set_long(pc, e);
+	set_dst(pc, dst, e);
+	set_addr(e, src);
+
+	emit(pc, e);
 }
 
 static void
@@ -537,72 +580,6 @@ emit_add_addr_imm(struct nv50_pc *pc, struct nv50_reg *dst,
 	emit(pc, e);
 }
 
-static struct nv50_reg *
-alloc_addr(struct nv50_pc *pc, struct nv50_reg *ref)
-{
-	struct nv50_reg *a_tgsi = NULL, *a = NULL;
-	int i;
-	uint8_t avail = ~pc->addr_alloc;
-
-	if (!ref) {
-		/* allocate for TGSI_FILE_ADDRESS */
-		while (avail) {
-			i = ffs(avail) - 1;
-
-			if (pc->r_addr[i].rhw < 0 ||
-			    pc->r_addr[i].acc != pc->insn_cur) {
-				pc->addr_alloc |= (1 << i);
-
-				pc->r_addr[i].rhw = -1;
-				pc->r_addr[i].index = i;
-				return &pc->r_addr[i];
-			}
-			avail &= ~(1 << i);
-		}
-		assert(0);
-		return NULL;
-	}
-
-	/* Allocate and set an address reg so we can access 'ref'.
-	 *
-	 * If and r_addr->index will be -1 or the hw index the value
-	 * value in rhw is relative to. If rhw < 0, the reg has not
-	 * been initialized or is in use for TGSI_FILE_ADDRESS.
-	 */
-	while (avail) { /* only consider regs that are not TGSI */
-		i = ffs(avail) - 1;
-		avail &= ~(1 << i);
-
-		if ((!a || a->rhw >= 0) && pc->r_addr[i].rhw < 0) {
-			/* prefer an usused reg with low hw index */
-			a = &pc->r_addr[i];
-			continue;
-		}
-		if (!a && pc->r_addr[i].acc != pc->insn_cur)
-			a = &pc->r_addr[i];
-
-		if (ref->hw - pc->r_addr[i].rhw >= 128)
-			continue;
-
-		if ((ref->acc >= 0 && pc->r_addr[i].index < 0) ||
-		    (ref->acc < 0 && pc->r_addr[i].index == ref->index)) {
-			pc->r_addr[i].acc = pc->insn_cur;
-			return &pc->r_addr[i];
-		}
-	}
-	assert(a);
-
-	if (ref->acc < 0)
-		a_tgsi = pc->addr[ref->index];
-
-	emit_add_addr_imm(pc, a, a_tgsi, (ref->hw & ~0x7f) * 4);
-
-	a->rhw = ref->hw & ~0x7f;
-	a->acc = pc->insn_cur;
-	a->index = a_tgsi ? ref->index : -1;
-	return a;
-}
-
 #define INTERP_LINEAR		0
 #define INTERP_FLAT		1
 #define INTERP_PERSPECTIVE	2
@@ -646,12 +623,12 @@ set_data(struct nv50_pc *pc, struct nv50_reg *src, unsigned m, unsigned s,
 	e->param.shift = s;
 	e->param.mask = m << (s % 32);
 
-	if (src->hw > 127)
-		set_addr(e, alloc_addr(pc, src));
+	if (src->hw < 0 || src->hw > 127) /* need (additional) address reg */
+		set_addr(e, get_address_reg(pc, src));
 	else
 	if (src->acc < 0) {
 		assert(src->type == P_CONST);
-		set_addr(e, pc->addr[src->index]);
+		set_addr(e, pc->addr[src->indirect[0]]);
 	}
 
 	e->inst[1] |= (((src->type == P_IMMD) ? 0 : 1) << 22);
@@ -683,6 +660,12 @@ emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 		if (src->type == P_ATTR) {
 			set_long(pc, e);
 			e->inst[1] |= 0x00200000;
+
+			if (src->vtx >= 0) {
+				/* indirect (vertex base + c) load from p[] */
+				e->inst[0] |= 0x01800000;
+				set_addr(e, get_address_reg(pc, src));
+			}
 		}
 
 		alloc_reg(pc, src);
@@ -780,7 +763,7 @@ set_src_0_restricted(struct nv50_pc *pc, struct nv50_reg *src,
 	struct nv50_reg *temp;
 
 	if (src->type != P_TEMP) {
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, e);
 		emit_mov(pc, temp, src);
 		src = temp;
 	}
@@ -797,9 +780,14 @@ set_src_0(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e)
 	if (src->type == P_ATTR) {
 		set_long(pc, e);
 		e->inst[1] |= 0x00200000;
+
+		if (src->vtx >= 0) {
+			e->inst[0] |= 0x01800000; /* src from p[] */
+			set_addr(e, get_address_reg(pc, src));
+		}
 	} else
 	if (src->type == P_CONST || src->type == P_IMMD) {
-		struct nv50_reg *temp = temp_temp(pc);
+		struct nv50_reg *temp = temp_temp(pc, e);
 
 		emit_mov(pc, temp, src);
 		src = temp;
@@ -815,19 +803,19 @@ static void
 set_src_1(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e)
 {
 	if (src->type == P_ATTR) {
-		struct nv50_reg *temp = temp_temp(pc);
+		struct nv50_reg *temp = temp_temp(pc, e);
 
 		emit_mov(pc, temp, src);
 		src = temp;
 	} else
 	if (src->type == P_CONST || src->type == P_IMMD) {
-		assert(!(e->inst[0] & 0x00800000));
-		if (e->inst[0] & 0x01000000) {
-			struct nv50_reg *temp = temp_temp(pc);
+		if (e->inst[0] & 0x01800000) {
+			struct nv50_reg *temp = temp_temp(pc, e);
 
 			emit_mov(pc, temp, src);
 			src = temp;
 		} else {
+			assert(!(e->inst[0] & 0x00800000));
 			set_data(pc, src, 0x7f, 16, e);
 			e->inst[0] |= 0x00800000;
 		}
@@ -845,19 +833,19 @@ set_src_2(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e)
 	set_long(pc, e);
 
 	if (src->type == P_ATTR) {
-		struct nv50_reg *temp = temp_temp(pc);
+		struct nv50_reg *temp = temp_temp(pc, e);
 
 		emit_mov(pc, temp, src);
 		src = temp;
 	} else
 	if (src->type == P_CONST || src->type == P_IMMD) {
-		assert(!(e->inst[0] & 0x01000000));
-		if (e->inst[0] & 0x00800000) {
-			struct nv50_reg *temp = temp_temp(pc);
+		if (e->inst[0] & 0x01800000) {
+			struct nv50_reg *temp = temp_temp(pc, e);
 
 			emit_mov(pc, temp, src);
 			src = temp;
 		} else {
+			assert(!(e->inst[0] & 0x01000000));
 			set_data(pc, src, 0x7f, 32+14, e);
 			e->inst[0] |= 0x01000000;
 		}
@@ -868,6 +856,26 @@ set_src_2(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e)
 }
 
 static void
+set_half_src(struct nv50_pc *pc, struct nv50_reg *src, int lh,
+	     struct nv50_program_exec *e, int pos)
+{
+	struct nv50_reg *r = src;
+
+	alloc_reg(pc, r);
+	if (r->type != P_TEMP) {
+		r = temp_temp(pc, e);
+		emit_mov(pc, r, src);
+	}
+
+	if (r->hw > (NV50_SU_MAX_TEMP / 2)) {
+		NOUVEAU_ERR("out of low GPRs\n");
+		abort();
+	}
+
+	e->inst[pos / 32] |= ((src->hw * 2) + lh) << (pos % 32);
+}
+
+static void
 emit_mov_from_pred(struct nv50_pc *pc, struct nv50_reg *dst, int pred)
 {
 	struct nv50_program_exec *e = exec(pc);
@@ -966,11 +974,125 @@ emit_arl(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src,
 
 	e->inst[0] |= dst->hw << 2;
 	e->inst[0] |= s << 16; /* shift left */
-	set_src_0_restricted(pc, src, e);
+	set_src_0(pc, src, e);
 
 	emit(pc, e);
 }
 
+static boolean
+address_reg_suitable(struct nv50_reg *a, struct nv50_reg *r)
+{
+	if (!r)
+		return FALSE;
+
+	if (r->vtx != a->vtx)
+		return FALSE;
+	if (r->vtx >= 0)
+		return (r->indirect[1] == a->indirect[1]);
+
+	if (r->hw < a->rhw || (r->hw - a->rhw) >= 128)
+		return FALSE;
+
+	if (a->index >= 0)
+		return (a->index == r->indirect[0]);
+	return (a->indirect[0] == r->indirect[0]);
+}
+
+static void
+load_vertex_base(struct nv50_pc *pc, struct nv50_reg *dst,
+		 struct nv50_reg *a, int shift)
+{
+	struct nv50_reg mem, *temp;
+
+	ctor_reg(&mem, P_ATTR, -1, dst->vtx);
+
+	assert(dst->type == P_ADDR);
+	if (!a) {
+		emit_arl(pc, dst, &mem, 0);
+		return;
+	}
+	temp = alloc_temp(pc, NULL);
+
+	if (shift) {
+		emit_mov_from_addr(pc, temp, a);
+		if (shift < 0)
+			emit_shl_imm(pc, temp, temp, shift);
+		emit_arl(pc, dst, temp, MAX2(shift, 0));
+	}
+	emit_mov(pc, temp, &mem);
+	set_addr(pc->p->exec_tail, dst);
+
+	emit_arl(pc, dst, temp, 0);
+	free_temp(pc, temp);
+}
+
+/* case (ref == NULL): allocate address register for TGSI_FILE_ADDRESS
+ * case (vtx >= 0, acc >= 0): load vertex base from a[vtx * 4] to $aX
+ * case (vtx >= 0, acc < 0): load vertex base from s[$aY + vtx * 4] to $aX
+ * case (vtx < 0, acc >= 0): memory address too high to encode
+ * case (vtx < 0, acc < 0): get source register for TGSI_FILE_ADDRESS
+ */
+static struct nv50_reg *
+get_address_reg(struct nv50_pc *pc, struct nv50_reg *ref)
+{
+	int i;
+	struct nv50_reg *a_ref, *a = NULL;
+
+	for (i = 0; i < NV50_SU_MAX_ADDR; ++i) {
+		if (pc->r_addr[i].acc == 0)
+			a = &pc->r_addr[i]; /* an unused address reg */
+		else
+		if (address_reg_suitable(&pc->r_addr[i], ref)) {
+			pc->r_addr[i].acc = pc->insn_cur;
+			return &pc->r_addr[i];
+		} else
+		if (!a && pc->r_addr[i].index < 0 &&
+		    pc->r_addr[i].acc < pc->insn_cur)
+			a = &pc->r_addr[i];
+	}
+	if (!a) {
+		/* We'll be able to spill address regs when this
+		 * mess is replaced with a proper compiler ...
+		 */
+		NOUVEAU_ERR("out of address regs\n");
+		abort();
+		return NULL;
+	}
+
+	/* initialize and reserve for this TGSI instruction */
+	a->rhw = 0;
+	a->index = a->indirect[0] = a->indirect[1] = -1;
+	a->acc = pc->insn_cur;
+
+	if (!ref) {
+		a->vtx = -1;
+		return a;
+	}
+	a->vtx = ref->vtx;
+
+	/* now put in the correct value ... */
+
+	if (ref->vtx >= 0) {
+		a->indirect[1] = ref->indirect[1];
+
+		/* For an indirect vertex index, we need to shift address right
+		 * by 2, the address register will contain vtx * 16, we need to
+		 * load from a[vtx * 4].
+		 */
+		load_vertex_base(pc, a, (ref->acc < 0) ?
+				 pc->addr[ref->indirect[1]] : NULL, -2);
+	} else {
+		assert(ref->acc < 0 || ref->indirect[0] < 0);
+
+		a->rhw = ref->hw & ~0x7f;
+		a->indirect[0] = ref->indirect[0];
+		a_ref = (ref->acc < 0) ? pc->addr[ref->indirect[0]] : NULL;
+
+		emit_add_addr_imm(pc, a, a_ref, a->rhw * 4);
+	}
+	return a;
+}
+
 #define NV50_MAX_F32 0x880
 #define NV50_MAX_S32 0x08c
 #define NV50_MAX_U32 0x084
@@ -1050,6 +1172,20 @@ emit_bitop2(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
 }
 
 static void
+emit_not(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[0] = 0xd0000000;
+	e->inst[1] = 0x0402c000;
+	set_long(pc, e);
+	set_dst(pc, dst, e);
+	set_src_1(pc, src, e);
+
+	emit(pc, e);
+}
+
+static void
 emit_shift(struct nv50_pc *pc, struct nv50_reg *dst,
 	   struct nv50_reg *src0, struct nv50_reg *src1, unsigned dir)
 {
@@ -1078,6 +1214,27 @@ emit_shift(struct nv50_pc *pc, struct nv50_reg *dst,
 }
 
 static void
+emit_shl_imm(struct nv50_pc *pc, struct nv50_reg *dst,
+	     struct nv50_reg *src, int s)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[0] = 0x30000000;
+	e->inst[1] = 0xc4100000;
+	if (s < 0) {
+		e->inst[1] |= 1 << 29;
+		s = -s;
+	}
+	e->inst[1] |= ((s & 0x7f) << 16);
+
+	set_long(pc, e);
+	set_dst(pc, dst, e);
+	set_src_0(pc, src, e);
+
+	emit(pc, e);
+}
+
+static void
 emit_mad(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
 	 struct nv50_reg *src1, struct nv50_reg *src2)
 {
@@ -1321,7 +1478,7 @@ emit_add_b32(struct nv50_pc *pc, struct nv50_reg *dst,
 
 	src1 = rsrc1;
 	if (src0->mod & rsrc1->mod & NV50_MOD_NEG) {
-		src1 = alloc_temp(pc, NULL);
+		src1 = temp_temp(pc, e);
 		emit_cvt(pc, src1, rsrc1, -1, CVT_S32_S32);
 	}
 
@@ -1350,9 +1507,43 @@ emit_add_b32(struct nv50_pc *pc, struct nv50_reg *dst,
 		e->inst[0] |= 1 << 22;
 
 	emit(pc, e);
+}
+
+static void
+emit_mad_u16(struct nv50_pc *pc, struct nv50_reg *dst,
+	     struct nv50_reg *src0, int lh_0, struct nv50_reg *src1, int lh_1,
+	     struct nv50_reg *src2)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[0] = 0x60000000;
+	if (!pc->allow32)
+		set_long(pc, e);
+	set_dst(pc, dst, e);
+
+	set_half_src(pc, src0, lh_0, e, 9);
+	set_half_src(pc, src1, lh_1, e, 16);
+	alloc_reg(pc, src2);
+	if (is_long(e) || (src2->type != P_TEMP) || (src2->hw != dst->hw))
+		set_src_2(pc, src2, e);
+
+	emit(pc, e);
+}
+
+static void
+emit_mul_u16(struct nv50_pc *pc, struct nv50_reg *dst,
+	     struct nv50_reg *src0, int lh_0, struct nv50_reg *src1, int lh_1)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[0] = 0x40000000;
+	set_long(pc, e);
+	set_dst(pc, dst, e);
+
+	set_half_src(pc, src0, lh_0, e, 9);
+	set_half_src(pc, src1, lh_1, e, 16);
 
-	if (src1 != rsrc1)
-		free_temp(pc, src1);
+	emit(pc, e);
 }
 
 static void
@@ -1362,6 +1553,9 @@ emit_sad(struct nv50_pc *pc, struct nv50_reg *dst,
 	struct nv50_program_exec *e = exec(pc);
 
 	e->inst[0] = 0x50000000;
+	if (!pc->allow32)
+		set_long(pc, e);
+	check_swap_src_0_1(pc, &src0, &src1);
 	set_dst(pc, dst, e);
 	set_src_0(pc, src0, e);
 	set_src_1(pc, src1, e);
@@ -1373,6 +1567,8 @@ emit_sad(struct nv50_pc *pc, struct nv50_reg *dst,
 		e->inst[1] |= 0x0c << 24;
 	else
 		e->inst[0] |= 0x81 << 8;
+
+	emit(pc, e);
 }
 
 static INLINE void
@@ -1422,10 +1618,10 @@ emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
 	if (mask & (1 << 2)) {
 		set_pred_wr(pc, 1, 0, pc->p->exec_tail);
 
-		tmp[1] = temp_temp(pc);
+		tmp[1] = temp_temp(pc, NULL);
 		emit_minmax(pc, NV50_MAX_F32, tmp[1], src[1], zero);
 
-		tmp[3] = temp_temp(pc);
+		tmp[3] = temp_temp(pc, NULL);
 		emit_minmax(pc, NV50_MAX_F32, tmp[3], src[3], neg128);
 		emit_minmax(pc, NV50_MIN_F32, tmp[3], tmp[3], pos128);
 
@@ -1524,6 +1720,18 @@ emit_ret(struct nv50_pc *pc, int pred, unsigned cc)
 	emit_control_flow(pc, 0x3, pred, cc);
 }
 
+static void
+emit_prim_cmd(struct nv50_pc *pc, unsigned cmd)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[0] = 0xf0000000 | (cmd << 9);
+	e->inst[1] = 0xc0000000;
+	set_long(pc, e);
+
+	emit(pc, e);
+}
+
 #define QOP_ADD 0
 #define QOP_SUBR 1
 #define QOP_SUB 2
@@ -1884,7 +2092,11 @@ convert_to_long(struct nv50_pc *pc, struct nv50_program_exec *e)
 	case 0x5:
 		/* SAD */
 		m = ~(0x81 << 8);
-		q = 0x0c << 24;
+		q = (0x0c << 24) | ((e->inst[0] & (0x7f << 2)) << 12);
+		break;
+	case 0x6:
+		/* MAD u16 */
+		q = (e->inst[0] & (0x7f << 2)) << 12;
 		break;
 	case 0x8:
 		/* INTERP (move centroid, perspective and flat bits) */
@@ -1964,8 +2176,11 @@ get_supported_mods(const struct tgsi_full_instruction *insn, int i)
 	case TGSI_OPCODE_IMAX:
 	case TGSI_OPCODE_IMIN:
 	case TGSI_OPCODE_ISHR:
+	case TGSI_OPCODE_NOT:
+	case TGSI_OPCODE_UMAD:
 	case TGSI_OPCODE_UMAX:
 	case TGSI_OPCODE_UMIN:
+	case TGSI_OPCODE_UMUL:
 	case TGSI_OPCODE_USHR:
 		return NV50_MOD_I32;
 	default:
@@ -2059,14 +2274,19 @@ tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst)
 	{
 		struct nv50_reg *r = pc->addr[dst->Register.Index * 4 + c];
 		if (!r) {
-			r = alloc_addr(pc, NULL);
-			pc->addr[dst->Register.Index * 4 + c] = r;
+			r = get_address_reg(pc, NULL);
+			r->index = dst->Register.Index * 4 + c;
+			pc->addr[r->index] = r;
 		}
 		assert(r);
 		return r;
 	}
 	case TGSI_FILE_NULL:
 		return NULL;
+	case TGSI_FILE_SYSTEM_VALUE:
+		assert(pc->sysval[dst->Register.Index].type == P_RESULT);
+		assert(c == 0);
+		return &pc->sysval[dst->Register.Index];
 	default:
 		break;
 	}
@@ -2096,6 +2316,18 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
 		switch (src->Register.File) {
 		case TGSI_FILE_INPUT:
 			r = &pc->attr[src->Register.Index * 4 + c];
+
+			if (!src->Dimension.Dimension)
+				break;
+			r = reg_instance(pc, r);
+			r->vtx = src->Dimension.Index;
+
+			if (!src->Dimension.Indirect)
+				break;
+			swz = tgsi_util_get_src_register_swizzle(
+				&src->DimIndirect, 0);
+			r->acc = -1;
+			r->indirect[1] = src->DimIndirect.Index * 4 + swz;
 			break;
 		case TGSI_FILE_TEMPORARY:
 			r = &pc->temp[src->Register.Index * 4 + c];
@@ -2109,12 +2341,12 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
 			 * use the index field to select the address reg.
 			 */
 			r = reg_instance(pc, NULL);
+			ctor_reg(r, P_CONST, -1, src->Register.Index * 4 + c);
+
 			swz = tgsi_util_get_src_register_swizzle(
-						 &src->Indirect, 0);
-			ctor_reg(r, P_CONST,
-				 src->Indirect.Index * 4 + swz,
-				 src->Register.Index * 4 + c);
+				&src->Indirect, 0);
 			r->acc = -1;
+			r->indirect[0] = src->Indirect.Index * 4 + swz;
 			break;
 		case TGSI_FILE_IMMEDIATE:
 			r = &pc->immd[src->Register.Index * 4 + c];
@@ -2125,6 +2357,10 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
 			r = pc->addr[src->Register.Index * 4 + c];
 			assert(r);
 			break;
+		case TGSI_FILE_SYSTEM_VALUE:
+			assert(c == 0);
+			r = &pc->sysval[src->Register.Index];
+			break;
 		default:
 			assert(0);
 			break;
@@ -2153,7 +2389,7 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
 	}
 
 	if ((r->mod & mod) != r->mod) {
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, NULL);
 		emit_cvt(pc, temp, r, -1, cvn);
 		r->mod = 0;
 		r = temp;
@@ -2161,7 +2397,7 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
 		r->mod |= mod & NV50_MOD_I32;
 
 	assert(r);
-	if (r->acc >= 0 && r != temp)
+	if (r->acc >= 0 && r->vtx < 0 && r != temp)
 		return reg_instance(pc, r); /* will clear r->mod */
 	return r;
 }
@@ -2341,7 +2577,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 
 	brdc = temp = pc->r_brdc;
 	if (brdc && brdc->type != P_TEMP) {
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, NULL);
 		if (sat)
 			brdc = temp;
 	} else
@@ -2350,7 +2586,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			if (!(mask & (1 << c)) || dst[c]->type == P_TEMP)
 				continue;
 			/* rdst[c] = dst[c]; */ /* done above */
-			dst[c] = temp_temp(pc);
+			dst[c] = temp_temp(pc, NULL);
 		}
 	}
 
@@ -2383,10 +2619,14 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		}
 		break;
 	case TGSI_OPCODE_ARL:
-		assert(src[0][0]);
-		temp = temp_temp(pc);
-		emit_cvt(pc, temp, src[0][0], -1, CVT_FLOOR | CVT_S32_F32);
-		emit_arl(pc, dst[0], temp, 4);
+		temp = temp_temp(pc, NULL);
+		for (c = 0; c < 4; c++) {
+			if (!(mask & (1 << c)))
+				continue;
+			emit_cvt(pc, temp, src[0][c], -1,
+				 CVT_FLOOR | CVT_S32_F32);
+			emit_arl(pc, dst[c], temp, 4);
+		}
 		break;
 	case TGSI_OPCODE_BGNLOOP:
 		pc->loop_brka[pc->loop_lvl] = emit_breakaddr(pc);
@@ -2441,7 +2681,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			if (!(mask &= 7))
 				break;
 			if (temp == dst[3])
-				temp = brdc = temp_temp(pc);
+				temp = brdc = temp_temp(pc, NULL);
 		}
 		emit_precossin(pc, temp, src[0][0]);
 		emit_flop(pc, NV50_FLOP_COS, brdc, temp);
@@ -2493,6 +2733,9 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		pc->if_insn[pc->if_lvl++] = pc->p->exec_tail;
 		terminate_mbb(pc);
 		break;
+	case TGSI_OPCODE_EMIT:
+		emit_prim_cmd(pc, 1);
+		break;
 	case TGSI_OPCODE_ENDIF:
 		pc->if_insn[--pc->if_lvl]->param.index = pc->p->exec_size;
 
@@ -2516,8 +2759,12 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		pc->loop_brka[pc->loop_lvl]->param.index = pc->p->exec_size;
 		terminate_mbb(pc);
 		break;
+	case TGSI_OPCODE_ENDPRIM:
+		emit_prim_cmd(pc, 2);
+		break;
 	case TGSI_OPCODE_ENDSUB:
 		assert(pc->in_subroutine);
+		terminate_mbb(pc);
 		pc->in_subroutine = FALSE;
 		break;
 	case TGSI_OPCODE_EX2:
@@ -2529,8 +2776,8 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		struct nv50_reg *t[2];
 
 		assert(!temp);
-		t[0] = temp_temp(pc);
-		t[1] = temp_temp(pc);
+		t[0] = temp_temp(pc, NULL);
+		t[1] = temp_temp(pc, NULL);
 
 		if (mask & 0x6)
 			emit_mov(pc, t[0], src[0][0]);
@@ -2575,7 +2822,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		}
 		break;
 	case TGSI_OPCODE_FRC:
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, NULL);
 		for (c = 0; c < 4; c++) {
 			if (!(mask & (1 << c)))
 				continue;
@@ -2639,9 +2886,9 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 	{
 		struct nv50_reg *t[2];
 
-		t[0] = temp_temp(pc);
+		t[0] = temp_temp(pc, NULL);
 		if (mask & (1 << 1))
-			t[1] = temp_temp(pc);
+			t[1] = temp_temp(pc, NULL);
 		else
 			t[1] = t[0];
 
@@ -2664,7 +2911,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 	}
 		break;
 	case TGSI_OPCODE_LRP:
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, NULL);
 		for (c = 0; c < 4; c++) {
 			if (!(mask & (1 << c)))
 				continue;
@@ -2707,10 +2954,19 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			emit_mul(pc, dst[c], src[0][c], src[1][c]);
 		}
 		break;
+	case TGSI_OPCODE_NOT:
+		for (c = 0; c < 4; c++) {
+			if (!(mask & (1 << c)))
+				continue;
+			emit_not(pc, dst[c], src[0][c]);
+		}
+		break;
 	case TGSI_OPCODE_POW:
 		emit_pow(pc, brdc, src[0][0], src[1][0]);
 		break;
 	case TGSI_OPCODE_RCP:
+		if (!sat && popcnt4(mask) == 1)
+			brdc = dst[ffs(mask) - 1];
 		emit_flop(pc, NV50_FLOP_RCP, brdc, src[0][0]);
 		break;
 	case TGSI_OPCODE_RET:
@@ -2719,6 +2975,8 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		emit_ret(pc, -1, 0);
 		break;
 	case TGSI_OPCODE_RSQ:
+		if (!sat && popcnt4(mask) == 1)
+			brdc = dst[ffs(mask) - 1];
 		src[0][0]->mod |= NV50_MOD_ABS;
 		emit_flop(pc, NV50_FLOP_RSQ, brdc, src[0][0]);
 		break;
@@ -2730,7 +2988,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		}
 		break;
 	case TGSI_OPCODE_SCS:
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, NULL);
 		if (mask & 3)
 			emit_precossin(pc, temp, src[0][0]);
 		if (mask & (1 << 0))
@@ -2759,7 +3017,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			if (!(mask &= 7))
 				break;
 			if (temp == dst[3])
-				temp = brdc = temp_temp(pc);
+				temp = brdc = temp_temp(pc, NULL);
 		}
 		emit_precossin(pc, temp, src[0][0]);
 		emit_flop(pc, NV50_FLOP_SIN, brdc, temp);
@@ -2847,8 +3105,41 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			emit_minmax(pc, 0x0a4, dst[c], src[0][c], src[1][c]);
 		}
 		break;
+	case TGSI_OPCODE_UMAD:
+	{
+		assert(!temp);
+		temp = temp_temp(pc, NULL);
+		for (c = 0; c < 4; c++) {
+			if (!(mask & (1 << c)))
+				continue;
+			emit_mul_u16(pc, temp, src[0][c], 0, src[1][c], 1);
+			emit_mad_u16(pc, temp, src[0][c], 1, src[1][c], 0,
+				     temp);
+			emit_shl_imm(pc, temp, temp, 16);
+			emit_mad_u16(pc, temp, src[0][c], 0, src[1][c], 0,
+				     temp);
+			emit_add_b32(pc, dst[c], temp, src[2][c]);
+		}
+	}
+		break;
+	case TGSI_OPCODE_UMUL:
+	{
+		assert(!temp);
+		temp = temp_temp(pc, NULL);
+		for (c = 0; c < 4; c++) {
+			if (!(mask & (1 << c)))
+				continue;
+			emit_mul_u16(pc, temp, src[0][c], 0, src[1][c], 1);
+			emit_mad_u16(pc, temp, src[0][c], 1, src[1][c], 0,
+				     temp);
+			emit_shl_imm(pc, temp, temp, 16);
+			emit_mad_u16(pc, dst[c], src[0][c], 0, src[1][c], 0,
+				     temp);
+		}
+	}
+		break;
 	case TGSI_OPCODE_XPD:
-		temp = temp_temp(pc);
+		temp = temp_temp(pc, NULL);
 		if (mask & (1 << 0)) {
 			emit_mul(pc, temp, src[0][2], src[1][1]);
 			emit_msb(pc, dst[0], src[0][1], src[1][2], temp);
@@ -2876,6 +3167,8 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			emit_nop(pc);
 
 		pc->p->exec_tail->inst[1] |= 1; /* set exit bit */
+
+		terminate_mbb(pc);
 		break;
 	default:
 		NOUVEAU_ERR("invalid opcode %d\n", inst->Instruction.Opcode);
@@ -2902,7 +3195,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		}
 	}
 
-	kill_temp_temp(pc);
+	kill_temp_temp(pc, NULL);
 	pc->reg_instance_nr = 0;
 
 	return TRUE;
@@ -3101,7 +3394,7 @@ nv50_tgsi_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 	if (is_scalar_op(insn.Instruction.Opcode)) {
 		pc->r_brdc = tgsi_broadcast_dst(pc, fd, deqs);
 		if (!pc->r_brdc)
-			pc->r_brdc = temp_temp(pc);
+			pc->r_brdc = temp_temp(pc, NULL);
 		return nv50_program_tx_insn(pc, &insn);
 	}
 	pc->r_brdc = NULL;
@@ -3171,17 +3464,40 @@ load_interpolant(struct nv50_pc *pc, struct nv50_reg *reg)
  * value of 0 for back-facing, and 0xffffffff for front-facing.
  */
 static void
-load_frontfacing(struct nv50_pc *pc, struct nv50_reg *a)
+load_frontfacing(struct nv50_pc *pc, struct nv50_reg *sv)
 {
-	struct nv50_reg *one = alloc_immd(pc, 1.0f);
+	struct nv50_reg *temp = alloc_temp(pc, NULL);
+	int r_pred = 0;
 
-	assert(a->rhw == -1);
-	alloc_reg(pc, a); /* do this before rhw is set */
-	a->rhw = 255;
-	load_interpolant(pc, a);
-	emit_bitop2(pc, a, a, one, TGSI_OPCODE_AND);
+	temp->rhw = 255;
+	emit_interp(pc, temp, NULL, INTERP_FLAT);
 
-	FREE(one);
+	emit_cvt(pc, sv, temp, r_pred, CVT_ABS | CVT_F32_S32);
+
+	emit_not(pc, temp, temp);
+	set_pred(pc, 0x2, r_pred, pc->p->exec_tail);
+	emit_cvt(pc, sv, temp, -1, CVT_F32_S32);
+	set_pred(pc, 0x2, r_pred, pc->p->exec_tail);
+
+	free_temp(pc, temp);
+}
+
+static void
+copy_semantic_info(struct nv50_program *p)
+{
+	unsigned i, id;
+
+	for (i = 0; i < p->cfg.in_nr; ++i) {
+		id = p->cfg.in[i].id;
+		p->cfg.in[i].sn = p->info.input_semantic_name[id];
+		p->cfg.in[i].si = p->info.input_semantic_index[id];
+	}
+
+	for (i = 0; i < p->cfg.out_nr; ++i) {
+		id = p->cfg.out[i].id;
+		p->cfg.out[i].sn = p->info.output_semantic_name[id];
+		p->cfg.out[i].si = p->info.output_semantic_index[id];
+	}
 }
 
 static boolean
@@ -3190,7 +3506,7 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 	struct tgsi_parse_context tp;
 	struct nv50_program *p = pc->p;
 	boolean ret = FALSE;
-	unsigned i, c, flat_nr = 0;
+	unsigned i, c, instance_id, vertex_id, flat_nr = 0;
 
 	tgsi_parse_init(&tp, pc->p->pipe.tokens);
 	while (!tgsi_parse_end_of_tokens(&tp)) {
@@ -3230,13 +3546,13 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 				switch (d->Semantic.Name) {
 				case TGSI_SEMANTIC_BCOLOR:
 					p->cfg.two_side[si].hw = first;
-					if (p->cfg.io_nr > first)
-						p->cfg.io_nr = first;
+					if (p->cfg.out_nr > first)
+						p->cfg.out_nr = first;
 					break;
 				case TGSI_SEMANTIC_PSIZE:
 					p->cfg.psiz = first;
-					if (p->cfg.io_nr > first)
-						p->cfg.io_nr = first;
+					if (p->cfg.out_nr > first)
+						p->cfg.out_nr = first;
 					break;
 				case TGSI_SEMANTIC_EDGEFLAG:
 					pc->edgeflag_out = first;
@@ -3276,6 +3592,37 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 					pc->interp_mode[i] = mode;
 			}
 				break;
+			case TGSI_FILE_SYSTEM_VALUE:
+				assert(d->Declaration.Semantic);
+				switch (d->Semantic.Name) {
+				case TGSI_SEMANTIC_FACE:
+					assert(p->type == PIPE_SHADER_FRAGMENT);
+					load_frontfacing(pc,
+							 &pc->sysval[first]);
+					break;
+				case TGSI_SEMANTIC_INSTANCEID:
+					assert(p->type == PIPE_SHADER_VERTEX);
+					instance_id = first;
+					p->cfg.regs[0] |= (1 << 4);
+					break;
+				case TGSI_SEMANTIC_PRIMID:
+					assert(p->type != PIPE_SHADER_VERTEX);
+					p->cfg.prim_id = first;
+					break;
+					/*
+				case TGSI_SEMANTIC_PRIMIDIN:
+					assert(p->type == PIPE_SHADER_GEOMETRY);
+					pc->sysval[first].hw = 6;
+					p->cfg.regs[0] |= (1 << 8);
+					break;
+				case TGSI_SEMANTIC_VERTEXID:
+					assert(p->type == PIPE_SHADER_VERTEX);
+					vertex_id = first;
+					p->cfg.regs[0] |= (1 << 12) | (1 << 0);
+					break;
+					*/
+				}
+				break;
 			case TGSI_FILE_ADDRESS:
 			case TGSI_FILE_CONSTANT:
 			case TGSI_FILE_SAMPLER:
@@ -3296,68 +3643,96 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 		}
 	}
 
-	if (p->type == PIPE_SHADER_VERTEX) {
+	if (p->type == PIPE_SHADER_VERTEX || p->type == PIPE_SHADER_GEOMETRY) {
 		int rid = 0;
 
-		for (i = 0; i < pc->attr_nr * 4; ++i) {
-			if (pc->attr[i].acc) {
-				pc->attr[i].hw = rid++;
-				p->cfg.attr[i / 32] |= 1 << (i % 32);
+		if (p->type == PIPE_SHADER_GEOMETRY) {
+			for (i = 0; i < pc->attr_nr; ++i) {
+				p->cfg.in[i].hw = rid;
+				p->cfg.in[i].id = i;
+
+				for (c = 0; c < 4; ++c) {
+					int n = i * 4 + c;
+					if (!pc->attr[n].acc)
+						continue;
+					pc->attr[n].hw = rid++;
+					p->cfg.in[i].mask |= 1 << c;
+				}
+			}
+		} else {
+			for (i = 0; i < pc->attr_nr * 4; ++i) {
+				if (pc->attr[i].acc) {
+					pc->attr[i].hw = rid++;
+					p->cfg.attr[i / 32] |= 1 << (i % 32);
+				}
 			}
+			if (p->cfg.regs[0] & (1 << 0))
+				pc->sysval[vertex_id].hw = rid++;
+			if (p->cfg.regs[0] & (1 << 4))
+				pc->sysval[instance_id].hw = rid++;
 		}
 
 		for (i = 0, rid = 0; i < pc->result_nr; ++i) {
-			p->cfg.io[i].hw = rid;
-			p->cfg.io[i].id = i;
+			p->cfg.out[i].hw = rid;
+			p->cfg.out[i].id = i;
 
 			for (c = 0; c < 4; ++c) {
 				int n = i * 4 + c;
 				if (!pc->result[n].acc)
 					continue;
 				pc->result[n].hw = rid++;
-				p->cfg.io[i].mask |= 1 << c;
+				p->cfg.out[i].mask |= 1 << c;
 			}
 		}
+		if (p->cfg.prim_id < 0x40) {
+			/* GP has to write to PrimitiveID */
+			ctor_reg(&pc->sysval[p->cfg.prim_id],
+				 P_RESULT, p->cfg.prim_id, rid);
+			p->cfg.prim_id = rid++;
+		}
 
 		for (c = 0; c < 2; ++c)
 			if (p->cfg.two_side[c].hw < 0x40)
-				p->cfg.two_side[c] = p->cfg.io[
+				p->cfg.two_side[c] = p->cfg.out[
 					p->cfg.two_side[c].hw];
 
 		if (p->cfg.psiz < 0x40)
-			p->cfg.psiz = p->cfg.io[p->cfg.psiz].hw;
+			p->cfg.psiz = p->cfg.out[p->cfg.psiz].hw;
+
+		copy_semantic_info(p);
 	} else
 	if (p->type == PIPE_SHADER_FRAGMENT) {
-		int rid, aid;
+		int rid, aid, base;
 		unsigned n = 0, m = pc->attr_nr - flat_nr;
 
 		pc->allow32 = TRUE;
 
-		int base = (TGSI_SEMANTIC_POSITION ==
-			    p->info.input_semantic_name[0]) ? 0 : 1;
+		base = (TGSI_SEMANTIC_POSITION ==
+			p->info.input_semantic_name[0]) ? 0 : 1;
 
 		/* non-flat interpolants have to be mapped to
 		 * the lower hardware IDs, so sort them:
 		 */
 		for (i = 0; i < pc->attr_nr; i++) {
 			if (pc->interp_mode[i] == INTERP_FLAT)
-				p->cfg.io[m++].id = i;
+				p->cfg.in[m++].id = i;
 			else {
 				if (!(pc->interp_mode[i] & INTERP_PERSPECTIVE))
-					p->cfg.io[n].linear = TRUE;
-				p->cfg.io[n++].id = i;
+					p->cfg.in[n].linear = TRUE;
+				p->cfg.in[n++].id = i;
 			}
 		}
+		copy_semantic_info(p);
 
 		if (!base) /* set w-coordinate mask from perspective interp */
-			p->cfg.io[0].mask |= p->cfg.regs[1] >> 24;
+			p->cfg.in[0].mask |= p->cfg.regs[1] >> 24;
 
 		aid = popcnt4( /* if fcrd isn't contained in cfg.io */
-			base ? (p->cfg.regs[1] >> 24) : p->cfg.io[0].mask);
+			base ? (p->cfg.regs[1] >> 24) : p->cfg.in[0].mask);
 
 		for (n = 0; n < pc->attr_nr; ++n) {
-			p->cfg.io[n].hw = rid = aid;
-			i = p->cfg.io[n].id;
+			p->cfg.in[n].hw = rid = aid;
+			i = p->cfg.in[n].id;
 
 			if (p->info.input_semantic_name[n] ==
 			    TGSI_SEMANTIC_FACE) {
@@ -3369,15 +3744,15 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 				if (!pc->attr[i * 4 + c].acc)
 					continue;
 				pc->attr[i * 4 + c].rhw = rid++;
-				p->cfg.io[n].mask |= 1 << c;
+				p->cfg.in[n].mask |= 1 << c;
 
 				load_interpolant(pc, &pc->attr[i * 4 + c]);
 			}
-			aid += popcnt4(p->cfg.io[n].mask);
+			aid += popcnt4(p->cfg.in[n].mask);
 		}
 
 		if (!base)
-			p->cfg.regs[1] |= p->cfg.io[0].mask << 24;
+			p->cfg.regs[1] |= p->cfg.in[0].mask << 24;
 
 		m = popcnt4(p->cfg.regs[1] >> 24);
 
@@ -3387,32 +3762,33 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 		p->cfg.regs[1] |= aid - m;
 
 		if (flat_nr) {
-			i = p->cfg.io[pc->attr_nr - flat_nr].hw;
+			i = p->cfg.in[pc->attr_nr - flat_nr].hw;
 			p->cfg.regs[1] |= (i - m) << 16;
 		} else
 			p->cfg.regs[1] |= p->cfg.regs[1] << 16;
 
 		/* mark color semantic for light-twoside */
-		n = 0x40;
-		for (i = 0; i < pc->attr_nr; i++) {
-			ubyte si, sn;
-
-			sn = p->info.input_semantic_name[p->cfg.io[i].id];
-			si = p->info.input_semantic_index[p->cfg.io[i].id];
-
-			if (sn == TGSI_SEMANTIC_COLOR) {
-				p->cfg.two_side[si] = p->cfg.io[i];
-
-				/* increase colour count */
-				p->cfg.regs[0] += popcnt4(
-					p->cfg.two_side[si].mask) << 16;
-
-				n = MIN2(n, p->cfg.io[i].hw - m);
+		n = 0x80;
+		for (i = 0; i < p->cfg.in_nr; i++) {
+			if (p->cfg.in[i].sn == TGSI_SEMANTIC_COLOR) {
+				n = MIN2(n, p->cfg.in[i].hw - m);
+				p->cfg.two_side[p->cfg.in[i].si] = p->cfg.in[i];
+
+				p->cfg.regs[0] += /* increase colour count */
+					popcnt4(p->cfg.in[i].mask) << 16;
 			}
 		}
-		if (n < 0x40)
+		if (n < 0x80)
 			p->cfg.regs[0] += n;
 
+		if (p->cfg.prim_id < 0x40) {
+			pc->sysval[p->cfg.prim_id].rhw = rid++;
+			emit_interp(pc, &pc->sysval[p->cfg.prim_id], NULL,
+				    INTERP_FLAT);
+			/* increase FP_INTERPOLANT_CTRL_COUNT */
+			p->cfg.regs[1] += 1;
+		}
+
 		/* Initialize FP results:
 		 * FragDepth is always first TGSI and last hw output
 		 */
@@ -3466,10 +3842,31 @@ free_nv50_pc(struct nv50_pc *pc)
 		FREE(pc->attr);
 	if (pc->temp)
 		FREE(pc->temp);
+	if (pc->sysval)
+		FREE(pc->sysval);
+	if (pc->insn_pos)
+		FREE(pc->insn_pos);
 
 	FREE(pc);
 }
 
+static INLINE uint32_t
+nv50_map_gs_output_prim(unsigned pprim)
+{
+	switch (pprim) {
+	case PIPE_PRIM_POINTS:
+		return NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE_POINTS;
+	case PIPE_PRIM_LINE_STRIP:
+		return NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE_LINE_STRIP;
+	case PIPE_PRIM_TRIANGLE_STRIP:
+		return NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE_TRIANGLE_STRIP;
+	default:
+		NOUVEAU_ERR("invalid GS_OUTPUT_PRIMITIVE: %u\n", pprim);
+		abort();
+		return 0;
+	}
+}
+
 static boolean
 ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
 {
@@ -3483,25 +3880,55 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
 	pc->param_nr = p->info.file_max[TGSI_FILE_CONSTANT] + 1;
 	pc->addr_nr = p->info.file_max[TGSI_FILE_ADDRESS] + 1;
 	assert(pc->addr_nr <= 2);
+	pc->sysval_nr = p->info.file_max[TGSI_FILE_SYSTEM_VALUE] + 1;
 
 	p->cfg.high_temp = 4;
 
 	p->cfg.two_side[0].hw = 0x40;
 	p->cfg.two_side[1].hw = 0x40;
+	p->cfg.prim_id = 0x40;
 
 	p->cfg.edgeflag_in = pc->edgeflag_out = 0xff;
 
+	for (i = 0; i < p->info.num_properties; ++i) {
+		unsigned *data = &p->info.properties[i].data[0];
+
+		switch (p->info.properties[i].name) {
+		case TGSI_PROPERTY_GS_OUTPUT_PRIM:
+			p->cfg.prim_type = nv50_map_gs_output_prim(data[0]);
+			break;
+		case TGSI_PROPERTY_GS_MAX_VERTICES:
+			p->cfg.vert_count = data[0];
+			break;
+		default:
+			break;
+		}
+	}
+
 	switch (p->type) {
 	case PIPE_SHADER_VERTEX:
 		p->cfg.psiz = 0x40;
 		p->cfg.clpd = 0x40;
-		p->cfg.io_nr = pc->result_nr;
+		p->cfg.out_nr = pc->result_nr;
+		break;
+	case PIPE_SHADER_GEOMETRY:
+		assert(p->cfg.prim_type);
+		assert(p->cfg.vert_count);
+
+		p->cfg.psiz = 0x80;
+		p->cfg.clpd = 0x80;
+		p->cfg.prim_id = 0x80;
+		p->cfg.out_nr = pc->result_nr;
+		p->cfg.in_nr = pc->attr_nr;
+
+		p->cfg.two_side[0].hw = 0x80;
+		p->cfg.two_side[1].hw = 0x80;
 		break;
 	case PIPE_SHADER_FRAGMENT:
 		rtype[0] = rtype[1] = P_TEMP;
 
 		p->cfg.regs[0] = 0x01000004;
-		p->cfg.io_nr = pc->attr_nr;
+		p->cfg.in_nr = pc->attr_nr;
 
 		if (p->info.writes_z) {
 			p->cfg.regs[2] |= 0x00000100;
@@ -3559,7 +3986,16 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
 			return FALSE;
 	}
 	for (i = 0; i < NV50_SU_MAX_ADDR; ++i)
-		ctor_reg(&pc->r_addr[i], P_ADDR, -256, i + 1);
+		ctor_reg(&pc->r_addr[i], P_ADDR, -1, i + 1);
+
+	if (pc->sysval_nr) {
+		pc->sysval = CALLOC(pc->sysval_nr, sizeof(struct nv50_reg *));
+		if (!pc->sysval)
+			return FALSE;
+		/* will only ever use SYSTEM_VALUE[i].x (hopefully) */
+		for (i = 0; i < pc->sysval_nr; ++i)
+			ctor_reg(&pc->sysval[i], rtype[0], i, -1);
+	}
 
 	return TRUE;
 }
@@ -3721,13 +4157,17 @@ nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p)
 
 	if (p->param_nr) {
 		unsigned cb;
-		uint32_t *map = pipe_buffer_map(pscreen, nv50->constbuf[p->type],
+		uint32_t *map = pipe_buffer_map(pscreen,
+						nv50->constbuf[p->type],
 						PIPE_BUFFER_USAGE_CPU_READ);
-
-		if (p->type == PIPE_SHADER_VERTEX)
+		switch (p->type) {
+		case PIPE_SHADER_GEOMETRY: cb = NV50_CB_PGP; break;
+		case PIPE_SHADER_FRAGMENT: cb = NV50_CB_PFP; break;
+		default:
 			cb = NV50_CB_PVP;
-		else
-			cb = NV50_CB_PFP;
+			assert(p->type == PIPE_SHADER_VERTEX);
+			break;
+		}
 
 		nv50_program_upload_data(nv50, map, 0, p->param_nr, cb);
 		pipe_buffer_unmap(pscreen, nv50->constbuf[p->type]);
@@ -3821,19 +4261,18 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
-	so = so_new(5, 8, 2);
+	so = so_new(5, 7, 2);
 	so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
-		      NOUVEAU_BO_HIGH, 0, 0);
+		  NOUVEAU_BO_HIGH, 0, 0);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
-		      NOUVEAU_BO_LOW, 0, 0);
+		  NOUVEAU_BO_LOW, 0, 0);
 	so_method(so, tesla, NV50TCL_VP_ATTR_EN_0, 2);
 	so_data  (so, p->cfg.attr[0]);
 	so_data  (so, p->cfg.attr[1]);
 	so_method(so, tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1);
 	so_data  (so, p->cfg.high_result);
-	so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 2);
-	so_data  (so, p->cfg.high_result); //8);
+	so_method(so, tesla, NV50TCL_VP_REG_ALLOC_TEMP, 1);
 	so_data  (so, p->cfg.high_temp);
 	so_method(so, tesla, NV50TCL_VP_START_ID, 1);
 	so_data  (so, 0); /* program start offset */
@@ -3877,12 +4316,51 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	so_ref(NULL, &so);
 }
 
-static void
+void
+nv50_geomprog_validate(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nv50_program *p = nv50->geomprog;
+	struct nouveau_stateobj *so;
+
+	if (!p->translated) {
+		nv50_program_validate(nv50, p);
+		if (!p->translated)
+			assert(0);
+	}
+
+	nv50_program_validate_data(nv50, p);
+	nv50_program_validate_code(nv50, p);
+
+	so = so_new(6, 7, 2);
+	so_method(so, tesla, NV50TCL_GP_ADDRESS_HIGH, 2);
+	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
+		  NOUVEAU_BO_HIGH, 0, 0);
+	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
+		  NOUVEAU_BO_LOW, 0, 0);
+	so_method(so, tesla, NV50TCL_GP_REG_ALLOC_TEMP, 1);
+	so_data  (so, p->cfg.high_temp);
+	so_method(so, tesla, NV50TCL_GP_REG_ALLOC_RESULT, 1);
+	so_data  (so, p->cfg.high_result);
+	so_method(so, tesla, NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE, 1);
+	so_data  (so, p->cfg.prim_type);
+	so_method(so, tesla, NV50TCL_GP_VERTEX_OUTPUT_COUNT, 1);
+	so_data  (so, p->cfg.vert_count);
+	so_method(so, tesla, NV50TCL_GP_START_ID, 1);
+	so_data  (so, 0);
+	so_ref(so, &nv50->state.geomprog);
+	so_ref(NULL, &so);
+}
+
+static uint32_t
 nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base)
 {
+	struct nv50_program *vp;
 	struct nv50_program *fp = nv50->fragprog;
-	struct nv50_program *vp = nv50->vertprog;
 	unsigned i, c, m = base;
+	uint32_t origin = 0x00000010;
+
+	vp = nv50->geomprog ? nv50->geomprog : nv50->vertprog;
 
 	/* XXX: this might not work correctly in all cases yet - we'll
 	 * just assume that an FP generic input that is not written in
@@ -3890,58 +4368,59 @@ nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base)
 	 */
 	memset(pntc, 0, 8 * sizeof(uint32_t));
 
-	for (i = 0; i < fp->cfg.io_nr; i++) {
-		uint8_t sn, si;
-		uint8_t j, k = fp->cfg.io[i].id;
-		unsigned n = popcnt4(fp->cfg.io[i].mask);
+	for (i = 0; i < fp->cfg.in_nr; i++) {
+		unsigned j, n = popcnt4(fp->cfg.in[i].mask);
 
-		if (fp->info.input_semantic_name[k] != TGSI_SEMANTIC_GENERIC) {
+		if (fp->cfg.in[i].sn != TGSI_SEMANTIC_GENERIC) {
 			m += n;
 			continue;
 		}
 
-		for (j = 0; j < vp->info.num_outputs; ++j) {
-			sn = vp->info.output_semantic_name[j];
-			si = vp->info.output_semantic_index[j];
-
-			if (sn == fp->info.input_semantic_name[k] &&
-			    si == fp->info.input_semantic_index[k])
+		for (j = 0; j < vp->cfg.out_nr; ++j)
+			if (vp->cfg.out[j].sn ==  fp->cfg.in[i].sn &&
+			    vp->cfg.out[j].si == fp->cfg.in[i].si)
 				break;
-		}
 
-		if (j < vp->info.num_outputs) {
-			ubyte mode =
-				nv50->rasterizer->pipe.sprite_coord_mode[si];
+		if (j < vp->cfg.out_nr) {
+			ubyte mode = nv50->rasterizer->pipe.sprite_coord_mode[
+				vp->cfg.out[j].si];
 
 			if (mode == PIPE_SPRITE_COORD_NONE) {
 				m += n;
 				continue;
-			}
+			} else
+			if (mode == PIPE_SPRITE_COORD_LOWER_LEFT)
+				origin = 0;
 		}
 
 		/* this is either PointCoord or replaced by sprite coords */
 		for (c = 0; c < 4; c++) {
-			if (!(fp->cfg.io[i].mask & (1 << c)))
+			if (!(fp->cfg.in[i].mask & (1 << c)))
 				continue;
 			pntc[m / 8] |= (c + 1) << ((m % 8) * 4);
 			++m;
 		}
 	}
+	return origin;
 }
 
 static int
-nv50_sreg4_map(uint32_t *p_map, int mid, uint32_t lin[4],
-	       struct nv50_sreg4 *fpi, struct nv50_sreg4 *vpo)
+nv50_vec4_map(uint32_t *map32, int mid, uint8_t zval, uint32_t lin[4],
+	      struct nv50_sreg4 *fpi, struct nv50_sreg4 *vpo)
 {
 	int c;
 	uint8_t mv = vpo->mask, mf = fpi->mask, oid = vpo->hw;
-	uint8_t *map = (uint8_t *)p_map;
+	uint8_t *map = (uint8_t *)map32;
 
 	for (c = 0; c < 4; ++c) {
 		if (mf & 1) {
 			if (fpi->linear == TRUE)
 				lin[mid / 32] |= 1 << (mid % 32);
-			map[mid++] = (mv & 1) ? oid : ((c == 3) ? 0x41 : 0x40);
+			if (mv & 1)
+				map[mid] = oid;
+			else
+				map[mid] = (c == 3) ? (zval + 1) : zval;
+			++mid;
 		}
 
 		oid += mv & 1;
@@ -3953,34 +4432,42 @@ nv50_sreg4_map(uint32_t *p_map, int mid, uint32_t lin[4],
 }
 
 void
-nv50_linkage_validate(struct nv50_context *nv50)
+nv50_fp_linkage_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nv50_program *vp = nv50->vertprog;
 	struct nv50_program *fp = nv50->fragprog;
 	struct nouveau_stateobj *so;
-	struct nv50_sreg4 dummy, *vpo;
+	struct nv50_sreg4 dummy;
 	int i, n, c, m = 0;
-	uint32_t map[16], lin[4], reg[5], pcrd[8];
+	uint32_t map[16], lin[4], reg[6], pcrd[8];
+	uint8_t zval = 0x40;
 
+	if (nv50->geomprog) {
+		vp = nv50->geomprog;
+		zval = 0x80;
+	}
 	memset(map, 0, sizeof(map));
 	memset(lin, 0, sizeof(lin));
 
 	reg[1] = 0x00000004; /* low and high clip distance map ids */
 	reg[2] = 0x00000000; /* layer index map id (disabled, GP only) */
 	reg[3] = 0x00000000; /* point size map id & enable */
+	reg[5] = 0x00000000; /* primitive ID map slot */
 	reg[0] = fp->cfg.regs[0]; /* colour semantic reg */
 	reg[4] = fp->cfg.regs[1]; /* interpolant info */
 
 	dummy.linear = FALSE;
 	dummy.mask = 0xf; /* map all components of HPOS */
-	m = nv50_sreg4_map(map, m, lin, &dummy, &vp->cfg.io[0]);
+	m = nv50_vec4_map(map, m, zval, lin, &dummy, &vp->cfg.out[0]);
 
 	dummy.mask = 0x0;
 
 	if (vp->cfg.clpd < 0x40) {
-		for (c = 0; c < vp->cfg.clpd_nr; ++c)
-			map[m++] = vp->cfg.clpd + c;
+		for (c = 0; c < vp->cfg.clpd_nr; ++c) {
+			map[m / 4] |= (vp->cfg.clpd + c) << ((m % 4) * 8);
+			++m;
+		}
 		reg[1] = (m << 8);
 	}
 
@@ -3988,35 +4475,37 @@ nv50_linkage_validate(struct nv50_context *nv50)
 
 	/* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */
 	if (nv50->rasterizer->pipe.light_twoside) {
-		vpo = &vp->cfg.two_side[0];
+		struct nv50_sreg4 *vpo = &vp->cfg.two_side[0];
+		struct nv50_sreg4 *fpi = &fp->cfg.two_side[0];
 
-		m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[0], &vpo[0]);
-		m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[1], &vpo[1]);
+		m = nv50_vec4_map(map, m, zval, lin, &fpi[0], &vpo[0]);
+		m = nv50_vec4_map(map, m, zval, lin, &fpi[1], &vpo[1]);
 	}
 
 	reg[0] += m - 4; /* adjust FFC0 id */
 	reg[4] |= m << 8; /* set mid where 'normal' FP inputs start */
 
-	for (i = 0; i < fp->cfg.io_nr; i++) {
-		ubyte sn = fp->info.input_semantic_name[fp->cfg.io[i].id];
-		ubyte si = fp->info.input_semantic_index[fp->cfg.io[i].id];
-
-		/* position must be mapped first */
-		assert(i == 0 || sn != TGSI_SEMANTIC_POSITION);
-
+	for (i = 0; i < fp->cfg.in_nr; i++) {
 		/* maybe even remove these from cfg.io */
-		if (sn == TGSI_SEMANTIC_POSITION || sn == TGSI_SEMANTIC_FACE)
+		if (fp->cfg.in[i].sn == TGSI_SEMANTIC_POSITION ||
+		    fp->cfg.in[i].sn == TGSI_SEMANTIC_FACE)
 			continue;
 
-		/* VP outputs and vp->cfg.io are in the same order */
-		for (n = 0; n < vp->info.num_outputs; ++n) {
-			if (vp->info.output_semantic_name[n] == sn &&
-			    vp->info.output_semantic_index[n] == si)
+		for (n = 0; n < vp->cfg.out_nr; ++n)
+			if (vp->cfg.out[n].sn == fp->cfg.in[i].sn &&
+			    vp->cfg.out[n].si == fp->cfg.in[i].si)
 				break;
-		}
-		vpo = (n < vp->info.num_outputs) ? &vp->cfg.io[n] : &dummy;
 
-		m = nv50_sreg4_map(map, m, lin, &fp->cfg.io[i], vpo);
+		m = nv50_vec4_map(map, m, zval, lin, &fp->cfg.in[i],
+				  (n < vp->cfg.out_nr) ?
+				  &vp->cfg.out[n] : &dummy);
+	}
+	/* PrimitiveID either is replaced by the system value, or
+	 * written by the geometry shader into an output register
+	 */
+	if (fp->cfg.prim_id < 0x40) {
+		map[m / 4] |= vp->cfg.prim_id << ((m % 4) * 8);
+		reg[5] = m++;
 	}
 
 	if (nv50->rasterizer->pipe.point_size_per_vertex) {
@@ -4024,14 +4513,28 @@ nv50_linkage_validate(struct nv50_context *nv50)
 		reg[3] = (m++ << 4) | 1;
 	}
 
-	/* now fill the stateobj */
-	so = so_new(6, 58, 0);
+	/* now fill the stateobj (at most 28 so_data)  */
+	so = so_new(10, 54, 0);
 
 	n = (m + 3) / 4;
-	so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
-	so_data  (so, m);
-	so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n);
-	so_datap (so, map, n);
+	assert(m <= 32);
+	if (vp->type == PIPE_SHADER_GEOMETRY) {
+		so_method(so, tesla, NV50TCL_GP_RESULT_MAP_SIZE, 1);
+		so_data  (so, m);
+		so_method(so, tesla, NV50TCL_GP_RESULT_MAP(0), n);
+		so_datap (so, map, n);
+	} else {
+		so_method(so, tesla, NV50TCL_VP_GP_BUILTIN_ATTR_EN, 1);
+		so_data  (so, vp->cfg.regs[0]);
+
+		so_method(so, tesla, NV50TCL_MAP_SEMANTIC_4, 1);
+		so_data  (so, reg[5]);
+
+		so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
+		so_data  (so, m);
+		so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n);
+		so_datap (so, map, n);
+	}
 
 	so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4);
 	so_datap (so, reg, 4);
@@ -4043,14 +4546,85 @@ nv50_linkage_validate(struct nv50_context *nv50)
 	so_datap (so, lin, 4);
 
 	if (nv50->rasterizer->pipe.point_sprite) {
-		nv50_pntc_replace(nv50, pcrd, (reg[4] >> 8) & 0xff);
+		so_method(so, tesla, NV50TCL_POINT_SPRITE_CTRL, 1);
+		so_data  (so,
+			  nv50_pntc_replace(nv50, pcrd, (reg[4] >> 8) & 0xff));
 
 		so_method(so, tesla, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8);
 		so_datap (so, pcrd, 8);
 	}
 
-        so_ref(so, &nv50->state.programs);
-        so_ref(NULL, &so);
+	so_method(so, tesla, NV50TCL_GP_ENABLE, 1);
+	so_data  (so, (vp->type == PIPE_SHADER_GEOMETRY) ? 1 : 0);
+
+	so_ref(so, &nv50->state.fp_linkage);
+	so_ref(NULL, &so);
+}
+
+static int
+construct_vp_gp_mapping(uint32_t *map32, int m,
+			struct nv50_program *vp, struct nv50_program *gp)
+{
+	uint8_t *map = (uint8_t *)map32;
+	int i, j, c;
+
+        for (i = 0; i < gp->cfg.in_nr; ++i) {
+                uint8_t oid, mv = 0, mg = gp->cfg.in[i].mask;
+
+                for (j = 0; j < vp->cfg.out_nr; ++j) {
+                        if (vp->cfg.out[j].sn == gp->cfg.in[i].sn &&
+                            vp->cfg.out[j].si == gp->cfg.in[i].si) {
+				mv = vp->cfg.out[j].mask;
+				oid = vp->cfg.out[j].hw;
+                                break;
+			}
+		}
+
+                for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) {
+			if (mg & mv & 1)
+				map[m++] = oid;
+			else
+			if (mg & 1)
+				map[m++] = (c == 3) ? 0x41 : 0x40;
+                        oid += mv & 1;
+                }
+        }
+	return m;
+}
+
+void
+nv50_gp_linkage_validate(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so;
+	struct nv50_program *vp = nv50->vertprog;
+	struct nv50_program *gp = nv50->geomprog;
+	uint32_t map[16];
+	int m = 0;
+
+	if (!gp) {
+		so_ref(NULL, &nv50->state.gp_linkage);
+		return;
+	}
+	memset(map, 0, sizeof(map));
+
+	m = construct_vp_gp_mapping(map, m, vp, gp);
+
+	so = so_new(3, 24 - 3, 0);
+
+	so_method(so, tesla, NV50TCL_VP_GP_BUILTIN_ATTR_EN, 1);
+	so_data  (so, vp->cfg.regs[0] | gp->cfg.regs[0]);
+
+	assert(m <= 32);
+	so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
+	so_data  (so, m);
+
+	m = (m + 3) / 4;
+	so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), m);
+	so_datap (so, map, m);
+
+	so_ref(so, &nv50->state.gp_linkage);
+	so_ref(NULL, &so);
 }
 
 void
diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h
index 461fec1d89c..1e3ad6bff05 100644
--- a/src/gallium/drivers/nv50/nv50_program.h
+++ b/src/gallium/drivers/nv50/nv50_program.h
@@ -16,11 +16,13 @@ struct nv50_program_exec {
 };
 
 struct nv50_sreg4 {
-	uint8_t hw;
-	uint8_t id; /* tgsi index, nv50 needs them sorted: flat ones last */
+	uint8_t hw; /* hw index, nv50 wants flat FP inputs last */
+	uint8_t id; /* tgsi index */
 
 	uint8_t mask;
 	boolean linear;
+
+	ubyte sn, si; /* semantic name & index */
 };
 
 struct nv50_program {
@@ -49,16 +51,24 @@ struct nv50_program {
 		uint32_t regs[4];
 
 		/* for VPs, io_nr doesn't count 'private' results (PSIZ etc.) */
-		unsigned io_nr;
-		struct nv50_sreg4 io[PIPE_MAX_SHADER_OUTPUTS];
+		unsigned in_nr, out_nr;
+		struct nv50_sreg4 in[PIPE_MAX_SHADER_INPUTS];
+		struct nv50_sreg4 out[PIPE_MAX_SHADER_OUTPUTS];
 
 		/* FP colour inputs, VP/GP back colour outputs */
 		struct nv50_sreg4 two_side[2];
 
-		/* VP only */
+		/* GP only */
+		unsigned vert_count;
+		uint8_t prim_type;
+
+		/* VP & GP only */
 		uint8_t clpd, clpd_nr;
 		uint8_t psiz;
 		uint8_t edgeflag_in;
+
+		/* FP & GP only */
+		uint8_t prim_id;
 	} cfg;
 };
 
diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c
index 5d9e18218ae..5a4ab3508b8 100644
--- a/src/gallium/drivers/nv50/nv50_query.c
+++ b/src/gallium/drivers/nv50/nv50_query.c
@@ -111,7 +111,7 @@ nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 
 	if (!q->ready) {
 		ret = nouveau_bo_map(q->bo, NOUVEAU_BO_RD |
-				     wait ? 0 : NOUVEAU_BO_NOWAIT);
+				     (wait ? 0 : NOUVEAU_BO_NOWAIT));
 		if (ret)
 			return false;
 		q->result = ((uint32_t *)q->bo->map)[1];
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 28e2b35deaa..73205652cc4 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -167,7 +167,7 @@ nv50_screen_destroy(struct pipe_screen *pscreen)
 	struct nv50_screen *screen = nv50_screen(pscreen);
 	unsigned i;
 
-	for (i = 0; i < 2; i++) {
+	for (i = 0; i < 3; i++) {
 		if (screen->constbuf_parm[i])
 			nouveau_bo_ref(NULL, &screen->constbuf_parm[i]);
 	}
@@ -329,7 +329,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_ref(NULL, &so);
 
 	/* Static tesla init */
-	so = so_new(40, 84, 20);
+	so = so_new(44, 90, 22);
 
 	so_method(so, screen->tesla, NV50TCL_COND_MODE, 1);
 	so_data  (so, NV50TCL_COND_MODE_ALWAYS);
@@ -352,10 +352,11 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_data  (so, 0xf);
 
 	/* max TIC (bits 4:8) & TSC (ignored) bindings, per program type */
-	so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(0), 1);
-	so_data  (so, 0x54);
-	so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(2), 1);
-	so_data  (so, 0x54);
+	for (i = 0; i < 3; ++i) {
+		so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(i), 1);
+		so_data  (so, 0x54);
+	}
+
 	/* origin is top left (set to 1 for bottom left) */
 	so_method(so, screen->tesla, NV50TCL_Y_ORIGIN_BOTTOM, 1);
 	so_data  (so, 0);
@@ -370,7 +371,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 		return NULL;
 	}
 
-	for (i = 0; i < 2; i++) {
+	for (i = 0; i < 3; i++) {
 		ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (128 * 4) * 4,
 				     &screen->constbuf_parm[i]);
 		if (ret) {
@@ -406,22 +407,33 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
 	so_data  (so, 0x00000001 | (NV50_CB_PMISC << 12));
 	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
+	so_data  (so, 0x00000021 | (NV50_CB_PMISC << 12));
+	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
 	so_data  (so, 0x00000031 | (NV50_CB_PMISC << 12));
 
 	so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
-	so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM |
-		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
-	so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM |
-		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
+	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
+	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
 	so_data  (so, (NV50_CB_PVP << 16) | 0x00000800);
 	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
 	so_data  (so, 0x00000101 | (NV50_CB_PVP << 12));
 
 	so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
-	so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM |
-		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
-	so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM |
-		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
+	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
+	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
+	so_data  (so, (NV50_CB_PGP << 16) | 0x00000800);
+	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
+	so_data  (so, 0x00000121 | (NV50_CB_PGP << 12));
+
+	so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
+	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
+	so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0,
+		  NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
 	so_data  (so, (NV50_CB_PFP << 16) | 0x00000800);
 	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
 	so_data  (so, 0x00000131 | (NV50_CB_PFP << 12));
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index a038a4e3c2a..7a155ca9c45 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -18,10 +18,10 @@ struct nv50_screen {
 	struct nouveau_notifier *sync;
 
 	struct nouveau_bo *constbuf_misc[1];
-	struct nouveau_bo *constbuf_parm[2];
+	struct nouveau_bo *constbuf_parm[PIPE_SHADER_TYPES];
 
 	struct nouveau_resource *immd_heap[1];
-	struct nouveau_resource *parm_heap[2];
+	struct nouveau_resource *parm_heap[PIPE_SHADER_TYPES];
 
 	struct nouveau_bo *tic;
 	struct nouveau_bo *tsc;
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 1f67df814b1..6ab33be663d 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -531,7 +531,7 @@ nv50_vp_state_delete(struct pipe_context *pipe, void *hwcso)
 	struct nv50_program *p = hwcso;
 
 	nv50_program_destroy(nv50, p);
-	FREE((void*)p->pipe.tokens);
+	FREE((void *)p->pipe.tokens);
 	FREE(p);
 }
 
@@ -563,7 +563,39 @@ nv50_fp_state_delete(struct pipe_context *pipe, void *hwcso)
 	struct nv50_program *p = hwcso;
 
 	nv50_program_destroy(nv50, p);
-	FREE((void*)p->pipe.tokens);
+	FREE((void *)p->pipe.tokens);
+	FREE(p);
+}
+
+static void *
+nv50_gp_state_create(struct pipe_context *pipe,
+		     const struct pipe_shader_state *cso)
+{
+	struct nv50_program *p = CALLOC_STRUCT(nv50_program);
+
+	p->pipe.tokens = tgsi_dup_tokens(cso->tokens);
+	p->type = PIPE_SHADER_GEOMETRY;
+	tgsi_scan_shader(p->pipe.tokens, &p->info);
+	return (void *)p;
+}
+
+static void
+nv50_gp_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
+
+	nv50->fragprog = hwcso;
+	nv50->dirty |= NV50_NEW_GEOMPROG;
+}
+
+static void
+nv50_gp_state_delete(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nv50_program *p = hwcso;
+
+	nv50_program_destroy(nv50, p);
+	FREE((void *)p->pipe.tokens);
 	FREE(p);
 }
 
@@ -585,17 +617,21 @@ nv50_set_clip_state(struct pipe_context *pipe,
 
 static void
 nv50_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
-			 const struct pipe_constant_buffer *buf )
+			 struct pipe_buffer *buf )
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
 
 	if (shader == PIPE_SHADER_VERTEX) {
-		nv50->constbuf[PIPE_SHADER_VERTEX] = buf->buffer;
+		nv50->constbuf[PIPE_SHADER_VERTEX] = buf;
 		nv50->dirty |= NV50_NEW_VERTPROG_CB;
 	} else
 	if (shader == PIPE_SHADER_FRAGMENT) {
-		nv50->constbuf[PIPE_SHADER_FRAGMENT] = buf->buffer;
+		nv50->constbuf[PIPE_SHADER_FRAGMENT] = buf;
 		nv50->dirty |= NV50_NEW_FRAGPROG_CB;
+	} else
+	if (shader == PIPE_SHADER_GEOMETRY) {
+		nv50->constbuf[PIPE_SHADER_GEOMETRY] = buf;
+		nv50->dirty |= NV50_NEW_GEOMPROG_CB;
 	}
 }
 
@@ -696,6 +732,10 @@ nv50_init_state_functions(struct nv50_context *nv50)
 	nv50->pipe.bind_fs_state = nv50_fp_state_bind;
 	nv50->pipe.delete_fs_state = nv50_fp_state_delete;
 
+	nv50->pipe.create_gs_state = nv50_gp_state_create;
+	nv50->pipe.bind_gs_state = nv50_gp_state_bind;
+	nv50->pipe.delete_gs_state = nv50_gp_state_delete;
+
 	nv50->pipe.set_blend_color = nv50_set_blend_color;
 	nv50->pipe.set_clip_state = nv50_set_clip_state;
 	nv50->pipe.set_constant_buffer = nv50_set_constant_buffer;
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index f83232f43cf..fcd07b59cd6 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -199,6 +199,8 @@ nv50_state_emit(struct nv50_context *nv50)
 			nv50->state.dirty |= NV50_NEW_VERTPROG;
 		if (nv50->state.fragprog)
 			nv50->state.dirty |= NV50_NEW_FRAGPROG;
+		if (nv50->state.geomprog)
+			nv50->state.dirty |= NV50_NEW_GEOMPROG;
 		if (nv50->state.rast)
 			nv50->state.dirty |= NV50_NEW_RASTERIZER;
 		if (nv50->state.blend_colour)
@@ -228,9 +230,14 @@ nv50_state_emit(struct nv50_context *nv50)
 		so_emit(chan, nv50->state.vertprog);
 	if (nv50->state.dirty & NV50_NEW_FRAGPROG)
 		so_emit(chan, nv50->state.fragprog);
+	if (nv50->state.dirty & NV50_NEW_GEOMPROG && nv50->state.geomprog)
+		so_emit(chan, nv50->state.geomprog);
 	if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
-				 NV50_NEW_RASTERIZER))
-		so_emit(chan, nv50->state.programs);
+				 NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER))
+		so_emit(chan, nv50->state.fp_linkage);
+	if ((nv50->state.dirty & (NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG))
+	    && nv50->state.gp_linkage)
+		so_emit(chan, nv50->state.gp_linkage);
 	if (nv50->state.dirty & NV50_NEW_RASTERIZER)
 		so_emit(chan, nv50->state.rast);
 	if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR)
@@ -291,9 +298,15 @@ nv50_state_validate(struct nv50_context *nv50)
 	if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB))
 		nv50_fragprog_validate(nv50);
 
+	if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB))
+		nv50_geomprog_validate(nv50);
+
 	if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
-			   NV50_NEW_RASTERIZER))
-		nv50_linkage_validate(nv50);
+			   NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER))
+		nv50_fp_linkage_validate(nv50);
+
+	if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_VERTPROG))
+		nv50_gp_linkage_validate(nv50);
 
 	if (nv50->dirty & NV50_NEW_RASTERIZER)
 		so_ref(nv50->rasterizer->so, &nv50->state.rast);
@@ -400,8 +413,9 @@ viewport_uptodate:
 		for (i = 0; i < PIPE_SHADER_TYPES; ++i)
 			nr += nv50->sampler_nr[i];
 
-		so = so_new(1+ 5 * PIPE_SHADER_TYPES, 1+ 19 * PIPE_SHADER_TYPES
-					+ nr * 8, PIPE_SHADER_TYPES * 2);
+		so = so_new(1 + 5 * PIPE_SHADER_TYPES,
+			    1 + 19 * PIPE_SHADER_TYPES + nr * 8,
+			    PIPE_SHADER_TYPES * 2);
 
 		nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX);
 		nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT);
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index bef548b7286..871536dca9b 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -155,7 +155,7 @@ static boolean
 nv50_validate_textures(struct nv50_context *nv50, struct nouveau_stateobj *so,
 		       unsigned p)
 {
-	static const unsigned p_remap[PIPE_SHADER_TYPES] = { 0, 2 };
+	static const unsigned p_remap[PIPE_SHADER_TYPES] = { 0, 2, 1 };
 
 	struct nouveau_grobj *eng2d = nv50->screen->eng2d;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index f2e510fba61..89a94d2fe81 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -55,6 +55,14 @@ nv50_prim(unsigned mode)
 	case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
 	case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
 	case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
+	case PIPE_PRIM_LINES_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY;
+	case PIPE_PRIM_LINE_STRIP_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY;
+	case PIPE_PRIM_TRIANGLES_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY;
+	case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY;
 	default:
 		break;
 	}
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index ffe066d5369..c14414fff6b 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -27,9 +27,9 @@
 
 static void r300_blitter_save_states(struct r300_context* r300)
 {
-    util_blitter_save_blend(r300->blitter, r300->blend_state);
-    util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state);
-    util_blitter_save_rasterizer(r300->blitter, r300->rs_state);
+    util_blitter_save_blend(r300->blitter, r300->blend_state.state);
+    util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
+    util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
     util_blitter_save_fragment_shader(r300->blitter, r300->fs);
     util_blitter_save_vertex_shader(r300->blitter, r300->vs);
 }
diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c
index 51fdb82ff34..92de297ef1d 100644
--- a/src/gallium/drivers/r300/r300_chipset.c
+++ b/src/gallium/drivers/r300/r300_chipset.c
@@ -33,6 +33,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
     /* Reasonable defaults */
     caps->num_vert_fpus = 4;
     caps->has_tcl = debug_get_bool_option("RADEON_NO_TCL", FALSE) ? FALSE : TRUE;
+    caps->is_r400 = FALSE;
     caps->is_r500 = FALSE;
     caps->high_second_pipe = FALSE;
 
@@ -123,6 +124,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x4A54:
             caps->family = CHIP_FAMILY_R420;
             caps->num_vert_fpus = 6;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x5548:
@@ -136,6 +138,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x5D57:
             caps->family = CHIP_FAMILY_R423;
             caps->num_vert_fpus = 6;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x554C:
@@ -147,6 +150,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x5D4A:
             caps->family = CHIP_FAMILY_R430;
             caps->num_vert_fpus = 6;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x5D4C:
@@ -157,6 +161,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x5D52:
             caps->family = CHIP_FAMILY_R480;
             caps->num_vert_fpus = 6;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x4B48:
@@ -166,6 +171,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x4B4C:
             caps->family = CHIP_FAMILY_R481;
             caps->num_vert_fpus = 6;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x5E4C:
@@ -182,6 +188,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x5E4D:
             caps->family = CHIP_FAMILY_RV410;
             caps->num_vert_fpus = 6;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x5954:
@@ -212,6 +219,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x791F:
             caps->family = CHIP_FAMILY_RS690;
             caps->has_tcl = FALSE;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x793F:
@@ -219,6 +227,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x7942:
             caps->family = CHIP_FAMILY_RS600;
             caps->has_tcl = FALSE;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x796C:
@@ -227,6 +236,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x796F:
             caps->family = CHIP_FAMILY_RS740;
             caps->has_tcl = FALSE;
+            caps->is_r400 = TRUE;
             break;
 
         case 0x7100:
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index 0633a8b8a72..28084864929 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -40,11 +40,18 @@ struct r300_capabilities {
     unsigned num_z_pipes;
     /* Whether or not TCL is physically present */
     boolean has_tcl;
+    /* Whether or not this is R400. The differences compared to their R3xx
+     * cousins are:
+     * - Extended fragment shader registers
+     * - Blend LTE/GTE thresholds */
+    boolean is_r400;
     /* Whether or not this is an RV515 or newer; R500s have many differences
      * that require extra consideration, compared to their R3xx cousins:
      * - Extra bit of width and height on texture sizes
      * - Blend color is split across two registers
-     * - Universal Shader (US) block used for fragment shaders */
+     * - Blend LTE/GTE thresholds
+     * - Universal Shader (US) block used for fragment shaders
+     * - FP16 blending and multisampling */
     boolean is_r500;
     /* Whether or not the second pixel pipe is accessed with the high bit */
     boolean high_second_pipe;
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index d5c2d63d393..5e4f6552c36 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -30,6 +30,7 @@
 
 #include "r300_blit.h"
 #include "r300_context.h"
+#include "r300_emit.h"
 #include "r300_flush.h"
 #include "r300_query.h"
 #include "r300_render.h"
@@ -69,11 +70,13 @@ static void r300_destroy_context(struct pipe_context* context)
         FREE(query);
     }
 
-    FREE(r300->blend_color_state);
+    FREE(r300->blend_color_state.state);
+    FREE(r300->clip_state.state);
     FREE(r300->rs_block);
-    FREE(r300->scissor_state);
+    FREE(r300->scissor_state.state);
     FREE(r300->vertex_info);
-    FREE(r300->viewport_state);
+    FREE(r300->viewport_state.state);
+    FREE(r300->ztop_state.state);
     FREE(r300);
 }
 
@@ -107,6 +110,35 @@ static void r300_flush_cb(void *data)
     cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
 }
 
+#define R300_INIT_ATOM(atomname, atomsize) \
+    r300->atomname##_state.name = #atomname; \
+    r300->atomname##_state.state = NULL; \
+    r300->atomname##_state.size = atomsize; \
+    r300->atomname##_state.emit = r300_emit_##atomname##_state; \
+    r300->atomname##_state.dirty = FALSE; \
+    insert_at_tail(&r300->atom_list, &r300->atomname##_state);
+
+static void r300_setup_atoms(struct r300_context* r300)
+{
+    /* Create the actual atom list.
+     *
+     * Each atom is examined and emitted in the order it appears here, which
+     * can affect performance and conformance if not handled with care.
+     *
+     * Some atoms never change size, others change every emit. This is just
+     * an upper bound on each atom, to keep the emission machinery from
+     * underallocating space. */
+    make_empty_list(&r300->atom_list);
+    R300_INIT_ATOM(ztop, 2);
+    R300_INIT_ATOM(blend, 8);
+    R300_INIT_ATOM(blend_color, 3);
+    R300_INIT_ATOM(clip, 29);
+    R300_INIT_ATOM(dsa, 8);
+    R300_INIT_ATOM(rs, 22);
+    R300_INIT_ATOM(scissor, 3);
+    R300_INIT_ATOM(viewport, 9);
+}
+
 struct pipe_context* r300_create_context(struct pipe_screen* screen,
                                          struct radeon_winsys* radeon_winsys)
 {
@@ -155,11 +187,15 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     r300->shader_hash_table = util_hash_table_create(r300_shader_key_hash,
         r300_shader_key_compare);
 
-    r300->blend_color_state = CALLOC_STRUCT(r300_blend_color_state);
+    r300_setup_atoms(r300);
+
+    r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
+    r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state);
     r300->rs_block = CALLOC_STRUCT(r300_rs_block);
-    r300->scissor_state = CALLOC_STRUCT(r300_scissor_state);
+    r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
     r300->vertex_info = CALLOC_STRUCT(r300_vertex_info);
-    r300->viewport_state = CALLOC_STRUCT(r300_viewport_state);
+    r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
+    r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
 
     /* Open up the OQ BO. */
     r300->oqbo = screen->buffer_create(screen, 4096,
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 232530b7dc4..682b9179c83 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -30,9 +30,28 @@
 #include "pipe/p_context.h"
 #include "pipe/p_inlines.h"
 
+struct r300_context;
+
 struct r300_fragment_shader;
 struct r300_vertex_shader;
 
+struct r300_atom {
+    /* List pointers. */
+    struct r300_atom *prev, *next;
+    /* Name, for debugging. */
+    const char* name;
+    /* Opaque state. */
+    void* state;
+    /* Emit the state to the context. */
+    void (*emit)(struct r300_context*, void*);
+    /* Upper bound on number of dwords to emit. */
+    unsigned size;
+    /* Whether this atom should be emitted. */
+    boolean dirty;
+    /* Another dirty flag that is never automatically cleared. */
+    boolean always_dirty;
+};
+
 struct r300_blend_state {
     uint32_t blend_control;       /* R300_RB3D_CBLEND: 0x4e04 */
     uint32_t alpha_blend_control; /* R300_RB3D_ABLEND: 0x4e08 */
@@ -62,11 +81,6 @@ struct r300_rs_state {
     /* Draw-specific rasterizer state */
     struct pipe_rasterizer_state rs;
 
-    /* Whether or not to enable the VTE. This is referenced at the very
-     * last moment during emission of VTE state, to decide whether or not
-     * the VTE should be used for transformation. */
-    boolean enable_vte;
-
     uint32_t vap_control_status;    /* R300_VAP_CNTL_STATUS: 0x2140 */
     uint32_t point_size;            /* R300_GA_POINT_SIZE: 0x421c */
     uint32_t point_minmax;          /* R300_GA_POINT_MINMAX: 0x4230 */
@@ -102,19 +116,6 @@ struct r300_sampler_state {
     unsigned min_lod, max_lod;
 };
 
-struct r300_scissor_regs {
-    uint32_t top_left;     /* R300_SC_SCISSORS_TL: 0x43e0 */
-    uint32_t bottom_right; /* R300_SC_SCISSORS_BR: 0x43e4 */
-
-    /* Whether everything is culled by scissoring. */
-    boolean empty_area;
-};
-
-struct r300_scissor_state {
-    struct r300_scissor_regs framebuffer;
-    struct r300_scissor_regs scissor;
-};
-
 struct r300_texture_state {
     uint32_t format0; /* R300_TX_FORMAT0: 0x4480 */
     uint32_t format1; /* R300_TX_FORMAT1: 0x44c0 */
@@ -135,24 +136,17 @@ struct r300_ztop_state {
     uint32_t z_buffer_top;      /* R300_ZB_ZTOP: 0x4f14 */
 };
 
-#define R300_NEW_BLEND           0x00000001
-#define R300_NEW_BLEND_COLOR     0x00000002
-#define R300_NEW_CLIP            0x00000004
-#define R300_NEW_DSA             0x00000008
 #define R300_NEW_FRAMEBUFFERS    0x00000010
 #define R300_NEW_FRAGMENT_SHADER 0x00000020
 #define R300_NEW_FRAGMENT_SHADER_CONSTANTS    0x00000040
-#define R300_NEW_RASTERIZER      0x00000080
 #define R300_NEW_RS_BLOCK        0x00000100
 #define R300_NEW_SAMPLER         0x00000200
 #define R300_ANY_NEW_SAMPLERS    0x0001fe00
-#define R300_NEW_SCISSOR         0x00020000
 #define R300_NEW_TEXTURE         0x00040000
 #define R300_ANY_NEW_TEXTURES    0x03fc0000
 #define R300_NEW_VERTEX_FORMAT   0x04000000
 #define R300_NEW_VERTEX_SHADER   0x08000000
 #define R300_NEW_VERTEX_SHADER_CONSTANTS    0x10000000
-#define R300_NEW_VIEWPORT        0x20000000
 #define R300_NEW_QUERY           0x40000000
 #define R300_NEW_KITCHEN_SINK    0x7fffffff
 
@@ -194,6 +188,12 @@ struct r300_query {
     struct r300_query* next;
 };
 
+enum r300_buffer_tiling {
+    R300_BUFFER_LINEAR = 0,
+    R300_BUFFER_TILED,
+    R300_BUFFER_SQUARETILED
+};
+
 struct r300_texture {
     /* Parent class */
     struct pipe_texture tex;
@@ -230,6 +230,9 @@ struct r300_texture {
 
     /* Registers carrying texture format data. */
     struct r300_texture_state state;
+
+    /* Buffer tiling */
+    enum r300_buffer_tiling microtile, macrotile;
 };
 
 struct r300_vertex_info {
@@ -273,38 +276,40 @@ struct r300_context {
     struct r300_vertex_info* vertex_info;
 
     /* Various CSO state objects. */
+    /* Beginning of atom list. */
+    struct r300_atom atom_list;
     /* Blend state. */
-    struct r300_blend_state* blend_state;
+    struct r300_atom blend_state;
     /* Blend color state. */
-    struct r300_blend_color_state* blend_color_state;
+    struct r300_atom blend_color_state;
     /* User clip planes. */
-    struct pipe_clip_state clip_state;
+    struct r300_atom clip_state;
     /* Shader constants. */
     struct r300_constant_buffer shader_constants[PIPE_SHADER_TYPES];
     /* Depth, stencil, and alpha state. */
-    struct r300_dsa_state* dsa_state;
+    struct r300_atom dsa_state;
     /* Fragment shader. */
     struct r300_fragment_shader* fs;
     /* Framebuffer state. We currently don't need our own version of this. */
     struct pipe_framebuffer_state framebuffer_state;
     /* Rasterizer state. */
-    struct r300_rs_state* rs_state;
+    struct r300_atom rs_state;
     /* RS block state. */
     struct r300_rs_block* rs_block;
     /* Sampler states. */
     struct r300_sampler_state* sampler_states[8];
     int sampler_count;
     /* Scissor state. */
-    struct r300_scissor_state* scissor_state;
+    struct r300_atom scissor_state;
     /* Texture states. */
     struct r300_texture* textures[8];
     int texture_count;
     /* Vertex shader. */
     struct r300_vertex_shader* vs;
     /* Viewport state. */
-    struct r300_viewport_state* viewport_state;
+    struct r300_atom viewport_state;
     /* ZTOP state. */
-    struct r300_ztop_state ztop_state;
+    struct r300_atom ztop_state;
 
     /* Vertex buffers for Gallium. */
     struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
@@ -317,6 +322,8 @@ struct r300_context {
     uint32_t dirty_state;
     /* Flag indicating whether or not the HW is dirty. */
     uint32_t dirty_hw;
+    /* Whether the TCL engine should be in bypass mode. */
+    boolean tcl_bypass;
 
     /** Combination of DBG_xxx flags */
     unsigned debug;
diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index d142fee0502..151f72b0fe4 100644
--- a/src/gallium/drivers/r300/r300_cs.h
+++ b/src/gallium/drivers/r300/r300_cs.h
@@ -52,7 +52,7 @@
 #define CS_LOCALS(context) \
     struct r300_context* const cs_context_copy = (context); \
     struct radeon_winsys* cs_winsys = cs_context_copy->winsys; \
-    int cs_count = 0;
+    int cs_count = 0; (void) cs_count;
 
 #define CHECK_CS(size) \
     assert(cs_winsys->check_cs(cs_winsys, (size)))
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index f8bfa714fef..9f93327e598 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -25,6 +25,7 @@
 
 #include "util/u_format.h"
 #include "util/u_math.h"
+#include "util/u_simple_list.h"
 
 #include "r300_context.h"
 #include "r300_cs.h"
@@ -36,11 +37,13 @@
 #include "r300_texture.h"
 #include "r300_vs.h"
 
-void r300_emit_blend_state(struct r300_context* r300,
-                           struct r300_blend_state* blend)
+void r300_emit_blend_state(struct r300_context* r300, void* state)
 {
+    struct r300_blend_state* blend = (struct r300_blend_state*)state;
     CS_LOCALS(r300);
+
     BEGIN_CS(8);
+    OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
     OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3);
     if (r300->framebuffer_state.nr_cbufs) {
         OUT_CS(blend->blend_control);
@@ -52,14 +55,13 @@ void r300_emit_blend_state(struct r300_context* r300,
         OUT_CS(0);
         /* XXX also disable fastfill here once it's supported */
     }
-    OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
     OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither);
     END_CS;
 }
 
-void r300_emit_blend_color_state(struct r300_context* r300,
-                                 struct r300_blend_color_state* bc)
+void r300_emit_blend_color_state(struct r300_context* r300, void* state)
 {
+    struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
 
@@ -76,9 +78,9 @@ void r300_emit_blend_color_state(struct r300_context* r300,
     }
 }
 
-void r300_emit_clip_state(struct r300_context* r300,
-                          struct pipe_clip_state* clip)
+void r300_emit_clip_state(struct r300_context* r300, void* state)
 {
+    struct pipe_clip_state* clip = (struct pipe_clip_state*)state;
     int i;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
@@ -106,13 +108,13 @@ void r300_emit_clip_state(struct r300_context* r300,
 
 }
 
-void r300_emit_dsa_state(struct r300_context* r300,
-                           struct r300_dsa_state* dsa)
+void r300_emit_dsa_state(struct r300_context* r300, void* state)
 {
+    struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
 
-    BEGIN_CS(r300screen->caps->is_r500 ? 10 : 8);
+    BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6);
     OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
 
     /* not needed since we use the 8bit alpha ref */
@@ -131,7 +133,6 @@ void r300_emit_dsa_state(struct r300_context* r300,
     }
 
     OUT_CS(dsa->stencil_ref_mask);
-    OUT_CS_REG(R300_ZB_ZTOP, r300->ztop_state.z_buffer_top);
 
     /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
     if (r300screen->caps->is_r500) {
@@ -145,6 +146,8 @@ static const float * get_shader_constant(
     struct rc_constant * constant,
     struct r300_constant_buffer * externals)
 {
+    struct r300_viewport_state* viewport =
+        (struct r300_viewport_state*)r300->viewport_state.state;
     static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
     struct pipe_texture *tex;
 
@@ -167,30 +170,28 @@ static const float * get_shader_constant(
 
                 /* Texture compare-fail value. */
                 /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient,
-                 * this is always (0,0,0,0). */
+                 * this is always (0,0,0,0), right? */
                 case RC_STATE_SHADOW_AMBIENT:
                     vec[3] = 0;
                     break;
 
                 case RC_STATE_R300_VIEWPORT_SCALE:
-                    if (r300->rs_state->enable_vte) {
-                        vec[0] = r300->viewport_state->xscale;
-                        vec[1] = r300->viewport_state->yscale;
-                        vec[2] = r300->viewport_state->zscale;
-                    } else {
+                    if (r300->tcl_bypass) {
                         vec[0] = 1;
                         vec[1] = 1;
                         vec[2] = 1;
+                    } else {
+                        vec[0] = viewport->xscale;
+                        vec[1] = viewport->yscale;
+                        vec[2] = viewport->zscale;
                     }
                     break;
 
                 case RC_STATE_R300_VIEWPORT_OFFSET:
-                    if (r300->rs_state->enable_vte) {
-                        vec[0] = r300->viewport_state->xoffset;
-                        vec[1] = r300->viewport_state->yoffset;
-                        vec[2] = r300->viewport_state->zoffset;
-                    } else {
-                        /* Zeros. */
+                    if (!r300->tcl_bypass) {
+                        vec[0] = viewport->xoffset;
+                        vec[1] = viewport->yoffset;
+                        vec[2] = viewport->zoffset;
                     }
                     break;
 
@@ -419,8 +420,10 @@ void r300_emit_fb_state(struct r300_context* r300,
 
         OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1);
         OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
-                     r300_translate_colorformat(tex->tex.format), 0,
-                     RADEON_GEM_DOMAIN_VRAM, 0);
+                     r300_translate_colorformat(tex->tex.format) |
+                     R300_COLOR_TILE(tex->macrotile) |
+                     R300_COLOR_MICROTILE(tex->microtile),
+                     0, RADEON_GEM_DOMAIN_VRAM, 0);
 
         OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i),
             r300_translate_out_fmt(surf->format));
@@ -443,8 +446,10 @@ void r300_emit_fb_state(struct r300_context* r300,
         OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format));
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
-        OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level], 0,
-                     RADEON_GEM_DOMAIN_VRAM, 0);
+        OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
+                     R300_DEPTHMACROTILE(tex->macrotile) |
+                     R300_DEPTHMICROTILE(tex->microtile),
+                     0, RADEON_GEM_DOMAIN_VRAM, 0);
     }
 
     END_CS;
@@ -576,8 +581,9 @@ void r300_emit_query_end(struct r300_context* r300)
         r300_emit_query_finish(r300, query);
 }
 
-void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs)
+void r300_emit_rs_state(struct r300_context* r300, void* state)
 {
+    struct r300_rs_state* rs = (struct r300_rs_state*)state;
     CS_LOCALS(r300);
 
     BEGIN_CS(22);
@@ -640,26 +646,47 @@ void r300_emit_rs_block_state(struct r300_context* r300,
     END_CS;
 }
 
-static void r300_emit_scissor_regs(struct r300_context* r300,
-                                   struct r300_scissor_regs* scissor)
+void r300_emit_scissor_state(struct r300_context* r300, void* state)
 {
+    unsigned minx, miny, maxx, maxy;
+    uint32_t top_left, bottom_right;
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state;
     CS_LOCALS(r300);
 
-    BEGIN_CS(3);
-    OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
-    OUT_CS(scissor->top_left);
-    OUT_CS(scissor->bottom_right);
-    END_CS;
-}
+    minx = miny = 0;
+    maxx = r300->framebuffer_state.width;
+    maxy = r300->framebuffer_state.height;
 
-void r300_emit_scissor_state(struct r300_context* r300,
-                             struct r300_scissor_state* scissor)
-{
-    if (r300->rs_state->rs.scissor) {
-        r300_emit_scissor_regs(r300, &scissor->scissor);
+    if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) {
+        minx = MAX2(minx, scissor->minx);
+        miny = MAX2(miny, scissor->miny);
+        maxx = MIN2(maxx, scissor->maxx);
+        maxy = MIN2(maxy, scissor->maxy);
+    }
+
+    if (r300screen->caps->is_r500) {
+        top_left =
+            (minx << R300_SCISSORS_X_SHIFT) |
+            (miny << R300_SCISSORS_Y_SHIFT);
+        bottom_right =
+            ((maxx - 1) << R300_SCISSORS_X_SHIFT) |
+            ((maxy - 1) << R300_SCISSORS_Y_SHIFT);
     } else {
-        r300_emit_scissor_regs(r300, &scissor->framebuffer);
+        /* Offset of 1440 in non-R500 chipsets. */
+        top_left =
+            ((minx + 1440) << R300_SCISSORS_X_SHIFT) |
+            ((miny + 1440) << R300_SCISSORS_Y_SHIFT);
+        bottom_right =
+            (((maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
+            (((maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
     }
+
+    BEGIN_CS(3);
+    OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
+    OUT_CS(top_left);
+    OUT_CS(bottom_right);
+    END_CS;
 }
 
 void r300_emit_texture(struct r300_context* r300,
@@ -695,8 +722,10 @@ void r300_emit_texture(struct r300_context* r300,
     OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1);
     OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2);
     OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1);
-    OUT_CS_RELOC(tex->buffer, 0,
-            RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0);
+    OUT_CS_RELOC(tex->buffer,
+                 R300_TXO_MACRO_TILE(tex->macrotile) |
+                 R300_TXO_MICRO_TILE(tex->microtile),
+                 RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0);
     END_CS;
 }
 
@@ -762,32 +791,6 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
     END_CS;
 }
 
-#if 0
-void r300_emit_draw_packet(struct r300_context* r300)
-{
-    CS_LOCALS(r300);
-
-    DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, "
-            "vertex size %d\n", r300->vbo,
-            r300->vertex_info->vinfo.size);
-    /* Set the pointer to our vertex buffer. The emitted values are this:
-     * PACKET3 [3D_LOAD_VBPNTR]
-     * COUNT   [1]
-     * FORMAT  [size | stride << 8]
-     * OFFSET  [offset into BO]
-     * VBPNTR  [relocated BO]
-     */
-    BEGIN_CS(7);
-    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
-    OUT_CS(1);
-    OUT_CS(r300->vertex_info->vinfo.size |
-            (r300->vertex_info->vinfo.size << 8));
-    OUT_CS(r300->vbo_offset);
-    OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0);
-    END_CS;
-}
-#endif
-
 void r300_emit_vertex_format_state(struct r300_context* r300)
 {
     int i;
@@ -912,26 +915,27 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300,
     END_CS;
 }
 
-void r300_emit_viewport_state(struct r300_context* r300,
-                              struct r300_viewport_state* viewport)
+void r300_emit_viewport_state(struct r300_context* r300, void* state)
 {
+    struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
     CS_LOCALS(r300);
 
-    BEGIN_CS(9);
-    OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
-    OUT_CS_32F(viewport->xscale);
-    OUT_CS_32F(viewport->xoffset);
-    OUT_CS_32F(viewport->yscale);
-    OUT_CS_32F(viewport->yoffset);
-    OUT_CS_32F(viewport->zscale);
-    OUT_CS_32F(viewport->zoffset);
-
-    if (r300->rs_state->enable_vte) {
-        OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
-    } else {
+    if (r300->tcl_bypass) {
+        BEGIN_CS(2);
         OUT_CS_REG(R300_VAP_VTE_CNTL, 0);
+        END_CS;
+    } else {
+        BEGIN_CS(9);
+        OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
+        OUT_CS_32F(viewport->xscale);
+        OUT_CS_32F(viewport->xoffset);
+        OUT_CS_32F(viewport->yscale);
+        OUT_CS_32F(viewport->yoffset);
+        OUT_CS_32F(viewport->zscale);
+        OUT_CS_32F(viewport->zoffset);
+        OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
+        END_CS;
     }
-    END_CS;
 }
 
 void r300_emit_texture_count(struct r300_context* r300)
@@ -955,6 +959,16 @@ void r300_emit_texture_count(struct r300_context* r300)
 
 }
 
+void r300_emit_ztop_state(struct r300_context* r300, void* state)
+{
+    struct r300_ztop_state* ztop = (struct r300_ztop_state*)state;
+    CS_LOCALS(r300);
+
+    BEGIN_CS(2);
+    OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top);
+    END_CS;
+}
+
 void r300_flush_textures(struct r300_context* r300)
 {
     CS_LOCALS(r300);
@@ -978,18 +992,24 @@ void r300_emit_dirty_state(struct r300_context* r300)
 {
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     struct r300_texture* tex;
-    int i, dirty_tex = 0;
+    struct r300_atom* atom;
+    unsigned i, dwords = 1024;
+    int dirty_tex = 0;
     boolean invalid = FALSE;
 
-    if (!(r300->dirty_state)) {
-        return;
+    /* Check the required number of dwords against the space remaining in the
+     * current CS object. If we need more, then flush. */
+
+    foreach(atom, &r300->atom_list) {
+        if (atom->dirty || atom->always_dirty) {
+            dwords += atom->size;
+        }
     }
 
-    /* Check size of CS. */
-    /* Make sure we have at least 8*1024 spare dwords. */
+    /* Make sure we have at least 2*1024 spare dwords. */
     /* XXX It would be nice to know the number of dwords we really need to
      * XXX emit. */
-    if (!r300->winsys->check_cs(r300->winsys, 8*1024)) {
+    if (!r300->winsys->check_cs(r300->winsys, dwords)) {
         r300->context.flush(&r300->context, 0, NULL);
     }
 
@@ -1029,10 +1049,12 @@ validate:
         }
     }
     /* ...occlusion query buffer... */
-    if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo,
-                0, RADEON_GEM_DOMAIN_GTT)) {
-        r300->context.flush(&r300->context, 0, NULL);
-        goto validate;
+    if (r300->dirty_state & R300_NEW_QUERY) {
+        if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo,
+                    0, RADEON_GEM_DOMAIN_GTT)) {
+            r300->context.flush(&r300->context, 0, NULL);
+            goto validate;
+        }
     }
     /* ...and vertex buffer. */
     if (r300->vbo) {
@@ -1060,24 +1082,11 @@ validate:
         r300->dirty_state &= ~R300_NEW_QUERY;
     }
 
-    if (r300->dirty_state & R300_NEW_BLEND) {
-        r300_emit_blend_state(r300, r300->blend_state);
-        r300->dirty_state &= ~R300_NEW_BLEND;
-    }
-
-    if (r300->dirty_state & R300_NEW_BLEND_COLOR) {
-        r300_emit_blend_color_state(r300, r300->blend_color_state);
-        r300->dirty_state &= ~R300_NEW_BLEND_COLOR;
-    }
-
-    if (r300->dirty_state & R300_NEW_CLIP) {
-        r300_emit_clip_state(r300, &r300->clip_state);
-        r300->dirty_state &= ~R300_NEW_CLIP;
-    }
-
-    if (r300->dirty_state & R300_NEW_DSA) {
-        r300_emit_dsa_state(r300, r300->dsa_state);
-        r300->dirty_state &= ~R300_NEW_DSA;
+    foreach(atom, &r300->atom_list) {
+        if (atom->dirty || atom->always_dirty) {
+            atom->emit(r300, atom->state);
+            atom->dirty = FALSE;
+        }
     }
 
     if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) {
@@ -1106,21 +1115,11 @@ validate:
         r300->dirty_state &= ~R300_NEW_FRAMEBUFFERS;
     }
 
-    if (r300->dirty_state & R300_NEW_RASTERIZER) {
-        r300_emit_rs_state(r300, r300->rs_state);
-        r300->dirty_state &= ~R300_NEW_RASTERIZER;
-    }
-
     if (r300->dirty_state & R300_NEW_RS_BLOCK) {
         r300_emit_rs_block_state(r300, r300->rs_block);
         r300->dirty_state &= ~R300_NEW_RS_BLOCK;
     }
 
-    if (r300->dirty_state & R300_NEW_SCISSOR) {
-        r300_emit_scissor_state(r300, r300->scissor_state);
-        r300->dirty_state &= ~R300_NEW_SCISSOR;
-    }
-
     /* Samplers and textures are tracked separately but emitted together. */
     if (r300->dirty_state &
             (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
@@ -1142,11 +1141,6 @@ validate:
         r300->dirty_state &= ~(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES);
     }
 
-    if (r300->dirty_state & R300_NEW_VIEWPORT) {
-        r300_emit_viewport_state(r300, r300->viewport_state);
-        r300->dirty_state &= ~R300_NEW_VIEWPORT;
-    }
-
     if (dirty_tex) {
         r300_flush_textures(r300);
     }
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 3797d3d332a..05a6bfeae86 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -31,17 +31,13 @@ struct r300_vertex_program_code;
 
 void r300_emit_aos(struct r300_context* r300, unsigned offset);
 
-void r300_emit_blend_state(struct r300_context* r300,
-                           struct r300_blend_state* blend);
+void r300_emit_blend_state(struct r300_context* r300, void* state);
 
-void r300_emit_blend_color_state(struct r300_context* r300,
-                                 struct r300_blend_color_state* bc);
+void r300_emit_blend_color_state(struct r300_context* r300, void* state);
 
-void r300_emit_clip_state(struct r300_context* r300,
-                          struct pipe_clip_state* clip);
+void r300_emit_clip_state(struct r300_context* r300, void* state);
 
-void r300_emit_dsa_state(struct r300_context* r300,
-                         struct r300_dsa_state* dsa);
+void r300_emit_dsa_state(struct r300_context* r300, void* state);
 
 void r300_emit_fragment_program_code(struct r300_context* r300,
                                      struct rX00_fragment_program_code* generic_code);
@@ -63,13 +59,12 @@ void r300_emit_query_begin(struct r300_context* r300,
 
 void r300_emit_query_end(struct r300_context* r300);
 
-void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs);
+void r300_emit_rs_state(struct r300_context* r300, void* state);
 
 void r300_emit_rs_block_state(struct r300_context* r300,
                               struct r300_rs_block* rs);
 
-void r300_emit_scissor_state(struct r300_context* r300,
-                             struct r300_scissor_state* scissor);
+void r300_emit_scissor_state(struct r300_context* r300, void* state);
 
 void r300_emit_texture(struct r300_context* r300,
                        struct r300_sampler_state* sampler,
@@ -89,11 +84,12 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300,
 void r300_emit_vertex_shader(struct r300_context* r300,
                              struct r300_vertex_shader* vs);
 
-void r300_emit_viewport_state(struct r300_context* r300,
-                              struct r300_viewport_state* viewport);
+void r300_emit_viewport_state(struct r300_context* r300, void* state);
 
 void r300_emit_texture_count(struct r300_context* r300);
 
+void r300_emit_ztop_state(struct r300_context* r300, void* state);
+
 void r300_flush_textures(struct r300_context* r300);
 
 /* Emit all dirty state. */
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index 14a08241fc4..59819cb1061 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -37,8 +37,10 @@ static void r300_flush(struct pipe_context* pipe,
 {
     struct r300_context *r300 = r300_context(pipe);
     struct r300_query *query;
+    struct r300_atom *atom;
 
     CS_LOCALS(r300);
+    (void) cs_count;
     /* We probably need to flush Draw, but we may have been called from
      * within Draw. This feels kludgy, but it might be the best thing.
      *
@@ -54,7 +56,15 @@ static void r300_flush(struct pipe_context* pipe,
         r300_emit_invariant_state(r300);
         r300->dirty_state = R300_NEW_KITCHEN_SINK;
         r300->dirty_hw = 0;
+
+        /* New kitchen sink, baby. */
+        foreach(atom, &r300->atom_list) {
+            if (atom->state) {
+                atom->dirty = TRUE;
+            }
+        }
     }
+
     /* reset flushed query */
     foreach(query, &r300->query_list) {
         query->flushed = TRUE;
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 034bfc15cf9..361813891fb 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -1619,18 +1619,20 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #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)
 #       define R300_TXO_ENDIAN_WORD_SWAP         (2 << 0)
 #       define R300_TXO_ENDIAN_HALFDW_SWAP       (3 << 0)
-#       define R300_TXO_MACRO_TILE               (1 << 2)
+#       define R300_TXO_MACRO_TILE_LINEAR        (0 << 2)
+#       define R300_TXO_MACRO_TILE_TILED         (1 << 2)
+#       define R300_TXO_MACRO_TILE(x)            ((x) << 2)
 #       define R300_TXO_MICRO_TILE_LINEAR        (0 << 3)
-#       define R300_TXO_MICRO_TILE               (1 << 3)
-#       define R300_TXO_MICRO_TILE_SQUARE        (2 << 3)
+#       define R300_TXO_MICRO_TILE_TILED         (1 << 3)
+#       define R300_TXO_MICRO_TILE_TILED_SQUARE  (2 << 3)
+#       define R300_TXO_MICRO_TILE(x)            ((x) << 3)
 #       define R300_TXO_OFFSET_MASK              0xffffffe0
 #       define R300_TXO_OFFSET_SHIFT             5
-	/* END: Guess from R200 */
 
 /* 32 bit chroma key */
 #define R300_TX_CHROMA_KEY_0                      0x4580
@@ -2283,9 +2285,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_COLORPITCH_MASK              0x00003FFE
 #       define R300_COLOR_TILE_DISABLE            (0 << 16)
 #       define R300_COLOR_TILE_ENABLE             (1 << 16)
+#       define R300_COLOR_TILE(x)                 ((x) << 16)
 #       define R300_COLOR_MICROTILE_DISABLE       (0 << 17)
 #       define R300_COLOR_MICROTILE_ENABLE        (1 << 17)
 #       define R300_COLOR_MICROTILE_ENABLE_SQUARE (2 << 17) /* Only available in 16-bit */
+#       define R300_COLOR_MICROTILE(x)            ((x) << 17)
 #       define R300_COLOR_ENDIAN_NO_SWAP          (0 << 19)
 #       define R300_COLOR_ENDIAN_WORD_SWAP        (1 << 19)
 #       define R300_COLOR_ENDIAN_DWORD_SWAP       (2 << 19)
@@ -2544,9 +2548,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_DEPTHPITCH_MASK              0x00003FFC
 #       define R300_DEPTHMACROTILE_DISABLE      (0 << 16)
 #       define R300_DEPTHMACROTILE_ENABLE       (1 << 16)
+#       define R300_DEPTHMACROTILE(x)           ((x) << 16)
 #       define R300_DEPTHMICROTILE_LINEAR       (0 << 17)
 #       define R300_DEPTHMICROTILE_TILED        (1 << 17)
 #       define R300_DEPTHMICROTILE_TILED_SQUARE (2 << 17)
+#       define R300_DEPTHMICROTILE(x)           ((x) << 17)
 #       define R300_DEPTHENDIAN_NO_SWAP         (0 << 18)
 #       define R300_DEPTHENDIAN_WORD_SWAP       (1 << 18)
 #       define R300_DEPTHENDIAN_DWORD_SWAP      (2 << 18)
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index a4ac9ad9a7b..710d850163f 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -26,6 +26,8 @@
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
 
+#include "indices/u_indices.h"
+
 #include "pipe/p_inlines.h"
 
 #include "util/u_memory.h"
@@ -69,16 +71,11 @@ uint32_t r300_translate_primitive(unsigned prim)
     }
 }
 
-static boolean r300_nothing_to_draw(struct r300_context *r300)
-{
-    return r300->rs_state->rs.scissor &&
-           r300->scissor_state->scissor.empty_area;
-}
-
 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
                                             unsigned mode)
 {
-    uint32_t color_control = r300->rs_state->color_control;
+    struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
+    uint32_t color_control = rs->color_control;
 
     /* By default (see r300_state.c:r300_create_rs_state) color_control is
      * initialized to provoking the first vertex.
@@ -98,7 +95,7 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
      * ~ C.
      */
 
-    if (r300->rs_state->rs.flatshade_first) {
+    if (rs->rs.flatshade_first) {
         switch (mode) {
             case PIPE_PRIM_TRIANGLE_FAN:
                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
@@ -119,6 +116,44 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
     return color_control;
 }
 
+static void r300_emit_draw_immediate(struct r300_context *r300,
+                                     unsigned mode,
+                                     unsigned start,
+                                     unsigned count)
+{
+    struct pipe_buffer* vbo = r300->vertex_buffer[0].buffer;
+    unsigned vertex_size = r300->vertex_buffer[0].stride / sizeof(float);
+    unsigned i;
+    uint32_t* map;
+    CS_LOCALS(r300);
+
+    map = (uint32_t*)pipe_buffer_map_range(r300->context.screen, vbo,
+            start * vertex_size, count * vertex_size,
+            PIPE_BUFFER_USAGE_CPU_READ);
+
+    BEGIN_CS(10 + count * vertex_size);
+    OUT_CS_REG(R300_GA_COLOR_CONTROL,
+            r300_provoking_vertex_fixes(r300, mode));
+    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
+    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
+    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
+    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
+            r300_translate_primitive(mode));
+    //debug_printf("r300: Immd %d verts, %d attrs\n", count, vertex_size);
+    for (i = 0; i < count * vertex_size; i++) {
+        if (i % vertex_size == 0) {
+            //debug_printf("r300: -- vert --\n");
+        }
+        //debug_printf("r300: 0x%08x\n", *map);
+        OUT_CS(*map);
+        map++;
+    }
+    END_CS;
+
+    pipe_buffer_unmap(r300->context.screen, vbo);
+}
+
 static void r300_emit_draw_arrays(struct r300_context *r300,
                                   unsigned mode,
                                   unsigned count)
@@ -212,17 +247,54 @@ validate:
     return TRUE;
 }
 
+static struct pipe_buffer* r300_translate_elts(struct r300_context* r300,
+                                               struct pipe_buffer* elts,
+                                               unsigned* size,
+                                               unsigned* mode,
+                                               unsigned* count)
+{
+    struct pipe_screen* screen = r300->context.screen;
+    struct pipe_buffer* new_elts;
+    void *in_map, *out_map;
+    unsigned out_prim, out_index_size, out_nr;
+    u_translate_func out_translate;
+
+    (void)u_index_translator(~0, *mode, *size, *count, PV_LAST, PV_LAST,
+        &out_prim, &out_index_size, &out_nr, &out_translate);
+
+    new_elts = screen->buffer_create(screen, 32,
+                                     PIPE_BUFFER_USAGE_INDEX |
+                                     PIPE_BUFFER_USAGE_CPU_WRITE |
+                                     PIPE_BUFFER_USAGE_GPU_READ,
+                                     out_index_size * out_nr);
+
+    in_map = pipe_buffer_map(screen, elts, PIPE_BUFFER_USAGE_CPU_READ);
+    out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
+
+    out_translate(in_map, *count, out_map);
+
+    pipe_buffer_unmap(screen, elts);
+    pipe_buffer_unmap(screen, new_elts);
+
+    *size = out_index_size;
+    *mode = out_prim;
+    *count = out_nr;
+
+    return new_elts;
+}
+
 /* This is the fast-path drawing & emission for HW TCL. */
 void r300_draw_range_elements(struct pipe_context* pipe,
-                                 struct pipe_buffer* indexBuffer,
-                                 unsigned indexSize,
-                                 unsigned minIndex,
-                                 unsigned maxIndex,
-                                 unsigned mode,
-                                 unsigned start,
-                                 unsigned count)
+                              struct pipe_buffer* indexBuffer,
+                              unsigned indexSize,
+                              unsigned minIndex,
+                              unsigned maxIndex,
+                              unsigned mode,
+                              unsigned start,
+                              unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct pipe_buffer* orgIndexBuffer = indexBuffer;
 
     if (!u_trim_pipe_prim(mode, &count)) {
         return;
@@ -235,23 +307,24 @@ void r300_draw_range_elements(struct pipe_context* pipe,
         return;
     }
 
-    if (r300_nothing_to_draw(r300)) {
-        return;
-    }
-
     r300_update_derived_state(r300);
 
     if (!r300_setup_vertex_buffers(r300)) {
         return;
     }
 
+    if (indexSize == 1) {
+        indexBuffer = r300_translate_elts(r300, indexBuffer,
+            &indexSize, &mode, &count);
+    }
+
     if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
                                   RADEON_GEM_DOMAIN_GTT, 0)) {
-        return;
+        goto cleanup;
     }
 
     if (!r300->winsys->validate(r300->winsys)) {
-        return;
+        goto cleanup;
     }
 
     r300_emit_dirty_state(r300);
@@ -260,6 +333,11 @@ void r300_draw_range_elements(struct pipe_context* pipe,
 
     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
                             mode, start, count);
+
+cleanup:
+    if (indexBuffer != orgIndexBuffer) {
+        pipe->screen->buffer_destroy(indexBuffer);
+    }
 }
 
 /* Simple helpers for context setup. Should probably be moved to util. */
@@ -273,7 +351,7 @@ void r300_draw_elements(struct pipe_context* pipe,
 }
 
 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
-                         unsigned start, unsigned count)
+                      unsigned start, unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
 
@@ -288,10 +366,6 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
         return;
     }
 
-    if (r300_nothing_to_draw(r300)) {
-        return;
-    }
-
     r300_update_derived_state(r300);
 
     if (!r300_setup_vertex_buffers(r300)) {
@@ -300,9 +374,12 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 
     r300_emit_dirty_state(r300);
 
-    r300_emit_aos(r300, start);
-
-    r300_emit_draw_arrays(r300, mode, count);
+    if (FALSE && count <= 4 && r300->vertex_buffer_count == 1) {
+        r300_emit_draw_immediate(r300, mode, start, count);
+    } else {
+        r300_emit_aos(r300, start);
+        r300_emit_draw_arrays(r300, mode, count);
+    }
 }
 
 /****************************************************************************
@@ -323,10 +400,6 @@ void r300_swtcl_draw_arrays(struct pipe_context* pipe,
         return;
     }
 
-    if (r300_nothing_to_draw(r300)) {
-        return;
-    }
-
     for (i = 0; i < r300->vertex_buffer_count; i++) {
         void* buf = pipe_buffer_map(pipe->screen,
                                     r300->vertex_buffer[i].buffer,
@@ -368,10 +441,6 @@ void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
         return;
     }
 
-    if (r300_nothing_to_draw(r300)) {
-        return;
-    }
-
     for (i = 0; i < r300->vertex_buffer_count; i++) {
         void* buf = pipe_buffer_map(pipe->screen,
                                     r300->vertex_buffer[i].buffer,
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index a145a7f18a5..60ad763cf48 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -317,8 +317,8 @@ static void r300_bind_blend_state(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
-    r300->blend_state = (struct r300_blend_state*)state;
-    r300->dirty_state |= R300_NEW_BLEND;
+    r300->blend_state.state = state;
+    r300->blend_state.dirty = TRUE;
 }
 
 /* Free blend state. */
@@ -340,20 +340,24 @@ static void r300_set_blend_color(struct pipe_context* pipe,
                                  const struct pipe_blend_color* color)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_screen* r300screen = r300_screen(pipe->screen);
+    struct r300_blend_color_state* state =
+        (struct r300_blend_color_state*)r300->blend_color_state.state;
     union util_color uc;
 
     util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
-    r300->blend_color_state->blend_color = uc.ui;
+    state->blend_color = uc.ui;
 
     /* XXX if FP16 blending is enabled, we should use the FP16 format */
-    r300->blend_color_state->blend_color_red_alpha =
+    state->blend_color_red_alpha =
         float_to_fixed10(color->color[0]) |
         (float_to_fixed10(color->color[3]) << 16);
-    r300->blend_color_state->blend_color_green_blue =
+    state->blend_color_green_blue =
         float_to_fixed10(color->color[2]) |
         (float_to_fixed10(color->color[1]) << 16);
 
-    r300->dirty_state |= R300_NEW_BLEND_COLOR;
+    r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2;
+    r300->blend_color_state.dirty = TRUE;
 }
 
 static void r300_set_clip_state(struct pipe_context* pipe,
@@ -362,12 +366,15 @@ static void r300_set_clip_state(struct pipe_context* pipe,
     struct r300_context* r300 = r300_context(pipe);
 
     if (r300_screen(pipe->screen)->caps->has_tcl) {
-        r300->clip_state = *state;
-        r300->dirty_state |= R300_NEW_CLIP;
+        memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
+        r300->clip_state.size = 29;
     } else {
         draw_flush(r300->draw);
         draw_set_clip_state(r300->draw, state);
+        r300->clip_state.size = 2;
     }
+
+    r300->clip_state.dirty = TRUE;
 }
 
 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
@@ -460,9 +467,11 @@ static void r300_bind_dsa_state(struct pipe_context* pipe,
                                 void* state)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_screen* r300screen = r300_screen(pipe->screen);
 
-    r300->dsa_state = (struct r300_dsa_state*)state;
-    r300->dirty_state |= R300_NEW_DSA;
+    r300->dsa_state.state = state;
+    r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6;
+    r300->dsa_state.dirty = TRUE;
 }
 
 /* Free DSA state. */
@@ -472,37 +481,11 @@ static void r300_delete_dsa_state(struct pipe_context* pipe,
     FREE(state);
 }
 
-static void r300_set_scissor_regs(const struct pipe_scissor_state* state,
-                                  struct r300_scissor_regs *scissor,
-                                  boolean is_r500)
-{
-    if (is_r500) {
-        scissor->top_left =
-            (state->minx << R300_SCISSORS_X_SHIFT) |
-            (state->miny << R300_SCISSORS_Y_SHIFT);
-        scissor->bottom_right =
-            ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) |
-            ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT);
-    } else {
-        /* Offset of 1440 in non-R500 chipsets. */
-        scissor->top_left =
-            ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) |
-            ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT);
-        scissor->bottom_right =
-            (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
-            (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
-    }
-
-    scissor->empty_area = state->minx >= state->maxx ||
-                          state->miny >= state->maxy;
-}
-
 static void
     r300_set_framebuffer_state(struct pipe_context* pipe,
                                const struct pipe_framebuffer_state* state)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct pipe_scissor_state scissor;
 
     if (r300->draw) {
         draw_flush(r300->draw);
@@ -510,19 +493,12 @@ static void
 
     r300->framebuffer_state = *state;
 
-    scissor.minx = scissor.miny = 0;
-    scissor.maxx = state->width;
-    scissor.maxy = state->height;
-    r300_set_scissor_regs(&scissor, &r300->scissor_state->framebuffer,
-                          r300_screen(r300->context.screen)->caps->is_r500);
-
     /* Don't rely on the order of states being set for the first time. */
-    if (!r300->rs_state || !r300->rs_state->rs.scissor) {
-        r300->dirty_state |= R300_NEW_SCISSOR;
-    }
     r300->dirty_state |= R300_NEW_FRAMEBUFFERS;
-    r300->dirty_state |= R300_NEW_BLEND;
-    r300->dirty_state |= R300_NEW_DSA;
+
+    r300->blend_state.dirty = TRUE;
+    r300->dsa_state.dirty = TRUE;
+    r300->scissor_state.dirty = TRUE;
 }
 
 /* Create fragment shader state. */
@@ -601,8 +577,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     /* Copy rasterizer state for Draw. */
     rs->rs = *state;
 
-    rs->enable_vte = !state->bypass_vs_clip_and_viewport;
-
 #ifdef PIPE_ARCH_LITTLE_ENDIAN
     rs->vap_control_status = R300_VC_NO_SWAP;
 #else
@@ -718,12 +692,20 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
         draw_set_rasterizer_state(r300->draw, &rs->rs);
     }
 
-    r300->rs_state = rs;
+    if (rs) {
+        r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport;
+    } else {
+        r300->tcl_bypass = FALSE;
+    }
+
+    r300->rs_state.state = rs;
+    r300->rs_state.dirty = TRUE;
+    /* XXX Why is this still needed, dammit!? */
+    r300->scissor_state.dirty = TRUE;
+    r300->viewport_state.dirty = TRUE;
+
     /* XXX Clean these up when we move to atom emits */
-    r300->dirty_state |= R300_NEW_RASTERIZER;
     r300->dirty_state |= R300_NEW_RS_BLOCK;
-    r300->dirty_state |= R300_NEW_SCISSOR;
-    r300->dirty_state |= R300_NEW_VIEWPORT;
     if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
         r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
@@ -862,49 +844,48 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
-    r300_set_scissor_regs(state, &r300->scissor_state->scissor,
-                          r300_screen(r300->context.screen)->caps->is_r500);
+    memcpy(r300->scissor_state.state, state,
+        sizeof(struct pipe_scissor_state));
 
-    /* Don't rely on the order of states being set for the first time. */
-    if (!r300->rs_state || r300->rs_state->rs.scissor) {
-        r300->dirty_state |= R300_NEW_SCISSOR;
-    }
+    r300->scissor_state.dirty = TRUE;
 }
 
 static void r300_set_viewport_state(struct pipe_context* pipe,
                                     const struct pipe_viewport_state* state)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_viewport_state* viewport =
+        (struct r300_viewport_state*)r300->viewport_state.state;
 
     /* Do the transform in HW. */
-    r300->viewport_state->vte_control = R300_VTX_W0_FMT;
+    viewport->vte_control = R300_VTX_W0_FMT;
 
     if (state->scale[0] != 1.0f) {
-        r300->viewport_state->xscale = state->scale[0];
-        r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA;
+        viewport->xscale = state->scale[0];
+        viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
     }
     if (state->scale[1] != 1.0f) {
-        r300->viewport_state->yscale = state->scale[1];
-        r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA;
+        viewport->yscale = state->scale[1];
+        viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
     }
     if (state->scale[2] != 1.0f) {
-        r300->viewport_state->zscale = state->scale[2];
-        r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA;
+        viewport->zscale = state->scale[2];
+        viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
     }
     if (state->translate[0] != 0.0f) {
-        r300->viewport_state->xoffset = state->translate[0];
-        r300->viewport_state->vte_control |= R300_VPORT_X_OFFSET_ENA;
+        viewport->xoffset = state->translate[0];
+        viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
     }
     if (state->translate[1] != 0.0f) {
-        r300->viewport_state->yoffset = state->translate[1];
-        r300->viewport_state->vte_control |= R300_VPORT_Y_OFFSET_ENA;
+        viewport->yoffset = state->translate[1];
+        viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
     }
     if (state->translate[2] != 0.0f) {
-        r300->viewport_state->zoffset = state->translate[2];
-        r300->viewport_state->vte_control |= R300_VPORT_Z_OFFSET_ENA;
+        viewport->zoffset = state->translate[2];
+        viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
     }
 
-    r300->dirty_state |= R300_NEW_VIEWPORT;
+    r300->viewport_state.dirty = TRUE;
     if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
         r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
@@ -1011,22 +992,22 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
 
 static void r300_set_constant_buffer(struct pipe_context *pipe,
                                      uint shader, uint index,
-                                     const struct pipe_constant_buffer *buf)
+                                     struct pipe_buffer *buf)
 {
     struct r300_context* r300 = r300_context(pipe);
     void *mapped;
 
-    if (buf == NULL || buf->buffer->size == 0 ||
-        (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
+    if (buf == NULL || buf->size == 0 ||
+        (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
     {
         r300->shader_constants[shader].count = 0;
         return;
     }
 
-    assert((buf->buffer->size % 4 * sizeof(float)) == 0);
-    memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size);
-    r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float));
-    pipe_buffer_unmap(pipe->screen, buf->buffer);
+    assert((buf->size % 4 * sizeof(float)) == 0);
+    memcpy(r300->shader_constants[shader].constants, mapped, buf->size);
+    r300->shader_constants[shader].count = buf->size / (4 * sizeof(float));
+    pipe_buffer_unmap(pipe->screen, buf);
 
     if (shader == PIPE_SHADER_VERTEX)
         r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 22660a52d9a..192846411ba 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -139,10 +139,10 @@ static void r300_vertex_psc(struct r300_context* r300)
 
     /* If TCL is bypassed, map vertex streams to equivalent VS output
      * locations. */
-    if (r300->rs_state->enable_vte) {
-        stream_tab = identity;
-    } else {
+    if (r300->tcl_bypass) {
         stream_tab = r300->vs->stream_loc_notcl;
+    } else {
+        stream_tab = identity;
     }
 
     /* Vertex shaders have no semantics on their inputs,
@@ -508,7 +508,8 @@ static boolean r300_dsa_alpha_test_enabled(struct r300_dsa_state* dsa)
 
 static void r300_update_ztop(struct r300_context* r300)
 {
-    r300->ztop_state.z_buffer_top = R300_ZTOP_ENABLE;
+    struct r300_ztop_state* ztop_state =
+        (struct r300_ztop_state*)r300->ztop_state.state;
 
     /* This is important enough that I felt it warranted a comment.
      *
@@ -530,31 +531,37 @@ static void r300_update_ztop(struct r300_context* r300)
      * 5) Depth writes in fragment shader
      * 6) Outstanding occlusion queries
      *
+     * This register causes stalls all the way from SC to CB when changed,
+     * but it is buffered on-chip so it does not hurt to write it if it has
+     * not changed.
+     *
      * ~C.
      */
 
     /* ZS writes */
-    if (r300_dsa_writes_depth_stencil(r300->dsa_state) &&
-           (r300_dsa_alpha_test_enabled(r300->dsa_state) ||   /* (1) */
-            r300->fs->info.uses_kill)) {                      /* (2) */
-        r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
-    } else if (r300_fragment_shader_writes_depth(r300->fs)) { /* (5) */
-        r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
-    } else if (r300->query_current) {                         /* (6) */
-        r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
+    if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) &&
+           (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||/* (1) */
+            r300->fs->info.uses_kill)) {                         /* (2) */
+        ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
+    } else if (r300_fragment_shader_writes_depth(r300->fs)) {    /* (5) */
+        ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
+    } else if (r300->query_current) {                            /* (6) */
+        ztop_state->z_buffer_top = R300_ZTOP_DISABLE;
+    } else {
+        ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
     }
+
+    r300->ztop_state.dirty = TRUE;
 }
 
 void r300_update_derived_state(struct r300_context* r300)
 {
+    /* XXX */
     if (r300->dirty_state &
         (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER |
-         R300_NEW_VERTEX_FORMAT)) {
+         R300_NEW_VERTEX_FORMAT) || r300->rs_state.dirty) {
         r300_update_derived_shader_state(r300);
     }
 
-    if (r300->dirty_state &
-            (R300_NEW_DSA | R300_NEW_FRAGMENT_SHADER | R300_NEW_QUERY)) {
-        r300_update_ztop(r300);
-    }
+    r300_update_ztop(r300);
 }
diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c
index f25f3ca217d..b0f309695c9 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.c
+++ b/src/gallium/drivers/r300/r300_state_invariant.c
@@ -115,10 +115,12 @@ void r300_emit_invariant_state(struct r300_context* r300)
     OUT_CS_REG(R300_SC_HYPERZ, 0x0000001C);
     OUT_CS_REG(R300_SC_EDGERULE, 0x2DA49525);
     OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0x00000000);
-    if (caps->is_r500) {
+
+    if (caps->family >= CHIP_FAMILY_RV350) {
         OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
         OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
     }
+
     OUT_CS_REG(R300_ZB_BW_CNTL, 0x00000000);
     OUT_CS_REG(R300_ZB_DEPTHCLEARVALUE, 0x00000000);
     OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0x00000000);
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index 9a96206a4dc..a9bbdd56d84 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -30,6 +30,18 @@
 #include "r300_texture.h"
 #include "r300_screen.h"
 
+#define TILE_WIDTH 0
+#define TILE_HEIGHT 1
+
+static const unsigned microblock_table[5][3][2] = {
+    /*linear  tiled   square-tiled */
+    {{32, 1}, {8, 4}, {0, 0}}, /*   8 bits per pixel */
+    {{16, 1}, {8, 2}, {4, 4}}, /*  16 bits per pixel */
+    {{ 8, 1}, {4, 2}, {0, 0}}, /*  32 bits per pixel */
+    {{ 4, 1}, {0, 0}, {2, 2}}, /*  64 bits per pixel */
+    {{ 2, 1}, {0, 0}, {0, 0}}  /* 128 bits per pixel */
+};
+
 static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500)
 {
     struct r300_texture_state* state = &tex->state;
@@ -92,33 +104,67 @@ unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
 }
 
 /**
+ * Return the width (dim==TILE_WIDTH) or height (dim==TILE_HEIGHT) of one tile
+ * of the given texture.
+ */
+static unsigned r300_texture_get_tile_size(struct r300_texture* tex, int dim)
+{
+    unsigned pixsize, tile_size;
+
+    pixsize = util_format_get_blocksize(tex->tex.format);
+    tile_size = microblock_table[util_logbase2(pixsize)][tex->microtile][dim] *
+                (tex->macrotile == R300_BUFFER_TILED ? 8 : 1);
+
+    assert(tile_size);
+    return tile_size;
+}
+
+/**
  * Return the stride, in bytes, of the texture images of the given texture
  * at the given level.
  */
 unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
 {
+    unsigned tile_width, width;
+
     if (tex->stride_override)
         return tex->stride_override;
 
+    /* Check the level. */
     if (level > tex->tex.last_level) {
         debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__,
             level, tex->tex.last_level);
         return 0;
     }
 
-    return align(util_format_get_stride(tex->tex.format, u_minify(tex->tex.width0, level)), 32);
+    tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH);
+    width = align(u_minify(tex->tex.width0, level), tile_width);
+
+    /* Should already be aligned except for S3TC. */
+    return align(util_format_get_stride(tex->tex.format, width), 32);
+}
+
+static unsigned r300_texture_get_nblocksy(struct r300_texture* tex,
+                                          unsigned level)
+{
+    unsigned height, tile_height;
+
+    tile_height = r300_texture_get_tile_size(tex, TILE_HEIGHT);
+    height = align(u_minify(tex->tex.height0, level), tile_height);
+
+    return util_format_get_nblocksy(tex->tex.format, height);
 }
 
 static void r300_setup_miptree(struct r300_texture* tex)
 {
     struct pipe_texture* base = &tex->tex;
-    int stride, size, layer_size;
-    int i;
+    unsigned stride, size, layer_size, nblocksy, i;
 
-    for (i = 0; i <= base->last_level; i++) {
-        unsigned nblocksy = util_format_get_nblocksy(base->format, u_minify(base->height0, i));
+    debug_printf("r300: Making miptree for texture, format %s\n", pf_name(base->format));
 
+    for (i = 0; i <= base->last_level; i++) {
         stride = r300_texture_get_stride(tex, i);
+        nblocksy = r300_texture_get_nblocksy(tex, i);
         layer_size = stride * nblocksy;
 
         if (base->target == PIPE_TEXTURE_CUBE)
@@ -132,9 +178,9 @@ static void r300_setup_miptree(struct r300_texture* tex)
         tex->pitch[i] = stride / util_format_get_blocksize(base->format);
 
         debug_printf("r300: Texture miptree: Level %d "
-                "(%dx%dx%d px, pitch %d bytes)\n",
+                "(%dx%dx%d px, pitch %d bytes) %d bytes total\n",
                 i, u_minify(base->width0, i), u_minify(base->height0, i),
-                u_minify(base->depth0, i), stride);
+                u_minify(base->depth0, i), stride, tex->size);
     }
 }
 
@@ -163,7 +209,7 @@ static struct pipe_texture*
     r300_setup_miptree(tex);
     r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500);
 
-    tex->buffer = screen->buffer_create(screen, 1024,
+    tex->buffer = screen->buffer_create(screen, 2048,
                                         PIPE_BUFFER_USAGE_PIXEL,
                                         tex->size);
 
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index f3ac6760db5..8e017939402 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -113,8 +113,8 @@ softpipe_destroy( struct pipe_context *pipe )
    }
 
    for (i = 0; i < Elements(softpipe->constants); i++) {
-      if (softpipe->constants[i].buffer) {
-         pipe_buffer_reference(&softpipe->constants[i].buffer, NULL);
+      if (softpipe->constants[i]) {
+         pipe_buffer_reference(&softpipe->constants[i], NULL);
       }
    }
 
@@ -256,6 +256,8 @@ softpipe_create( struct pipe_screen *screen )
    softpipe->pipe.draw_arrays = softpipe_draw_arrays;
    softpipe->pipe.draw_elements = softpipe_draw_elements;
    softpipe->pipe.draw_range_elements = softpipe_draw_range_elements;
+   softpipe->pipe.draw_arrays_instanced = softpipe_draw_arrays_instanced;
+   softpipe->pipe.draw_elements_instanced = softpipe_draw_elements_instanced;
 
    softpipe->pipe.clear = softpipe_clear;
    softpipe->pipe.flush = softpipe_flush;
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 73fa744f9d4..da673c57ada 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -63,7 +63,7 @@ struct softpipe_context {
    /** Other rendering state */
    struct pipe_blend_color blend_color;
    struct pipe_clip_state clip;
-   struct pipe_constant_buffer constants[PIPE_SHADER_TYPES];
+   struct pipe_buffer *constants[PIPE_SHADER_TYPES];
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index 03d35fb3cb5..03b58d2fb72 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -52,18 +52,18 @@ softpipe_map_constant_buffers(struct softpipe_context *sp)
    uint i, vssize, gssize;
 
    for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      if (sp->constants[i].buffer && sp->constants[i].buffer->size)
-         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer,
+      if (sp->constants[i] && sp->constants[i]->size)
+         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i],
                                                   PIPE_BUFFER_USAGE_CPU_READ);
    }
 
-   if (sp->constants[PIPE_SHADER_VERTEX].buffer)
-      vssize = sp->constants[PIPE_SHADER_VERTEX].buffer->size;
+   if (sp->constants[PIPE_SHADER_VERTEX])
+      vssize = sp->constants[PIPE_SHADER_VERTEX]->size;
    else
       vssize = 0;
 
-   if (sp->constants[PIPE_SHADER_GEOMETRY].buffer)
-      gssize = sp->constants[PIPE_SHADER_GEOMETRY].buffer->size;
+   if (sp->constants[PIPE_SHADER_GEOMETRY])
+      gssize = sp->constants[PIPE_SHADER_GEOMETRY]->size;
    else
       gssize = 0;
 
@@ -91,26 +91,48 @@ softpipe_unmap_constant_buffers(struct softpipe_context *sp)
    draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_GEOMETRY, NULL, 0);
 
    for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      if (sp->constants[i].buffer && sp->constants[i].buffer->size)
-         ws->buffer_unmap(ws, sp->constants[i].buffer);
+      if (sp->constants[i] && sp->constants[i]->size)
+         ws->buffer_unmap(ws, sp->constants[i]);
       sp->mapped_constants[i] = NULL;
    }
 }
 
 
+/**
+ * Draw vertex arrays, with optional indexing.
+ * Basically, map the vertex buffers (and drawing surfaces), then hand off
+ * the drawing to the 'draw' module.
+ */
+static void
+softpipe_draw_range_elements_instanced(struct pipe_context *pipe,
+                                       struct pipe_buffer *indexBuffer,
+                                       unsigned indexSize,
+                                       unsigned minIndex,
+                                       unsigned maxIndex,
+                                       unsigned mode,
+                                       unsigned start,
+                                       unsigned count,
+                                       unsigned startInstance,
+                                       unsigned instanceCount);
+
+
 void
 softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode,
                      unsigned start, unsigned count)
 {
-   softpipe_draw_elements(pipe, NULL, 0, mode, start, count);
+   softpipe_draw_range_elements_instanced(pipe,
+                                          NULL,
+                                          0,
+                                          0,
+                                          0xffffffff,
+                                          mode,
+                                          start,
+                                          count,
+                                          0,
+                                          1);
 }
 
 
-/**
- * Draw vertex arrays, with optional indexing.
- * Basically, map the vertex buffers (and drawing surfaces), then hand off
- * the drawing to the 'draw' module.
- */
 void
 softpipe_draw_range_elements(struct pipe_context *pipe,
                              struct pipe_buffer *indexBuffer,
@@ -119,6 +141,91 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
                              unsigned max_index,
                              unsigned mode, unsigned start, unsigned count)
 {
+   softpipe_draw_range_elements_instanced(pipe,
+                                          indexBuffer,
+                                          indexSize,
+                                          min_index,
+                                          max_index,
+                                          mode,
+                                          start,
+                                          count,
+                                          0,
+                                          1);
+}
+
+
+void
+softpipe_draw_elements(struct pipe_context *pipe,
+                       struct pipe_buffer *indexBuffer,
+                       unsigned indexSize,
+                       unsigned mode, unsigned start, unsigned count)
+{
+   softpipe_draw_range_elements_instanced(pipe,
+                                          indexBuffer,
+                                          indexSize,
+                                          0,
+                                          0xffffffff,
+                                          mode,
+                                          start,
+                                          count,
+                                          0,
+                                          1);
+}
+
+void
+softpipe_draw_arrays_instanced(struct pipe_context *pipe,
+                               unsigned mode,
+                               unsigned start,
+                               unsigned count,
+                               unsigned startInstance,
+                               unsigned instanceCount)
+{
+   softpipe_draw_range_elements_instanced(pipe,
+                                          NULL,
+                                          0,
+                                          0,
+                                          0xffffffff,
+                                          mode,
+                                          start,
+                                          count,
+                                          startInstance,
+                                          instanceCount);
+}
+
+void
+softpipe_draw_elements_instanced(struct pipe_context *pipe,
+                                 struct pipe_buffer *indexBuffer,
+                                 unsigned indexSize,
+                                 unsigned mode,
+                                 unsigned start,
+                                 unsigned count,
+                                 unsigned startInstance,
+                                 unsigned instanceCount)
+{
+   softpipe_draw_range_elements_instanced(pipe,
+                                          indexBuffer,
+                                          indexSize,
+                                          0,
+                                          0xffffffff,
+                                          mode,
+                                          start,
+                                          count,
+                                          startInstance,
+                                          instanceCount);
+}
+
+static void
+softpipe_draw_range_elements_instanced(struct pipe_context *pipe,
+                                       struct pipe_buffer *indexBuffer,
+                                       unsigned indexSize,
+                                       unsigned minIndex,
+                                       unsigned maxIndex,
+                                       unsigned mode,
+                                       unsigned start,
+                                       unsigned count,
+                                       unsigned startInstance,
+                                       unsigned instanceCount)
+{
    struct softpipe_context *sp = softpipe_context(pipe);
    struct draw_context *draw = sp->draw;
    unsigned i;
@@ -128,45 +235,48 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
 
    sp->reduced_api_prim = u_reduced_prim(mode);
 
-   if (sp->dirty)
-      softpipe_update_derived( sp );
+   if (sp->dirty) {
+      softpipe_update_derived(sp);
+   }
 
    softpipe_map_transfers(sp);
    softpipe_map_constant_buffers(sp);
 
-   /*
-    * Map vertex buffers
-    */
+   /* Map vertex buffers */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
-      void *buf
-         = pipe_buffer_map(pipe->screen,
-                                    sp->vertex_buffer[i].buffer,
-                                    PIPE_BUFFER_USAGE_CPU_READ);
+      void *buf;
+
+      buf = pipe_buffer_map(pipe->screen,
+                            sp->vertex_buffer[i].buffer,
+                            PIPE_BUFFER_USAGE_CPU_READ);
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
 
    /* Map index buffer, if present */
    if (indexBuffer) {
-      void *mapped_indexes
-         = pipe_buffer_map(pipe->screen, indexBuffer,
-                                    PIPE_BUFFER_USAGE_CPU_READ);
-      draw_set_mapped_element_buffer_range(draw, indexSize,
-                                           min_index,
-                                           max_index,
+      void *mapped_indexes;
+
+      mapped_indexes = pipe_buffer_map(pipe->screen,
+                                       indexBuffer,
+                                       PIPE_BUFFER_USAGE_CPU_READ);
+      draw_set_mapped_element_buffer_range(draw,
+                                           indexSize,
+                                           minIndex,
+                                           maxIndex,
                                            mapped_indexes);
-   }
-   else {
+   } else {
       /* no index/element buffer */
-      draw_set_mapped_element_buffer_range(draw, 0, start,
-                                           start + count - 1, NULL);
+      draw_set_mapped_element_buffer_range(draw,
+                                           0,
+                                           start,
+                                           start + count - 1,
+                                           NULL);
    }
 
    /* draw! */
-   draw_arrays(draw, mode, start, count);
+   draw_arrays_instanced(draw, mode, start, count, startInstance, instanceCount);
 
-   /*
-    * unmap vertex/index buffers - will cause draw module to flush
-    */
+   /* unmap vertex/index buffers - will cause draw module to flush */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
       pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
@@ -176,22 +286,8 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
       pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
-
    /* Note: leave drawing surfaces mapped */
    softpipe_unmap_constant_buffers(sp);
 
    sp->dirty_render_cache = TRUE;
 }
-
-
-void
-softpipe_draw_elements(struct pipe_context *pipe,
-                       struct pipe_buffer *indexBuffer,
-                       unsigned indexSize,
-                       unsigned mode, unsigned start, unsigned count)
-{
-   softpipe_draw_range_elements( pipe, indexBuffer,
-                                 indexSize,
-                                 0, 0xffffffff,
-                                 mode, start, count );
-}
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 9b18dac67bd..7f244c4fd49 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -139,7 +139,7 @@ void softpipe_set_clip_state( struct pipe_context *,
 
 void softpipe_set_constant_buffer(struct pipe_context *,
                                   uint shader, uint index,
-                                  const struct pipe_constant_buffer *buf);
+                                  struct pipe_buffer *buf);
 
 void *softpipe_create_fs_state(struct pipe_context *,
                                const struct pipe_shader_state *);
@@ -200,6 +200,24 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
                              unsigned mode, unsigned start, unsigned count);
 
 void
+softpipe_draw_arrays_instanced(struct pipe_context *pipe,
+                               unsigned mode,
+                               unsigned start,
+                               unsigned count,
+                               unsigned startInstance,
+                               unsigned instanceCount);
+
+void
+softpipe_draw_elements_instanced(struct pipe_context *pipe,
+                                 struct pipe_buffer *indexBuffer,
+                                 unsigned indexSize,
+                                 unsigned mode,
+                                 unsigned start,
+                                 unsigned count,
+                                 unsigned startInstance,
+                                 unsigned instanceCount);
+
+void
 softpipe_map_transfers(struct softpipe_context *sp);
 
 void
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index aa12bb215a8..b7ed4441b43 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -159,7 +159,7 @@ softpipe_delete_vs_state(struct pipe_context *pipe, void *vs)
 void
 softpipe_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
-                             const struct pipe_constant_buffer *buf)
+                             struct pipe_buffer *buf)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
@@ -169,8 +169,7 @@ softpipe_set_constant_buffer(struct pipe_context *pipe,
    draw_flush(softpipe->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&softpipe->constants[shader].buffer,
-			 buf ? buf->buffer : NULL);
+   pipe_buffer_reference(&softpipe->constants[shader], buf);
 
    softpipe->dirty |= SP_NEW_CONSTANTS;
 }
diff --git a/src/gallium/drivers/svga/svga_pipe_constants.c b/src/gallium/drivers/svga/svga_pipe_constants.c
index 10e7a121892..ca2c7c49d72 100644
--- a/src/gallium/drivers/svga/svga_pipe_constants.c
+++ b/src/gallium/drivers/svga/svga_pipe_constants.c
@@ -49,7 +49,7 @@ struct svga_constbuf
 
 static void svga_set_constant_buffer(struct pipe_context *pipe,
                                      uint shader, uint index,
-                                     const struct pipe_constant_buffer *buf)
+                                     struct pipe_buffer *buf)
 {
    struct svga_context *svga = svga_context(pipe);
 
@@ -57,7 +57,7 @@ static void svga_set_constant_buffer(struct pipe_context *pipe,
    assert(index == 0);
 
    pipe_buffer_reference( &svga->curr.cb[shader],
-                          buf->buffer );
+                          buf );
 
    if (shader == PIPE_SHADER_FRAGMENT)
       svga->dirty |= SVGA_NEW_FS_CONST_BUFFER;
diff --git a/src/gallium/drivers/svga/svga_state_fs.c b/src/gallium/drivers/svga/svga_state_fs.c
index bba80a93a56..ec2886348b9 100644
--- a/src/gallium/drivers/svga/svga_state_fs.c
+++ b/src/gallium/drivers/svga/svga_state_fs.c
@@ -72,7 +72,7 @@ static enum pipe_error compile_fs( struct svga_context *svga,
                                    struct svga_shader_result **out_result )
 {
    struct svga_shader_result *result;
-   enum pipe_error ret;
+   enum pipe_error ret = PIPE_ERROR;
 
    result = svga_translate_fragment_program( fs, key );
    if (result == NULL) {
@@ -81,8 +81,10 @@ static enum pipe_error compile_fs( struct svga_context *svga,
    }
 
    result->id = util_bitmask_add(svga->fs_bm);
-   if(result->id == UTIL_BITMASK_INVALID_INDEX)
+   if(result->id == UTIL_BITMASK_INVALID_INDEX) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
       goto fail;
+   }
 
    ret = SVGA3D_DefineShader(svga->swc, 
                              result->id,
diff --git a/src/gallium/drivers/svga/svga_state_vs.c b/src/gallium/drivers/svga/svga_state_vs.c
index ae1e77e7d44..e7e6c084321 100644
--- a/src/gallium/drivers/svga/svga_state_vs.c
+++ b/src/gallium/drivers/svga/svga_state_vs.c
@@ -71,7 +71,7 @@ static enum pipe_error compile_vs( struct svga_context *svga,
                                    struct svga_shader_result **out_result )
 {
    struct svga_shader_result *result;
-   enum pipe_error ret = PIPE_OK;
+   enum pipe_error ret = PIPE_ERROR;
 
    result = svga_translate_vertex_program( vs, key );
    if (result == NULL) {
@@ -80,8 +80,10 @@ static enum pipe_error compile_vs( struct svga_context *svga,
    }
 
    result->id = util_bitmask_add(svga->vs_bm);
-   if(result->id == UTIL_BITMASK_INVALID_INDEX)
+   if(result->id == UTIL_BITMASK_INVALID_INDEX) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
       goto fail;
+   }
 
    ret = SVGA3D_DefineShader(svga->swc, 
                              result->id,
@@ -200,10 +202,12 @@ static int update_zero_stride( struct svga_context *svga,
 
          key.output_stride = 4 * sizeof(float);
          key.nr_elements = 1;
+         key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
          key.element[0].input_format = vel->src_format;
          key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
          key.element[0].input_buffer = vel->vertex_buffer_index;
          key.element[0].input_offset = vel->src_offset;
+         key.element[0].instance_divisor = vel->instance_divisor;
          key.element[0].output_offset = const_idx * 4 * sizeof(float);
 
          translate_key_sanitize(&key);
@@ -222,7 +226,7 @@ static int update_zero_stride( struct svga_context *svga,
          translate->set_buffer(translate, vel->vertex_buffer_index,
                                mapped_buffer,
                                vbuffer->stride);
-         translate->run(translate, 0, 1,
+         translate->run(translate, 0, 1, 0,
                         svga->curr.zero_stride_constants);
 
          pipe_buffer_unmap(svga->pipe.screen,
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index 075e4f9a0b2..5a9f0fc6901 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -812,13 +812,13 @@ trace_context_set_clip_state(struct pipe_context *_pipe,
 static INLINE void
 trace_context_set_constant_buffer(struct pipe_context *_pipe,
                                   uint shader, uint index,
-                                  const struct pipe_constant_buffer *buffer)
+                                  struct pipe_buffer *buffer)
 {
    struct trace_context *tr_ctx = trace_context(_pipe);
    struct pipe_context *pipe = tr_ctx->pipe;
 
    if (buffer)
-      trace_screen_user_buffer_update(_pipe->screen, buffer->buffer);
+      trace_screen_user_buffer_update(_pipe->screen, buffer);
 
    trace_dump_call_begin("pipe_context", "set_constant_buffer");
 
@@ -827,10 +827,11 @@ trace_context_set_constant_buffer(struct pipe_context *_pipe,
    trace_dump_arg(uint, index);
    trace_dump_arg(constant_buffer, buffer);
 
+   /* XXX hmm? */
    if (buffer) {
-      struct pipe_constant_buffer _buffer;
-      _buffer.buffer = trace_buffer_unwrap(tr_ctx, buffer->buffer);
-      pipe->set_constant_buffer(pipe, shader, index, &_buffer);
+      struct pipe_buffer *_buffer;
+      _buffer = trace_buffer_unwrap(tr_ctx, buffer);
+      pipe->set_constant_buffer(pipe, shader, index, _buffer);
    } else {
       pipe->set_constant_buffer(pipe, shader, index, buffer);
    }
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 86237e03bcc..32f61f8c944 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -229,7 +229,7 @@ void trace_dump_clip_state(const struct pipe_clip_state *state)
 }
 
 
-void trace_dump_constant_buffer(const struct pipe_constant_buffer *state)
+void trace_dump_constant_buffer(const struct pipe_buffer *state)
 {
    if (!trace_dumping_enabled_locked())
       return;
@@ -241,7 +241,7 @@ void trace_dump_constant_buffer(const struct pipe_constant_buffer *state)
 
    trace_dump_struct_begin("pipe_constant_buffer");
 
-   trace_dump_member(buffer_ptr, state, buffer);
+   trace_dump_reference(&state->reference);
 
    trace_dump_struct_end();
 }
diff --git a/src/gallium/drivers/trace/tr_dump_state.h b/src/gallium/drivers/trace/tr_dump_state.h
index 07ad6fbb205..c7860fd6e18 100644
--- a/src/gallium/drivers/trace/tr_dump_state.h
+++ b/src/gallium/drivers/trace/tr_dump_state.h
@@ -47,7 +47,7 @@ void trace_dump_scissor_state(const struct pipe_scissor_state *state);
 
 void trace_dump_clip_state(const struct pipe_clip_state *state);
 
-void trace_dump_constant_buffer(const struct pipe_constant_buffer *state);
+void trace_dump_constant_buffer(const struct pipe_buffer *state);
 
 void trace_dump_token(const struct tgsi_token *token);
 
diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h
index f7368bb95b3..18ebd0c9483 100644
--- a/src/gallium/include/pipe/p_compiler.h
+++ b/src/gallium/include/pipe/p_compiler.h
@@ -52,45 +52,15 @@
 #endif /* _MSC_VER */
 
 
-#if defined(_MSC_VER)
-
-typedef __int8             int8_t;
-typedef unsigned __int8    uint8_t;
-typedef __int16            int16_t;
-typedef unsigned __int16   uint16_t;
-#ifndef __eglplatform_h_
-typedef __int32            int32_t;
-#endif
-typedef unsigned __int32   uint32_t;
-typedef __int64            int64_t;
-typedef unsigned __int64   uint64_t;
-
-#if defined(_WIN64)
-typedef __int64            intptr_t;
-typedef unsigned __int64   uintptr_t;
-#else
-typedef __int32            intptr_t;
-typedef unsigned __int32   uintptr_t;
-#endif
-
-#define INT64_C(__val) __val##i64
-#define UINT64_C(__val) __val##ui64
-
-#ifndef __cplusplus
-#define false   0
-#define true    1
-#define bool    _Bool
-typedef int     _Bool;
-#define __bool_true_false_are_defined   1
-#endif /* !__cplusplus */
-
-#else
+/*
+ * Alternative stdint.h and stdbool.h headers are supplied in include/c99 for
+ * systems that lack it.
+ */
 #ifndef __STDC_LIMIT_MACROS
 #define __STDC_LIMIT_MACROS 1
 #endif
 #include <stdint.h>
 #include <stdbool.h>
-#endif
 
 
 #ifndef __HAIKU__
@@ -99,11 +69,7 @@ typedef unsigned short     ushort;
 #endif
 typedef unsigned char      ubyte;
 
-#if 0
-#define boolean bool
-#else
 typedef unsigned char boolean;
-#endif
 #ifndef TRUE
 #define TRUE  true
 #endif
@@ -135,6 +101,17 @@ typedef unsigned char boolean;
 #  endif
 #endif
 
+
+/* Function visibility */
+#ifndef PUBLIC
+#  if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 303
+#    define PUBLIC __attribute__((visibility("default")))
+#  else
+#    define PUBLIC
+#  endif
+#endif
+
+
 /* The __FUNCTION__ gcc variable is generally only used for debugging.
  * If we're not using gcc, define __FUNCTION__ as a cpp symbol here.
  */
@@ -162,22 +139,33 @@ typedef unsigned char boolean;
 
 
 
+/* Macros for data alignment. */
 #if defined(__GNUC__)
-#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME##___aligned[SIZE] __attribute__(( aligned( 16 ) ))
-#define ALIGN16_ASSIGN(NAME) NAME##___aligned
-#define ALIGN16_ATTRIB  __attribute__(( aligned( 16 ) ))
-#define ALIGN8_ATTRIB  __attribute__(( aligned( 8 ) ))
+
+/* See http://gcc.gnu.org/onlinedocs/gcc-4.4.2/gcc/Type-Attributes.html */
+#define PIPE_ALIGN_TYPE(_alignment, _type) _type __attribute__((aligned(_alignment)))
+
+/* See http://gcc.gnu.org/onlinedocs/gcc-4.4.2/gcc/Variable-Attributes.html */
+#define PIPE_ALIGN_VAR(_alignment) __attribute__((aligned(_alignment)))
+
 #if (__GNUC__ > 4 || (__GNUC__ == 4 &&__GNUC_MINOR__>1)) && !defined(PIPE_ARCH_X86_64)
-#define ALIGN_STACK __attribute__((force_align_arg_pointer))
+#define PIPE_ALIGN_STACK __attribute__((force_align_arg_pointer))
 #else
-#define ALIGN_STACK
+#define PIPE_ALIGN_STACK
 #endif
+
+#elif defined(_MSC_VER)
+
+/* See http://msdn.microsoft.com/en-us/library/83ythb65.aspx */
+#define PIPE_ALIGN_TYPE(_alignment, _type) __declspec(align(_alignment)) _type
+#define PIPE_ALIGN_VAR(_alignment) __declspec(align(_alignment))
+
+#define PIPE_ALIGN_STACK
+
 #else
-#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME##___unaligned[SIZE + 1]
-#define ALIGN16_ASSIGN(NAME) align16(NAME##___unaligned)
-#define ALIGN16_ATTRIB
-#define ALIGN8_ATTRIB
-#define ALIGN_STACK
+
+#error "Unsupported compiler"
+
 #endif
 
 
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index d2f8085b421..0b8f6da2f4a 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -69,6 +69,22 @@ struct pipe_context {
                           unsigned indexSize,
                           unsigned mode, unsigned start, unsigned count);
 
+   void (*draw_arrays_instanced)(struct pipe_context *pipe,
+                                 unsigned mode,
+                                 unsigned start,
+                                 unsigned count,
+                                 unsigned startInstance,
+                                 unsigned instanceCount);
+
+   void (*draw_elements_instanced)(struct pipe_context *pipe,
+                                   struct pipe_buffer *indexBuffer,
+                                   unsigned indexSize,
+                                   unsigned mode,
+                                   unsigned start,
+                                   unsigned count,
+                                   unsigned startInstance,
+                                   unsigned instanceCount);
+
    /* XXX: this is (probably) a temporary entrypoint, as the range
     * information should be available from the vertex_buffer state.
     * Using this to quickly evaluate a specialized path in the draw
@@ -170,7 +186,7 @@ struct pipe_context {
 
    void (*set_constant_buffer)( struct pipe_context *,
                                 uint shader, uint index,
-                                const struct pipe_constant_buffer *buf );
+                                struct pipe_buffer *buf );
 
    void (*set_framebuffer_state)( struct pipe_context *,
                                   const struct pipe_framebuffer_state * );
diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h
index f0a4de5df33..b8e001a6b01 100644
--- a/src/gallium/include/pipe/p_screen.h
+++ b/src/gallium/include/pipe/p_screen.h
@@ -266,6 +266,11 @@ struct pipe_screen {
 
    void (*video_surface_destroy)( struct pipe_video_surface *vsfc );
 
+   /**
+    * Do any special operations to ensure buffer size is correct
+    */
+   void (*update_buffer)( struct pipe_screen *ws,
+                          void *context_private );
 
    /**
     * Do any special operations to ensure frontbuffer contents are
diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
index 550e2abc32a..b489b044667 100644
--- a/src/gallium/include/pipe/p_shader_tokens.h
+++ b/src/gallium/include/pipe/p_shader_tokens.h
@@ -121,17 +121,18 @@ struct tgsi_declaration_range
    unsigned Last    : 16; /**< UINT */
 };
 
-#define TGSI_SEMANTIC_POSITION  0
-#define TGSI_SEMANTIC_COLOR     1
-#define TGSI_SEMANTIC_BCOLOR    2 /**< back-face color */
-#define TGSI_SEMANTIC_FOG       3
-#define TGSI_SEMANTIC_PSIZE     4
-#define TGSI_SEMANTIC_GENERIC   5
-#define TGSI_SEMANTIC_NORMAL    6
-#define TGSI_SEMANTIC_FACE      7
-#define TGSI_SEMANTIC_EDGEFLAG  8
-#define TGSI_SEMANTIC_PRIMID    9
-#define TGSI_SEMANTIC_COUNT    10 /**< number of semantic values */
+#define TGSI_SEMANTIC_POSITION   0
+#define TGSI_SEMANTIC_COLOR      1
+#define TGSI_SEMANTIC_BCOLOR     2  /**< back-face color */
+#define TGSI_SEMANTIC_FOG        3
+#define TGSI_SEMANTIC_PSIZE      4
+#define TGSI_SEMANTIC_GENERIC    5
+#define TGSI_SEMANTIC_NORMAL     6
+#define TGSI_SEMANTIC_FACE       7
+#define TGSI_SEMANTIC_EDGEFLAG   8
+#define TGSI_SEMANTIC_PRIMID     9
+#define TGSI_SEMANTIC_INSTANCEID 10
+#define TGSI_SEMANTIC_COUNT      11 /**< number of semantic values */
 
 struct tgsi_declaration_semantic
 {
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 60e96b98deb..fdd29ed4492 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -66,10 +66,6 @@ extern "C" {
 #define PIPE_MAX_TEXTURE_LEVELS   16
 
 
-/* fwd decls */
-struct pipe_surface;
-
-
 /**
  * The driver will certainly subclass this to include actual memory
  * management information.
@@ -178,15 +174,6 @@ struct pipe_clip_state
 };
 
 
-/**
- * Constants for vertex/fragment shaders
- */
-struct pipe_constant_buffer
-{
-   struct pipe_buffer *buffer;
-};
-
-
 struct pipe_shader_state
 {
    const struct tgsi_token *tokens;
@@ -376,6 +363,11 @@ struct pipe_vertex_element
    /** Offset of this attribute, in bytes, from the start of the vertex */
    unsigned src_offset;
 
+   /** Instance data rate divisor. 0 means this is per-vertex data,
+    *  n means per-instance data used for n consecutive instances (n > 0).
+    */
+   unsigned instance_divisor;
+
    /** Which vertex_buffer (as given to pipe->set_vertex_buffer()) does
     * this attribute live in?
     */
diff --git a/src/gallium/include/state_tracker/drm_api.h b/src/gallium/include/state_tracker/drm_api.h
index 4d1259e1ee7..bb928928c90 100644
--- a/src/gallium/include/state_tracker/drm_api.h
+++ b/src/gallium/include/state_tracker/drm_api.h
@@ -28,6 +28,8 @@ struct drm_create_screen_arg {
 
 struct drm_api
 {
+        const char *name;
+
 	/**
 	 * Special buffer functions
 	 */
diff --git a/src/gallium/state_trackers/dri/dri_drawable.c b/src/gallium/state_trackers/dri/dri_drawable.c
index 1058dd38c25..0fdfa96b35a 100644
--- a/src/gallium/state_trackers/dri/dri_drawable.c
+++ b/src/gallium/state_trackers/dri/dri_drawable.c
@@ -180,6 +180,7 @@ dri_get_buffers(__DRIdrawable * dPriv)
 
       switch (buffers[i].attachment) {
       case __DRI_BUFFER_FRONT_LEFT:
+	 continue;
       case __DRI_BUFFER_FAKE_FRONT_LEFT:
 	 index = ST_SURFACE_FRONT_LEFT;
 	 format = drawable->color_format;
@@ -268,6 +269,14 @@ void dri2_set_tex_buffer(__DRIcontext *pDRICtx, GLint target,
 }
 
 void
+dri_update_buffer(struct pipe_screen *screen, void *context_private)
+{
+   struct dri_context *ctx = (struct dri_context *)context_private;
+
+   dri_get_buffers(ctx->dPriv);
+}
+
+void
 dri_flush_frontbuffer(struct pipe_screen *screen,
 		      struct pipe_surface *surf, void *context_private)
 {
@@ -364,6 +373,7 @@ dri_create_buffer(__DRIscreen * sPriv,
    /* TODO incase of double buffer visual, delay fake creation */
    i = 0;
    drawable->attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
+   drawable->attachments[i++] = __DRI_BUFFER_FAKE_FRONT_LEFT;
 
    if (visual->doubleBufferMode)
       drawable->attachments[i++] = __DRI_BUFFER_BACK_LEFT;
diff --git a/src/gallium/state_trackers/dri/dri_drawable.h b/src/gallium/state_trackers/dri/dri_drawable.h
index 80bb5d7f617..8bc59cb4c3d 100644
--- a/src/gallium/state_trackers/dri/dri_drawable.h
+++ b/src/gallium/state_trackers/dri/dri_drawable.h
@@ -81,6 +81,9 @@ dri_create_buffer(__DRIscreen * sPriv,
 		  const __GLcontextModes * visual, boolean isPixmap);
 
 void
+dri_update_buffer(struct pipe_screen *screen, void *context_private);
+
+void
 dri_flush_frontbuffer(struct pipe_screen *screen,
 		      struct pipe_surface *surf, void *context_private);
 
diff --git a/src/gallium/state_trackers/dri/dri_screen.c b/src/gallium/state_trackers/dri/dri_screen.c
index bb12baf2815..793db087ee1 100644
--- a/src/gallium/state_trackers/dri/dri_screen.c
+++ b/src/gallium/state_trackers/dri/dri_screen.c
@@ -308,6 +308,7 @@ dri_init_screen2(__DRIscreen * sPriv)
    }
 
    /* We need to hook in here */
+   screen->pipe_screen->update_buffer = dri_update_buffer;
    screen->pipe_screen->flush_frontbuffer = dri_flush_frontbuffer;
 
    driParseOptionInfo(&screen->optionCache,
diff --git a/src/gallium/state_trackers/egl_g3d/Makefile b/src/gallium/state_trackers/egl_g3d/Makefile
new file mode 100644
index 00000000000..213eb3e815b
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/Makefile
@@ -0,0 +1,72 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+common_INCLUDES = \
+	-I. \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/auxiliary \
+	-I$(TOP)/src/egl/main \
+	-I$(TOP)/include
+
+common_SOURCES = $(wildcard common/*.c)
+common_OBJECTS = $(common_SOURCES:.c=.o)
+
+
+x11_INCLUDES = \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/src/glx/x11 \
+	-I$(TOP)/src/mesa \
+	$(shell pkg-config --cflags-only-I libdrm)
+
+x11_SOURCES = $(wildcard x11/*.c) $(TOP)/src/glx/x11/dri2.c
+x11_OBJECTS = $(x11_SOURCES:.c=.o)
+
+
+kms_INCLUDES = $(shell pkg-config --cflags-only-I libdrm)
+kms_SOURCES = $(wildcard kms/*.c)
+kms_OBJECTS = $(kms_SOURCES:.c=.o)
+
+
+ALL_INCLUDES = $(common_INCLUDES) $(x11_INCLUDES) $(kms_INCLUDES)
+ALL_SOURCES = $(common_SOURCES) $(x11_SOURCES) $(kms_SOURCES)
+ALL_OBJECTS = $(common_OBJECTS) $(x11_OBJECTS) $(kms_OBJECTS)
+
+##### TARGETS #####
+
+EGL_DISPLAYS_MODS = $(foreach dpy, $(EGL_DISPLAYS), libegl$(dpy).a)
+
+default: depend $(EGL_DISPLAYS_MODS)
+
+
+libeglx11.a: $(x11_OBJECTS) $(common_OBJECTS) Makefile
+	$(MKLIB) -o eglx11 -static $(x11_OBJECTS) $(common_OBJECTS)
+
+libeglkms.a: $(kms_OBJECTS) $(common_OBJECTS) Makefile
+	$(MKLIB) -o eglkms -static $(kms_OBJECTS) $(common_OBJECTS)
+
+depend: 
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(ALL_INCLUDES) $(ALL_SOURCES) 2> /dev/null
+
+clean:
+	rm -f $(ALL_OBJECTS)
+	rm -f $(EGL_DISPLAYS_MODS)
+	rm -f depend depend.bak
+
+# Dummy target
+install:
+	@echo -n ""
+
+##### RULES #####
+
+$(common_OBJECTS): %.o: %.c
+	$(CC) -c $(common_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
+
+$(x11_OBJECTS): %.o: %.c
+	$(CC) -c $(common_INCLUDES) $(x11_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
+
+$(kms_OBJECTS): %.o: %.c
+	$(CC) -c $(common_INCLUDES) $(kms_INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
+
+sinclude depend
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_g3d.c b/src/gallium/state_trackers/egl_g3d/common/egl_g3d.c
new file mode 100644
index 00000000000..51da8e19f5e
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/common/egl_g3d.c
@@ -0,0 +1,1147 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 <assert.h>
+#include <string.h>
+#include "pipe/p_screen.h"
+#include "util/u_memory.h"
+#include "egldriver.h"
+#include "eglcurrent.h"
+#include "eglconfigutil.h"
+#include "egllog.h"
+
+#include "native.h"
+#include "egl_g3d.h"
+#include "egl_st.h"
+
+/**
+ * Validate the draw/read surfaces of the context.
+ */
+static void
+egl_g3d_validate_context(_EGLDisplay *dpy, _EGLContext *ctx)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct pipe_screen *screen = gdpy->native->screen;
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   EGLint num_surfaces;
+   EGLint s, i;
+
+   /* validate draw and/or read buffers */
+   num_surfaces = (gctx->base.ReadSurface == gctx->base.DrawSurface) ? 1 : 2;
+   for (s = 0; s < num_surfaces; s++) {
+      struct pipe_texture *textures[NUM_NATIVE_ATTACHMENTS];
+      struct egl_g3d_surface *gsurf;
+      struct egl_g3d_buffer *gbuf;
+
+      if (s == 0) {
+         gsurf = egl_g3d_surface(gctx->base.DrawSurface);
+         gbuf = &gctx->draw;
+      }
+      else {
+         gsurf = egl_g3d_surface(gctx->base.ReadSurface);
+         gbuf = &gctx->read;
+      }
+
+      if (!gctx->force_validate) {
+         unsigned int seq_num;
+
+         gsurf->native->validate(gsurf->native,
+               gbuf->native_atts, gbuf->num_atts,
+               &seq_num, NULL, NULL, NULL);
+         /* skip validation */
+         if (gsurf->sequence_number == seq_num)
+            continue;
+      }
+
+      gsurf->native->validate(gsurf->native,
+            gbuf->native_atts, gbuf->num_atts,
+            &gsurf->sequence_number, textures,
+            &gsurf->base.Width, &gsurf->base.Height);
+      for (i = 0; i < gbuf->num_atts; i++) {
+         struct pipe_texture *pt = textures[i];
+         struct pipe_surface *ps;
+
+         if (pt) {
+            ps = screen->get_tex_surface(screen, pt, 0, 0, 0,
+                  PIPE_BUFFER_USAGE_GPU_READ |
+                  PIPE_BUFFER_USAGE_GPU_WRITE);
+            gctx->stapi->st_set_framebuffer_surface(gbuf->st_fb,
+                  gbuf->st_atts[i], ps);
+
+            if (gbuf->native_atts[i] == gsurf->render_att)
+               pipe_surface_reference(&gsurf->render_surface, ps);
+
+            pipe_surface_reference(&ps, NULL);
+            pipe_texture_reference(&pt, NULL);
+         }
+      }
+
+      gctx->stapi->st_resize_framebuffer(gbuf->st_fb,
+            gsurf->base.Width, gsurf->base.Height);
+   }
+
+   gctx->force_validate = EGL_FALSE;
+
+}
+
+/**
+ * Create a st_framebuffer.
+ */
+static struct st_framebuffer *
+create_framebuffer(_EGLContext *ctx, _EGLSurface *surf)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+   struct egl_g3d_config *gconf = egl_g3d_config(gsurf->base.Config);
+
+   return gctx->stapi->st_create_framebuffer(&gconf->native->mode,
+         gconf->native->color_format, gconf->native->depth_format,
+         gconf->native->stencil_format,
+         gsurf->base.Width, gsurf->base.Height, &gsurf->base);
+}
+
+/**
+ * Update the attachments of draw/read surfaces.
+ */
+static void
+egl_g3d_route_context(_EGLDisplay *dpy, _EGLContext *ctx)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   const uint st_att_map[NUM_NATIVE_ATTACHMENTS] = {
+      ST_SURFACE_FRONT_LEFT,
+      ST_SURFACE_BACK_LEFT,
+      ST_SURFACE_FRONT_RIGHT,
+      ST_SURFACE_BACK_RIGHT,
+   };
+   EGLint s, i;
+
+   /* route draw and read buffers' attachments */
+   for (s = 0; s < 2; s++) {
+      struct egl_g3d_surface *gsurf;
+      struct egl_g3d_buffer *gbuf;
+
+      if (s == 0) {
+         gsurf = egl_g3d_surface(gctx->base.DrawSurface);
+         gbuf = &gctx->draw;
+      }
+      else {
+         gsurf = egl_g3d_surface(gctx->base.ReadSurface);
+         gbuf = &gctx->read;
+      }
+
+      gbuf->native_atts[0] = gsurf->render_att;
+      gbuf->num_atts = 1;
+
+      for (i = 0; i < gbuf->num_atts; i++)
+         gbuf->st_atts[i] = st_att_map[gbuf->native_atts[i]];
+
+      /* FIXME OpenGL defaults to draw the front or back buffer when the
+       * context is single-buffered or double-buffered respectively.  In EGL,
+       * however, the buffer to be drawn is determined by the surface, instead
+       * of the context.  As a result, rendering to a pixmap surface with a
+       * double-buffered context does not work as expected.
+       *
+       * gctx->stapi->st_draw_front_buffer(gctx->st_ctx, natt ==
+       *    NATIVE_ATTACHMENT_FRONT_LEFT);
+       */
+
+      /*
+       * FIXME If the back buffer is asked for here, and the front buffer is
+       * later needed by the client API (e.g. glDrawBuffer is called to draw
+       * the front buffer), it will create a new pipe texture and draw there.
+       * One fix is to ask for both buffers here, but it would be a waste if
+       * the front buffer is never used.  A better fix is to add a callback to
+       * the pipe screen with context private (just like flush_frontbuffer).
+       */
+   }
+}
+
+/**
+ * Reallocate the context's framebuffers after draw/read surfaces change.
+ */
+static EGLBoolean
+egl_g3d_realloc_context(_EGLDisplay *dpy, _EGLContext *ctx)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   struct egl_g3d_surface *gdraw = egl_g3d_surface(gctx->base.DrawSurface);
+   struct egl_g3d_surface *gread = egl_g3d_surface(gctx->base.ReadSurface);
+
+   /* unreference the old framebuffers */
+   if (gctx->draw.st_fb) {
+      EGLBoolean is_equal = (gctx->draw.st_fb == gctx->read.st_fb);
+      void *priv;
+
+      priv = gctx->stapi->st_framebuffer_private(gctx->draw.st_fb);
+      if (!gdraw || priv != (void *) &gdraw->base) {
+         gctx->stapi->st_unreference_framebuffer(gctx->draw.st_fb);
+         gctx->draw.st_fb = NULL;
+         gctx->draw.num_atts = 0;
+      }
+
+      if (is_equal) {
+         gctx->read.st_fb = NULL;
+         gctx->draw.num_atts = 0;
+      }
+      else {
+         priv = gctx->stapi->st_framebuffer_private(gctx->read.st_fb);
+         if (!gread || priv != (void *) &gread->base) {
+            gctx->stapi->st_unreference_framebuffer(gctx->read.st_fb);
+            gctx->read.st_fb = NULL;
+            gctx->draw.num_atts = 0;
+         }
+      }
+   }
+
+   if (!gdraw)
+      return EGL_TRUE;
+
+   /* create the draw fb */
+   if (!gctx->draw.st_fb) {
+      gctx->draw.st_fb = create_framebuffer(&gctx->base, &gdraw->base);
+      if (!gctx->draw.st_fb)
+         return EGL_FALSE;
+   }
+
+   /* create the read fb */
+   if (!gctx->read.st_fb) {
+      if (gread != gdraw) {
+         gctx->read.st_fb = create_framebuffer(&gctx->base, &gread->base);
+         if (!gctx->read.st_fb) {
+            gctx->stapi->st_unreference_framebuffer(gctx->draw.st_fb);
+            gctx->draw.st_fb = NULL;
+            return EGL_FALSE;
+         }
+      }
+      else {
+         /* there is no st_reference_framebuffer... */
+         gctx->read.st_fb = gctx->draw.st_fb;
+      }
+   }
+
+   egl_g3d_route_context(dpy, &gctx->base);
+   gctx->force_validate = EGL_TRUE;
+
+   return EGL_TRUE;
+}
+
+/**
+ * Return the current context of the given API.
+ */
+static struct egl_g3d_context *
+egl_g3d_get_current_context(EGLint api)
+{
+   _EGLThreadInfo *t = _eglGetCurrentThread();
+   EGLint api_index = _eglConvertApiToIndex(api);
+   return egl_g3d_context(t->CurrentContexts[api_index]);
+}
+
+/**
+ * Return the state tracker for the given context.
+ */
+static const struct egl_g3d_st *
+egl_g3d_choose_st(_EGLDriver *drv, _EGLContext *ctx)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   const struct egl_g3d_st *stapi;
+   EGLint idx = -1;
+
+   switch (ctx->ClientAPI) {
+   case EGL_OPENGL_ES_API:
+      switch (ctx->ClientVersion) {
+      case 1:
+         idx = EGL_G3D_ST_OPENGL_ES;
+         break;
+      case 2:
+         idx = EGL_G3D_ST_OPENGL_ES2;
+         break;
+      default:
+         _eglLog(_EGL_WARNING, "unknown client version %d",
+               ctx->ClientVersion);
+         break;
+      }
+      break;
+   case EGL_OPENVG_API:
+      idx = EGL_G3D_ST_OPENVG;
+      break;
+   case EGL_OPENGL_API:
+      idx = EGL_G3D_ST_OPENGL;
+      break;
+   default:
+      _eglLog(_EGL_WARNING, "unknown client API 0x%04x", ctx->ClientAPI);
+      break;
+   }
+
+   stapi = (idx >= 0) ? gdrv->stapis[idx] : NULL;
+   return stapi;
+}
+
+/**
+ * Return an API mask that consists of the state trackers that supports the
+ * given mode.
+ *
+ * FIXME add st_is_mode_supported()?
+ */
+static EGLint
+get_mode_api_mask(const __GLcontextModes *mode, EGLint api_mask)
+{
+   EGLint check;
+
+   /* OpenGL ES 1.x and 2.x are checked together */
+   check = EGL_OPENGL_ES_BIT | EGL_OPENGL_ES2_BIT;
+   if (api_mask & check) {
+      /* this is required by EGL, not by OpenGL ES */
+      if (mode->drawableType & GLX_WINDOW_BIT && !mode->doubleBufferMode)
+         api_mask &= ~check;
+   }
+
+   check = EGL_OPENVG_BIT;
+   if (api_mask & check) {
+      /* vega st needs the depth/stencil rb */
+      if (!mode->depthBits && !mode->stencilBits)
+         api_mask &= ~check;
+   }
+
+   return api_mask;
+}
+
+#ifdef EGL_MESA_screen_surface
+
+static void
+egl_g3d_add_screens(_EGLDriver *drv, _EGLDisplay *dpy)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   const struct native_connector **native_connectors;
+   EGLint num_connectors, i;
+
+   native_connectors =
+      gdpy->native->modeset->get_connectors(gdpy->native, &num_connectors, NULL);
+   if (!num_connectors) {
+      if (native_connectors)
+         free(native_connectors);
+      return;
+   }
+
+   for (i = 0; i < num_connectors; i++) {
+      const struct native_connector *nconn = native_connectors[i];
+      struct egl_g3d_screen *gscr;
+      const struct native_mode **native_modes;
+      EGLint num_modes, j;
+
+      /* TODO support for hotplug */
+      native_modes =
+         gdpy->native->modeset->get_modes(gdpy->native, nconn, &num_modes);
+      if (!num_modes) {
+         if (native_modes)
+            free(native_modes);
+         continue;
+      }
+
+      gscr = CALLOC_STRUCT(egl_g3d_screen);
+      if (!gscr) {
+         free(native_modes);
+         continue;
+      }
+
+      _eglInitScreen(&gscr->base);
+
+      for (j = 0; j < num_modes; j++) {
+         const struct native_mode *nmode = native_modes[j];
+         _EGLMode *mode;
+
+         mode = _eglAddNewMode(&gscr->base, nmode->width, nmode->height,
+               nmode->refresh_rate, nmode->desc);
+         if (!mode)
+            break;
+         /* gscr->native_modes and gscr->base.Modes should be consistent */
+         assert(mode == &gscr->base.Modes[j]);
+      }
+
+      gscr->native = nconn;
+      gscr->native_modes = native_modes;
+
+      _eglAddScreen(dpy, &gscr->base);
+   }
+
+   free(native_connectors);
+}
+
+#endif /* EGL_MESA_screen_surface */
+
+/**
+ * Add configs to display and return the next config ID.
+ */
+static EGLint
+egl_g3d_add_configs(_EGLDriver *drv, _EGLDisplay *dpy, EGLint id)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   const struct native_config **native_configs;
+   int num_configs, i;
+
+   native_configs = gdpy->native->get_configs(gdpy->native,
+         &num_configs);
+   if (!num_configs) {
+      if (native_configs)
+         free(native_configs);
+      return id;
+   }
+
+   for (i = 0; i < num_configs; i++) {
+      EGLint api_mask;
+      struct egl_g3d_config *gconf;
+      EGLBoolean valid;
+
+      api_mask = get_mode_api_mask(&native_configs[i]->mode, gdrv->api_mask);
+      if (!api_mask) {
+         _eglLog(_EGL_DEBUG, "no state tracker supports config 0x%x",
+               native_configs[i]->mode.visualID);
+         continue;
+      }
+
+      gconf = CALLOC_STRUCT(egl_g3d_config);
+      if (!gconf)
+         continue;
+
+      _eglInitConfig(&gconf->base, id);
+      valid = _eglConfigFromContextModesRec(&gconf->base,
+            &native_configs[i]->mode, api_mask, api_mask);
+      if (valid) {
+#ifdef EGL_MESA_screen_surface
+         /* check if scanout surface bit is set */
+         if (native_configs[i]->scanout_bit) {
+            EGLint val = GET_CONFIG_ATTRIB(&gconf->base, EGL_SURFACE_TYPE);
+            val |= EGL_SCREEN_BIT_MESA;
+            SET_CONFIG_ATTRIB(&gconf->base, EGL_SURFACE_TYPE, val);
+         }
+#endif
+         valid = _eglValidateConfig(&gconf->base, EGL_FALSE);
+      }
+      if (!valid) {
+         _eglLog(_EGL_DEBUG, "skip invalid config 0x%x",
+               native_configs[i]->mode.visualID);
+         free(gconf);
+         continue;
+      }
+
+      gconf->native = native_configs[i];
+      _eglAddConfig(dpy, &gconf->base);
+      id++;
+   }
+
+   free(native_configs);
+   return id;
+}
+
+/**
+ * Flush the front buffer of the context's draw surface.
+ */
+static void
+egl_g3d_flush_frontbuffer(struct pipe_screen *screen,
+                          struct pipe_surface *surf, void *context_private)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(context_private);
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(gctx->base.DrawSurface);
+
+   if (gsurf)
+      gsurf->native->flush_frontbuffer(gsurf->native);
+}
+
+/**
+ * Re-validate the context.
+ */
+static void
+egl_g3d_update_buffer(struct pipe_screen *screen, void *context_private)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(context_private);
+
+   /**
+    * It is likely that the surface has changed when this function is called.
+    * Set force_validate to skip an unnecessary check.
+    */
+   gctx->force_validate = EGL_TRUE;
+   egl_g3d_validate_context(gctx->base.Display, &gctx->base);
+}
+
+static EGLBoolean
+egl_g3d_terminate(_EGLDriver *drv, _EGLDisplay *dpy)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   EGLint i;
+
+   _eglReleaseDisplayResources(drv, dpy);
+   _eglCleanupDisplay(dpy);
+
+   if (dpy->Screens) {
+      for (i = 0; i < dpy->NumScreens; i++) {
+         struct egl_g3d_screen *gscr = egl_g3d_screen(dpy->Screens[i]);
+         free(gscr->native_modes);
+         free(gscr);
+      }
+      free(dpy->Screens);
+   }
+
+   if (gdpy->native)
+      gdpy->native->destroy(gdpy->native);
+
+   free(gdpy);
+   dpy->DriverData = NULL;
+
+   return EGL_TRUE;
+}
+
+static EGLBoolean
+egl_g3d_initialize(_EGLDriver *drv, _EGLDisplay *dpy,
+                   EGLint *major, EGLint *minor)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   struct egl_g3d_display *gdpy;
+
+   gdpy = CALLOC_STRUCT(egl_g3d_display);
+   if (!gdpy) {
+      _eglError(EGL_BAD_ALLOC, "eglInitialize");
+      goto fail;
+   }
+   dpy->DriverData = gdpy;
+
+   gdpy->native = native_create_display(dpy->NativeDisplay);
+   if (!gdpy->native) {
+      _eglError(EGL_NOT_INITIALIZED, "eglInitialize(no usable display)");
+      goto fail;
+   }
+
+   gdpy->native->screen->flush_frontbuffer = egl_g3d_flush_frontbuffer;
+   gdpy->native->screen->update_buffer = egl_g3d_update_buffer;
+
+   dpy->ClientAPIsMask = gdrv->api_mask;
+
+   if (egl_g3d_add_configs(drv, dpy, 1) == 1) {
+      _eglError(EGL_NOT_INITIALIZED, "eglInitialize(unable to add configs)");
+      goto fail;
+   }
+
+#ifdef EGL_MESA_screen_surface
+   /* enable MESA_screen_surface */
+   if (gdpy->native->modeset) {
+      dpy->Extensions.MESA_screen_surface = EGL_TRUE;
+      egl_g3d_add_screens(drv, dpy);
+   }
+#endif
+
+   *major = 1;
+   *minor = 4;
+
+   return EGL_TRUE;
+
+fail:
+   if (gdpy)
+      egl_g3d_terminate(drv, dpy);
+   return EGL_FALSE;
+}
+
+static _EGLContext *
+egl_g3d_create_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
+                       _EGLContext *share, const EGLint *attribs)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_context *gshare = egl_g3d_context(share);
+   struct egl_g3d_config *gconf = egl_g3d_config(conf);
+   struct egl_g3d_context *gctx;
+   const __GLcontextModes *mode;
+
+   gctx = CALLOC_STRUCT(egl_g3d_context);
+   if (!gctx) {
+      _eglError(EGL_BAD_ALLOC, "eglCreateContext");
+      return NULL;
+   }
+
+   if (!_eglInitContext(drv, &gctx->base, conf, attribs)) {
+      free(gctx);
+      return NULL;
+   }
+
+   gctx->stapi = egl_g3d_choose_st(drv, &gctx->base);
+   if (!gctx->stapi) {
+      free(gctx);
+      return NULL;
+   }
+
+   mode = &gconf->native->mode;
+   gctx->pipe =
+      gdpy->native->create_context(gdpy->native, (void *) &gctx->base);
+   if (!gctx->pipe) {
+      free(gctx);
+      return NULL;
+   }
+
+   gctx->st_ctx = gctx->stapi->st_create_context(gctx->pipe, mode,
+                        (gshare) ? gshare->st_ctx : NULL);
+   if (!gctx->st_ctx) {
+      gctx->pipe->destroy(gctx->pipe);
+      free(gctx);
+      return NULL;
+   }
+
+   return &gctx->base;
+}
+
+static EGLBoolean
+egl_g3d_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+
+   if (_eglIsContextBound(&gctx->base))
+      return EGL_TRUE;
+
+   egl_g3d_realloc_context(dpy, &gctx->base);
+
+   /* it will destroy pipe context */
+   gctx->stapi->st_destroy_context(gctx->st_ctx);
+
+   free(gctx);
+
+   return EGL_TRUE;
+}
+
+static EGLBoolean
+init_surface_geometry(_EGLSurface *surf)
+{
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+
+   return gsurf->native->validate(gsurf->native, NULL, 0,
+         &gsurf->sequence_number, NULL,
+         &gsurf->base.Width, &gsurf->base.Height);
+}
+
+static _EGLSurface *
+egl_g3d_create_window_surface(_EGLDriver *drv, _EGLDisplay *dpy,
+                              _EGLConfig *conf, EGLNativeWindowType win,
+                              const EGLint *attribs)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_config *gconf = egl_g3d_config(conf);
+   struct egl_g3d_surface *gsurf;
+
+   gsurf = CALLOC_STRUCT(egl_g3d_surface);
+   if (!gsurf) {
+      _eglError(EGL_BAD_ALLOC, "eglCreateWindowSurface");
+      return NULL;
+   }
+
+   if (!_eglInitSurface(drv, &gsurf->base, EGL_WINDOW_BIT, conf, attribs)) {
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->native =
+      gdpy->native->create_window_surface(gdpy->native, win, gconf->native);
+   if (!gsurf->native) {
+      free(gsurf);
+      return NULL;
+   }
+
+   if (!init_surface_geometry(&gsurf->base)) {
+      gsurf->native->destroy(gsurf->native);
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->render_att = (gsurf->base.RenderBuffer == EGL_SINGLE_BUFFER ||
+                        !gconf->native->mode.doubleBufferMode) ?
+      NATIVE_ATTACHMENT_FRONT_LEFT : NATIVE_ATTACHMENT_BACK_LEFT;
+
+   return &gsurf->base;
+}
+
+static _EGLSurface *
+egl_g3d_create_pixmap_surface(_EGLDriver *drv, _EGLDisplay *dpy,
+                              _EGLConfig *conf, EGLNativePixmapType pix,
+                              const EGLint *attribs)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_config *gconf = egl_g3d_config(conf);
+   struct egl_g3d_surface *gsurf;
+
+   gsurf = CALLOC_STRUCT(egl_g3d_surface);
+   if (!gsurf) {
+      _eglError(EGL_BAD_ALLOC, "eglCreatePixmapSurface");
+      return NULL;
+   }
+
+   if (!_eglInitSurface(drv, &gsurf->base, EGL_PIXMAP_BIT, conf, attribs)) {
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->native =
+      gdpy->native->create_pixmap_surface(gdpy->native, pix, gconf->native);
+   if (!gsurf->native) {
+      free(gsurf);
+      return NULL;
+   }
+
+   if (!init_surface_geometry(&gsurf->base)) {
+      gsurf->native->destroy(gsurf->native);
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->render_att = NATIVE_ATTACHMENT_FRONT_LEFT;
+
+   return &gsurf->base;
+}
+
+static _EGLSurface *
+egl_g3d_create_pbuffer_surface(_EGLDriver *drv, _EGLDisplay *dpy,
+                               _EGLConfig *conf, const EGLint *attribs)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_config *gconf = egl_g3d_config(conf);
+   struct egl_g3d_surface *gsurf;
+
+   gsurf = CALLOC_STRUCT(egl_g3d_surface);
+   if (!gsurf) {
+      _eglError(EGL_BAD_ALLOC, "eglCreatePbufferSurface");
+      return NULL;
+   }
+
+   if (!_eglInitSurface(drv, &gsurf->base, EGL_PBUFFER_BIT, conf, attribs)) {
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->native =
+      gdpy->native->create_pbuffer_surface(gdpy->native, gconf->native,
+            gsurf->base.Width, gsurf->base.Height);
+   if (!gsurf->native) {
+      free(gsurf);
+      return NULL;
+   }
+
+   if (!init_surface_geometry(&gsurf->base)) {
+      gsurf->native->destroy(gsurf->native);
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->render_att = (!gconf->native->mode.doubleBufferMode) ?
+      NATIVE_ATTACHMENT_FRONT_LEFT : NATIVE_ATTACHMENT_BACK_LEFT;
+
+   return &gsurf->base;
+}
+
+static EGLBoolean
+egl_g3d_destroy_surface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
+{
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+
+   if (_eglIsSurfaceBound(&gsurf->base))
+         return EGL_TRUE;
+
+   pipe_surface_reference(&gsurf->render_surface, NULL);
+   gsurf->native->destroy(gsurf->native);
+   free(gsurf);
+   return EGL_TRUE;
+}
+
+static EGLBoolean
+egl_g3d_make_current(_EGLDriver *drv, _EGLDisplay *dpy,
+                     _EGLSurface *draw, _EGLSurface *read, _EGLContext *ctx)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   struct egl_g3d_context *old_gctx;
+   EGLint api;
+   EGLBoolean ok = EGL_TRUE;
+
+   /* find the old context */
+   api = (gctx) ? gctx->base.ClientAPI : eglQueryAPI();
+   old_gctx = egl_g3d_get_current_context(api);
+   if (old_gctx && !_eglIsContextLinked(&old_gctx->base))
+         old_gctx = NULL;
+
+   if (!_eglMakeCurrent(drv, dpy, draw, read, ctx))
+      return EGL_FALSE;
+
+   if (old_gctx) {
+      /* flush old context */
+      old_gctx->stapi->st_flush(old_gctx->st_ctx,
+            PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL);
+
+      /*
+       * The old context is no longer current, and egl_g3d_realloc_context()
+       * should be called to destroy the framebuffers.  However, it is possible
+       * that it will be made current again with the same draw/read surfaces.
+       * It might be better to keep it around.
+       */
+   }
+
+   if (gctx) {
+      struct egl_g3d_surface *gdraw = egl_g3d_surface(draw);
+
+      ok = egl_g3d_realloc_context(dpy, &gctx->base);
+      if (ok) {
+         ok = gctx->stapi->st_make_current(gctx->st_ctx,
+               gctx->draw.st_fb, gctx->read.st_fb);
+         if (ok) {
+            egl_g3d_validate_context(dpy, &gctx->base);
+            if (gdraw->base.Type == EGL_WINDOW_BIT) {
+               gctx->base.WindowRenderBuffer =
+                  (gdraw->render_att == NATIVE_ATTACHMENT_FRONT_LEFT) ?
+                  EGL_SINGLE_BUFFER : EGL_BACK_BUFFER;
+            }
+         }
+      }
+   }
+   else if (old_gctx) {
+      ok = old_gctx->stapi->st_make_current(NULL, NULL, NULL);
+      old_gctx->base.WindowRenderBuffer = EGL_NONE;
+   }
+
+   return ok;
+}
+
+static EGLBoolean
+egl_g3d_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
+{
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+   _EGLContext *ctx = _eglGetCurrentContext();
+   struct egl_g3d_context *gctx = NULL;
+
+   /* no-op for pixmap or pbuffer surface */
+   if (gsurf->base.Type == EGL_PIXMAP_BIT ||
+       gsurf->base.Type == EGL_PBUFFER_BIT)
+      return EGL_TRUE;
+
+   /* or when the surface is single-buffered */
+   if (gsurf->render_att == NATIVE_ATTACHMENT_FRONT_LEFT)
+      return EGL_TRUE;
+
+   if (ctx && ctx->DrawSurface == surf)
+      gctx = egl_g3d_context(ctx);
+
+   /* flush if the surface is current */
+   if (gctx)
+      gctx->stapi->st_notify_swapbuffers(gctx->draw.st_fb);
+
+   /*
+    * We drew on the back buffer, unless there was no back buffer.
+    * In that case, we drew on the front buffer.  Either case, we call
+    * swap_buffers.
+    */
+   if (!gsurf->native->swap_buffers(gsurf->native))
+      return EGL_FALSE;
+
+   if (gctx) {
+      struct egl_g3d_config *gconf = egl_g3d_config(gsurf->base.Config);
+
+      /* force validation if the swap method is not copy */
+      if (gconf->native->mode.swapMethod != GLX_SWAP_COPY_OML) {
+         gctx->force_validate = EGL_TRUE;
+         egl_g3d_validate_context(dpy, &gctx->base);
+      }
+   }
+
+   return EGL_TRUE;
+}
+
+static EGLBoolean
+egl_g3d_wait_client(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
+{
+   struct egl_g3d_context *gctx = egl_g3d_context(ctx);
+   gctx->stapi->st_finish(gctx->st_ctx);
+   return EGL_TRUE;
+}
+
+static EGLBoolean
+egl_g3d_wait_native(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine)
+{
+   _EGLSurface *surf = _eglGetCurrentSurface(EGL_DRAW);
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+
+   if (engine != EGL_CORE_NATIVE_ENGINE)
+      return _eglError(EGL_BAD_PARAMETER, "eglWaitNative");
+
+   if (gsurf)
+      gsurf->native->wait(gsurf->native);
+
+   return EGL_TRUE;
+}
+
+static _EGLProc
+egl_g3d_get_proc_address(_EGLDriver *drv, const char *procname)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   _EGLProc proc;
+   EGLint i;
+
+   for (i = 0; i < NUM_EGL_G3D_STS; i++) {
+      const struct egl_g3d_st *stapi = gdrv->stapis[i];
+      if (stapi) {
+         proc = (_EGLProc) stapi->st_get_proc_address(procname);
+         if (proc)
+            return proc;
+      }
+   }
+
+   return (_EGLProc) NULL;
+}
+
+static EGLBoolean
+egl_g3d_bind_tex_image(_EGLDriver *drv, _EGLDisplay *dpy,
+                       _EGLSurface *surf, EGLint buffer)
+{
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+   struct egl_g3d_context *gctx;
+   enum pipe_format target_format;
+   int target;
+
+   if (!gsurf || gsurf->base.Type != EGL_PBUFFER_BIT)
+      return _eglError(EGL_BAD_SURFACE, "eglBindTexImage");
+   if (buffer != EGL_BACK_BUFFER)
+      return _eglError(EGL_BAD_PARAMETER, "eglBindTexImage");
+   if (gsurf->base.BoundToTexture)
+      return _eglError(EGL_BAD_ACCESS, "eglBindTexImage");
+
+   switch (gsurf->base.TextureFormat) {
+   case EGL_TEXTURE_RGB:
+      target_format = PIPE_FORMAT_R8G8B8_UNORM;
+      break;
+   case EGL_TEXTURE_RGBA:
+      target_format = PIPE_FORMAT_A8R8G8B8_UNORM;
+      break;
+   default:
+      return _eglError(EGL_BAD_MATCH, "eglBindTexImage");
+   }
+
+   switch (gsurf->base.TextureTarget) {
+   case EGL_TEXTURE_2D:
+      target = ST_TEXTURE_2D;
+      break;
+   default:
+      return _eglError(EGL_BAD_MATCH, "eglBindTexImage");
+   }
+
+   /* flush properly if the surface is bound */
+   if (gsurf->base.Binding) {
+      gctx = egl_g3d_context(gsurf->base.Binding);
+      gctx->stapi->st_flush(gctx->st_ctx,
+            PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL);
+   }
+
+   /* XXX change to EGL_OPENGL_ES_API once OpenGL ES is merged */
+   gctx = egl_g3d_get_current_context(EGL_OPENGL_API);
+   if (gctx) {
+      if (!gsurf->render_surface)
+         return EGL_FALSE;
+
+      gctx->stapi->st_bind_texture_surface(gsurf->render_surface,
+            target, gsurf->base.MipmapLevel, target_format);
+      gsurf->base.BoundToTexture = EGL_TRUE;
+   }
+
+   return EGL_TRUE;
+}
+
+static EGLBoolean
+egl_g3d_release_tex_image(_EGLDriver *drv, _EGLDisplay *dpy,
+                          _EGLSurface *surf, EGLint buffer)
+{
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+
+   if (!gsurf || gsurf->base.Type != EGL_PBUFFER_BIT ||
+       !gsurf->base.BoundToTexture)
+      return _eglError(EGL_BAD_SURFACE, "eglReleaseTexImage");
+   if (buffer != EGL_BACK_BUFFER)
+      return _eglError(EGL_BAD_PARAMETER, "eglReleaseTexImage");
+
+   if (gsurf->render_surface) {
+      _EGLThreadInfo *t = _eglGetCurrentThread();
+      /* XXX change to EGL_OPENGL_ES_API once OpenGL ES is merged */
+      struct egl_g3d_context *gctx = egl_g3d_context(
+            t->CurrentContexts[_eglConvertApiToIndex(EGL_OPENGL_API)]);
+
+      /* what if the context the surface binds to is no longer current? */
+      if (gctx)
+         gctx->stapi->st_unbind_texture_surface(gsurf->render_surface,
+               ST_TEXTURE_2D, gsurf->base.MipmapLevel);
+   }
+
+   gsurf->base.BoundToTexture = EGL_FALSE;
+
+   return EGL_TRUE;
+}
+
+#ifdef EGL_MESA_screen_surface
+
+static _EGLSurface *
+egl_g3d_create_screen_surface(_EGLDriver *drv, _EGLDisplay *dpy,
+                              _EGLConfig *conf, const EGLint *attribs)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_config *gconf = egl_g3d_config(conf);
+   struct egl_g3d_surface *gsurf;
+
+   gsurf = CALLOC_STRUCT(egl_g3d_surface);
+   if (!gsurf) {
+      _eglError(EGL_BAD_ALLOC, "eglCreatePbufferSurface");
+      return NULL;
+   }
+
+   if (!_eglInitSurface(drv, &gsurf->base,
+            EGL_SCREEN_BIT_MESA, conf, attribs)) {
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->native =
+      gdpy->native->modeset->create_scanout_surface(gdpy->native,
+            gconf->native, gsurf->base.Width, gsurf->base.Height);
+   if (!gsurf->native) {
+      free(gsurf);
+      return NULL;
+   }
+
+   gsurf->render_att = (!gconf->native->mode.doubleBufferMode) ?
+      NATIVE_ATTACHMENT_FRONT_LEFT : NATIVE_ATTACHMENT_BACK_LEFT;
+
+   return &gsurf->base;
+}
+
+static EGLBoolean
+egl_g3d_show_screen_surface(_EGLDriver *drv, _EGLDisplay *dpy,
+                            _EGLScreen *scr, _EGLSurface *surf,
+                            _EGLMode *mode)
+{
+   struct egl_g3d_display *gdpy = egl_g3d_display(dpy);
+   struct egl_g3d_screen *gscr = egl_g3d_screen(scr);
+   struct egl_g3d_surface *gsurf = egl_g3d_surface(surf);
+   struct native_surface *nsurf;
+   const struct native_mode *nmode;
+   EGLBoolean changed;
+
+   if (gsurf) {
+      EGLint idx;
+
+      if (!mode)
+         return _eglError(EGL_BAD_MATCH, "eglShowSurfaceMESA");
+      if (gsurf->base.Type != EGL_SCREEN_BIT_MESA)
+         return _eglError(EGL_BAD_SURFACE, "eglShowScreenSurfaceMESA");
+      if (gsurf->base.Width < mode->Width || gsurf->base.Height < mode->Height)
+         return _eglError(EGL_BAD_MATCH,
+               "eglShowSurfaceMESA(surface smaller than mode size)");
+
+      /* find the index of the mode */
+      for (idx = 0; idx < gscr->base.NumModes; idx++)
+         if (mode == &gscr->base.Modes[idx])
+            break;
+      if (idx >= gscr->base.NumModes) {
+         return _eglError(EGL_BAD_MODE_MESA,
+               "eglShowSurfaceMESA(unknown mode)");
+      }
+
+      nsurf = gsurf->native;
+      nmode = gscr->native_modes[idx];
+   }
+   else {
+      if (mode)
+         return _eglError(EGL_BAD_MATCH, "eglShowSurfaceMESA");
+
+      /* disable the screen */
+      nsurf = NULL;
+      nmode = NULL;
+   }
+
+   /* TODO surface panning by CRTC choosing */
+   changed = gdpy->native->modeset->program(gdpy->native, 0, nsurf,
+         gscr->base.OriginX, gscr->base.OriginY, &gscr->native, 1, nmode);
+   if (changed) {
+      gscr->base.CurrentSurface = &gsurf->base;
+      gscr->base.CurrentMode = mode;
+   }
+
+   return changed;
+}
+
+#endif /* EGL_MESA_screen_surface */
+
+static void
+egl_g3d_unload(_EGLDriver *drv)
+{
+   struct egl_g3d_driver *gdrv = egl_g3d_driver(drv);
+   free(gdrv);
+}
+
+_EGLDriver *
+_eglMain(const char *args)
+{
+   static char driver_name[64];
+   struct egl_g3d_driver *gdrv;
+   EGLint i;
+
+   snprintf(driver_name, sizeof(driver_name),
+         "Gallium/%s", native_get_name());
+
+   gdrv = CALLOC_STRUCT(egl_g3d_driver);
+   if (!gdrv)
+      return NULL;
+
+   _eglInitDriverFallbacks(&gdrv->base);
+
+   gdrv->base.API.Initialize = egl_g3d_initialize;
+   gdrv->base.API.Terminate = egl_g3d_terminate;
+   gdrv->base.API.CreateContext = egl_g3d_create_context;
+   gdrv->base.API.DestroyContext = egl_g3d_destroy_context;
+   gdrv->base.API.CreateWindowSurface = egl_g3d_create_window_surface;
+   gdrv->base.API.CreatePixmapSurface = egl_g3d_create_pixmap_surface;
+   gdrv->base.API.CreatePbufferSurface = egl_g3d_create_pbuffer_surface;
+   gdrv->base.API.DestroySurface = egl_g3d_destroy_surface;
+   gdrv->base.API.MakeCurrent = egl_g3d_make_current;
+   gdrv->base.API.SwapBuffers = egl_g3d_swap_buffers;
+   gdrv->base.API.WaitClient = egl_g3d_wait_client;
+   gdrv->base.API.WaitNative = egl_g3d_wait_native;
+   gdrv->base.API.GetProcAddress = egl_g3d_get_proc_address;
+
+   gdrv->base.API.BindTexImage = egl_g3d_bind_tex_image;
+   gdrv->base.API.ReleaseTexImage = egl_g3d_release_tex_image;
+
+#ifdef EGL_MESA_screen_surface
+   gdrv->base.API.CreateScreenSurfaceMESA = egl_g3d_create_screen_surface;
+   gdrv->base.API.ShowScreenSurfaceMESA = egl_g3d_show_screen_surface;
+#endif
+
+   gdrv->base.Name = driver_name;
+   gdrv->base.Unload = egl_g3d_unload;
+
+   for (i = 0; i < NUM_EGL_G3D_STS; i++) {
+      gdrv->stapis[i] = egl_g3d_get_st(i);
+      if (gdrv->stapis[i])
+         gdrv->api_mask |= gdrv->stapis[i]->api_bit;
+   }
+
+   if (gdrv->api_mask)
+      _eglLog(_EGL_DEBUG, "Driver API mask: 0x%x", gdrv->api_mask);
+   else
+      _eglLog(_EGL_WARNING, "No supported client API");
+
+   return &gdrv->base;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_g3d.h b/src/gallium/state_trackers/egl_g3d/common/egl_g3d.h
new file mode 100644
index 00000000000..4c8b8dfe9e0
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/common/egl_g3d.h
@@ -0,0 +1,127 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 _EGL_G3D_H_
+#define _EGL_G3D_H_
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_context.h"
+#include "pipe/p_format.h"
+#include "egldriver.h"
+#include "egldisplay.h"
+#include "eglcontext.h"
+#include "eglsurface.h"
+#include "eglconfig.h"
+#include "eglscreen.h"
+#include "eglmode.h"
+
+#include "native.h"
+#include "egl_st.h"
+
+struct egl_g3d_driver {
+   _EGLDriver base;
+   const struct egl_g3d_st *stapis[NUM_EGL_G3D_STS];
+   EGLint api_mask;
+};
+
+struct egl_g3d_display {
+   struct native_display *native;
+};
+
+struct egl_g3d_buffer {
+   struct st_framebuffer *st_fb;
+   EGLint num_atts;
+   enum native_attachment native_atts[NUM_NATIVE_ATTACHMENTS];
+   uint st_atts[NUM_NATIVE_ATTACHMENTS];
+};
+
+struct egl_g3d_context {
+   _EGLContext base;
+
+   const struct egl_g3d_st *stapi;
+   struct pipe_context *pipe;
+
+   struct st_context *st_ctx;
+   EGLBoolean force_validate;
+   struct egl_g3d_buffer draw, read;
+};
+
+struct egl_g3d_surface {
+   _EGLSurface base;
+   struct native_surface *native;
+   enum native_attachment render_att;
+   struct pipe_surface *render_surface;
+   unsigned int sequence_number;
+};
+
+struct egl_g3d_config {
+   _EGLConfig base;
+   const struct native_config *native;
+};
+
+struct egl_g3d_screen {
+   _EGLScreen base;
+   const struct native_connector *native;
+   const struct native_mode **native_modes;
+};
+
+static INLINE struct egl_g3d_driver *
+egl_g3d_driver(_EGLDriver *drv)
+{
+   return (struct egl_g3d_driver *) drv;
+}
+
+static INLINE struct egl_g3d_display *
+egl_g3d_display(_EGLDisplay *dpy)
+{
+   /* note that it is not direct casting */
+   return (struct egl_g3d_display *) dpy->DriverData;
+}
+
+static INLINE struct egl_g3d_context *
+egl_g3d_context(_EGLContext *ctx)
+{
+   return (struct egl_g3d_context *) ctx;
+}
+
+static INLINE struct egl_g3d_surface *
+egl_g3d_surface(_EGLSurface *surf)
+{
+   return (struct egl_g3d_surface *) surf;
+}
+
+static INLINE struct egl_g3d_config *
+egl_g3d_config(_EGLConfig *conf)
+{
+   return (struct egl_g3d_config *) conf;
+}
+
+static INLINE struct egl_g3d_screen *
+egl_g3d_screen(_EGLScreen *scr)
+{
+   return (struct egl_g3d_screen *) scr;
+}
+
+#endif /* _EGL_G3D_H_ */
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_st.c b/src/gallium/state_trackers/egl_g3d/common/egl_st.c
new file mode 100644
index 00000000000..a88ff911cd5
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/common/egl_st.c
@@ -0,0 +1,131 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 <dlfcn.h>
+#include "pipe/p_compiler.h"
+#include "util/u_memory.h"
+#include "egllog.h"
+#include "EGL/egl.h" /* for EGL_api_BIT */
+
+#include "egl_st.h"
+
+#ifndef HAVE_DLADDR
+#define HAVE_DLADDR 1
+#endif
+
+#if HAVE_DLADDR
+
+static const char *
+egl_g3d_st_names[] = {
+#define ST_PUBLIC(name, ...) #name,
+#include "st_public_tmp.h"
+   NULL
+};
+
+static boolean
+egl_g3d_fill_st(struct egl_g3d_st *stapi, void *sym)
+{
+   st_proc *procs = (st_proc *) stapi;
+   void *handle;
+   Dl_info info;
+   const char **name;
+
+   if (!dladdr(sym, &info))
+      return FALSE;
+   handle = dlopen(info.dli_fname, RTLD_LAZY | RTLD_LOCAL | RTLD_NODELETE);
+   if (!handle)
+      return FALSE;
+
+   for (name = egl_g3d_st_names; *name; name++) {
+      st_proc proc = (st_proc) dlsym(handle, *name);
+      if (!proc) {
+         _eglLog(_EGL_WARNING, "%s is missing in %s", *name, info.dli_fname);
+         memset(stapi, 0, sizeof(*stapi));
+         dlclose(handle);
+         return FALSE;
+      }
+      *procs++ = proc;
+   }
+
+   dlclose(handle);
+   return TRUE;
+}
+
+#else /* HAVE_DLADDR */
+
+static boolean
+egl_g3d_fill_st(struct egl_g3d_st *stapi, void *sym)
+{
+#define ST_PUBLIC(name, ...) stapi->name = name;
+#include "st_public_tmp.h"
+   return TRUE;
+}
+
+#endif /* HAVE_DLADDR */
+
+static boolean
+egl_g3d_init_st(struct egl_g3d_st *stapi, const char *api)
+{
+   void *handle, *sym;
+   boolean res = FALSE;
+
+   /* already initialized */
+   if (stapi->st_notify_swapbuffers != NULL)
+      return TRUE;
+
+   handle = dlopen(NULL, RTLD_LAZY | RTLD_LOCAL);
+   if (!handle)
+      return FALSE;
+
+   sym = dlsym(handle, api);
+   if (sym && egl_g3d_fill_st(stapi, sym))
+      res = TRUE;
+
+   dlclose(handle);
+   return res;
+}
+
+static struct {
+   const char *symbol;
+   EGLint api_bit;
+} egl_g3d_st_info[NUM_EGL_G3D_STS] = {
+   { "st_api_OpenGL_ES1",  EGL_OPENGL_ES_BIT },
+   { "st_api_OpenVG",      EGL_OPENVG_BIT },
+   { "st_api_OpenGL_ES2",  EGL_OPENGL_ES2_BIT },
+   { "st_api_OpenGL",      EGL_OPENGL_BIT },
+};
+
+const struct egl_g3d_st *
+egl_g3d_get_st(enum egl_g3d_st_api api)
+{
+   static struct egl_g3d_st all_trackers[NUM_EGL_G3D_STS];
+
+   if (egl_g3d_init_st(&all_trackers[api], egl_g3d_st_info[api].symbol)) {
+      all_trackers[api].api_bit = egl_g3d_st_info[api].api_bit;
+      return &all_trackers[api];
+   }
+   else {
+      return NULL;
+   }
+}
diff --git a/src/gallium/state_trackers/egl_g3d/common/egl_st.h b/src/gallium/state_trackers/egl_g3d/common/egl_st.h
new file mode 100644
index 00000000000..8fb464bd3d7
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/common/egl_st.h
@@ -0,0 +1,73 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 _EGL_ST_H_
+#define _EGL_ST_H_
+
+#include "GL/gl.h" /* for GL types */
+#include "GL/internal/glcore.h"  /* for __GLcontextModes */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+
+/* avoid calling st functions directly */
+#if 1
+
+#define ST_SURFACE_FRONT_LEFT   0
+#define ST_SURFACE_BACK_LEFT    1
+#define ST_SURFACE_FRONT_RIGHT  2
+#define ST_SURFACE_BACK_RIGHT   3
+
+#define ST_TEXTURE_2D    0x2
+
+struct st_context;
+struct st_framebuffer;
+typedef void (*st_proc)();
+
+#else
+#include "state_tracker/st_public.h"
+#endif
+
+/* remember to update egl_g3d_get_st() when update the enums */
+enum egl_g3d_st_api {
+   EGL_G3D_ST_OPENGL_ES = 0,
+   EGL_G3D_ST_OPENVG,
+   EGL_G3D_ST_OPENGL_ES2,
+   EGL_G3D_ST_OPENGL,
+
+   NUM_EGL_G3D_STS
+};
+
+struct egl_g3d_st {
+#define ST_PUBLIC(name, ret, ...) ret (*name)(__VA_ARGS__);
+#include "st_public_tmp.h"
+   /* fields must be added here */
+   EGLint api_bit;
+};
+
+const struct egl_g3d_st *
+egl_g3d_get_st(enum egl_g3d_st_api api);
+
+#endif /* _EGL_ST_H_ */
diff --git a/src/gallium/state_trackers/egl_g3d/common/native.h b/src/gallium/state_trackers/egl_g3d/common/native.h
new file mode 100644
index 00000000000..76f0e0c78ac
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/common/native.h
@@ -0,0 +1,220 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 _NATIVE_H_
+#define _NATIVE_H_
+
+#include "EGL/egl.h"  /* for EGL native types */
+#include "GL/gl.h" /* for GL types needed by __GLcontextModes */
+#include "GL/internal/glcore.h"  /* for __GLcontextModes */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+/**
+ * Only color buffers are listed.  The others are allocated privately through,
+ * for example, st_renderbuffer_alloc_storage().
+ */
+enum native_attachment {
+   NATIVE_ATTACHMENT_FRONT_LEFT,
+   NATIVE_ATTACHMENT_BACK_LEFT,
+   NATIVE_ATTACHMENT_FRONT_RIGHT,
+   NATIVE_ATTACHMENT_BACK_RIGHT,
+
+   NUM_NATIVE_ATTACHMENTS
+};
+
+struct native_surface {
+   void (*destroy)(struct native_surface *nsurf);
+
+   /**
+    * Swap the front and back buffers so that the back buffer is visible.  It
+    * is no-op if the surface is single-buffered.  The contents of the back
+    * buffer after swapping may or may not be preserved.
+    */
+   boolean (*swap_buffers)(struct native_surface *nsurf);
+
+   /**
+    * Make the front buffer visible.  In some native displays, changes to the
+    * front buffer might not be visible immediately and require manual flush.
+    */
+   boolean (*flush_frontbuffer)(struct native_surface *nsurf);
+
+   /**
+    * Validate the buffers of the surface.  The returned textures are owned by
+    * the caller.  A sequence number is also returned.  The caller can use it
+    * to check if anything has changed since the last call. Any of the pointers
+    * may be NULL and it indicates the caller has no interest in those values.
+    *
+    * If this function is called multiple times with different attachments,
+    * those not listed in the latest call might be destroyed.  This behavior
+    * might change in the future.
+    */
+   boolean (*validate)(struct native_surface *nsurf,
+                       const enum native_attachment *natts,
+                       unsigned num_natts,
+                       unsigned int *seq_num, struct pipe_texture **textures,
+                       int *width, int *height);
+
+   /**
+    * Wait until all native commands affecting the surface has been executed.
+    */
+   void (*wait)(struct native_surface *nsurf);
+};
+
+struct native_config {
+   /* __GLcontextModes should go away some day */
+   __GLcontextModes mode;
+   enum pipe_format color_format;
+   enum pipe_format depth_format;
+   enum pipe_format stencil_format;
+
+   /* treat it as an additional flag to mode.drawableType */
+   boolean scanout_bit;
+};
+
+struct native_connector {
+   int dummy;
+};
+
+struct native_mode {
+   const char *desc;
+   int width, height;
+   int refresh_rate;
+};
+
+struct native_display_modeset;
+
+/**
+ * A pipe winsys abstracts the OS.  A pipe screen abstracts the graphcis
+ * hardware.  A native display consists of a pipe winsys, a pipe screen, and
+ * the native display server.
+ */
+struct native_display {
+   /**
+    * The pipe screen of the native display.
+    *
+    * Note that the "flush_frontbuffer" and "update_buffer" callbacks will be
+    * overridden.
+    */
+   struct pipe_screen *screen;
+
+   void (*destroy)(struct native_display *ndpy);
+
+   /**
+    * Get the supported configs.  The configs are owned by the display, but
+    * the returned array should be free()ed.
+    *
+    * The configs will be converted to EGL config by
+    * _eglConfigFromContextModesRec and validated by _eglValidateConfig.
+    * Those failing to pass the test will be skipped.
+    */
+   const struct native_config **(*get_configs)(struct native_display *ndpy,
+                                               int *num_configs);
+
+   /**
+    * Create a pipe context.
+    */
+   struct pipe_context *(*create_context)(struct native_display *ndpy,
+                                          void *context_private);
+
+   /**
+    * Create a window surface.  Required unless no config has GLX_WINDOW_BIT
+    * set.
+    */
+   struct native_surface *(*create_window_surface)(struct native_display *ndpy,
+                                                   EGLNativeWindowType win,
+                                                   const struct native_config *nconf);
+
+   /**
+    * Create a pixmap surface.  Required unless no config has GLX_PIXMAP_BIT
+    * set.
+    */
+   struct native_surface *(*create_pixmap_surface)(struct native_display *ndpy,
+                                                   EGLNativePixmapType pix,
+                                                   const struct native_config *nconf);
+
+   /**
+    * Create a pbuffer surface.  Required unless no config has GLX_PBUFFER_BIT
+    * set.
+    */
+   struct native_surface *(*create_pbuffer_surface)(struct native_display *ndpy,
+                                                    const struct native_config *nconf,
+                                                    uint width, uint height);
+
+   const struct native_display_modeset *modeset;
+};
+
+/**
+ * Mode setting interface of the native display.  It exposes the mode setting
+ * capabilities of the underlying graphics hardware.
+ */
+struct native_display_modeset {
+   /**
+    * Get the available physical connectors and the number of CRTCs.
+    */
+   const struct native_connector **(*get_connectors)(struct native_display *ndpy,
+                                                     int *num_connectors,
+                                                     int *num_crtcs);
+
+   /**
+    * Get the current supported modes of a connector.  The returned modes may
+    * change every time this function is called and those from previous calls
+    * might become invalid.
+    */
+   const struct native_mode **(*get_modes)(struct native_display *ndpy,
+                                           const struct native_connector *nconn,
+                                           int *num_modes);
+
+   /**
+    * Create a scan-out surface.  Required unless no config has
+    * GLX_SCREEN_BIT_MESA set.
+    */
+   struct native_surface *(*create_scanout_surface)(struct native_display *ndpy,
+                                                    const struct native_config *nconf,
+                                                    uint width, uint height);
+
+   /**
+    * Program the CRTC to output the surface to the given connectors with the
+    * given mode.  When surface is not given, the CRTC is disabled.
+    *
+    * This interface does not export a way to query capabilities of the CRTCs.
+    * The native display usually needs to dynamically map the index to a CRTC
+    * that supports the given connectors.
+    */
+   boolean (*program)(struct native_display *ndpy, int crtc_idx,
+                      struct native_surface *nsurf, uint x, uint y,
+                      const struct native_connector **nconns, int num_nconns,
+                      const struct native_mode *nmode);
+};
+
+const char *
+native_get_name(void);
+
+struct native_display *
+native_create_display(EGLNativeDisplayType dpy);
+
+#endif /* _NATIVE_H_ */
diff --git a/src/gallium/state_trackers/egl_g3d/common/st_public_tmp.h b/src/gallium/state_trackers/egl_g3d/common/st_public_tmp.h
new file mode 100644
index 00000000000..507a0ec4027
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/common/st_public_tmp.h
@@ -0,0 +1,20 @@
+ST_PUBLIC(st_create_context,              struct st_context *,       struct pipe_context *pipe, const __GLcontextModes *visual, struct st_context *share)
+ST_PUBLIC(st_destroy_context,             void,                      struct st_context *st)
+ST_PUBLIC(st_copy_context_state,          void,                      struct st_context *dst, struct st_context *src, uint mask)
+ST_PUBLIC(st_create_framebuffer,          struct st_framebuffer *,   const __GLcontextModes *visual, enum pipe_format colorFormat, enum pipe_format depthFormat, enum pipe_format stencilFormat, uint width, uint height, void *privateData)
+ST_PUBLIC(st_resize_framebuffer,          void,                      struct st_framebuffer *stfb, uint width, uint height)
+ST_PUBLIC(st_set_framebuffer_surface,     void,                      struct st_framebuffer *stfb, uint surfIndex, struct pipe_surface *surf)
+ST_PUBLIC(st_get_framebuffer_dimensions,  void,                      struct st_framebuffer *stfb, uint *width, uint *height)
+ST_PUBLIC(st_get_framebuffer_surface,     int,                       struct st_framebuffer *stfb, uint surfIndex, struct pipe_surface **surface)
+ST_PUBLIC(st_get_framebuffer_texture,     int,                       struct st_framebuffer *stfb, uint surfIndex, struct pipe_texture **texture)
+ST_PUBLIC(st_framebuffer_private,         void *,                    struct st_framebuffer *stfb)
+ST_PUBLIC(st_unreference_framebuffer,     void,                      struct st_framebuffer *stfb)
+ST_PUBLIC(st_make_current,                GLboolean,                 struct st_context *st, struct st_framebuffer *draw, struct st_framebuffer *read)
+ST_PUBLIC(st_get_current,                 struct st_context *,       void)
+ST_PUBLIC(st_flush,                       void,                      struct st_context *st, uint pipeFlushFlags, struct pipe_fence_handle **fence)
+ST_PUBLIC(st_finish,                      void,                      struct st_context *st)
+ST_PUBLIC(st_notify_swapbuffers,          void,                      struct st_framebuffer *stfb)
+ST_PUBLIC(st_bind_texture_surface,        int,                       struct pipe_surface *ps, int target, int level, enum pipe_format format)
+ST_PUBLIC(st_unbind_texture_surface,      int,                       struct pipe_surface *ps, int target, int level)
+ST_PUBLIC(st_get_proc_address,            st_proc,                   const char *procname)
+#undef ST_PUBLIC
diff --git a/src/gallium/state_trackers/egl_g3d/kms/native_kms.c b/src/gallium/state_trackers/egl_g3d/kms/native_kms.c
new file mode 100644
index 00000000000..65829fc7b3d
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/kms/native_kms.c
@@ -0,0 +1,855 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 <string.h>
+
+#include "pipe/p_screen.h"
+#include "pipe/p_context.h"
+#include "util/u_debug.h"
+#include "util/u_memory.h"
+#include "egllog.h"
+
+#include "native_kms.h"
+
+static boolean
+kms_surface_validate(struct native_surface *nsurf,
+                     const enum native_attachment *natts,
+                     unsigned num_natts,
+                     unsigned int *seq_num, struct pipe_texture **textures,
+                     int *width, int *height)
+{
+   struct kms_surface *ksurf = kms_surface(nsurf);
+   struct kms_display *kdpy = ksurf->kdpy;
+   struct pipe_screen *screen = kdpy->base.screen;
+   struct pipe_texture templ, *ptex;
+   int i;
+
+   if (num_natts) {
+      if (textures)
+         memset(textures, 0, sizeof(*textures) * num_natts);
+
+      memset(&templ, 0, sizeof(templ));
+      templ.target = PIPE_TEXTURE_2D;
+      templ.last_level = 0;
+      templ.width0 = ksurf->width;
+      templ.height0 = ksurf->height;
+      templ.depth0 = 1;
+      templ.format = ksurf->color_format;
+      templ.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
+      if (ksurf->type == KMS_SURFACE_TYPE_SCANOUT)
+         templ.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY;
+   }
+
+   /* create textures */
+   for (i = 0; i < num_natts; i++) {
+      enum native_attachment natt = natts[i];
+
+      ptex = ksurf->textures[natt];
+      if (!ptex) {
+         ptex = screen->texture_create(screen, &templ);
+         ksurf->textures[natt] = ptex;
+      }
+
+      if (textures)
+         pipe_texture_reference(&textures[i], ptex);
+   }
+
+   if (seq_num)
+      *seq_num = ksurf->sequence_number;
+   if (width)
+      *width = ksurf->width;
+   if (height)
+      *height = ksurf->height;
+
+   return TRUE;
+}
+
+/**
+ * Add textures as DRM framebuffers.
+ */
+static boolean
+kms_surface_init_framebuffers(struct native_surface *nsurf, boolean need_back)
+{
+   struct kms_surface *ksurf = kms_surface(nsurf);
+   struct kms_display *kdpy = ksurf->kdpy;
+   int num_framebuffers = (need_back) ? 2 : 1;
+   int i, err;
+
+   for (i = 0; i < num_framebuffers; i++) {
+      struct kms_framebuffer *fb;
+      enum native_attachment natt;
+      unsigned int handle, stride;
+      uint block_bits;
+
+      if (i == 0) {
+         fb = &ksurf->front_fb;
+         natt = NATIVE_ATTACHMENT_FRONT_LEFT;
+      }
+      else {
+         fb = &ksurf->back_fb;
+         natt = NATIVE_ATTACHMENT_BACK_LEFT;
+      }
+
+      if (!fb->texture) {
+         /* make sure the texture has been allocated */
+         kms_surface_validate(&ksurf->base, &natt, 1, NULL, NULL, NULL, NULL);
+         if (!ksurf->textures[natt])
+            return FALSE;
+
+         pipe_texture_reference(&fb->texture, ksurf->textures[natt]);
+      }
+
+      /* already initialized */
+      if (fb->buffer_id)
+         continue;
+
+      /* TODO detect the real value */
+      fb->is_passive = TRUE;
+
+      if (!kdpy->api->local_handle_from_texture(kdpy->api,
+               kdpy->base.screen, fb->texture, &stride, &handle))
+         return FALSE;
+
+      block_bits = util_format_get_blocksizebits(ksurf->color_format);
+      err = drmModeAddFB(kdpy->fd, ksurf->width, ksurf->height,
+            block_bits, block_bits, stride, handle, &fb->buffer_id);
+      if (err) {
+         fb->buffer_id = 0;
+         return FALSE;
+      }
+   }
+
+   return TRUE;
+}
+
+static boolean
+kms_surface_flush_frontbuffer(struct native_surface *nsurf)
+{
+#ifdef DRM_MODE_FEATURE_DIRTYFB
+   struct kms_surface *ksurf = kms_surface(nsurf);
+   struct kms_display *kdpy = ksurf->kdpy;
+
+   /* pbuffer is private */
+   if (ksurf->type == KMS_SURFACE_TYPE_PBUFFER)
+      return TRUE;
+
+   if (ksurf->front_fb.is_passive)
+      drmModeDirtyFB(kdpy->fd, ksurf->front_fb.buffer_id, NULL, 0);
+#endif
+
+   return TRUE;
+}
+
+static boolean
+kms_surface_swap_buffers(struct native_surface *nsurf)
+{
+   struct kms_surface *ksurf = kms_surface(nsurf);
+   struct kms_crtc *kcrtc = &ksurf->current_crtc;
+   struct kms_display *kdpy = ksurf->kdpy;
+   struct kms_framebuffer tmp_fb;
+   struct pipe_texture *tmp_texture;
+   int err;
+
+   /* pbuffer is private */
+   if (ksurf->type == KMS_SURFACE_TYPE_PBUFFER)
+      return TRUE;
+
+   if (!ksurf->back_fb.buffer_id) {
+      if (!kms_surface_init_framebuffers(&ksurf->base, TRUE))
+         return FALSE;
+   }
+
+   if (ksurf->is_shown && kcrtc->crtc) {
+      err = drmModeSetCrtc(kdpy->fd, kcrtc->crtc->crtc_id,
+            ksurf->back_fb.buffer_id, kcrtc->crtc->x, kcrtc->crtc->y,
+            kcrtc->connectors, kcrtc->num_connectors, &kcrtc->crtc->mode);
+      if (err)
+         return FALSE;
+   }
+
+   /* swap the buffers */
+   tmp_fb = ksurf->front_fb;
+   ksurf->front_fb = ksurf->back_fb;
+   ksurf->back_fb = tmp_fb;
+
+   tmp_texture = ksurf->textures[NATIVE_ATTACHMENT_FRONT_LEFT];
+   ksurf->textures[NATIVE_ATTACHMENT_FRONT_LEFT] =
+      ksurf->textures[NATIVE_ATTACHMENT_BACK_LEFT];
+   ksurf->textures[NATIVE_ATTACHMENT_BACK_LEFT] = tmp_texture;
+
+   /* the front/back textures are swapped */
+   ksurf->sequence_number++;
+
+   return TRUE;
+}
+
+static void
+kms_surface_wait(struct native_surface *nsurf)
+{
+   /* no-op */
+}
+
+static void
+kms_surface_destroy(struct native_surface *nsurf)
+{
+   struct kms_surface *ksurf = kms_surface(nsurf);
+   int i;
+
+   if (ksurf->current_crtc.crtc)
+         drmModeFreeCrtc(ksurf->current_crtc.crtc);
+
+   if (ksurf->front_fb.buffer_id)
+      drmModeRmFB(ksurf->kdpy->fd, ksurf->front_fb.buffer_id);
+   pipe_texture_reference(&ksurf->front_fb.texture, NULL);
+
+   if (ksurf->back_fb.buffer_id)
+      drmModeRmFB(ksurf->kdpy->fd, ksurf->back_fb.buffer_id);
+   pipe_texture_reference(&ksurf->back_fb.texture, NULL);
+
+   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
+      struct pipe_texture *ptex = ksurf->textures[i];
+      pipe_texture_reference(&ptex, NULL);
+   }
+
+   free(ksurf);
+}
+
+static struct kms_surface *
+kms_display_create_surface(struct native_display *ndpy,
+                           enum kms_surface_type type,
+                           const struct native_config *nconf,
+                           uint width, uint height)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   struct kms_config *kconf = kms_config(nconf);
+   struct kms_surface *ksurf;
+
+   ksurf = CALLOC_STRUCT(kms_surface);
+   if (!ksurf)
+      return NULL;
+
+   ksurf->kdpy = kdpy;
+   ksurf->type = type;
+   ksurf->color_format = kconf->base.color_format;
+   ksurf->width = width;
+   ksurf->height = height;
+
+   ksurf->base.destroy = kms_surface_destroy;
+   ksurf->base.swap_buffers = kms_surface_swap_buffers;
+   ksurf->base.flush_frontbuffer = kms_surface_flush_frontbuffer;
+   ksurf->base.validate = kms_surface_validate;
+   ksurf->base.wait = kms_surface_wait;
+
+   return ksurf;
+}
+
+/**
+ * Choose a CRTC that supports all given connectors.
+ */
+static uint32_t
+kms_display_choose_crtc(struct native_display *ndpy,
+                        uint32_t *connectors, int num_connectors)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   int idx;
+
+   for (idx = 0; idx < kdpy->resources->count_crtcs; idx++) {
+      boolean found_crtc = TRUE;
+      int i, j;
+
+      for (i = 0; i < num_connectors; i++) {
+         drmModeConnectorPtr connector;
+         int encoder_idx = -1;
+
+         connector = drmModeGetConnector(kdpy->fd, connectors[i]);
+         if (!connector) {
+            found_crtc = FALSE;
+            break;
+         }
+
+         /* find an encoder the CRTC supports */
+         for (j = 0; j < connector->count_encoders; j++) {
+            drmModeEncoderPtr encoder =
+               drmModeGetEncoder(kdpy->fd, connector->encoders[j]);
+            if (encoder->possible_crtcs & (1 << idx)) {
+               encoder_idx = j;
+               break;
+            }
+            drmModeFreeEncoder(encoder);
+         }
+
+         drmModeFreeConnector(connector);
+         if (encoder_idx < 0) {
+            found_crtc = FALSE;
+            break;
+         }
+      }
+
+      if (found_crtc)
+         break;
+   }
+
+   if (idx >= kdpy->resources->count_crtcs) {
+      _eglLog(_EGL_WARNING,
+            "failed to find a CRTC that supports the given %d connectors",
+            num_connectors);
+      return 0;
+   }
+
+   return kdpy->resources->crtcs[idx];
+}
+
+/**
+ * Remember the original CRTC status and set the CRTC
+ */
+static boolean
+kms_display_set_crtc(struct native_display *ndpy, int crtc_idx,
+                     uint32_t buffer_id, uint32_t x, uint32_t y,
+                     uint32_t *connectors, int num_connectors,
+                     drmModeModeInfoPtr mode)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   struct kms_crtc *kcrtc = &kdpy->saved_crtcs[crtc_idx];
+   uint32_t crtc_id;
+   int err;
+
+   if (kcrtc->crtc) {
+      crtc_id = kcrtc->crtc->crtc_id;
+   }
+   else {
+      int count = 0, i;
+
+      /*
+       * Choose the CRTC once.  It could be more dynamic, but let's keep it
+       * simple for now.
+       */
+      crtc_id = kms_display_choose_crtc(&kdpy->base,
+            connectors, num_connectors);
+
+      /* save the original CRTC status */
+      kcrtc->crtc = drmModeGetCrtc(kdpy->fd, crtc_id);
+      if (!kcrtc->crtc)
+         return FALSE;
+
+      for (i = 0; i < kdpy->num_connectors; i++) {
+         struct kms_connector *kconn = &kdpy->connectors[i];
+         drmModeConnectorPtr connector = kconn->connector;
+         drmModeEncoderPtr encoder;
+
+         encoder = drmModeGetEncoder(kdpy->fd, connector->encoder_id);
+         if (encoder) {
+            if (encoder->crtc_id == crtc_id) {
+               kcrtc->connectors[count++] = connector->connector_id;
+               if (count >= Elements(kcrtc->connectors))
+                  break;
+            }
+            drmModeFreeEncoder(encoder);
+         }
+      }
+
+      kcrtc->num_connectors = count;
+   }
+
+   err = drmModeSetCrtc(kdpy->fd, crtc_id, buffer_id, x, y,
+         connectors, num_connectors, mode);
+   if (err) {
+      drmModeFreeCrtc(kcrtc->crtc);
+      kcrtc->crtc = NULL;
+      kcrtc->num_connectors = 0;
+
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+static boolean
+kms_display_program(struct native_display *ndpy, int crtc_idx,
+                    struct native_surface *nsurf, uint x, uint y,
+                    const struct native_connector **nconns, int num_nconns,
+                    const struct native_mode *nmode)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   struct kms_surface *ksurf = kms_surface(nsurf);
+   const struct kms_mode *kmode = kms_mode(nmode);
+   uint32_t connector_ids[32];
+   uint32_t buffer_id;
+   drmModeModeInfo mode_tmp, *mode;
+   int i;
+
+   if (num_nconns > Elements(connector_ids)) {
+      _eglLog(_EGL_WARNING, "too many connectors (%d)", num_nconns);
+      num_nconns = Elements(connector_ids);
+   }
+
+   if (ksurf) {
+      if (!kms_surface_init_framebuffers(&ksurf->base, FALSE))
+         return FALSE;
+
+      buffer_id = ksurf->front_fb.buffer_id;
+      /* the mode argument of drmModeSetCrtc is not constified */
+      mode_tmp = kmode->mode;
+      mode = &mode_tmp;
+   }
+   else {
+      /* disable the CRTC */
+      buffer_id = 0;
+      mode = NULL;
+      num_nconns = 0;
+   }
+
+   for (i = 0; i < num_nconns; i++) {
+      struct kms_connector *kconn = kms_connector(nconns[i]);
+      connector_ids[i] = kconn->connector->connector_id;
+   }
+
+   if (!kms_display_set_crtc(&kdpy->base, crtc_idx, buffer_id, x, y,
+            connector_ids, num_nconns, mode)) {
+      _eglLog(_EGL_WARNING, "failed to set CRTC %d", crtc_idx);
+
+      return FALSE;
+   }
+
+   if (kdpy->shown_surfaces[crtc_idx])
+      kdpy->shown_surfaces[crtc_idx]->is_shown = FALSE;
+   kdpy->shown_surfaces[crtc_idx] = ksurf;
+
+   /* remember the settings for buffer swapping */
+   if (ksurf) {
+      uint32_t crtc_id = kdpy->saved_crtcs[crtc_idx].crtc->crtc_id;
+      struct kms_crtc *kcrtc = &ksurf->current_crtc;
+
+      if (kcrtc->crtc)
+         drmModeFreeCrtc(kcrtc->crtc);
+      kcrtc->crtc = drmModeGetCrtc(kdpy->fd, crtc_id);
+
+      assert(num_nconns < Elements(kcrtc->connectors));
+      memcpy(kcrtc->connectors, connector_ids,
+            sizeof(*connector_ids) * num_nconns);
+      kcrtc->num_connectors = num_nconns;
+
+      ksurf->is_shown = TRUE;
+   }
+
+   return TRUE;
+}
+
+static const struct native_mode **
+kms_display_get_modes(struct native_display *ndpy,
+                      const struct native_connector *nconn,
+                      int *num_modes)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   struct kms_connector *kconn = kms_connector(nconn);
+   const struct native_mode **nmodes_return;
+   int count, i;
+
+   /* delete old data */
+   if (kconn->connector) {
+      drmModeFreeConnector(kconn->connector);
+      free(kconn->kms_modes);
+
+      kconn->connector = NULL;
+      kconn->kms_modes = NULL;
+      kconn->num_modes = 0;
+   }
+
+   /* detect again */
+   kconn->connector = drmModeGetConnector(kdpy->fd, kconn->connector_id);
+   if (!kconn->connector)
+      return NULL;
+
+   count = kconn->connector->count_modes;
+   kconn->kms_modes = calloc(count, sizeof(*kconn->kms_modes));
+   if (!kconn->kms_modes) {
+      drmModeFreeConnector(kconn->connector);
+      kconn->connector = NULL;
+
+      return NULL;
+   }
+
+   for (i = 0; i < count; i++) {
+      struct kms_mode *kmode = &kconn->kms_modes[i];
+      drmModeModeInfoPtr mode = &kconn->connector->modes[i];
+
+      kmode->mode = *mode;
+
+      kmode->base.desc = kmode->mode.name;
+      kmode->base.width = kmode->mode.hdisplay;
+      kmode->base.height = kmode->mode.vdisplay;
+      kmode->base.refresh_rate = kmode->mode.vrefresh / 1000;
+   }
+
+   nmodes_return = malloc(count * sizeof(*nmodes_return));
+   if (nmodes_return) {
+      for (i = 0; i < count; i++)
+         nmodes_return[i] = &kconn->kms_modes[i].base;
+      if (num_modes)
+         *num_modes = count;
+   }
+
+   return nmodes_return;
+}
+
+static const struct native_connector **
+kms_display_get_connectors(struct native_display *ndpy, int *num_connectors,
+                           int *num_crtc)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   const struct native_connector **connectors;
+   int i;
+
+   if (!kdpy->connectors) {
+      kdpy->connectors =
+         calloc(kdpy->resources->count_connectors, sizeof(*kdpy->connectors));
+      if (!kdpy->connectors)
+         return NULL;
+
+      for (i = 0; i < kdpy->resources->count_connectors; i++) {
+         struct kms_connector *kconn = &kdpy->connectors[i];
+
+         kconn->connector_id = kdpy->resources->connectors[i];
+         /* kconn->connector is allocated when the modes are asked */
+      }
+
+      kdpy->num_connectors = kdpy->resources->count_connectors;
+   }
+
+   connectors = malloc(kdpy->num_connectors * sizeof(*connectors));
+   if (connectors) {
+      for (i = 0; i < kdpy->num_connectors; i++)
+         connectors[i] = &kdpy->connectors[i].base;
+      if (num_connectors)
+         *num_connectors = kdpy->num_connectors;
+   }
+
+   if (num_crtc)
+      *num_crtc = kdpy->resources->count_crtcs;
+
+   return connectors;
+}
+
+static struct native_surface *
+kms_display_create_scanout_surface(struct native_display *ndpy,
+                                   const struct native_config *nconf,
+                                   uint width, uint height)
+{
+   struct kms_surface *ksurf;
+
+   ksurf = kms_display_create_surface(ndpy,
+         KMS_SURFACE_TYPE_SCANOUT, nconf, width, height);
+   return &ksurf->base;
+}
+
+static struct native_surface *
+kms_display_create_pbuffer_surface(struct native_display *ndpy,
+                                   const struct native_config *nconf,
+                                   uint width, uint height)
+{
+   struct kms_surface *ksurf;
+
+   ksurf = kms_display_create_surface(ndpy,
+         KMS_SURFACE_TYPE_PBUFFER, nconf, width, height);
+   return &ksurf->base;
+}
+
+static struct pipe_context *
+kms_display_create_context(struct native_display *ndpy, void *context_private)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   struct pipe_context *pctx;
+
+   pctx = kdpy->api->create_context(kdpy->api, kdpy->base.screen);
+   if (pctx)
+      pctx->priv = context_private;
+   return pctx;
+}
+
+static boolean
+kms_display_is_format_supported(struct native_display *ndpy,
+                                enum pipe_format fmt, boolean is_color)
+{
+   return ndpy->screen->is_format_supported(ndpy->screen,
+         fmt, PIPE_TEXTURE_2D,
+         (is_color) ? PIPE_TEXTURE_USAGE_RENDER_TARGET :
+         PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0);
+}
+
+static const struct native_config **
+kms_display_get_configs(struct native_display *ndpy, int *num_configs)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   const struct native_config **configs;
+
+   /* first time */
+   if (!kdpy->config) {
+      struct native_config *nconf;
+      enum pipe_format format;
+
+      kdpy->config = calloc(1, sizeof(*kdpy->config));
+      if (!kdpy->config)
+         return NULL;
+
+      nconf = &kdpy->config->base;
+
+      /* always double-buffered */
+      nconf->mode.doubleBufferMode = TRUE;
+
+      format = PIPE_FORMAT_A8R8G8B8_UNORM;
+      if (!kms_display_is_format_supported(&kdpy->base, format, TRUE)) {
+         format = PIPE_FORMAT_B8G8R8A8_UNORM;
+         if (!kms_display_is_format_supported(&kdpy->base, format, TRUE))
+            format = PIPE_FORMAT_NONE;
+      }
+      if (format == PIPE_FORMAT_NONE)
+         return NULL;
+
+      nconf->color_format = format;
+      nconf->mode.redBits = 8;
+      nconf->mode.greenBits = 8;
+      nconf->mode.blueBits = 8;
+      nconf->mode.alphaBits = 8;
+      nconf->mode.rgbBits = 32;
+
+      format = PIPE_FORMAT_S8Z24_UNORM;
+      if (!kms_display_is_format_supported(&kdpy->base, format, FALSE)) {
+         format = PIPE_FORMAT_Z24S8_UNORM;
+         if (!kms_display_is_format_supported(&kdpy->base, format, FALSE))
+            format = PIPE_FORMAT_NONE;
+      }
+      if (format != PIPE_FORMAT_NONE) {
+         nconf->depth_format = format;
+         nconf->stencil_format = format;
+
+         nconf->mode.depthBits = 24;
+         nconf->mode.stencilBits = 8;
+         nconf->mode.haveDepthBuffer = TRUE;
+         nconf->mode.haveStencilBuffer = TRUE;
+      }
+
+      nconf->scanout_bit = TRUE;
+      nconf->mode.drawableType = GLX_PBUFFER_BIT;
+      nconf->mode.swapMethod = GLX_SWAP_EXCHANGE_OML;
+
+      nconf->mode.visualID = 0;
+      nconf->mode.visualType = EGL_NONE;
+
+      nconf->mode.renderType = GLX_RGBA_BIT;
+      nconf->mode.rgbMode = TRUE;
+      nconf->mode.xRenderable = FALSE;
+   }
+
+   configs = malloc(sizeof(*configs));
+   if (configs) {
+      configs[0] = &kdpy->config->base;
+      if (num_configs)
+         *num_configs = 1;
+   }
+
+   return configs;
+}
+
+static void
+kms_display_destroy(struct native_display *ndpy)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   int i;
+
+   if (kdpy->config)
+      free(kdpy->config);
+
+   if (kdpy->connectors) {
+      for (i = 0; i < kdpy->num_connectors; i++) {
+         struct kms_connector *kconn = &kdpy->connectors[i];
+         if (kconn->connector) {
+            drmModeFreeConnector(kconn->connector);
+            free(kconn->kms_modes);
+         }
+      }
+      free(kdpy->connectors);
+   }
+
+   if (kdpy->shown_surfaces)
+      free(kdpy->shown_surfaces);
+
+   if (kdpy->saved_crtcs) {
+      for (i = 0; i < kdpy->resources->count_crtcs; i++) {
+         struct kms_crtc *kcrtc = &kdpy->saved_crtcs[i];
+
+         if (kcrtc->crtc) {
+            /* restore crtc */
+            drmModeSetCrtc(kdpy->fd, kcrtc->crtc->crtc_id,
+                  kcrtc->crtc->buffer_id, kcrtc->crtc->x, kcrtc->crtc->y,
+                  kcrtc->connectors, kcrtc->num_connectors,
+                  &kcrtc->crtc->mode);
+
+            drmModeFreeCrtc(kcrtc->crtc);
+         }
+      }
+      free(kdpy->saved_crtcs);
+   }
+
+   if (kdpy->resources)
+      drmModeFreeResources(kdpy->resources);
+
+   if (kdpy->base.screen)
+      kdpy->base.screen->destroy(kdpy->base.screen);
+
+   if (kdpy->fd >= 0)
+      drmClose(kdpy->fd);
+
+   if (kdpy->api)
+      kdpy->api->destroy(kdpy->api);
+   free(kdpy);
+}
+
+/**
+ * Initialize KMS and pipe screen.
+ */
+static boolean
+kms_display_init_screen(struct native_display *ndpy)
+{
+   struct kms_display *kdpy = kms_display(ndpy);
+   struct drm_create_screen_arg arg;
+   int fd;
+
+   fd = drmOpen(kdpy->api->name, NULL);
+   if (fd < 0) {
+      _eglLog(_EGL_WARNING, "failed to open DRM device");
+      return FALSE;
+   }
+
+#if 0
+   if (drmSetMaster(fd)) {
+      _eglLog(_EGL_WARNING, "failed to become DRM master");
+      return FALSE;
+   }
+#endif
+
+   memset(&arg, 0, sizeof(arg));
+   arg.mode = DRM_CREATE_NORMAL;
+   kdpy->base.screen = kdpy->api->create_screen(kdpy->api, fd, &arg);
+   if (!kdpy->base.screen) {
+      _eglLog(_EGL_WARNING, "failed to create DRM screen");
+      drmClose(fd);
+      return FALSE;
+   }
+
+   kdpy->fd = fd;
+
+   return TRUE;
+}
+
+static struct native_display_modeset kms_display_modeset = {
+   .get_connectors = kms_display_get_connectors,
+   .get_modes = kms_display_get_modes,
+   .create_scanout_surface = kms_display_create_scanout_surface,
+   .program = kms_display_program
+};
+
+static struct native_display *
+kms_create_display(EGLNativeDisplayType dpy, struct drm_api *api)
+{
+   struct kms_display *kdpy;
+
+   kdpy = CALLOC_STRUCT(kms_display);
+   if (!kdpy)
+      return NULL;
+
+   kdpy->api = api;
+   if (!kdpy->api) {
+      _eglLog(_EGL_WARNING, "failed to create DRM API");
+      free(kdpy);
+      return NULL;
+   }
+
+   kdpy->fd = -1;
+   if (!kms_display_init_screen(&kdpy->base)) {
+      kms_display_destroy(&kdpy->base);
+      return NULL;
+   }
+
+   /* resources are fixed, unlike crtc, connector, or encoder */
+   kdpy->resources = drmModeGetResources(kdpy->fd);
+   if (!kdpy->resources) {
+      kms_display_destroy(&kdpy->base);
+      return NULL;
+   }
+
+   kdpy->saved_crtcs =
+      calloc(kdpy->resources->count_crtcs, sizeof(*kdpy->saved_crtcs));
+   if (!kdpy->saved_crtcs) {
+      kms_display_destroy(&kdpy->base);
+      return NULL;
+   }
+
+   kdpy->shown_surfaces =
+      calloc(kdpy->resources->count_crtcs, sizeof(*kdpy->shown_surfaces));
+   if (!kdpy->shown_surfaces) {
+      kms_display_destroy(&kdpy->base);
+      return NULL;
+   }
+
+   kdpy->base.destroy = kms_display_destroy;
+   kdpy->base.get_configs = kms_display_get_configs;
+   kdpy->base.create_context = kms_display_create_context;
+   kdpy->base.create_pbuffer_surface = kms_display_create_pbuffer_surface;
+
+   kdpy->base.modeset = &kms_display_modeset;
+
+   return &kdpy->base;
+}
+
+/* the api is destroyed with the native display */
+static struct drm_api *drm_api;
+
+const char *
+native_get_name(void)
+{
+   static char kms_name[32];
+
+   if (!drm_api)
+      drm_api = drm_api_create();
+
+   if (drm_api)
+      snprintf(kms_name, sizeof(kms_name), "KMS/%s", drm_api->name);
+   else
+      snprintf(kms_name, sizeof(kms_name), "KMS");
+
+   return kms_name;
+}
+
+struct native_display *
+native_create_display(EGLNativeDisplayType dpy)
+{
+   struct native_display *ndpy = NULL;
+
+   if (!drm_api)
+      drm_api = drm_api_create();
+
+   if (drm_api)
+      ndpy = kms_create_display(dpy, drm_api);
+
+   return ndpy;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/kms/native_kms.h b/src/gallium/state_trackers/egl_g3d/kms/native_kms.h
new file mode 100644
index 00000000000..095186e3cf3
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/kms/native_kms.h
@@ -0,0 +1,139 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 _NATIVE_KMS_H_
+#define _NATIVE_KMS_H_
+
+#include <xf86drm.h>
+#include <xf86drmMode.h>
+
+#include "pipe/p_compiler.h"
+#include "util/u_format.h"
+#include "pipe/p_state.h"
+#include "state_tracker/drm_api.h"
+
+#include "common/native.h"
+
+enum kms_surface_type {
+   KMS_SURFACE_TYPE_PBUFFER,
+   KMS_SURFACE_TYPE_SCANOUT
+};
+
+struct kms_config;
+struct kms_connector;
+struct kms_mode;
+
+struct kms_crtc {
+   drmModeCrtcPtr crtc;
+   uint32_t connectors[32];
+   int num_connectors;
+};
+
+struct kms_display {
+   struct native_display base;
+
+   int fd;
+   struct drm_api *api;
+   drmModeResPtr resources;
+   struct kms_config *config;
+
+   struct kms_connector *connectors;
+   int num_connectors;
+
+   struct kms_surface **shown_surfaces;
+   /* save the original settings of the CRTCs */
+   struct kms_crtc *saved_crtcs;
+};
+
+struct kms_framebuffer {
+   struct pipe_texture *texture;
+   boolean is_passive;
+
+   uint32_t buffer_id;
+};
+
+struct kms_surface {
+   struct native_surface base;
+   enum kms_surface_type type;
+   enum pipe_format color_format;
+   struct kms_display *kdpy;
+   int width, height;
+
+   struct pipe_texture *textures[NUM_NATIVE_ATTACHMENTS];
+   unsigned int sequence_number;
+   struct kms_framebuffer front_fb, back_fb;
+
+   boolean is_shown;
+   struct kms_crtc current_crtc;
+};
+
+struct kms_config {
+   struct native_config base;
+};
+
+struct kms_connector {
+   struct native_connector base;
+
+   uint32_t connector_id;
+   drmModeConnectorPtr connector;
+   struct kms_mode *kms_modes;
+   int num_modes;
+};
+
+struct kms_mode {
+   struct native_mode base;
+   drmModeModeInfo mode;
+};
+
+static INLINE struct kms_display *
+kms_display(const struct native_display *ndpy)
+{
+   return (struct kms_display *) ndpy;
+}
+
+static INLINE struct kms_surface *
+kms_surface(const struct native_surface *nsurf)
+{
+   return (struct kms_surface *) nsurf;
+}
+
+static INLINE struct kms_config *
+kms_config(const struct native_config *nconf)
+{
+   return (struct kms_config *) nconf;
+}
+
+static INLINE struct kms_connector *
+kms_connector(const struct native_connector *nconn)
+{
+   return (struct kms_connector *) nconn;
+}
+
+static INLINE struct kms_mode *
+kms_mode(const struct native_mode *nmode)
+{
+   return (struct kms_mode *) nmode;
+}
+
+#endif /* _NATIVE_KMS_H_ */
diff --git a/src/gallium/state_trackers/egl_g3d/x11/glxinit.c b/src/gallium/state_trackers/egl_g3d/x11/glxinit.c
new file mode 100644
index 00000000000..c955a908b93
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/glxinit.c
@@ -0,0 +1,573 @@
+/**
+ * GLX initialization.  Code based on glxext.c, glx_query.c, and
+ * glcontextmodes.c under src/glx/x11/.  The major difference is that no DRI
+ * related code here.
+ *
+ */
+
+#include <assert.h>
+#include <X11/Xlib.h>
+#include <X11/Xproto.h>
+#include <X11/extensions/Xext.h>
+#include <X11/extensions/extutil.h>
+#include <sys/time.h>
+
+#include "x11_screen.h"
+#include "glxinit.h"
+
+typedef struct GLXGenericGetString
+{
+   CARD8 reqType;
+   CARD8 glxCode;
+   CARD16 length B16;
+   CARD32 for_whom B32;
+   CARD32 name B32;
+} xGLXGenericGetStringReq;
+
+#define sz_xGLXGenericGetStringReq 12
+#define X_GLXGenericGetString 0
+
+/* Extension required boiler plate */
+
+static char *__glXExtensionName = GLX_EXTENSION_NAME;
+static XExtensionInfo *__glXExtensionInfo = NULL;
+
+static /* const */ XExtensionHooks __glXExtensionHooks = { NULL };
+static
+XEXT_GENERATE_FIND_DISPLAY(__glXFindDisplay, __glXExtensionInfo,
+                           __glXExtensionName, &__glXExtensionHooks,
+                           __GLX_NUMBER_EVENTS, NULL)
+
+static GLint
+_gl_convert_from_x_visual_type(int visualType)
+{
+#define NUM_VISUAL_TYPES   6
+   static const int glx_visual_types[NUM_VISUAL_TYPES] = {
+      GLX_STATIC_GRAY, GLX_GRAY_SCALE,
+      GLX_STATIC_COLOR, GLX_PSEUDO_COLOR,
+      GLX_TRUE_COLOR, GLX_DIRECT_COLOR
+   };
+
+   return ((unsigned) visualType < NUM_VISUAL_TYPES)
+      ? glx_visual_types[visualType] : GLX_NONE;
+}
+
+_X_HIDDEN char *
+__glXQueryServerString(Display * dpy, int opcode, CARD32 screen, CARD32 name)
+{
+   xGLXGenericGetStringReq *req;
+   xGLXSingleReply reply;
+   int length;
+   int numbytes;
+   char *buf;
+   CARD32 for_whom = screen;
+   CARD32 glxCode = X_GLXQueryServerString;
+
+
+   LockDisplay(dpy);
+
+
+   /* All of the GLX protocol requests for getting a string from the server
+    * look the same.  The exact meaning of the for_whom field is usually
+    * either the screen number (for glXQueryServerString) or the context tag
+    * (for GLXSingle).
+    */
+
+   GetReq(GLXGenericGetString, req);
+   req->reqType = opcode;
+   req->glxCode = glxCode;
+   req->for_whom = for_whom;
+   req->name = name;
+
+   _XReply(dpy, (xReply *) & reply, 0, False);
+
+   length = reply.length * 4;
+   numbytes = reply.size;
+
+   buf = (char *) Xmalloc(numbytes);
+   if (buf != NULL) {
+      _XRead(dpy, buf, numbytes);
+      length -= numbytes;
+   }
+
+   _XEatData(dpy, length);
+
+   UnlockDisplay(dpy);
+   SyncHandle();
+
+   return buf;
+}
+
+/************************************************************************/
+/*
+** Free the per screen configs data as well as the array of
+** __glXScreenConfigs.
+*/
+static void
+FreeScreenConfigs(__GLXdisplayPrivate * priv)
+{
+   __GLXscreenConfigs *psc;
+   GLint i, screens;
+
+   /* Free screen configuration information */
+   psc = priv->screenConfigs;
+   screens = ScreenCount(priv->dpy);
+   for (i = 0; i < screens; i++, psc++) {
+      if (psc->configs) {
+         x11_context_modes_destroy(psc->configs);
+         psc->configs = NULL;   /* NOTE: just for paranoia */
+      }
+      if (psc->visuals) {
+         x11_context_modes_destroy(psc->visuals);
+         psc->visuals = NULL;   /* NOTE: just for paranoia */
+      }
+      Xfree((char *) psc->serverGLXexts);
+   }
+   XFree((char *) priv->screenConfigs);
+   priv->screenConfigs = NULL;
+}
+
+/************************************************************************/
+
+/*
+** Query the version of the GLX extension.  This procedure works even if
+** the client extension is not completely set up.
+*/
+static Bool
+QueryVersion(Display * dpy, int opcode, int *major, int *minor)
+{
+   xGLXQueryVersionReq *req;
+   xGLXQueryVersionReply reply;
+
+   /* Send the glXQueryVersion request */
+   LockDisplay(dpy);
+   GetReq(GLXQueryVersion, req);
+   req->reqType = opcode;
+   req->glxCode = X_GLXQueryVersion;
+   req->majorVersion = GLX_MAJOR_VERSION;
+   req->minorVersion = GLX_MINOR_VERSION;
+   _XReply(dpy, (xReply *) & reply, 0, False);
+   UnlockDisplay(dpy);
+   SyncHandle();
+
+   if (reply.majorVersion != GLX_MAJOR_VERSION) {
+      /*
+       ** The server does not support the same major release as this
+       ** client.
+       */
+      return GL_FALSE;
+   }
+   *major = reply.majorVersion;
+   *minor = min(reply.minorVersion, GLX_MINOR_VERSION);
+   return GL_TRUE;
+}
+
+_X_HIDDEN void
+__glXInitializeVisualConfigFromTags(__GLcontextModes * config, int count,
+                                    const INT32 * bp, Bool tagged_only,
+                                    Bool fbconfig_style_tags)
+{
+   int i;
+
+   if (!tagged_only) {
+      /* Copy in the first set of properties */
+      config->visualID = *bp++;
+
+      config->visualType = _gl_convert_from_x_visual_type(*bp++);
+
+      config->rgbMode = *bp++;
+
+      config->redBits = *bp++;
+      config->greenBits = *bp++;
+      config->blueBits = *bp++;
+      config->alphaBits = *bp++;
+      config->accumRedBits = *bp++;
+      config->accumGreenBits = *bp++;
+      config->accumBlueBits = *bp++;
+      config->accumAlphaBits = *bp++;
+
+      config->doubleBufferMode = *bp++;
+      config->stereoMode = *bp++;
+
+      config->rgbBits = *bp++;
+      config->depthBits = *bp++;
+      config->stencilBits = *bp++;
+      config->numAuxBuffers = *bp++;
+      config->level = *bp++;
+
+      count -= __GLX_MIN_CONFIG_PROPS;
+   }
+
+   /*
+    ** Additional properties may be in a list at the end
+    ** of the reply.  They are in pairs of property type
+    ** and property value.
+    */
+
+#define FETCH_OR_SET(tag) \
+    config-> tag = ( fbconfig_style_tags ) ? *bp++ : 1
+
+   for (i = 0; i < count; i += 2) {
+      switch (*bp++) {
+      case GLX_RGBA:
+         FETCH_OR_SET(rgbMode);
+         break;
+      case GLX_BUFFER_SIZE:
+         config->rgbBits = *bp++;
+         break;
+      case GLX_LEVEL:
+         config->level = *bp++;
+         break;
+      case GLX_DOUBLEBUFFER:
+         FETCH_OR_SET(doubleBufferMode);
+         break;
+      case GLX_STEREO:
+         FETCH_OR_SET(stereoMode);
+         break;
+      case GLX_AUX_BUFFERS:
+         config->numAuxBuffers = *bp++;
+         break;
+      case GLX_RED_SIZE:
+         config->redBits = *bp++;
+         break;
+      case GLX_GREEN_SIZE:
+         config->greenBits = *bp++;
+         break;
+      case GLX_BLUE_SIZE:
+         config->blueBits = *bp++;
+         break;
+      case GLX_ALPHA_SIZE:
+         config->alphaBits = *bp++;
+         break;
+      case GLX_DEPTH_SIZE:
+         config->depthBits = *bp++;
+         break;
+      case GLX_STENCIL_SIZE:
+         config->stencilBits = *bp++;
+         break;
+      case GLX_ACCUM_RED_SIZE:
+         config->accumRedBits = *bp++;
+         break;
+      case GLX_ACCUM_GREEN_SIZE:
+         config->accumGreenBits = *bp++;
+         break;
+      case GLX_ACCUM_BLUE_SIZE:
+         config->accumBlueBits = *bp++;
+         break;
+      case GLX_ACCUM_ALPHA_SIZE:
+         config->accumAlphaBits = *bp++;
+         break;
+      case GLX_VISUAL_CAVEAT_EXT:
+         config->visualRating = *bp++;
+         break;
+      case GLX_X_VISUAL_TYPE:
+         config->visualType = *bp++;
+         break;
+      case GLX_TRANSPARENT_TYPE:
+         config->transparentPixel = *bp++;
+         break;
+      case GLX_TRANSPARENT_INDEX_VALUE:
+         config->transparentIndex = *bp++;
+         break;
+      case GLX_TRANSPARENT_RED_VALUE:
+         config->transparentRed = *bp++;
+         break;
+      case GLX_TRANSPARENT_GREEN_VALUE:
+         config->transparentGreen = *bp++;
+         break;
+      case GLX_TRANSPARENT_BLUE_VALUE:
+         config->transparentBlue = *bp++;
+         break;
+      case GLX_TRANSPARENT_ALPHA_VALUE:
+         config->transparentAlpha = *bp++;
+         break;
+      case GLX_VISUAL_ID:
+         config->visualID = *bp++;
+         break;
+      case GLX_DRAWABLE_TYPE:
+         config->drawableType = *bp++;
+         break;
+      case GLX_RENDER_TYPE:
+         config->renderType = *bp++;
+         break;
+      case GLX_X_RENDERABLE:
+         config->xRenderable = *bp++;
+         break;
+      case GLX_FBCONFIG_ID:
+         config->fbconfigID = *bp++;
+         break;
+      case GLX_MAX_PBUFFER_WIDTH:
+         config->maxPbufferWidth = *bp++;
+         break;
+      case GLX_MAX_PBUFFER_HEIGHT:
+         config->maxPbufferHeight = *bp++;
+         break;
+      case GLX_MAX_PBUFFER_PIXELS:
+         config->maxPbufferPixels = *bp++;
+         break;
+      case GLX_OPTIMAL_PBUFFER_WIDTH_SGIX:
+         config->optimalPbufferWidth = *bp++;
+         break;
+      case GLX_OPTIMAL_PBUFFER_HEIGHT_SGIX:
+         config->optimalPbufferHeight = *bp++;
+         break;
+      case GLX_VISUAL_SELECT_GROUP_SGIX:
+         config->visualSelectGroup = *bp++;
+         break;
+      case GLX_SWAP_METHOD_OML:
+         config->swapMethod = *bp++;
+         break;
+      case GLX_SAMPLE_BUFFERS_SGIS:
+         config->sampleBuffers = *bp++;
+         break;
+      case GLX_SAMPLES_SGIS:
+         config->samples = *bp++;
+         break;
+      case GLX_BIND_TO_TEXTURE_RGB_EXT:
+         config->bindToTextureRgb = *bp++;
+         break;
+      case GLX_BIND_TO_TEXTURE_RGBA_EXT:
+         config->bindToTextureRgba = *bp++;
+         break;
+      case GLX_BIND_TO_MIPMAP_TEXTURE_EXT:
+         config->bindToMipmapTexture = *bp++;
+         break;
+      case GLX_BIND_TO_TEXTURE_TARGETS_EXT:
+         config->bindToTextureTargets = *bp++;
+         break;
+      case GLX_Y_INVERTED_EXT:
+         config->yInverted = *bp++;
+         break;
+      case None:
+         i = count;
+         break;
+      default:
+         break;
+      }
+   }
+
+   config->renderType =
+      (config->rgbMode) ? GLX_RGBA_BIT : GLX_COLOR_INDEX_BIT;
+
+   config->haveAccumBuffer = ((config->accumRedBits +
+                               config->accumGreenBits +
+                               config->accumBlueBits +
+                               config->accumAlphaBits) > 0);
+   config->haveDepthBuffer = (config->depthBits > 0);
+   config->haveStencilBuffer = (config->stencilBits > 0);
+}
+
+static __GLcontextModes *
+createConfigsFromProperties(Display * dpy, int nvisuals, int nprops,
+                            int screen, GLboolean tagged_only)
+{
+   INT32 buf[__GLX_TOTAL_CONFIG], *props;
+   unsigned prop_size;
+   __GLcontextModes *modes, *m;
+   int i;
+
+   if (nprops == 0)
+      return NULL;
+
+   /* FIXME: Is the __GLX_MIN_CONFIG_PROPS test correct for FBconfigs? */
+
+   /* Check number of properties */
+   if (nprops < __GLX_MIN_CONFIG_PROPS || nprops > __GLX_MAX_CONFIG_PROPS)
+      return NULL;
+
+   /* Allocate memory for our config structure */
+   modes = x11_context_modes_create(nvisuals);
+   if (!modes)
+      return NULL;
+
+   prop_size = nprops * __GLX_SIZE_INT32;
+   if (prop_size <= sizeof(buf))
+      props = buf;
+   else
+      props = Xmalloc(prop_size);
+
+   /* Read each config structure and convert it into our format */
+   m = modes;
+   for (i = 0; i < nvisuals; i++) {
+      _XRead(dpy, (char *) props, prop_size);
+      /* Older X servers don't send this so we default it here. */
+      m->drawableType = GLX_WINDOW_BIT;
+      __glXInitializeVisualConfigFromTags(m, nprops, props,
+                                     tagged_only, GL_TRUE);
+      m->screen = screen;
+      m = m->next;
+   }
+
+   if (props != buf)
+      Xfree(props);
+
+   return modes;
+}
+
+static GLboolean
+getVisualConfigs(Display * dpy, __GLXdisplayPrivate * priv, int screen)
+{
+   xGLXGetVisualConfigsReq *req;
+   __GLXscreenConfigs *psc;
+   xGLXGetVisualConfigsReply reply;
+
+   LockDisplay(dpy);
+
+   psc = priv->screenConfigs + screen;
+   psc->visuals = NULL;
+   GetReq(GLXGetVisualConfigs, req);
+   req->reqType = priv->majorOpcode;
+   req->glxCode = X_GLXGetVisualConfigs;
+   req->screen = screen;
+
+   if (!_XReply(dpy, (xReply *) & reply, 0, False))
+      goto out;
+
+   psc->visuals = createConfigsFromProperties(dpy,
+                                              reply.numVisuals,
+                                              reply.numProps,
+                                              screen, GL_FALSE);
+
+ out:
+   UnlockDisplay(dpy);
+   return psc->visuals != NULL;
+}
+
+static GLboolean
+getFBConfigs(Display * dpy, __GLXdisplayPrivate * priv, int screen)
+{
+   xGLXGetFBConfigsReq *fb_req;
+   xGLXGetFBConfigsSGIXReq *sgi_req;
+   xGLXVendorPrivateWithReplyReq *vpreq;
+   xGLXGetFBConfigsReply reply;
+   __GLXscreenConfigs *psc;
+
+   psc = priv->screenConfigs + screen;
+   psc->serverGLXexts =
+      __glXQueryServerString(dpy, priv->majorOpcode, screen, GLX_EXTENSIONS);
+
+   LockDisplay(dpy);
+
+   psc->configs = NULL;
+   if (atof(priv->serverGLXversion) >= 1.3) {
+      GetReq(GLXGetFBConfigs, fb_req);
+      fb_req->reqType = priv->majorOpcode;
+      fb_req->glxCode = X_GLXGetFBConfigs;
+      fb_req->screen = screen;
+   }
+   else if (strstr(psc->serverGLXexts, "GLX_SGIX_fbconfig") != NULL) {
+      GetReqExtra(GLXVendorPrivateWithReply,
+                  sz_xGLXGetFBConfigsSGIXReq +
+                  sz_xGLXVendorPrivateWithReplyReq, vpreq);
+      sgi_req = (xGLXGetFBConfigsSGIXReq *) vpreq;
+      sgi_req->reqType = priv->majorOpcode;
+      sgi_req->glxCode = X_GLXVendorPrivateWithReply;
+      sgi_req->vendorCode = X_GLXvop_GetFBConfigsSGIX;
+      sgi_req->screen = screen;
+   }
+   else
+      goto out;
+
+   if (!_XReply(dpy, (xReply *) & reply, 0, False))
+      goto out;
+
+   psc->configs = createConfigsFromProperties(dpy,
+                                              reply.numFBConfigs,
+                                              reply.numAttribs * 2,
+                                              screen, GL_TRUE);
+
+ out:
+   UnlockDisplay(dpy);
+   return psc->configs != NULL;
+}
+
+static GLboolean
+AllocAndFetchScreenConfigs(Display * dpy, __GLXdisplayPrivate * priv)
+{
+   __GLXscreenConfigs *psc;
+   GLint i, screens;
+
+   /*
+    ** First allocate memory for the array of per screen configs.
+    */
+   screens = ScreenCount(dpy);
+   psc = (__GLXscreenConfigs *) Xmalloc(screens * sizeof(__GLXscreenConfigs));
+   if (!psc) {
+      return GL_FALSE;
+   }
+   memset(psc, 0, screens * sizeof(__GLXscreenConfigs));
+   priv->screenConfigs = psc;
+
+   priv->serverGLXversion =
+      __glXQueryServerString(dpy, priv->majorOpcode, 0, GLX_VERSION);
+   if (priv->serverGLXversion == NULL) {
+      FreeScreenConfigs(priv);
+      return GL_FALSE;
+   }
+
+   for (i = 0; i < screens; i++, psc++) {
+      getFBConfigs(dpy, priv, i);
+      getVisualConfigs(dpy, priv, i);
+      psc->scr = i;
+      psc->dpy = dpy;
+   }
+
+   SyncHandle();
+
+   return GL_TRUE;
+}
+
+_X_HIDDEN void
+__glXRelease(__GLXdisplayPrivate *dpyPriv)
+{
+   FreeScreenConfigs(dpyPriv);
+
+   if (dpyPriv->serverGLXvendor) {
+      Xfree((char *) dpyPriv->serverGLXvendor);
+      dpyPriv->serverGLXvendor = NULL;
+   }
+   if (dpyPriv->serverGLXversion) {
+      Xfree((char *) dpyPriv->serverGLXversion);
+      dpyPriv->serverGLXversion = NULL;
+   }
+
+   Xfree(dpyPriv);
+}
+
+_X_HIDDEN __GLXdisplayPrivate *
+__glXInitialize(Display * dpy)
+{
+   XExtDisplayInfo *info = __glXFindDisplay(dpy);
+   __GLXdisplayPrivate *dpyPriv;
+   int major, minor;
+
+   if (!XextHasExtension(info))
+      return NULL;
+
+   /* See if the versions are compatible */
+   if (!QueryVersion(dpy, info->codes->major_opcode, &major, &minor))
+      return NULL;
+
+   dpyPriv = (__GLXdisplayPrivate *) Xcalloc(1, sizeof(__GLXdisplayPrivate));
+   if (!dpyPriv)
+      return NULL;
+
+   /*
+    ** Init the display private and then read in the screen config
+    ** structures from the server.
+    */
+   dpyPriv->majorOpcode = info->codes->major_opcode;
+   dpyPriv->majorVersion = major;
+   dpyPriv->minorVersion = minor;
+   dpyPriv->dpy = dpy;
+
+   dpyPriv->serverGLXvendor = NULL;
+   dpyPriv->serverGLXversion = NULL;
+
+   if (!AllocAndFetchScreenConfigs(dpy, dpyPriv)) {
+      Xfree(dpyPriv);
+      return NULL;
+   }
+
+   return dpyPriv;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/x11/glxinit.h b/src/gallium/state_trackers/egl_g3d/x11/glxinit.h
new file mode 100644
index 00000000000..515a8252220
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/glxinit.h
@@ -0,0 +1,14 @@
+#ifndef GLXINIT_INCLUDED
+#define GLXINIT_INCLUDED
+
+#include <X11/Xlib.h>
+
+#ifndef GLX_DIRECT_RENDERING
+#define GLX_DIRECT_RENDERING
+#endif
+#include "glxclient.h"
+
+extern void
+__glXRelease(__GLXdisplayPrivate *dpyPriv);
+
+#endif /* GLXINIT_INCLUDED */
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_dri2.c b/src/gallium/state_trackers/egl_g3d/x11/native_dri2.c
new file mode 100644
index 00000000000..f497d8c1c77
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/native_dri2.c
@@ -0,0 +1,690 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 "util/u_memory.h"
+#include "util/u_math.h"
+#include "util/u_format.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "state_tracker/drm_api.h"
+#include "egllog.h"
+
+#include "native_x11.h"
+#include "x11_screen.h"
+
+enum dri2_surface_type {
+   DRI2_SURFACE_TYPE_WINDOW,
+   DRI2_SURFACE_TYPE_PIXMAP,
+   DRI2_SURFACE_TYPE_PBUFFER
+};
+
+struct dri2_display {
+   struct native_display base;
+   Display *dpy;
+   boolean own_dpy;
+
+   struct drm_api *api;
+   struct x11_screen *xscr;
+   int xscr_number;
+
+   struct dri2_config *configs;
+   int num_configs;
+};
+
+struct dri2_surface {
+   struct native_surface base;
+   Drawable drawable;
+   enum dri2_surface_type type;
+   enum pipe_format color_format;
+   struct dri2_display *dri2dpy;
+
+   struct pipe_texture *pbuffer_textures[NUM_NATIVE_ATTACHMENTS];
+   boolean have_back, have_fake;
+   int width, height;
+   unsigned int sequence_number;
+};
+
+struct dri2_config {
+   struct native_config base;
+};
+
+static INLINE struct dri2_display *
+dri2_display(const struct native_display *ndpy)
+{
+   return (struct dri2_display *) ndpy;
+}
+
+static INLINE struct dri2_surface *
+dri2_surface(const struct native_surface *nsurf)
+{
+   return (struct dri2_surface *) nsurf;
+}
+
+static INLINE struct dri2_config *
+dri2_config(const struct native_config *nconf)
+{
+   return (struct dri2_config *) nconf;
+}
+
+static boolean
+dri2_surface_flush_frontbuffer(struct native_surface *nsurf)
+{
+   struct dri2_surface *dri2surf = dri2_surface(nsurf);
+   struct dri2_display *dri2dpy = dri2surf->dri2dpy;
+
+   /* pbuffer is private */
+   if (dri2surf->type == DRI2_SURFACE_TYPE_PBUFFER)
+      return TRUE;
+
+   /* copy to real front buffer */
+   if (dri2surf->have_fake)
+      x11_drawable_copy_buffers(dri2dpy->xscr, dri2surf->drawable,
+            0, 0, dri2surf->width, dri2surf->height,
+            DRI2BufferFakeFrontLeft, DRI2BufferFrontLeft);
+
+   return TRUE;
+}
+
+static boolean
+dri2_surface_swap_buffers(struct native_surface *nsurf)
+{
+   struct dri2_surface *dri2surf = dri2_surface(nsurf);
+   struct dri2_display *dri2dpy = dri2surf->dri2dpy;
+
+   /* pbuffer is private */
+   if (dri2surf->type == DRI2_SURFACE_TYPE_PBUFFER)
+      return TRUE;
+
+   /* copy to front buffer */
+   if (dri2surf->have_back)
+      x11_drawable_copy_buffers(dri2dpy->xscr, dri2surf->drawable,
+            0, 0, dri2surf->width, dri2surf->height,
+            DRI2BufferBackLeft, DRI2BufferFrontLeft);
+
+   /* and update fake front buffer */
+   if (dri2surf->have_fake)
+      x11_drawable_copy_buffers(dri2dpy->xscr, dri2surf->drawable,
+            0, 0, dri2surf->width, dri2surf->height,
+            DRI2BufferFrontLeft, DRI2BufferFakeFrontLeft);
+
+   return TRUE;
+}
+
+static boolean
+dri2_surface_validate(struct native_surface *nsurf,
+                             const enum native_attachment *natts,
+                             unsigned num_natts,
+                             unsigned int *seq_num,
+                             struct pipe_texture **textures,
+                             int *width, int *height)
+{
+   struct dri2_surface *dri2surf = dri2_surface(nsurf);
+   struct dri2_display *dri2dpy = dri2surf->dri2dpy;
+   unsigned int dri2atts[NUM_NATIVE_ATTACHMENTS];
+   EGLint texture_indices[NUM_NATIVE_ATTACHMENTS];
+   struct pipe_texture templ;
+   struct x11_drawable_buffer *xbufs;
+   int num_ins, num_outs, i;
+
+   if (num_natts) {
+      memset(&templ, 0, sizeof(templ));
+      templ.target = PIPE_TEXTURE_2D;
+      templ.last_level = 0;
+      templ.width0 = dri2surf->width;
+      templ.height0 = dri2surf->height;
+      templ.depth0 = 1;
+      templ.format = dri2surf->color_format;
+      templ.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+      if (textures)
+         memset(textures, 0, sizeof(*textures) * num_natts);
+   }
+
+   /* create textures for pbuffer */
+   if (dri2surf->type == DRI2_SURFACE_TYPE_PBUFFER) {
+      struct pipe_screen *screen = dri2dpy->base.screen;
+
+      for (i = 0; i < num_natts; i++) {
+         enum native_attachment natt = natts[i];
+         struct pipe_texture *ptex = dri2surf->pbuffer_textures[natt];
+
+         if (!ptex) {
+            ptex = screen->texture_create(screen, &templ);
+            dri2surf->pbuffer_textures[natt] = ptex;
+         }
+
+         if (textures)
+            pipe_texture_reference(&textures[i], ptex);
+      }
+
+      if (seq_num)
+         *seq_num = dri2surf->sequence_number;
+      if (width)
+         *width = dri2surf->width;
+      if (height)
+         *height = dri2surf->height;
+
+      return TRUE;
+   }
+
+   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++)
+      texture_indices[i] = -1;
+
+   /* prepare the attachments */
+   num_ins = num_natts;
+   for (i = 0; i < num_natts; i++) {
+      unsigned int dri2att;
+
+      switch (natts[i]) {
+      case NATIVE_ATTACHMENT_FRONT_LEFT:
+         dri2att = DRI2BufferFrontLeft;
+         break;
+      case NATIVE_ATTACHMENT_BACK_LEFT:
+         dri2att = DRI2BufferBackLeft;
+         break;
+      case NATIVE_ATTACHMENT_FRONT_RIGHT:
+         dri2att = DRI2BufferFrontRight;
+         break;
+      case NATIVE_ATTACHMENT_BACK_RIGHT:
+         dri2att = DRI2BufferBackRight;
+         break;
+      default:
+         assert(0);
+         dri2att = 0;
+         break;
+      }
+      dri2atts[i] = dri2att;
+      texture_indices[natts[i]] = i;
+   }
+
+   dri2surf->have_back = FALSE;
+   dri2surf->have_fake = FALSE;
+
+   /* remember old geometry */
+   templ.width0 = dri2surf->width;
+   templ.height0 = dri2surf->height;
+
+   xbufs = x11_drawable_get_buffers(dri2dpy->xscr, dri2surf->drawable,
+                                    &dri2surf->width, &dri2surf->height,
+                                    dri2atts, FALSE, num_ins, &num_outs);
+   if (!xbufs)
+      return FALSE;
+
+   if (templ.width0 != dri2surf->width || templ.height0 != dri2surf->height) {
+      /* are there cases where the buffers change and the geometry doesn't? */
+      dri2surf->sequence_number++;
+
+      templ.width0 = dri2surf->width;
+      templ.height0 = dri2surf->height;
+   }
+
+   for (i = 0; i < num_outs; i++) {
+      struct x11_drawable_buffer *xbuf = &xbufs[i];
+      const char *desc;
+      enum native_attachment natt;
+
+      switch (xbuf->attachment) {
+      case DRI2BufferFrontLeft:
+         natt = NATIVE_ATTACHMENT_FRONT_LEFT;
+         desc = "DRI2 Front Buffer";
+         break;
+      case DRI2BufferFakeFrontLeft:
+         natt = NATIVE_ATTACHMENT_FRONT_LEFT;
+         desc = "DRI2 Fake Front Buffer";
+         dri2surf->have_fake = TRUE;
+         break;
+      case DRI2BufferBackLeft:
+         natt = NATIVE_ATTACHMENT_BACK_LEFT;
+         desc = "DRI2 Back Buffer";
+         dri2surf->have_back = TRUE;
+         break;
+      default:
+         desc = NULL;
+         break;
+      }
+
+      if (!desc || texture_indices[natt] < 0 ||
+          (textures && textures[texture_indices[natt]])) {
+         if (!desc)
+            _eglLog(_EGL_WARNING, "unknown buffer %d", xbuf->attachment);
+         else if (texture_indices[natt] < 0)
+            _eglLog(_EGL_WARNING, "unexpected buffer %d", xbuf->attachment);
+         else if (textures && textures[texture_indices[natt]])
+            _eglLog(_EGL_WARNING, "both real and fake front buffers are listed");
+         continue;
+      }
+
+      if (textures) {
+         struct pipe_texture *ptex =
+            dri2dpy->api->texture_from_shared_handle(dri2dpy->api,
+                  dri2dpy->base.screen, &templ,
+                  desc, xbuf->pitch, xbuf->name);
+         if (ptex) {
+            /* the caller owns the textures */
+            textures[texture_indices[natt]] = ptex;
+         }
+      }
+   }
+
+   free(xbufs);
+
+   if (seq_num)
+      *seq_num = dri2surf->sequence_number;
+   if (width)
+      *width = dri2surf->width;
+   if (height)
+      *height = dri2surf->height;
+
+   return TRUE;
+}
+
+static void
+dri2_surface_wait(struct native_surface *nsurf)
+{
+   struct dri2_surface *dri2surf = dri2_surface(nsurf);
+   struct dri2_display *dri2dpy = dri2surf->dri2dpy;
+
+   if (dri2surf->have_fake) {
+      x11_drawable_copy_buffers(dri2dpy->xscr, dri2surf->drawable,
+            0, 0, dri2surf->width, dri2surf->height,
+            DRI2BufferFrontLeft, DRI2BufferFakeFrontLeft);
+   }
+}
+
+static void
+dri2_surface_destroy(struct native_surface *nsurf)
+{
+   struct dri2_surface *dri2surf = dri2_surface(nsurf);
+   int i;
+
+   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
+      struct pipe_texture *ptex = dri2surf->pbuffer_textures[i];
+      pipe_texture_reference(&ptex, NULL);
+   }
+
+   if (dri2surf->drawable)
+      x11_drawable_enable_dri2(dri2surf->dri2dpy->xscr,
+            dri2surf->drawable, FALSE);
+   free(dri2surf);
+}
+
+static struct dri2_surface *
+dri2_display_create_surface(struct native_display *ndpy,
+                            enum dri2_surface_type type,
+                            Drawable drawable,
+                            const struct native_config *nconf)
+{
+   struct dri2_display *dri2dpy = dri2_display(ndpy);
+   struct dri2_config *dri2conf = dri2_config(nconf);
+   struct dri2_surface *dri2surf;
+
+   dri2surf = CALLOC_STRUCT(dri2_surface);
+   if (!dri2surf)
+      return NULL;
+
+   if (drawable)
+      x11_drawable_enable_dri2(dri2dpy->xscr, drawable, TRUE);
+
+   dri2surf->dri2dpy = dri2dpy;
+   dri2surf->type = type;
+   dri2surf->drawable = drawable;
+   dri2surf->color_format = dri2conf->base.color_format;
+
+   dri2surf->base.destroy = dri2_surface_destroy;
+   dri2surf->base.swap_buffers = dri2_surface_swap_buffers;
+   dri2surf->base.flush_frontbuffer = dri2_surface_flush_frontbuffer;
+   dri2surf->base.validate = dri2_surface_validate;
+   dri2surf->base.wait = dri2_surface_wait;
+
+   return dri2surf;
+}
+
+static struct native_surface *
+dri2_display_create_window_surface(struct native_display *ndpy,
+                                   EGLNativeWindowType win,
+                                   const struct native_config *nconf)
+{
+   struct dri2_surface *dri2surf;
+
+   dri2surf = dri2_display_create_surface(ndpy, DRI2_SURFACE_TYPE_WINDOW,
+         (Drawable) win, nconf);
+   return (dri2surf) ? &dri2surf->base : NULL;
+}
+
+static struct native_surface *
+dri2_display_create_pixmap_surface(struct native_display *ndpy,
+                                   EGLNativePixmapType pix,
+                                   const struct native_config *nconf)
+{
+   struct dri2_surface *dri2surf;
+
+   dri2surf = dri2_display_create_surface(ndpy, DRI2_SURFACE_TYPE_PIXMAP,
+         (Drawable) pix, nconf);
+   return (dri2surf) ? &dri2surf->base : NULL;
+}
+
+static struct native_surface *
+dri2_display_create_pbuffer_surface(struct native_display *ndpy,
+                                    const struct native_config *nconf,
+                                    uint width, uint height)
+{
+   struct dri2_surface *dri2surf;
+
+   dri2surf = dri2_display_create_surface(ndpy, DRI2_SURFACE_TYPE_PBUFFER,
+         (Drawable) None, nconf);
+   if (dri2surf) {
+      dri2surf->width = width;
+      dri2surf->height = height;
+   }
+   return (dri2surf) ? &dri2surf->base : NULL;
+}
+
+static struct pipe_context *
+dri2_display_create_context(struct native_display *ndpy, void *context_private)
+{
+   struct dri2_display *dri2dpy = dri2_display(ndpy);
+   struct pipe_context *pctx;
+
+   pctx = dri2dpy->api->create_context(dri2dpy->api, dri2dpy->base.screen);
+   if (pctx)
+      pctx->priv = context_private;
+   return pctx;
+}
+
+static int
+choose_color_format(const __GLcontextModes *mode, enum pipe_format formats[32])
+{
+   int count = 0;
+
+   switch (mode->rgbBits) {
+   case 32:
+      formats[count++] = PIPE_FORMAT_A8R8G8B8_UNORM;
+      formats[count++] = PIPE_FORMAT_B8G8R8A8_UNORM;
+      break;
+   case 24:
+      formats[count++] = PIPE_FORMAT_X8R8G8B8_UNORM;
+      formats[count++] = PIPE_FORMAT_B8G8R8X8_UNORM;
+      formats[count++] = PIPE_FORMAT_A8R8G8B8_UNORM;
+      formats[count++] = PIPE_FORMAT_B8G8R8A8_UNORM;
+      break;
+   case 16:
+      formats[count++] = PIPE_FORMAT_R5G6B5_UNORM;
+      break;
+   default:
+      break;
+   }
+
+   return count;
+}
+
+static int
+choose_depth_stencil_format(const __GLcontextModes *mode,
+                            enum pipe_format formats[32])
+{
+   int count = 0;
+
+   switch (mode->depthBits) {
+   case 32:
+      formats[count++] = PIPE_FORMAT_Z32_UNORM;
+      break;
+   case 24:
+      if (mode->stencilBits) {
+         formats[count++] = PIPE_FORMAT_S8Z24_UNORM;
+         formats[count++] = PIPE_FORMAT_Z24S8_UNORM;
+      }
+      else {
+         formats[count++] = PIPE_FORMAT_X8Z24_UNORM;
+         formats[count++] = PIPE_FORMAT_Z24X8_UNORM;
+      }
+      break;
+   case 16:
+      formats[count++] = PIPE_FORMAT_Z16_UNORM;
+      break;
+   default:
+      break;
+   }
+
+   return count;
+}
+
+static boolean
+is_format_supported(struct pipe_screen *screen,
+                    enum pipe_format fmt, boolean is_color)
+{
+   return screen->is_format_supported(screen, fmt, PIPE_TEXTURE_2D,
+         (is_color) ? PIPE_TEXTURE_USAGE_RENDER_TARGET :
+         PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0);
+}
+
+static boolean
+dri2_display_convert_config(struct native_display *ndpy,
+                            const __GLcontextModes *mode,
+                            struct native_config *nconf)
+{
+   enum pipe_format formats[32];
+   int num_formats, i;
+
+   if (!(mode->renderType & GLX_RGBA_BIT) || !mode->rgbMode)
+      return FALSE;
+
+   /* skip single-buffered configs */
+   if (!mode->doubleBufferMode)
+      return FALSE;
+
+   nconf->mode = *mode;
+   nconf->mode.renderType = GLX_RGBA_BIT;
+   nconf->mode.rgbMode = TRUE;
+   /* pbuffer is allocated locally and is always supported */
+   nconf->mode.drawableType |= GLX_PBUFFER_BIT;
+   /* the swap method is always copy */
+   nconf->mode.swapMethod = GLX_SWAP_COPY_OML;
+
+   /* fix up */
+   nconf->mode.rgbBits =
+      nconf->mode.redBits + nconf->mode.greenBits +
+      nconf->mode.blueBits + nconf->mode.alphaBits;
+   if (!(nconf->mode.drawableType & GLX_WINDOW_BIT)) {
+      nconf->mode.visualID = 0;
+      nconf->mode.visualType = GLX_NONE;
+   }
+   if (!(nconf->mode.drawableType & GLX_PBUFFER_BIT)) {
+      nconf->mode.bindToTextureRgb = FALSE;
+      nconf->mode.bindToTextureRgba = FALSE;
+   }
+
+   nconf->color_format = PIPE_FORMAT_NONE;
+   nconf->depth_format = PIPE_FORMAT_NONE;
+   nconf->stencil_format = PIPE_FORMAT_NONE;
+
+   /* choose color format */
+   num_formats = choose_color_format(mode, formats);
+   for (i = 0; i < num_formats; i++) {
+      if (is_format_supported(ndpy->screen, formats[i], TRUE)) {
+         nconf->color_format = formats[i];
+         break;
+      }
+   }
+   if (nconf->color_format == PIPE_FORMAT_NONE)
+      return FALSE;
+
+   /* choose depth/stencil format */
+   num_formats = choose_depth_stencil_format(mode, formats);
+   for (i = 0; i < num_formats; i++) {
+      if (is_format_supported(ndpy->screen, formats[i], FALSE)) {
+         nconf->depth_format = formats[i];
+         nconf->stencil_format = formats[i];
+         break;
+      }
+   }
+   if ((nconf->mode.depthBits && nconf->depth_format == PIPE_FORMAT_NONE) ||
+       (nconf->mode.stencilBits && nconf->stencil_format == PIPE_FORMAT_NONE))
+      return FALSE;
+
+   return TRUE;
+}
+
+static const struct native_config **
+dri2_display_get_configs(struct native_display *ndpy, int *num_configs)
+{
+   struct dri2_display *dri2dpy = dri2_display(ndpy);
+   const struct native_config **configs;
+   int i;
+
+   /* first time */
+   if (!dri2dpy->configs) {
+      const __GLcontextModes *modes;
+      int num_modes, count;
+
+      modes = x11_screen_get_glx_configs(dri2dpy->xscr);
+      if (!modes)
+         return NULL;
+      num_modes = x11_context_modes_count(modes);
+
+      dri2dpy->configs = calloc(num_modes, sizeof(*dri2dpy->configs));
+      if (!dri2dpy->configs)
+         return NULL;
+
+      count = 0;
+      for (i = 0; i < num_modes; i++) {
+         struct native_config *nconf = &dri2dpy->configs[count].base;
+         if (dri2_display_convert_config(&dri2dpy->base, modes, nconf))
+            count++;
+         modes = modes->next;
+      }
+
+      dri2dpy->num_configs = count;
+   }
+
+   configs = malloc(dri2dpy->num_configs * sizeof(*configs));
+   if (configs) {
+      for (i = 0; i < dri2dpy->num_configs; i++)
+         configs[i] = (const struct native_config *) &dri2dpy->configs[i];
+      if (num_configs)
+         *num_configs = dri2dpy->num_configs;
+   }
+
+   return configs;
+}
+
+static void
+dri2_display_destroy(struct native_display *ndpy)
+{
+   struct dri2_display *dri2dpy = dri2_display(ndpy);
+
+   if (dri2dpy->configs)
+      free(dri2dpy->configs);
+
+   if (dri2dpy->base.screen)
+      dri2dpy->base.screen->destroy(dri2dpy->base.screen);
+
+   if (dri2dpy->xscr)
+      x11_screen_destroy(dri2dpy->xscr);
+   if (dri2dpy->own_dpy)
+      XCloseDisplay(dri2dpy->dpy);
+   if (dri2dpy->api && dri2dpy->api->destroy)
+      dri2dpy->api->destroy(dri2dpy->api);
+   free(dri2dpy);
+}
+
+/**
+ * Initialize DRI2 and pipe screen.
+ */
+static boolean
+dri2_display_init_screen(struct native_display *ndpy)
+{
+   struct dri2_display *dri2dpy = dri2_display(ndpy);
+   const char *driver = dri2dpy->api->name;
+   struct drm_create_screen_arg arg;
+   int fd;
+
+   if (!x11_screen_support(dri2dpy->xscr, X11_SCREEN_EXTENSION_DRI2) ||
+       !x11_screen_support(dri2dpy->xscr, X11_SCREEN_EXTENSION_GLX)) {
+      _eglLog(_EGL_WARNING, "GLX/DRI2 is not supported");
+      return FALSE;
+   }
+
+   fd = x11_screen_enable_dri2(dri2dpy->xscr, driver);
+   if (fd < 0)
+      return FALSE;
+
+   memset(&arg, 0, sizeof(arg));
+   arg.mode = DRM_CREATE_NORMAL;
+   dri2dpy->base.screen = dri2dpy->api->create_screen(dri2dpy->api, fd, &arg);
+   if (!dri2dpy->base.screen) {
+      _eglLog(_EGL_WARNING, "failed to create DRM screen");
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+struct native_display *
+x11_create_dri2_display(EGLNativeDisplayType dpy, struct drm_api *api)
+{
+   struct dri2_display *dri2dpy;
+
+   dri2dpy = CALLOC_STRUCT(dri2_display);
+   if (!dri2dpy)
+      return NULL;
+
+   dri2dpy->api = api;
+   if (!dri2dpy->api) {
+      _eglLog(_EGL_WARNING, "failed to create DRM API");
+      free(dri2dpy);
+      return NULL;
+   }
+
+   dri2dpy->dpy = dpy;
+   if (!dri2dpy->dpy) {
+      dri2dpy->dpy = XOpenDisplay(NULL);
+      if (!dri2dpy->dpy) {
+         dri2_display_destroy(&dri2dpy->base);
+         return NULL;
+      }
+      dri2dpy->own_dpy = TRUE;
+   }
+
+   dri2dpy->xscr_number = DefaultScreen(dri2dpy->dpy);
+   dri2dpy->xscr = x11_screen_create(dri2dpy->dpy, dri2dpy->xscr_number);
+   if (!dri2dpy->xscr) {
+      dri2_display_destroy(&dri2dpy->base);
+      return NULL;
+   }
+
+   if (!dri2_display_init_screen(&dri2dpy->base)) {
+      dri2_display_destroy(&dri2dpy->base);
+      return NULL;
+   }
+
+   dri2dpy->base.destroy = dri2_display_destroy;
+   dri2dpy->base.get_configs = dri2_display_get_configs;
+   dri2dpy->base.create_context = dri2_display_create_context;
+   dri2dpy->base.create_window_surface = dri2_display_create_window_surface;
+   dri2dpy->base.create_pixmap_surface = dri2_display_create_pixmap_surface;
+   dri2dpy->base.create_pbuffer_surface = dri2_display_create_pbuffer_surface;
+
+   return &dri2dpy->base;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_x11.c b/src/gallium/state_trackers/egl_g3d/x11/native_x11.c
new file mode 100644
index 00000000000..583ce3d3293
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/native_x11.c
@@ -0,0 +1,72 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 <string.h>
+#include "util/u_debug.h"
+#include "state_tracker/drm_api.h"
+#include "egllog.h"
+
+#include "native_x11.h"
+
+static struct drm_api *api;
+
+const char *
+native_get_name(void)
+{
+   static char x11_name[32];
+
+   if (!api)
+      api = drm_api_create();
+
+   if (api)
+      snprintf(x11_name, sizeof(x11_name), "X11/%s", api->name);
+   else
+      snprintf(x11_name, sizeof(x11_name), "X11");
+
+   return x11_name;
+}
+
+struct native_display *
+native_create_display(EGLNativeDisplayType dpy)
+{
+   struct native_display *ndpy = NULL;
+   boolean force_sw;
+
+   if (!api)
+      api = drm_api_create();
+
+   force_sw = debug_get_bool_option("EGL_SOFTWARE", FALSE);
+   if (api && !force_sw) {
+      ndpy = x11_create_dri2_display(dpy, api);
+   }
+
+   if (!ndpy) {
+      EGLint level = (force_sw) ? _EGL_INFO : _EGL_WARNING;
+
+      _eglLog(level, "use software fallback");
+      ndpy = x11_create_ximage_display(dpy, TRUE);
+   }
+
+   return ndpy;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_x11.h b/src/gallium/state_trackers/egl_g3d/x11/native_x11.h
new file mode 100644
index 00000000000..622ddac5df6
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/native_x11.h
@@ -0,0 +1,37 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 _NATIVE_X11_H_
+#define _NATIVE_X11_H_
+
+#include "state_tracker/drm_api.h"
+#include "common/native.h"
+
+struct native_display *
+x11_create_ximage_display(EGLNativeDisplayType dpy, boolean use_xshm);
+
+struct native_display *
+x11_create_dri2_display(EGLNativeDisplayType dpy, struct drm_api *api);
+
+#endif /* _NATIVE_X11_H_ */
diff --git a/src/gallium/state_trackers/egl_g3d/x11/native_ximage.c b/src/gallium/state_trackers/egl_g3d/x11/native_ximage.c
new file mode 100644
index 00000000000..24a50df7a0a
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/native_ximage.c
@@ -0,0 +1,676 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 <assert.h>
+#include <sys/ipc.h>
+#include <sys/types.h>
+#include <sys/shm.h>
+#include <X11/Xlib.h>
+#include <X11/Xutil.h>
+#include <X11/extensions/XShm.h>
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "util/u_format.h"
+#include "pipe/p_compiler.h"
+#include "pipe/internal/p_winsys_screen.h"
+#include "softpipe/sp_winsys.h"
+#include "egllog.h"
+
+#include "sw_winsys.h"
+#include "native_x11.h"
+#include "x11_screen.h"
+
+enum ximage_surface_type {
+   XIMAGE_SURFACE_TYPE_WINDOW,
+   XIMAGE_SURFACE_TYPE_PIXMAP,
+   XIMAGE_SURFACE_TYPE_PBUFFER
+};
+
+struct ximage_display {
+   struct native_display base;
+   Display *dpy;
+   boolean own_dpy;
+
+   struct x11_screen *xscr;
+   int xscr_number;
+
+   boolean use_xshm;
+
+   struct pipe_winsys *winsys;
+   struct ximage_config *configs;
+   int num_configs;
+};
+
+struct ximage_buffer {
+   XImage *ximage;
+
+   struct pipe_texture *texture;
+   struct pipe_transfer *transfer;
+   XShmSegmentInfo *shm_info;
+   boolean xshm_attached;
+};
+
+struct ximage_surface {
+   struct native_surface base;
+   Drawable drawable;
+   enum ximage_surface_type type;
+   enum pipe_format color_format;
+   XVisualInfo visual;
+   struct ximage_display *xdpy;
+
+   int width, height;
+   GC gc;
+
+   struct ximage_buffer buffers[NUM_NATIVE_ATTACHMENTS];
+   unsigned int sequence_number;
+};
+
+struct ximage_config {
+   struct native_config base;
+   const XVisualInfo *visual;
+};
+
+static INLINE struct ximage_display *
+ximage_display(const struct native_display *ndpy)
+{
+   return (struct ximage_display *) ndpy;
+}
+
+static INLINE struct ximage_surface *
+ximage_surface(const struct native_surface *nsurf)
+{
+   return (struct ximage_surface *) nsurf;
+}
+
+static INLINE struct ximage_config *
+ximage_config(const struct native_config *nconf)
+{
+   return (struct ximage_config *) nconf;
+}
+
+static void
+ximage_surface_free_buffer(struct native_surface *nsurf,
+                           enum native_attachment which)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   struct ximage_buffer *xbuf = &xsurf->buffers[which];
+   struct pipe_screen *screen = xsurf->xdpy->base.screen;
+
+   if (xbuf->transfer) {
+      screen->tex_transfer_destroy(xbuf->transfer);
+      xbuf->transfer = NULL;
+   }
+   pipe_texture_reference(&xbuf->texture, NULL);
+
+   if (xbuf->shm_info) {
+      if (xbuf->xshm_attached)
+         XShmDetach(xsurf->xdpy->dpy, xbuf->shm_info);
+      if (xbuf->shm_info->shmaddr != (void *) -1)
+         shmdt(xbuf->shm_info->shmaddr);
+      if (xbuf->shm_info->shmid != -1)
+         shmctl(xbuf->shm_info->shmid, IPC_RMID, 0);
+
+      xbuf->shm_info->shmaddr = (void *) -1;
+      xbuf->shm_info->shmid = -1;
+   }
+}
+
+static boolean
+ximage_surface_alloc_buffer(struct native_surface *nsurf,
+                            enum native_attachment which)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   struct ximage_buffer *xbuf = &xsurf->buffers[which];
+   struct pipe_screen *screen = xsurf->xdpy->base.screen;
+   struct pipe_texture templ;
+
+   /* free old data */
+   if (xbuf->texture)
+      ximage_surface_free_buffer(&xsurf->base, which);
+
+   memset(&templ, 0, sizeof(templ));
+   templ.target = PIPE_TEXTURE_2D;
+   templ.format = xsurf->color_format;
+   templ.width0 = xsurf->width;
+   templ.height0 = xsurf->height;
+   templ.depth0 = 1;
+   templ.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+   if (xbuf->shm_info) {
+      struct pipe_buffer *pbuf;
+      unsigned stride, size;
+      void *addr = NULL;
+
+      stride = util_format_get_stride(xsurf->color_format, xsurf->width);
+      /* alignment should depend on visual? */
+      stride = align(stride, 4);
+      size = stride * xsurf->height;
+
+      /* create and attach shm object */
+      xbuf->shm_info->shmid = shmget(IPC_PRIVATE, size, 0755);
+      if (xbuf->shm_info->shmid != -1) {
+         xbuf->shm_info->shmaddr =
+            shmat(xbuf->shm_info->shmid, NULL, 0);
+         if (xbuf->shm_info->shmaddr != (void *) -1) {
+            if (XShmAttach(xsurf->xdpy->dpy, xbuf->shm_info)) {
+               addr = xbuf->shm_info->shmaddr;
+               xbuf->xshm_attached = TRUE;
+            }
+         }
+      }
+
+      if (addr) {
+         pbuf = screen->user_buffer_create(screen, addr, size);
+         if (pbuf) {
+            xbuf->texture =
+               screen->texture_blanket(screen, &templ, &stride, pbuf);
+            pipe_buffer_reference(&pbuf, NULL);
+         }
+      }
+   }
+   else {
+      xbuf->texture = screen->texture_create(screen, &templ);
+   }
+
+   if (xbuf->texture) {
+      xbuf->transfer = screen->get_tex_transfer(screen, xbuf->texture,
+            0, 0, 0, PIPE_TRANSFER_READ, 0, 0, xsurf->width, xsurf->height);
+      if (!xbuf->transfer)
+         pipe_texture_reference(&xbuf->texture, NULL);
+   }
+
+   /* clean up the buffer if allocation failed */
+   if (!xbuf->texture)
+      ximage_surface_free_buffer(&xsurf->base, which);
+
+   return (xbuf->texture != NULL);
+}
+
+static boolean
+ximage_surface_draw_buffer(struct native_surface *nsurf,
+                           enum native_attachment which)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   struct ximage_buffer *xbuf = &xsurf->buffers[which];
+   struct pipe_screen *screen = xsurf->xdpy->base.screen;
+
+   if (xsurf->type == XIMAGE_SURFACE_TYPE_PBUFFER)
+      return TRUE;
+
+   assert(xsurf->drawable && xbuf->ximage && xbuf->texture);
+
+   xbuf->ximage->data = screen->transfer_map(screen, xbuf->transfer);
+
+   if (xbuf->shm_info)
+      XShmPutImage(xsurf->xdpy->dpy, xsurf->drawable, xsurf->gc,
+            xbuf->ximage, 0, 0, 0, 0, xsurf->width, xsurf->height, False);
+   else
+      XPutImage(xsurf->xdpy->dpy, xsurf->drawable, xsurf->gc,
+            xbuf->ximage, 0, 0, 0, 0, xsurf->width, xsurf->height);
+
+   xbuf->ximage->data = NULL;
+   screen->transfer_unmap(screen, xbuf->transfer);
+
+   XSync(xsurf->xdpy->dpy, FALSE);
+
+   return TRUE;
+}
+
+static boolean
+ximage_surface_flush_frontbuffer(struct native_surface *nsurf)
+{
+   return ximage_surface_draw_buffer(nsurf, NATIVE_ATTACHMENT_FRONT_LEFT);
+}
+
+static boolean
+ximage_surface_swap_buffers(struct native_surface *nsurf)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   struct ximage_buffer *xfront, *xback, xtmp;
+
+   xfront = &xsurf->buffers[NATIVE_ATTACHMENT_FRONT_LEFT];
+   xback = &xsurf->buffers[NATIVE_ATTACHMENT_BACK_LEFT];
+
+   /* draw the back buffer directly if there is no front buffer */
+   if (!xfront->texture)
+      return ximage_surface_draw_buffer(nsurf, NATIVE_ATTACHMENT_BACK_LEFT);
+
+   /* swap the buffers */
+   xtmp = *xfront;
+   *xfront = *xback;
+   *xback = xtmp;
+
+   /* the front/back textures are swapped */
+   xsurf->sequence_number++;
+
+   return ximage_surface_draw_buffer(nsurf, NATIVE_ATTACHMENT_FRONT_LEFT);
+}
+
+static void
+ximage_surface_update_geometry(struct native_surface *nsurf)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   Status ok;
+   Window root;
+   int x, y;
+   unsigned int w, h, border, depth;
+
+   /* pbuffer has fixed geometry */
+   if (xsurf->type == XIMAGE_SURFACE_TYPE_PBUFFER)
+      return;
+
+   ok = XGetGeometry(xsurf->xdpy->dpy, xsurf->drawable,
+         &root, &x, &y, &w, &h, &border, &depth);
+   if (ok) {
+      xsurf->width = w;
+      xsurf->height = h;
+   }
+}
+
+static boolean
+ximage_surface_validate(struct native_surface *nsurf,
+                               const enum native_attachment *natts,
+                               unsigned num_natts,
+                               unsigned int *seq_num,
+                               struct pipe_texture **textures,
+                               int *width, int *height)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   boolean new_buffers = FALSE, error = FALSE;
+   unsigned i;
+
+   ximage_surface_update_geometry(&xsurf->base);
+
+   if (textures)
+      memset(textures, 0, sizeof(*textures) * num_natts);
+
+   for (i = 0; i < num_natts; i++) {
+      enum native_attachment natt = natts[i];
+      struct ximage_buffer *xbuf = &xsurf->buffers[natt];
+
+      if (!xbuf)
+         continue;
+
+      /* reallocate the texture */
+      if (!xbuf->texture ||
+          xsurf->width != xbuf->texture->width0 ||
+          xsurf->height != xbuf->texture->height0) {
+         new_buffers = TRUE;
+         if (ximage_surface_alloc_buffer(&xsurf->base, natt)) {
+            /* update ximage */
+            if (xbuf->ximage) {
+               xbuf->ximage->width = xbuf->transfer->width;
+               xbuf->ximage->height = xbuf->transfer->height;
+               xbuf->ximage->bytes_per_line = xbuf->transfer->stride;
+            }
+         }
+      }
+
+      /* allocation failed */
+      if (!xbuf->texture) {
+         unsigned j;
+         for (j = 0; j < i; j++)
+            pipe_texture_reference(&textures[j], NULL);
+         for (j = i; j < num_natts; j++)
+            textures[j] = NULL;
+         error = TRUE;
+         break;
+      }
+
+      if (textures)
+         pipe_texture_reference(&textures[i], xbuf->texture);
+   }
+
+   /* increase the sequence number so that caller knows */
+   if (new_buffers)
+      xsurf->sequence_number++;
+
+   if (seq_num)
+      *seq_num = xsurf->sequence_number;
+   if (width)
+      *width = xsurf->width;
+   if (height)
+      *height = xsurf->height;
+
+   return !error;
+}
+
+static void
+ximage_surface_wait(struct native_surface *nsurf)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   XSync(xsurf->xdpy->dpy, FALSE);
+   /* TODO XGetImage and update the front texture */
+}
+
+static void
+ximage_surface_destroy(struct native_surface *nsurf)
+{
+   struct ximage_surface *xsurf = ximage_surface(nsurf);
+   int i;
+
+   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
+      struct ximage_buffer *xbuf = &xsurf->buffers[i];
+      ximage_surface_free_buffer(&xsurf->base, i);
+      /* xbuf->shm_info is owned by xbuf->ximage? */
+      if (xbuf->ximage) {
+         XDestroyImage(xbuf->ximage);
+         xbuf->ximage = NULL;
+      }
+   }
+
+   if (xsurf->type != XIMAGE_SURFACE_TYPE_PBUFFER)
+      XFreeGC(xsurf->xdpy->dpy, xsurf->gc);
+   free(xsurf);
+}
+
+static struct ximage_surface *
+ximage_display_create_surface(struct native_display *ndpy,
+                              enum ximage_surface_type type,
+                              Drawable drawable,
+                              const struct native_config *nconf)
+{
+   struct ximage_display *xdpy = ximage_display(ndpy);
+   struct ximage_config *xconf = ximage_config(nconf);
+   struct ximage_surface *xsurf;
+   int i;
+
+   xsurf = CALLOC_STRUCT(ximage_surface);
+   if (!xsurf)
+      return NULL;
+
+   xsurf->xdpy = xdpy;
+   xsurf->type = type;
+   xsurf->color_format = xconf->base.color_format;
+   xsurf->drawable = drawable;
+
+   if (xsurf->type != XIMAGE_SURFACE_TYPE_PBUFFER) {
+      xsurf->drawable = drawable;
+      xsurf->visual = *xconf->visual;
+
+      xsurf->gc = XCreateGC(xdpy->dpy, xsurf->drawable, 0, NULL);
+      if (!xsurf->gc) {
+         free(xsurf);
+         return NULL;
+      }
+
+      for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
+         struct ximage_buffer *xbuf = &xsurf->buffers[i];
+
+         if (xdpy->use_xshm) {
+            xbuf->shm_info = calloc(1, sizeof(*xbuf->shm_info));
+            if (xbuf->shm_info) {
+               /* initialize shm info */
+               xbuf->shm_info->shmid = -1;
+               xbuf->shm_info->shmaddr = (void *) -1;
+               xbuf->shm_info->readOnly = TRUE;
+
+               xbuf->ximage = XShmCreateImage(xsurf->xdpy->dpy,
+                     xsurf->visual.visual,
+                     xsurf->visual.depth,
+                     ZPixmap, NULL,
+                     xbuf->shm_info,
+                     0, 0);
+            }
+         }
+         else {
+            xbuf->ximage = XCreateImage(xsurf->xdpy->dpy,
+                  xsurf->visual.visual,
+                  xsurf->visual.depth,
+                  ZPixmap, 0,   /* format, offset */
+                  NULL,         /* data */
+                  0, 0,         /* size */
+                  8,            /* bitmap_pad */
+                  0);           /* bytes_per_line */
+         }
+
+         if (!xbuf->ximage) {
+            XFreeGC(xdpy->dpy, xsurf->gc);
+            free(xsurf);
+            return NULL;
+         }
+      }
+   }
+
+   xsurf->base.destroy = ximage_surface_destroy;
+   xsurf->base.swap_buffers = ximage_surface_swap_buffers;
+   xsurf->base.flush_frontbuffer = ximage_surface_flush_frontbuffer;
+   xsurf->base.validate = ximage_surface_validate;
+   xsurf->base.wait = ximage_surface_wait;
+
+   return xsurf;
+}
+
+static struct native_surface *
+ximage_display_create_window_surface(struct native_display *ndpy,
+                                     EGLNativeWindowType win,
+                                     const struct native_config *nconf)
+{
+   struct ximage_surface *xsurf;
+
+   xsurf = ximage_display_create_surface(ndpy, XIMAGE_SURFACE_TYPE_WINDOW,
+         (Drawable) win, nconf);
+   return (xsurf) ? &xsurf->base : NULL;
+}
+
+static struct native_surface *
+ximage_display_create_pixmap_surface(struct native_display *ndpy,
+                                     EGLNativePixmapType pix,
+                                     const struct native_config *nconf)
+{
+   struct ximage_surface *xsurf;
+
+   xsurf = ximage_display_create_surface(ndpy, XIMAGE_SURFACE_TYPE_PIXMAP,
+         (Drawable) pix, nconf);
+   return (xsurf) ? &xsurf->base : NULL;
+}
+
+static struct native_surface *
+ximage_display_create_pbuffer_surface(struct native_display *ndpy,
+                                      const struct native_config *nconf,
+                                      uint width, uint height)
+{
+   struct ximage_surface *xsurf;
+
+   xsurf = ximage_display_create_surface(ndpy, XIMAGE_SURFACE_TYPE_PBUFFER,
+         (Drawable) None, nconf);
+   if (xsurf) {
+      xsurf->width = width;
+      xsurf->height = height;
+   }
+   return (xsurf) ? &xsurf->base : NULL;
+}
+
+static struct pipe_context *
+ximage_display_create_context(struct native_display *ndpy,
+                              void *context_private)
+{
+   struct pipe_context *pctx = softpipe_create(ndpy->screen);
+   if (pctx)
+      pctx->priv = context_private;
+   return pctx;
+}
+
+static enum pipe_format
+choose_format(const XVisualInfo *vinfo)
+{
+   enum pipe_format fmt;
+   /* TODO elaborate the formats */
+   switch (vinfo->depth) {
+   case 32:
+      fmt = PIPE_FORMAT_A8R8G8B8_UNORM;
+      break;
+   case 24:
+      fmt = PIPE_FORMAT_X8R8G8B8_UNORM;
+      break;
+   case 16:
+      fmt = PIPE_FORMAT_R5G6B5_UNORM;
+      break;
+   default:
+      fmt = PIPE_FORMAT_NONE;
+      break;
+   }
+
+   return fmt;
+}
+
+static const struct native_config **
+ximage_display_get_configs(struct native_display *ndpy, int *num_configs)
+{
+   struct ximage_display *xdpy = ximage_display(ndpy);
+   const struct native_config **configs;
+   int i;
+
+   /* first time */
+   if (!xdpy->configs) {
+      const XVisualInfo *visuals;
+      int num_visuals, count, j;
+
+      visuals = x11_screen_get_visuals(xdpy->xscr, &num_visuals);
+      if (!visuals)
+         return NULL;
+
+      /*
+       * Create two configs for each visual.
+       * One with depth/stencil buffer; one without
+       */
+      xdpy->configs = calloc(num_visuals * 2, sizeof(*xdpy->configs));
+      if (!xdpy->configs)
+         return NULL;
+
+      count = 0;
+      for (i = 0; i < num_visuals; i++) {
+         for (j = 0; j < 2; j++) {
+            struct ximage_config *xconf = &xdpy->configs[count];
+            __GLcontextModes *mode = &xconf->base.mode;
+
+            xconf->visual = &visuals[i];
+            xconf->base.color_format = choose_format(xconf->visual);
+            if (xconf->base.color_format == PIPE_FORMAT_NONE)
+               continue;
+
+            x11_screen_convert_visual(xdpy->xscr, xconf->visual, mode);
+            /* support double buffer mode */
+            mode->doubleBufferMode = TRUE;
+
+            xconf->base.depth_format = PIPE_FORMAT_NONE;
+            xconf->base.stencil_format = PIPE_FORMAT_NONE;
+            /* create the second config with depth/stencil buffer */
+            if (j == 1) {
+               xconf->base.depth_format = PIPE_FORMAT_S8Z24_UNORM;
+               xconf->base.stencil_format = PIPE_FORMAT_S8Z24_UNORM;
+               mode->depthBits = 24;
+               mode->stencilBits = 8;
+               mode->haveDepthBuffer = TRUE;
+               mode->haveStencilBuffer = TRUE;
+            }
+
+            mode->maxPbufferWidth = 4096;
+            mode->maxPbufferHeight = 4096;
+            mode->maxPbufferPixels = 4096 * 4096;
+            mode->drawableType =
+               GLX_WINDOW_BIT | GLX_PIXMAP_BIT | GLX_PBUFFER_BIT;
+            mode->swapMethod = GLX_SWAP_EXCHANGE_OML;
+
+            if (mode->alphaBits)
+               mode->bindToTextureRgba = TRUE;
+            else
+               mode->bindToTextureRgb = TRUE;
+
+            count++;
+         }
+      }
+
+      xdpy->num_configs = count;
+   }
+
+   configs = malloc(xdpy->num_configs * sizeof(*configs));
+   if (configs) {
+      for (i = 0; i < xdpy->num_configs; i++)
+         configs[i] = (const struct native_config *) &xdpy->configs[i];
+      if (num_configs)
+         *num_configs = xdpy->num_configs;
+   }
+   return configs;
+}
+
+static void
+ximage_display_destroy(struct native_display *ndpy)
+{
+   struct ximage_display *xdpy = ximage_display(ndpy);
+
+   if (xdpy->configs)
+      free(xdpy->configs);
+
+   xdpy->base.screen->destroy(xdpy->base.screen);
+   free(xdpy->winsys);
+
+   x11_screen_destroy(xdpy->xscr);
+   if (xdpy->own_dpy)
+      XCloseDisplay(xdpy->dpy);
+   free(xdpy);
+}
+
+struct native_display *
+x11_create_ximage_display(EGLNativeDisplayType dpy, boolean use_xshm)
+{
+   struct ximage_display *xdpy;
+
+   xdpy = CALLOC_STRUCT(ximage_display);
+   if (!xdpy)
+      return NULL;
+
+   xdpy->dpy = dpy;
+   if (!xdpy->dpy) {
+      xdpy->dpy = XOpenDisplay(NULL);
+      if (!xdpy->dpy) {
+         free(xdpy);
+         return NULL;
+      }
+      xdpy->own_dpy = TRUE;
+   }
+
+   xdpy->xscr_number = DefaultScreen(xdpy->dpy);
+   xdpy->xscr = x11_screen_create(xdpy->dpy, xdpy->xscr_number);
+   if (!xdpy->xscr) {
+      free(xdpy);
+      return NULL;
+   }
+
+   xdpy->use_xshm =
+      (use_xshm && x11_screen_support(xdpy->xscr, X11_SCREEN_EXTENSION_XSHM));
+
+   xdpy->winsys = create_sw_winsys();
+   xdpy->base.screen = softpipe_create_screen(xdpy->winsys);
+
+   xdpy->base.destroy = ximage_display_destroy;
+
+   xdpy->base.get_configs = ximage_display_get_configs;
+   xdpy->base.create_context = ximage_display_create_context;
+   xdpy->base.create_window_surface = ximage_display_create_window_surface;
+   xdpy->base.create_pixmap_surface = ximage_display_create_pixmap_surface;
+   xdpy->base.create_pbuffer_surface = ximage_display_create_pbuffer_surface;
+
+   return &xdpy->base;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.c b/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.c
new file mode 100644
index 00000000000..6ee3ede38cb
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.c
@@ -0,0 +1,231 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, 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 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.
+ * 
+ **************************************************************************/
+
+/**
+ * Totally software-based winsys layer.
+ * Note that the one winsys function that we can't implement here
+ * is flush_frontbuffer().
+ * Whoever uses this code will have to provide that.
+ *
+ * Authors: Brian Paul
+ */
+
+
+#include "pipe/internal/p_winsys_screen.h"
+#include "pipe/p_state.h"
+#include "pipe/p_inlines.h"
+#include "util/u_format.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "sw_winsys.h"
+
+
+
+/** Subclass of pipe_winsys */
+struct sw_pipe_winsys
+{
+   struct pipe_winsys Base;
+   /* no extra fields for now */
+};
+
+
+/** subclass of pipe_buffer */
+struct sw_pipe_buffer
+{
+   struct pipe_buffer Base;
+   boolean UserBuffer;  /** Is this a user-space buffer? */
+   void *Data;
+   void *Mapped;
+};
+
+
+/** cast wrapper */
+static INLINE struct sw_pipe_buffer *
+sw_pipe_buffer(struct pipe_buffer *b)
+{
+   return (struct sw_pipe_buffer *) b;
+}
+
+
+static const char *
+get_name(struct pipe_winsys *pws)
+{
+   return "software";
+}
+
+
+/** Create new pipe_buffer and allocate storage of given size */
+static struct pipe_buffer *
+buffer_create(struct pipe_winsys *pws, 
+              unsigned alignment, 
+              unsigned usage,
+              unsigned size)
+{
+   struct sw_pipe_buffer *buffer = CALLOC_STRUCT(sw_pipe_buffer);
+   if (!buffer)
+      return NULL;
+
+   pipe_reference_init(&buffer->Base.reference, 1);
+   buffer->Base.alignment = alignment;
+   buffer->Base.usage = usage;
+   buffer->Base.size = size;
+
+   /* align to 16-byte multiple for Cell */
+   buffer->Data = align_malloc(size, MAX2(alignment, 16));
+
+   return &buffer->Base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
+{
+   struct sw_pipe_buffer *buffer = CALLOC_STRUCT(sw_pipe_buffer);
+   if (!buffer)
+      return NULL;
+
+   pipe_reference_init(&buffer->Base.reference, 1);
+   buffer->Base.size = bytes;
+   buffer->UserBuffer = TRUE;
+   buffer->Data = ptr;
+
+   return &buffer->Base;
+}
+
+
+static void *
+buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buf, unsigned flags)
+{
+   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
+   buffer->Mapped = buffer->Data;
+   return buffer->Mapped;
+}
+
+
+static void
+buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf)
+{
+   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
+   buffer->Mapped = NULL;
+}
+
+
+static void
+buffer_destroy(struct pipe_buffer *buf)
+{
+   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
+
+   if (buffer->Data && !buffer->UserBuffer) {
+      align_free(buffer->Data);
+      buffer->Data = NULL;
+   }
+
+   free(buffer);
+}
+
+
+static struct pipe_buffer *
+surface_buffer_create(struct pipe_winsys *winsys,
+                      unsigned width, unsigned height,
+                      enum pipe_format format, 
+                      unsigned usage,
+                      unsigned tex_usage,
+                      unsigned *stride)
+{
+   const unsigned alignment = 64;
+   unsigned nblocksy;
+
+   nblocksy = util_format_get_nblocksy(format, height);
+   *stride = align(util_format_get_stride(format, width), alignment);
+
+   return winsys->buffer_create(winsys, alignment,
+                                usage,
+                                *stride * nblocksy);
+}
+
+
+static void
+fence_reference(struct pipe_winsys *sws, struct pipe_fence_handle **ptr,
+                struct pipe_fence_handle *fence)
+{
+   /* no-op */
+}
+
+
+static int
+fence_signalled(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
+                unsigned flag)
+{
+   /* no-op */
+   return 0;
+}
+
+
+static int
+fence_finish(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
+             unsigned flag)
+{
+   /* no-op */
+   return 0;
+}
+
+
+/**
+ * Create/return a new pipe_winsys object.
+ */
+struct pipe_winsys *
+create_sw_winsys(void)
+{
+   struct sw_pipe_winsys *ws = CALLOC_STRUCT(sw_pipe_winsys);
+   if (!ws)
+      return NULL;
+
+   /* Fill in this struct with callbacks that pipe will need to
+    * communicate with the window system, buffer manager, etc. 
+    */
+   ws->Base.buffer_create = buffer_create;
+   ws->Base.user_buffer_create = user_buffer_create;
+   ws->Base.buffer_map = buffer_map;
+   ws->Base.buffer_unmap = buffer_unmap;
+   ws->Base.buffer_destroy = buffer_destroy;
+
+   ws->Base.surface_buffer_create = surface_buffer_create;
+
+   ws->Base.fence_reference = fence_reference;
+   ws->Base.fence_signalled = fence_signalled;
+   ws->Base.fence_finish = fence_finish;
+
+   ws->Base.flush_frontbuffer = NULL; /* not implemented here! */
+
+   ws->Base.get_name = get_name;
+
+   return &ws->Base;
+}
diff --git a/src/mesa/state_tracker/st_api.c b/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.h
index fc0e9a23164..f96c5a14b0a 100644
--- a/src/mesa/state_tracker/st_api.c
+++ b/src/gallium/state_trackers/egl_g3d/x11/sw_winsys.h
@@ -26,8 +26,15 @@
  **************************************************************************/
 
 
-/**
- * Just a global symbol for EGL to look for to identify the supported
- * graphics API.
- */
-int st_api_OpenGL = 1;
+#ifndef SW_WINSYS_H
+#define SW_WINSYS_H
+
+
+struct pipe_winsys;
+
+
+extern struct pipe_winsys *
+create_sw_winsys(void);
+
+
+#endif /* SW_WINSYS_H */
diff --git a/src/gallium/state_trackers/egl_g3d/x11/x11_screen.c b/src/gallium/state_trackers/egl_g3d/x11/x11_screen.c
new file mode 100644
index 00000000000..1e98943242a
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/x11_screen.c
@@ -0,0 +1,402 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 <unistd.h>
+#include <fcntl.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <X11/Xlibint.h>
+#include <X11/extensions/XShm.h>
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "xf86drm.h"
+#include "egllog.h"
+
+#include "x11_screen.h"
+#include "dri2.h"
+#include "glxinit.h"
+
+struct x11_screen {
+   Display *dpy;
+   int number;
+
+   __GLXdisplayPrivate *glx_dpy;
+
+   int dri_major, dri_minor;
+   char *dri_driver;
+   char *dri_device;
+   int dri_fd;
+
+   XVisualInfo *visuals;
+   int num_visuals;
+};
+
+
+/**
+ * Create a X11 screen.
+ */
+struct x11_screen *
+x11_screen_create(Display *dpy, int screen)
+{
+   struct x11_screen *xscr;
+
+   if (screen >= ScreenCount(dpy))
+      return NULL;
+
+   xscr = CALLOC_STRUCT(x11_screen);
+   if (xscr) {
+      xscr->dpy = dpy;
+      xscr->number = screen;
+
+      xscr->dri_major = -1;
+      xscr->dri_fd = -1;
+   }
+   return xscr;
+}
+
+/**
+ * Destroy a X11 screen.
+ */
+void
+x11_screen_destroy(struct x11_screen *xscr)
+{
+   if (xscr->dri_fd >= 0)
+      close(xscr->dri_fd);
+   if (xscr->dri_driver)
+      Xfree(xscr->dri_driver);
+   if (xscr->dri_device)
+      Xfree(xscr->dri_device);
+
+   if (xscr->glx_dpy)
+      __glXRelease(xscr->glx_dpy);
+   if (xscr->visuals)
+      XFree(xscr->visuals);
+   free(xscr);
+}
+
+static boolean
+x11_screen_init_dri2(struct x11_screen *xscr)
+{
+   if (xscr->dri_major < 0) {
+      int eventBase, errorBase;
+
+      if (!DRI2QueryExtension(xscr->dpy, &eventBase, &errorBase) ||
+          !DRI2QueryVersion(xscr->dpy, &xscr->dri_major, &xscr->dri_minor))
+         xscr->dri_major = -1;
+   }
+   return (xscr->dri_major >= 0);
+}
+
+static boolean
+x11_screen_init_glx(struct x11_screen *xscr)
+{
+   if (!xscr->glx_dpy)
+      xscr->glx_dpy = __glXInitialize(xscr->dpy);
+   return (xscr->glx_dpy != NULL);
+}
+
+/**
+ * Return true if the screen supports the extension.
+ */
+boolean
+x11_screen_support(struct x11_screen *xscr, enum x11_screen_extension ext)
+{
+   boolean supported = FALSE;
+
+   switch (ext) {
+   case X11_SCREEN_EXTENSION_XSHM:
+      supported = XShmQueryExtension(xscr->dpy);
+      break;
+   case X11_SCREEN_EXTENSION_GLX:
+      supported = x11_screen_init_glx(xscr);
+      break;
+   case X11_SCREEN_EXTENSION_DRI2:
+      supported = x11_screen_init_dri2(xscr);
+      break;
+   default:
+      break;
+   }
+
+   return supported;
+}
+
+/**
+ * Return the X visuals.
+ */
+const XVisualInfo *
+x11_screen_get_visuals(struct x11_screen *xscr, int *num_visuals)
+{
+   if (!xscr->visuals) {
+      XVisualInfo vinfo_template;
+      vinfo_template.screen = xscr->number;
+      xscr->visuals = XGetVisualInfo(xscr->dpy, VisualScreenMask,
+            &vinfo_template, &xscr->num_visuals);
+   }
+
+   if (num_visuals)
+      *num_visuals = xscr->num_visuals;
+   return xscr->visuals;
+}
+
+void
+x11_screen_convert_visual(struct x11_screen *xscr, const XVisualInfo *visual,
+                          __GLcontextModes *mode)
+{
+   int r, g, b, a;
+   int visual_type;
+
+   r = util_bitcount(visual->red_mask);
+   g = util_bitcount(visual->green_mask);
+   b = util_bitcount(visual->blue_mask);
+   a = visual->depth - (r + g + b);
+#if defined(__cplusplus) || defined(c_plusplus)
+   visual_type = visual->c_class;
+#else
+   visual_type = visual->class;
+#endif
+
+   /* convert to GLX visual type */
+   switch (visual_type) {
+   case TrueColor:
+      visual_type = GLX_TRUE_COLOR;
+      break;
+   case DirectColor:
+      visual_type = GLX_DIRECT_COLOR;
+      break;
+   case PseudoColor:
+      visual_type = GLX_PSEUDO_COLOR;
+      break;
+   case StaticColor:
+      visual_type = GLX_STATIC_COLOR;
+      break;
+   case GrayScale:
+      visual_type = GLX_GRAY_SCALE;
+      break;
+   case StaticGray:
+      visual_type = GLX_STATIC_GRAY;
+      break;
+   default:
+      visual_type = GLX_NONE;
+      break;
+   }
+
+   mode->rgbBits = r + g + b + a;
+   mode->redBits = r;
+   mode->greenBits = g;
+   mode->blueBits = b;
+   mode->alphaBits = a;
+   mode->visualID = visual->visualid;
+   mode->visualType = visual_type;
+
+   /* sane defaults */
+   mode->renderType = GLX_RGBA_BIT;
+   mode->rgbMode = TRUE;
+   mode->visualRating = GLX_SLOW_CONFIG;
+   mode->xRenderable = TRUE;
+}
+
+/**
+ * Return the GLX fbconfigs.
+ */
+const __GLcontextModes *
+x11_screen_get_glx_configs(struct x11_screen *xscr)
+{
+   return (x11_screen_init_glx(xscr))
+      ? xscr->glx_dpy->screenConfigs[xscr->number].configs
+      : NULL;
+}
+
+/**
+ * Return the GLX visuals.
+ */
+const __GLcontextModes *
+x11_screen_get_glx_visuals(struct x11_screen *xscr)
+{
+   return (x11_screen_init_glx(xscr))
+      ? xscr->glx_dpy->screenConfigs[xscr->number].visuals
+      : NULL;
+}
+
+static boolean
+x11_screen_is_driver_equal(struct x11_screen *xscr, const char *driver)
+{
+   return (strcmp(xscr->dri_driver, driver) == 0);
+}
+
+/**
+ * Enable DRI2 and returns the file descriptor of the DRM device.  The file
+ * descriptor will be closed automatically when the screen is destoryed.
+ */
+int
+x11_screen_enable_dri2(struct x11_screen *xscr, const char *driver)
+{
+   if (xscr->dri_fd < 0) {
+      int fd;
+      drm_magic_t magic;
+
+      /* get the driver name and the device name first */
+      if (!xscr->dri_driver) {
+         if (!DRI2Connect(xscr->dpy, RootWindow(xscr->dpy, xscr->number),
+                  &xscr->dri_driver, &xscr->dri_device)) {
+            xscr->dri_driver = xscr->dri_device = NULL;
+            return -1;
+         }
+      }
+
+      if (!x11_screen_is_driver_equal(xscr, driver)) {
+         _eglLog(_EGL_WARNING, "Driver mismatch: %s != %s",
+               xscr->dri_driver, driver);
+         return -1;
+      }
+
+      fd = open(xscr->dri_device, O_RDWR);
+      if (fd < 0) {
+         _eglLog(_EGL_WARNING, "failed to open %s", xscr->dri_device);
+         return -1;
+      }
+
+      memset(&magic, 0, sizeof(magic));
+      if (drmGetMagic(fd, &magic)) {
+         _eglLog(_EGL_WARNING, "failed to get magic");
+         close(fd);
+         return -1;
+      }
+
+      if (!DRI2Authenticate(xscr->dpy,
+               RootWindow(xscr->dpy, xscr->number), magic)) {
+         _eglLog(_EGL_WARNING, "failed to authenticate magic");
+         close(fd);
+         return -1;
+      }
+
+      xscr->dri_fd = fd;
+   }
+
+   return xscr->dri_fd;
+}
+
+/**
+ * Create/Destroy the DRI drawable.
+ */
+void
+x11_drawable_enable_dri2(struct x11_screen *xscr,
+                         Drawable drawable, boolean on)
+{
+   if (on)
+      DRI2CreateDrawable(xscr->dpy, drawable);
+   else
+      DRI2DestroyDrawable(xscr->dpy, drawable);
+}
+
+/**
+ * Copy between buffers of the DRI2 drawable.
+ */
+void
+x11_drawable_copy_buffers(struct x11_screen *xscr, Drawable drawable,
+                          int x, int y, int width, int height,
+                          int src_buf, int dst_buf)
+{
+   XRectangle rect;
+   XserverRegion region;
+
+   rect.x = x;
+   rect.y = y;
+   rect.width = width;
+   rect.height = height;
+
+   region = XFixesCreateRegion(xscr->dpy, &rect, 1);
+   DRI2CopyRegion(xscr->dpy, drawable, region, dst_buf, src_buf);
+   XFixesDestroyRegion(xscr->dpy, region);
+}
+
+/**
+ * Get the buffers of the DRI2 drawable.  The returned array should be freed.
+ */
+struct x11_drawable_buffer *
+x11_drawable_get_buffers(struct x11_screen *xscr, Drawable drawable,
+                         int *width, int *height, unsigned int *attachments,
+                         boolean with_format, int num_ins, int *num_outs)
+{
+   DRI2Buffer *dri2bufs;
+
+   if (with_format)
+      dri2bufs = DRI2GetBuffersWithFormat(xscr->dpy, drawable, width, height,
+            attachments, num_ins, num_outs);
+   else
+      dri2bufs = DRI2GetBuffers(xscr->dpy, drawable, width, height,
+            attachments, num_ins, num_outs);
+
+   return (struct x11_drawable_buffer *) dri2bufs;
+}
+
+/**
+ * Create a mode list of the given size.
+ */
+__GLcontextModes *
+x11_context_modes_create(unsigned count)
+{
+   const size_t size = sizeof(__GLcontextModes);
+   __GLcontextModes *base = NULL;
+   __GLcontextModes **next;
+   unsigned i;
+
+   next = &base;
+   for (i = 0; i < count; i++) {
+      *next = (__GLcontextModes *) calloc(1, size);
+      if (*next == NULL) {
+         x11_context_modes_destroy(base);
+         base = NULL;
+         break;
+      }
+      next = &((*next)->next);
+   }
+
+   return base;
+}
+
+/**
+ * Destroy a mode list.
+ */
+void
+x11_context_modes_destroy(__GLcontextModes *modes)
+{
+   while (modes != NULL) {
+      __GLcontextModes *next = modes->next;
+      free(modes);
+      modes = next;
+   }
+}
+
+/**
+ * Return the number of the modes in the mode list.
+ */
+unsigned
+x11_context_modes_count(const __GLcontextModes *modes)
+{
+   const __GLcontextModes *mode;
+   int count = 0;
+   for (mode = modes; mode; mode = mode->next)
+      count++;
+   return count;
+}
diff --git a/src/gallium/state_trackers/egl_g3d/x11/x11_screen.h b/src/gallium/state_trackers/egl_g3d/x11/x11_screen.h
new file mode 100644
index 00000000000..86e8e0501a3
--- /dev/null
+++ b/src/gallium/state_trackers/egl_g3d/x11/x11_screen.h
@@ -0,0 +1,99 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.8
+ *
+ * Copyright (C) 2009-2010 Chia-I Wu <[email protected]>
+ *
+ * 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 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
+ * BRIAN PAUL 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 _X11_SCREEN_H_
+#define _X11_SCREEN_H_
+
+#include <X11/Xlib.h>
+#include <X11/Xutil.h>
+#include <X11/extensions/dri2tokens.h>
+#include "pipe/p_compiler.h"
+#include "common/native.h"
+
+enum x11_screen_extension {
+   X11_SCREEN_EXTENSION_XSHM,
+   X11_SCREEN_EXTENSION_GLX,
+   X11_SCREEN_EXTENSION_DRI2,
+};
+
+/* the same as DRI2Buffer */
+struct x11_drawable_buffer {
+   unsigned int attachment;
+   unsigned int name;
+   unsigned int pitch;
+   unsigned int cpp;
+   unsigned int flags;
+};
+
+struct x11_screen;
+
+struct x11_screen *
+x11_screen_create(Display *dpy, int screen);
+
+void
+x11_screen_destroy(struct x11_screen *xscr);
+
+boolean
+x11_screen_support(struct x11_screen *xscr, enum x11_screen_extension ext);
+
+const XVisualInfo *
+x11_screen_get_visuals(struct x11_screen *xscr, int *num_visuals);
+
+void
+x11_screen_convert_visual(struct x11_screen *xscr, const XVisualInfo *visual,
+                          __GLcontextModes *mode);
+
+const __GLcontextModes *
+x11_screen_get_glx_configs(struct x11_screen *xscr);
+
+const __GLcontextModes *
+x11_screen_get_glx_visuals(struct x11_screen *xscr);
+
+int
+x11_screen_enable_dri2(struct x11_screen *xscr, const char *driver);
+
+__GLcontextModes *
+x11_context_modes_create(unsigned count);
+
+void
+x11_context_modes_destroy(__GLcontextModes *modes);
+
+unsigned
+x11_context_modes_count(const __GLcontextModes *modes);
+
+void
+x11_drawable_enable_dri2(struct x11_screen *xscr,
+                         Drawable drawable, boolean on);
+
+void
+x11_drawable_copy_buffers(struct x11_screen *xscr, Drawable drawable,
+                          int x, int y, int width, int height,
+                          int src_buf, int dst_buf);
+
+struct x11_drawable_buffer *
+x11_drawable_get_buffers(struct x11_screen *xscr, Drawable drawable,
+                         int *width, int *height, unsigned int *attachments,
+                         boolean with_format, int num_ins, int *num_outs);
+
+#endif /* _X11_SCREEN_H_ */
diff --git a/src/gallium/state_trackers/glx/xlib/glx_api.c b/src/gallium/state_trackers/glx/xlib/glx_api.c
index 228ac9a20e9..3caf56e924b 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_api.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_api.c
@@ -1007,7 +1007,7 @@ choose_visual( Display *dpy, int screen, const int *list, GLboolean fbConfig )
 }
 
 
-XVisualInfo *
+PUBLIC XVisualInfo *
 glXChooseVisual( Display *dpy, int screen, int *list )
 {
    XMesaVisual xmvis;
@@ -1029,7 +1029,7 @@ glXChooseVisual( Display *dpy, int screen, int *list )
 }
 
 
-GLXContext
+PUBLIC GLXContext
 glXCreateContext( Display *dpy, XVisualInfo *visinfo,
                   GLXContext share_list, Bool direct )
 {
@@ -1084,7 +1084,7 @@ static XMesaBuffer MakeCurrent_PrevReadBuffer = 0;
 
 
 /* GLX 1.3 and later */
-Bool
+PUBLIC Bool
 glXMakeContextCurrent( Display *dpy, GLXDrawable draw,
                        GLXDrawable read, GLXContext ctx )
 {
@@ -1180,21 +1180,21 @@ glXMakeContextCurrent( Display *dpy, GLXDrawable draw,
 }
 
 
-Bool
+PUBLIC Bool
 glXMakeCurrent( Display *dpy, GLXDrawable drawable, GLXContext ctx )
 {
    return glXMakeContextCurrent( dpy, drawable, drawable, ctx );
 }
 
 
-GLXContext
+PUBLIC GLXContext
 glXGetCurrentContext(void)
 {
    return GetCurrentContext();
 }
 
 
-Display *
+PUBLIC Display *
 glXGetCurrentDisplay(void)
 {
    GLXContext glxCtx = glXGetCurrentContext();
@@ -1203,14 +1203,14 @@ glXGetCurrentDisplay(void)
 }
 
 
-Display *
+PUBLIC Display *
 glXGetCurrentDisplayEXT(void)
 {
    return glXGetCurrentDisplay();
 }
 
 
-GLXDrawable
+PUBLIC GLXDrawable
 glXGetCurrentDrawable(void)
 {
    GLXContext gc = glXGetCurrentContext();
@@ -1218,7 +1218,7 @@ glXGetCurrentDrawable(void)
 }
 
 
-GLXDrawable
+PUBLIC GLXDrawable
 glXGetCurrentReadDrawable(void)
 {
    GLXContext gc = glXGetCurrentContext();
@@ -1226,14 +1226,14 @@ glXGetCurrentReadDrawable(void)
 }
 
 
-GLXDrawable
+PUBLIC GLXDrawable
 glXGetCurrentReadDrawableSGI(void)
 {
    return glXGetCurrentReadDrawable();
 }
 
 
-GLXPixmap
+PUBLIC GLXPixmap
 glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap )
 {
    XMesaVisual v;
@@ -1258,7 +1258,7 @@ glXCreateGLXPixmap( Display *dpy, XVisualInfo *visinfo, Pixmap pixmap )
 
 /*** GLX_MESA_pixmap_colormap ***/
 
-GLXPixmap
+PUBLIC GLXPixmap
 glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo,
                         Pixmap pixmap, Colormap cmap )
 {
@@ -1282,7 +1282,7 @@ glXCreateGLXPixmapMESA( Display *dpy, XVisualInfo *visinfo,
 }
 
 
-void
+PUBLIC void
 glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap )
 {
    XMesaBuffer b = XMesaFindBuffer(dpy, pixmap);
@@ -1295,7 +1295,7 @@ glXDestroyGLXPixmap( Display *dpy, GLXPixmap pixmap )
 }
 
 
-void
+PUBLIC void
 glXCopyContext( Display *dpy, GLXContext src, GLXContext dst,
                 unsigned long mask )
 {
@@ -1309,7 +1309,7 @@ glXCopyContext( Display *dpy, GLXContext src, GLXContext dst,
 }
 
 
-Bool
+PUBLIC Bool
 glXQueryExtension( Display *dpy, int *errorBase, int *eventBase )
 {
    int op, ev, err;
@@ -1324,7 +1324,7 @@ glXQueryExtension( Display *dpy, int *errorBase, int *eventBase )
 }
 
 
-void
+PUBLIC void
 glXDestroyContext( Display *dpy, GLXContext ctx )
 {
    GLXContext glxCtx = ctx;
@@ -1340,7 +1340,7 @@ glXDestroyContext( Display *dpy, GLXContext ctx )
 }
 
 
-Bool
+PUBLIC Bool
 glXIsDirect( Display *dpy, GLXContext ctx )
 {
    GLXContext glxCtx = ctx;
@@ -1350,7 +1350,7 @@ glXIsDirect( Display *dpy, GLXContext ctx )
 
 
 
-void
+PUBLIC void
 glXSwapBuffers( Display *dpy, GLXDrawable drawable )
 {
    XMesaBuffer buffer = XMesaFindBuffer( dpy, drawable );
@@ -1377,7 +1377,7 @@ glXSwapBuffers( Display *dpy, GLXDrawable drawable )
 
 /*** GLX_MESA_copy_sub_buffer ***/
 
-void
+PUBLIC void
 glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable,
                            int x, int y, int width, int height )
 {
@@ -1391,7 +1391,7 @@ glXCopySubBufferMESA( Display *dpy, GLXDrawable drawable,
 }
 
 
-Bool
+PUBLIC Bool
 glXQueryVersion( Display *dpy, int *maj, int *min )
 {
    (void) dpy;
@@ -1608,7 +1608,7 @@ get_config( XMesaVisual xmvis, int attrib, int *value, GLboolean fbconfig )
 }
 
 
-int
+PUBLIC int
 glXGetConfig( Display *dpy, XVisualInfo *visinfo,
                    int attrib, int *value )
 {
@@ -1638,7 +1638,7 @@ glXGetConfig( Display *dpy, XVisualInfo *visinfo,
 }
 
 
-void
+PUBLIC void
 glXWaitGL( void )
 {
    XMesaContext xmesa = XMesaGetCurrentContext();
@@ -1647,7 +1647,7 @@ glXWaitGL( void )
 
 
 
-void
+PUBLIC void
 glXWaitX( void )
 {
    XMesaContext xmesa = XMesaGetCurrentContext();
@@ -1664,7 +1664,7 @@ get_extensions( void )
 
 
 /* GLX 1.1 and later */
-const char *
+PUBLIC const char *
 glXQueryExtensionsString( Display *dpy, int screen )
 {
    (void) dpy;
@@ -1675,7 +1675,7 @@ glXQueryExtensionsString( Display *dpy, int screen )
 
 
 /* GLX 1.1 and later */
-const char *
+PUBLIC const char *
 glXQueryServerString( Display *dpy, int screen, int name )
 {
    static char version[1000];
@@ -1700,7 +1700,7 @@ glXQueryServerString( Display *dpy, int screen, int name )
 
 
 /* GLX 1.1 and later */
-const char *
+PUBLIC const char *
 glXGetClientString( Display *dpy, int name )
 {
    static char version[1000];
@@ -1728,7 +1728,7 @@ glXGetClientString( Display *dpy, int name )
  */
 
 
-int
+PUBLIC int
 glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config,
                            int attribute, int *value )
 {
@@ -1743,7 +1743,7 @@ glXGetFBConfigAttrib( Display *dpy, GLXFBConfig config,
 }
 
 
-GLXFBConfig *
+PUBLIC GLXFBConfig *
 glXGetFBConfigs( Display *dpy, int screen, int *nelements )
 {
    XVisualInfo *visuals, visTemplate;
@@ -1769,7 +1769,7 @@ glXGetFBConfigs( Display *dpy, int screen, int *nelements )
 }
 
 
-GLXFBConfig *
+PUBLIC GLXFBConfig *
 glXChooseFBConfig( Display *dpy, int screen,
                         const int *attribList, int *nitems )
 {
@@ -1798,7 +1798,7 @@ glXChooseFBConfig( Display *dpy, int screen,
 }
 
 
-XVisualInfo *
+PUBLIC XVisualInfo *
 glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config )
 {
    if (dpy && config) {
@@ -1820,7 +1820,7 @@ glXGetVisualFromFBConfig( Display *dpy, GLXFBConfig config )
 }
 
 
-GLXWindow
+PUBLIC GLXWindow
 glXCreateWindow( Display *dpy, GLXFBConfig config, Window win,
                       const int *attribList )
 {
@@ -1840,7 +1840,7 @@ glXCreateWindow( Display *dpy, GLXFBConfig config, Window win,
 }
 
 
-void
+PUBLIC void
 glXDestroyWindow( Display *dpy, GLXWindow window )
 {
    XMesaBuffer b = XMesaFindBuffer(dpy, (Drawable) window);
@@ -1851,7 +1851,7 @@ glXDestroyWindow( Display *dpy, GLXWindow window )
 
 
 /* XXX untested */
-GLXPixmap
+PUBLIC GLXPixmap
 glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap,
                       const int *attribList )
 {
@@ -1961,7 +1961,7 @@ glXCreatePixmap( Display *dpy, GLXFBConfig config, Pixmap pixmap,
 }
 
 
-void
+PUBLIC void
 glXDestroyPixmap( Display *dpy, GLXPixmap pixmap )
 {
    XMesaBuffer b = XMesaFindBuffer(dpy, (Drawable)pixmap);
@@ -1971,7 +1971,7 @@ glXDestroyPixmap( Display *dpy, GLXPixmap pixmap )
 }
 
 
-GLXPbuffer
+PUBLIC GLXPbuffer
 glXCreatePbuffer( Display *dpy, GLXFBConfig config,
                        const int *attribList )
 {
@@ -2034,7 +2034,7 @@ glXCreatePbuffer( Display *dpy, GLXFBConfig config,
 }
 
 
-void
+PUBLIC void
 glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf )
 {
    XMesaBuffer b = XMesaFindBuffer(dpy, pbuf);
@@ -2044,7 +2044,7 @@ glXDestroyPbuffer( Display *dpy, GLXPbuffer pbuf )
 }
 
 
-void
+PUBLIC void
 glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute,
                        unsigned int *value )
 {
@@ -2090,7 +2090,7 @@ glXQueryDrawable( Display *dpy, GLXDrawable draw, int attribute,
 }
 
 
-GLXContext
+PUBLIC GLXContext
 glXCreateNewContext( Display *dpy, GLXFBConfig config,
                           int renderType, GLXContext shareList, Bool direct )
 {
@@ -2124,7 +2124,7 @@ glXCreateNewContext( Display *dpy, GLXFBConfig config,
 }
 
 
-int
+PUBLIC int
 glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value )
 {
    GLXContext glxCtx = ctx;
@@ -2153,7 +2153,7 @@ glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value )
 }
 
 
-void
+PUBLIC void
 glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask )
 {
    XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
@@ -2162,7 +2162,7 @@ glXSelectEvent( Display *dpy, GLXDrawable drawable, unsigned long mask )
 }
 
 
-void
+PUBLIC void
 glXGetSelectedEvent( Display *dpy, GLXDrawable drawable,
                           unsigned long *mask )
 {
@@ -2177,7 +2177,7 @@ glXGetSelectedEvent( Display *dpy, GLXDrawable drawable,
 
 /*** GLX_SGI_swap_control ***/
 
-int
+PUBLIC int
 glXSwapIntervalSGI(int interval)
 {
    (void) interval;
@@ -2190,7 +2190,7 @@ glXSwapIntervalSGI(int interval)
 
 static unsigned int FrameCounter = 0;
 
-int
+PUBLIC int
 glXGetVideoSyncSGI(unsigned int *count)
 {
    /* this is a bogus implementation */
@@ -2198,7 +2198,7 @@ glXGetVideoSyncSGI(unsigned int *count)
    return 0;
 }
 
-int
+PUBLIC int
 glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count)
 {
    if (divisor <= 0 || remainder < 0)
@@ -2215,7 +2215,7 @@ glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count)
 
 /*** GLX_SGI_make_current_read ***/
 
-Bool
+PUBLIC Bool
 glXMakeCurrentReadSGI(Display *dpy, GLXDrawable draw, GLXDrawable read, GLXContext ctx)
 {
    return glXMakeContextCurrent( dpy, draw, read, ctx );
@@ -2233,7 +2233,7 @@ glXGetCurrentReadDrawableSGI(void)
 /*** GLX_SGIX_video_source ***/
 #if defined(_VL_H)
 
-GLXVideoSourceSGIX
+PUBLIC GLXVideoSourceSGIX
 glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath path, int nodeClass, VLNode drainNode)
 {
    (void) dpy;
@@ -2245,7 +2245,7 @@ glXCreateGLXVideoSourceSGIX(Display *dpy, int screen, VLServer server, VLPath pa
    return 0;
 }
 
-void
+PUBLIC void
 glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src)
 {
    (void) dpy;
@@ -2257,21 +2257,21 @@ glXDestroyGLXVideoSourceSGIX(Display *dpy, GLXVideoSourceSGIX src)
 
 /*** GLX_EXT_import_context ***/
 
-void
+PUBLIC void
 glXFreeContextEXT(Display *dpy, GLXContext context)
 {
    (void) dpy;
    (void) context;
 }
 
-GLXContextID
+PUBLIC GLXContextID
 glXGetContextIDEXT(const GLXContext context)
 {
    (void) context;
    return 0;
 }
 
-GLXContext
+PUBLIC GLXContext
 glXImportContextEXT(Display *dpy, GLXContextID contextID)
 {
    (void) dpy;
@@ -2279,7 +2279,7 @@ glXImportContextEXT(Display *dpy, GLXContextID contextID)
    return 0;
 }
 
-int
+PUBLIC int
 glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int *value)
 {
    (void) dpy;
@@ -2293,20 +2293,20 @@ glXQueryContextInfoEXT(Display *dpy, GLXContext context, int attribute, int *val
 
 /*** GLX_SGIX_fbconfig ***/
 
-int
+PUBLIC int
 glXGetFBConfigAttribSGIX(Display *dpy, GLXFBConfigSGIX config, int attribute, int *value)
 {
    return glXGetFBConfigAttrib(dpy, config, attribute, value);
 }
 
-GLXFBConfigSGIX *
+PUBLIC GLXFBConfigSGIX *
 glXChooseFBConfigSGIX(Display *dpy, int screen, int *attrib_list, int *nelements)
 {
    return (GLXFBConfig *) glXChooseFBConfig(dpy, screen, attrib_list, nelements);
 }
 
 
-GLXPixmap
+PUBLIC GLXPixmap
 glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pixmap)
 {
    XMesaVisual xmvis = (XMesaVisual) config;
@@ -2315,7 +2315,7 @@ glXCreateGLXPixmapWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, Pixmap pi
 }
 
 
-GLXContext
+PUBLIC GLXContext
 glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_type, GLXContext share_list, Bool direct)
 {
    XMesaVisual xmvis = (XMesaVisual) config;
@@ -2344,14 +2344,14 @@ glXCreateContextWithConfigSGIX(Display *dpy, GLXFBConfigSGIX config, int render_
 }
 
 
-XVisualInfo *
+PUBLIC XVisualInfo *
 glXGetVisualFromFBConfigSGIX(Display *dpy, GLXFBConfigSGIX config)
 {
    return glXGetVisualFromFBConfig(dpy, config);
 }
 
 
-GLXFBConfigSGIX
+PUBLIC GLXFBConfigSGIX
 glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis)
 {
    XMesaVisual xmvis = find_glx_visual(dpy, vis);
@@ -2367,7 +2367,7 @@ glXGetFBConfigFromVisualSGIX(Display *dpy, XVisualInfo *vis)
 
 /*** GLX_SGIX_pbuffer ***/
 
-GLXPbufferSGIX
+PUBLIC GLXPbufferSGIX
 glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config,
                              unsigned int width, unsigned int height,
                              int *attribList)
@@ -2406,7 +2406,7 @@ glXCreateGLXPbufferSGIX(Display *dpy, GLXFBConfigSGIX config,
 }
 
 
-void
+PUBLIC void
 glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf)
 {
    XMesaBuffer xmbuf = XMesaFindBuffer(dpy, pbuf);
@@ -2416,7 +2416,7 @@ glXDestroyGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf)
 }
 
 
-int
+PUBLIC int
 glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigned int *value)
 {
    const XMesaBuffer xmbuf = XMesaFindBuffer(dpy, pbuf);
@@ -2449,7 +2449,7 @@ glXQueryGLXPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuf, int attribute, unsigne
 }
 
 
-void
+PUBLIC void
 glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask)
 {
    XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
@@ -2460,7 +2460,7 @@ glXSelectEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long mask)
 }
 
 
-void
+PUBLIC void
 glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask)
 {
    XMesaBuffer xmbuf = XMesaFindBuffer(dpy, drawable);
@@ -2476,7 +2476,7 @@ glXGetSelectedEventSGIX(Display *dpy, GLXDrawable drawable, unsigned long *mask)
 
 /*** GLX_SGI_cushion ***/
 
-void
+PUBLIC void
 glXCushionSGI(Display *dpy, Window win, float cushion)
 {
    (void) dpy;
@@ -2488,7 +2488,7 @@ glXCushionSGI(Display *dpy, Window win, float cushion)
 
 /*** GLX_SGIX_video_resize ***/
 
-int
+PUBLIC int
 glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window)
 {
    (void) dpy;
@@ -2498,7 +2498,7 @@ glXBindChannelToWindowSGIX(Display *dpy, int screen, int channel , Window window
    return 0;
 }
 
-int
+PUBLIC int
 glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, int h)
 {
    (void) dpy;
@@ -2511,7 +2511,7 @@ glXChannelRectSGIX(Display *dpy, int screen, int channel, int x, int y, int w, i
    return 0;
 }
 
-int
+PUBLIC int
 glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, int *w, int *h)
 {
    (void) dpy;
@@ -2524,7 +2524,7 @@ glXQueryChannelRectSGIX(Display *dpy, int screen, int channel, int *x, int *y, i
    return 0;
 }
 
-int
+PUBLIC int
 glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *dy, int *dw, int *dh)
 {
    (void) dpy;
@@ -2537,7 +2537,7 @@ glXQueryChannelDeltasSGIX(Display *dpy, int screen, int channel, int *dx, int *d
    return 0;
 }
 
-int
+PUBLIC int
 glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype)
 {
    (void) dpy;
@@ -2552,7 +2552,7 @@ glXChannelRectSyncSGIX(Display *dpy, int screen, int channel, GLenum synctype)
 /*** GLX_SGIX_dmbuffer **/
 
 #if defined(_DM_BUFFER_H_)
-Bool
+PUBLIC Bool
 glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params, DMbuffer dmbuffer)
 {
    (void) dpy;
@@ -2566,7 +2566,7 @@ glXAssociateDMPbufferSGIX(Display *dpy, GLXPbufferSGIX pbuffer, DMparams *params
 
 /*** GLX_SGIX_swap_group ***/
 
-void
+PUBLIC void
 glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member)
 {
    (void) dpy;
@@ -2578,7 +2578,7 @@ glXJoinSwapGroupSGIX(Display *dpy, GLXDrawable drawable, GLXDrawable member)
 
 /*** GLX_SGIX_swap_barrier ***/
 
-void
+PUBLIC void
 glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier)
 {
    (void) dpy;
@@ -2586,7 +2586,7 @@ glXBindSwapBarrierSGIX(Display *dpy, GLXDrawable drawable, int barrier)
    (void) barrier;
 }
 
-Bool
+PUBLIC Bool
 glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max)
 {
    (void) dpy;
@@ -2599,7 +2599,7 @@ glXQueryMaxSwapBarriersSGIX(Display *dpy, int screen, int *max)
 
 /*** GLX_SUN_get_transparent_index ***/
 
-Status
+PUBLIC Status
 glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *pTransparent)
 {
    (void) dpy;
@@ -2617,7 +2617,7 @@ glXGetTransparentIndexSUN(Display *dpy, Window overlay, Window underlay, long *p
  * Release the depth, stencil, accum buffers attached to a GLXDrawable
  * (a window or pixmap) prior to destroying the GLXDrawable.
  */
-Bool
+PUBLIC Bool
 glXReleaseBuffersMESA( Display *dpy, GLXDrawable d )
 {
    XMesaBuffer b = XMesaFindBuffer(dpy, d);
@@ -2630,7 +2630,7 @@ glXReleaseBuffersMESA( Display *dpy, GLXDrawable d )
 
 /*** GLX_EXT_texture_from_pixmap ***/
 
-void
+PUBLIC void
 glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer,
                         const int *attrib_list)
 {
@@ -2639,7 +2639,7 @@ glXBindTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer,
       XMesaBindTexImage(dpy, b, buffer, attrib_list);
 }
 
-void
+PUBLIC void
 glXReleaseTexImageEXT(Display *dpy, GLXDrawable drawable, int buffer)
 {
    XMesaBuffer b = XMesaFindBuffer(dpy, drawable);
diff --git a/src/gallium/state_trackers/glx/xlib/glx_getproc.c b/src/gallium/state_trackers/glx/xlib/glx_getproc.c
index ca7d88c9227..84d47b12edc 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_getproc.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_getproc.c
@@ -34,6 +34,7 @@
 #include <string.h>
 #include "GL/glx.h"
 #include "glapi/glapi.h"
+#include "pipe/p_compiler.h"
 
 
 struct name_address_pair {
@@ -208,6 +209,7 @@ glXGetProcAddressARB(const GLubyte *procName)
 
 
 /* GLX 1.4 */
+PUBLIC
 void (*glXGetProcAddress(const GLubyte *procName))()
 {
    return glXGetProcAddressARB(procName);
diff --git a/src/gallium/state_trackers/glx/xlib/glx_usefont.c b/src/gallium/state_trackers/glx/xlib/glx_usefont.c
index acc64df62b6..16e5ce642f3 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_usefont.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_usefont.c
@@ -33,6 +33,7 @@
 #include "main/context.h"
 #include "main/imports.h"
 #include <GL/glx.h>
+#include "pipe/p_compiler.h"
 
 
 /* Some debugging info.  */
@@ -210,7 +211,7 @@ isvalid(XFontStruct * fs, unsigned int which)
 }
 
 
-void
+PUBLIC void
 glXUseXFont(Font font, int first, int count, int listbase)
 {
    Display *dpy;
diff --git a/src/gallium/state_trackers/python/SConscript b/src/gallium/state_trackers/python/SConscript
index d4fdd43688f..8498a90812e 100644
--- a/src/gallium/state_trackers/python/SConscript
+++ b/src/gallium/state_trackers/python/SConscript
@@ -28,14 +28,27 @@ if 'python' in env['statetrackers']:
             'X11',
         ])
 
+    sources = [
+        'gallium.i',
+        'st_device.c',
+        'st_sample.c',
+    ]
+
+    drivers = [
+        trace
+    ]
+
+    if 'llvmpipe' in env['drivers']:
+        env.Tool('llvm')
+        sources += ['st_llvmpipe_winsys.c']
+        drivers += [llvmpipe]
+    else:
+        sources += ['st_softpipe_winsys.c']
+        drivers += [softpipe]
+
     pyst = env.ConvenienceLibrary(
         target = 'pyst',
-        source = [
-            'gallium.i',
-            'st_device.c',
-            'st_sample.c',
-            'st_softpipe_winsys.c',
-        ],
+        source = sources,
     )
 
     env['no_import_lib'] = 1
@@ -45,5 +58,5 @@ if 'python' in env['statetrackers']:
         source = [
             'st_hardpipe_winsys.c',
         ],
-        LIBS = [pyst, softpipe, trace] + gallium + env['LIBS'],
+        LIBS = [pyst] + drivers + gallium + env['LIBS'],
     )
diff --git a/src/gallium/state_trackers/python/gallium.i b/src/gallium/state_trackers/python/gallium.i
index 96b13c2258e..68797228459 100644
--- a/src/gallium/state_trackers/python/gallium.i
+++ b/src/gallium/state_trackers/python/gallium.i
@@ -76,7 +76,6 @@
 %rename(BlendColor) pipe_blend_color;
 %rename(Blend) pipe_blend_state;
 %rename(Clip) pipe_clip_state;
-%rename(ConstantBuffer) pipe_constant_buffer;
 %rename(Depth) pipe_depth_state;
 %rename(Stencil) pipe_stencil_state;
 %rename(Alpha) pipe_alpha_state;
diff --git a/src/gallium/state_trackers/python/p_context.i b/src/gallium/state_trackers/python/p_context.i
index 84ce1a41e6d..ce893dad453 100644
--- a/src/gallium/state_trackers/python/p_context.i
+++ b/src/gallium/state_trackers/python/p_context.i
@@ -142,10 +142,7 @@ struct st_context {
    void set_constant_buffer(unsigned shader, unsigned index,
                             struct pipe_buffer *buffer ) 
    {
-      struct pipe_constant_buffer state;
-      memset(&state, 0, sizeof(state));
-      state.buffer = buffer;
-      $self->pipe->set_constant_buffer($self->pipe, shader, index, &state);
+      $self->pipe->set_constant_buffer($self->pipe, shader, index, buffer);
    }
 
    void set_framebuffer(const struct pipe_framebuffer_state *state ) 
diff --git a/src/gallium/state_trackers/python/p_device.i b/src/gallium/state_trackers/python/p_device.i
index 2dc995adb07..0eba488a078 100644
--- a/src/gallium/state_trackers/python/p_device.i
+++ b/src/gallium/state_trackers/python/p_device.i
@@ -87,6 +87,10 @@ struct st_device {
                             enum pipe_texture_target target,
                             unsigned tex_usage, 
                             unsigned geom_flags ) {
+      /* We can't really display surfaces with the python statetracker so mask
+       * out that usage */
+      tex_usage &= ~PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+
       return $self->screen->is_format_supported( $self->screen, 
                                                  format, 
                                                  target, 
@@ -110,6 +114,11 @@ struct st_device {
          unsigned tex_usage = 0
       ) {
       struct pipe_texture templat;
+
+      /* We can't really display surfaces with the python statetracker so mask
+       * out that usage */
+      tex_usage &= ~PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+
       memset(&templat, 0, sizeof(templat));
       templat.format = format;
       templat.width0 = width;
@@ -118,6 +127,7 @@ struct st_device {
       templat.last_level = last_level;
       templat.target = target;
       templat.tex_usage = tex_usage;
+
       return $self->screen->texture_create($self->screen, &templat);
    }
    
diff --git a/src/gallium/state_trackers/python/retrace/interpreter.py b/src/gallium/state_trackers/python/retrace/interpreter.py
index a68709f5cf3..b61d47d6456 100755
--- a/src/gallium/state_trackers/python/retrace/interpreter.py
+++ b/src/gallium/state_trackers/python/retrace/interpreter.py
@@ -94,7 +94,7 @@ struct_factories = {
     "pipe_blend_color": gallium.BlendColor,
     "pipe_blend_state": gallium.Blend,
     #"pipe_clip_state": gallium.Clip,
-    #"pipe_constant_buffer": gallium.ConstantBuffer,
+    #"pipe_buffer": gallium.Buffer,
     "pipe_depth_state": gallium.Depth,
     "pipe_stencil_state": gallium.Stencil,
     "pipe_alpha_state": gallium.Alpha,
@@ -462,10 +462,10 @@ class Context(Object):
         sys.stdout.flush()
 
     def set_constant_buffer(self, shader, index, buffer):
-        if buffer is not None and buffer.buffer is not None:
-            self.real.set_constant_buffer(shader, index, buffer.buffer)
+        if buffer is not None:
+            self.real.set_constant_buffer(shader, index, buffer)
 
-            self.dump_constant_buffer(buffer.buffer)
+            self.dump_constant_buffer(buffer)
 
     def set_framebuffer_state(self, state):
         _state = gallium.Framebuffer()
diff --git a/src/gallium/state_trackers/python/samples/gs.py b/src/gallium/state_trackers/python/samples/gs.py
index 1ceead5f17c..a07cf557f2f 100644
--- a/src/gallium/state_trackers/python/samples/gs.py
+++ b/src/gallium/state_trackers/python/samples/gs.py
@@ -136,10 +136,10 @@ def test(dev):
     cbuf = dev.texture_create(
         PIPE_FORMAT_X8R8G8B8_UNORM,
         width, height,
-        tex_usage=PIPE_TEXTURE_USAGE_DISPLAY_TARGET,
+        tex_usage=PIPE_TEXTURE_USAGE_RENDER_TARGET,
     ).get_surface()
     zbuf = dev.texture_create(
-        PIPE_FORMAT_Z16_UNORM,
+        PIPE_FORMAT_Z32_UNORM,
         width, height,
         tex_usage=PIPE_TEXTURE_USAGE_DEPTH_STENCIL,
     ).get_surface()
diff --git a/src/gallium/state_trackers/python/samples/tri.py b/src/gallium/state_trackers/python/samples/tri.py
index af80426dc6c..e5e168bdc8d 100644
--- a/src/gallium/state_trackers/python/samples/tri.py
+++ b/src/gallium/state_trackers/python/samples/tri.py
@@ -136,10 +136,10 @@ def test(dev):
     cbuf = dev.texture_create(
         PIPE_FORMAT_X8R8G8B8_UNORM, 
         width, height,
-        tex_usage=PIPE_TEXTURE_USAGE_DISPLAY_TARGET,
+        tex_usage=PIPE_TEXTURE_USAGE_RENDER_TARGET,
     ).get_surface()
     zbuf = dev.texture_create(
-        PIPE_FORMAT_Z16_UNORM, 
+        PIPE_FORMAT_Z32_UNORM,
         width, height,
         tex_usage=PIPE_TEXTURE_USAGE_DEPTH_STENCIL,
     ).get_surface()
diff --git a/src/gallium/state_trackers/python/st_llvmpipe_winsys.c b/src/gallium/state_trackers/python/st_llvmpipe_winsys.c
new file mode 100644
index 00000000000..0096b18c994
--- /dev/null
+++ b/src/gallium/state_trackers/python/st_llvmpipe_winsys.c
@@ -0,0 +1,148 @@
+/**************************************************************************
+ * 
+ * Copyright 2010 VMware, 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 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.
+ * 
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Llvmpipe support. 
+ * 
+ * @author Jose Fonseca
+ */
+
+
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+#include "pipe/p_inlines.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "llvmpipe/lp_winsys.h"
+#include "st_winsys.h"
+
+
+static boolean
+llvmpipe_ws_is_displaytarget_format_supported( struct llvmpipe_winsys *ws,
+      enum pipe_format format )
+{
+   return FALSE;
+}
+
+
+static void *
+llvmpipe_ws_displaytarget_map(struct llvmpipe_winsys *ws,
+                              struct llvmpipe_displaytarget *dt,
+                              unsigned flags )
+{
+   assert(0);
+   return NULL;
+}
+
+
+static void
+llvmpipe_ws_displaytarget_unmap(struct llvmpipe_winsys *ws,
+                                struct llvmpipe_displaytarget *dt )
+{
+   assert(0);
+}
+
+
+static void
+llvmpipe_ws_displaytarget_destroy(struct llvmpipe_winsys *winsys,
+                                  struct llvmpipe_displaytarget *dt)
+{
+   assert(0);
+}
+
+
+static struct llvmpipe_displaytarget *
+llvmpipe_ws_displaytarget_create(struct llvmpipe_winsys *winsys,
+                                 enum pipe_format format,
+                                 unsigned width, unsigned height,
+                                 unsigned alignment,
+                                 unsigned *stride)
+{
+   return NULL;
+}
+
+
+static void
+llvmpipe_ws_displaytarget_display(struct llvmpipe_winsys *winsys,
+                                  struct llvmpipe_displaytarget *dt,
+                                  void *context_private)
+{
+   assert(0);
+}
+
+
+static void
+llvmpipe_ws_destroy(struct llvmpipe_winsys *winsys)
+{
+   FREE(winsys);
+}
+
+
+static struct pipe_screen *
+st_llvmpipe_screen_create(void)
+{
+   static struct llvmpipe_winsys *winsys;
+   struct pipe_screen *screen;
+
+   winsys = CALLOC_STRUCT(llvmpipe_winsys);
+   if (!winsys)
+      goto no_winsys;
+
+   winsys->destroy = llvmpipe_ws_destroy;
+   winsys->is_displaytarget_format_supported = llvmpipe_ws_is_displaytarget_format_supported;
+   winsys->displaytarget_create = llvmpipe_ws_displaytarget_create;
+   winsys->displaytarget_map = llvmpipe_ws_displaytarget_map;
+   winsys->displaytarget_unmap = llvmpipe_ws_displaytarget_unmap;
+   winsys->displaytarget_display = llvmpipe_ws_displaytarget_display;
+   winsys->displaytarget_destroy = llvmpipe_ws_displaytarget_destroy;
+
+   screen = llvmpipe_create_screen(winsys);
+   if (!screen)
+      goto no_screen;
+
+   return screen;
+
+no_screen:
+   FREE(winsys);
+no_winsys:
+   return NULL;
+}
+
+
+static struct pipe_context *
+st_llvmpipe_context_create(struct pipe_screen *screen)
+{
+   return llvmpipe_create(screen);
+}
+
+
+const struct st_winsys st_softpipe_winsys = {
+   &st_llvmpipe_screen_create,
+   &st_llvmpipe_context_create,
+};
diff --git a/src/gallium/state_trackers/python/tests/regress/fragment-shader/.gitignore b/src/gallium/state_trackers/python/tests/regress/fragment-shader/.gitignore
new file mode 100644
index 00000000000..e33609d251c
--- /dev/null
+++ b/src/gallium/state_trackers/python/tests/regress/fragment-shader/.gitignore
@@ -0,0 +1 @@
+*.png
diff --git a/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py b/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py
index eed6cdd1e64..8d3bf9d4d7e 100644
--- a/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py
+++ b/src/gallium/state_trackers/python/tests/regress/fragment-shader/fragment-shader.py
@@ -114,7 +114,7 @@ def test(dev, name):
     cbuf = dev.texture_create(
         PIPE_FORMAT_X8R8G8B8_UNORM,
         width, height,
-        tex_usage=PIPE_TEXTURE_USAGE_DISPLAY_TARGET,
+        tex_usage=PIPE_TEXTURE_USAGE_RENDER_TARGET,
     ).get_surface()
     fb = Framebuffer()
     fb.width = width
diff --git a/src/gallium/state_trackers/python/tests/regress/vertex-shader/.gitignore b/src/gallium/state_trackers/python/tests/regress/vertex-shader/.gitignore
new file mode 100644
index 00000000000..e33609d251c
--- /dev/null
+++ b/src/gallium/state_trackers/python/tests/regress/vertex-shader/.gitignore
@@ -0,0 +1 @@
+*.png
diff --git a/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py
index 41bebd0604a..01bf5a3210d 100644
--- a/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py
+++ b/src/gallium/state_trackers/python/tests/regress/vertex-shader/vertex-shader.py
@@ -114,7 +114,7 @@ def test(dev, name):
     cbuf = dev.texture_create(
         PIPE_FORMAT_X8R8G8B8_UNORM,
         width, height,
-        tex_usage=PIPE_TEXTURE_USAGE_DISPLAY_TARGET,
+        tex_usage=PIPE_TEXTURE_USAGE_RENDER_TARGET,
     ).get_surface()
     fb = Framebuffer()
     fb.width = width
diff --git a/src/gallium/state_trackers/vega/api_filters.c b/src/gallium/state_trackers/vega/api_filters.c
index 2f984fb7b9a..8f69ee01099 100644
--- a/src/gallium/state_trackers/vega/api_filters.c
+++ b/src/gallium/state_trackers/vega/api_filters.c
@@ -147,22 +147,22 @@ static void setup_constant_buffer(struct vg_context *ctx, const void *buffer,
                                   VGint param_bytes)
 {
    struct pipe_context *pipe = ctx->pipe;
-   struct pipe_constant_buffer *cbuf = &ctx->filter.buffer;
+   struct pipe_buffer **cbuf = &ctx->filter.buffer;
 
    /* We always need to get a new buffer, to keep the drivers simple and
     * avoid gratuitous rendering synchronization. */
-   pipe_buffer_reference(&cbuf->buffer, NULL);
+   pipe_buffer_reference(cbuf, NULL);
 
-   cbuf->buffer = pipe_buffer_create(pipe->screen, 16,
-                                     PIPE_BUFFER_USAGE_CONSTANT,
-                                     param_bytes);
+   *cbuf = pipe_buffer_create(pipe->screen, 16,
+                              PIPE_BUFFER_USAGE_CONSTANT,
+                              param_bytes);
 
-   if (cbuf->buffer) {
-      st_no_flush_pipe_buffer_write(ctx, cbuf->buffer,
+   if (*cbuf) {
+      st_no_flush_pipe_buffer_write(ctx, *cbuf,
                                     0, param_bytes, buffer);
    }
 
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf);
+   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
 }
 
 static void setup_samplers(struct vg_context *ctx, struct filter_info *info)
diff --git a/src/gallium/state_trackers/vega/mask.c b/src/gallium/state_trackers/vega/mask.c
index 42300bb6d57..3e260e70736 100644
--- a/src/gallium/state_trackers/vega/mask.c
+++ b/src/gallium/state_trackers/vega/mask.c
@@ -217,7 +217,7 @@ static void setup_mask_framebuffer(struct pipe_surface *surf,
 static void setup_mask_operation(VGMaskOperation operation)
 {
    struct vg_context *ctx = vg_current_context();
-   struct pipe_constant_buffer *cbuf = &ctx->mask.cbuf;
+   struct pipe_buffer **cbuf = &ctx->mask.cbuf;
    const VGint param_bytes = 4 * sizeof(VGfloat);
    const VGfloat ones[4] = {1.f, 1.f, 1.f, 1.f};
    void *shader = 0;
@@ -225,17 +225,17 @@ static void setup_mask_operation(VGMaskOperation operation)
    /* We always need to get a new buffer, to keep the drivers simple and
     * avoid gratuitous rendering synchronization.
     */
-   pipe_buffer_reference(&cbuf->buffer, NULL);
+   pipe_buffer_reference(cbuf, NULL);
 
-   cbuf->buffer = pipe_buffer_create(ctx->pipe->screen, 1,
-                                     PIPE_BUFFER_USAGE_CONSTANT,
-                                     param_bytes);
-   if (cbuf->buffer) {
-      st_no_flush_pipe_buffer_write(ctx, cbuf->buffer,
+   *cbuf = pipe_buffer_create(ctx->pipe->screen, 1,
+                              PIPE_BUFFER_USAGE_CONSTANT,
+                              param_bytes);
+   if (*cbuf) {
+      st_no_flush_pipe_buffer_write(ctx, *cbuf,
                                     0, param_bytes, ones);
    }
 
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf);
+   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
    switch (operation) {
    case VG_UNION_MASK: {
       if (!ctx->mask.union_fs) {
@@ -320,22 +320,22 @@ static void setup_mask_samplers(struct pipe_texture *umask)
 static void setup_mask_fill(const VGfloat color[4])
 {
    struct vg_context *ctx = vg_current_context();
-   struct pipe_constant_buffer *cbuf = &ctx->mask.cbuf;
+   struct pipe_buffer **cbuf = &ctx->mask.cbuf;
    const VGint param_bytes = 4 * sizeof(VGfloat);
 
    /* We always need to get a new buffer, to keep the drivers simple and
     * avoid gratuitous rendering synchronization.
     */
-   pipe_buffer_reference(&cbuf->buffer, NULL);
+   pipe_buffer_reference(cbuf, NULL);
 
-   cbuf->buffer = pipe_buffer_create(ctx->pipe->screen, 1,
-                                     PIPE_BUFFER_USAGE_CONSTANT,
-                                     param_bytes);
-   if (cbuf->buffer) {
-      st_no_flush_pipe_buffer_write(ctx, cbuf->buffer, 0, param_bytes, color);
+   *cbuf = pipe_buffer_create(ctx->pipe->screen, 1,
+                              PIPE_BUFFER_USAGE_CONSTANT,
+                              param_bytes);
+   if (*cbuf) {
+      st_no_flush_pipe_buffer_write(ctx, *cbuf, 0, param_bytes, color);
    }
 
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf);
+   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
    cso_set_fragment_shader_handle(ctx->cso_context,
                                   shaders_cache_fill(ctx->sc,
                                                      VEGA_SOLID_FILL_SHADER));
diff --git a/src/gallium/state_trackers/vega/paint.c b/src/gallium/state_trackers/vega/paint.c
index cc73771d358..d8f6299b2db 100644
--- a/src/gallium/state_trackers/vega/paint.c
+++ b/src/gallium/state_trackers/vega/paint.c
@@ -77,7 +77,8 @@ struct vg_paint {
       struct pipe_sampler_state sampler;
    } pattern;
 
-   struct pipe_constant_buffer cbuf;
+   /* XXX next 3 all unneded? */
+   struct pipe_buffer *cbuf;
    struct pipe_shader_state fs_state;
    void *fs;
 };
diff --git a/src/gallium/state_trackers/vega/polygon.c b/src/gallium/state_trackers/vega/polygon.c
index b6d282d803b..d385ee567fa 100644
--- a/src/gallium/state_trackers/vega/polygon.c
+++ b/src/gallium/state_trackers/vega/polygon.c
@@ -293,6 +293,7 @@ static void draw_polygon(struct vg_context *ctx,
 
    /* tell pipe about the vertex attributes */
    velement.src_offset = 0;
+   velement.instance_divisor = 0;
    velement.vertex_buffer_index = 0;
    velement.src_format = PIPE_FORMAT_R32G32_FLOAT;
    velement.nr_components = COMPONENTS;
diff --git a/src/gallium/state_trackers/vega/shader.c b/src/gallium/state_trackers/vega/shader.c
index d9074a377b3..bd5ae79e551 100644
--- a/src/gallium/state_trackers/vega/shader.c
+++ b/src/gallium/state_trackers/vega/shader.c
@@ -51,7 +51,7 @@ struct shader {
    VGImageMode image_mode;
 
    float constants[MAX_CONSTANTS];
-   struct pipe_constant_buffer cbuf;
+   struct pipe_buffer *cbuf;
    struct pipe_shader_state fs_state;
    void *fs;
 };
@@ -96,25 +96,25 @@ static void setup_constant_buffer(struct shader *shader)
 {
    struct vg_context *ctx = shader->context;
    struct pipe_context *pipe = shader->context->pipe;
-   struct pipe_constant_buffer *cbuf = &shader->cbuf;
+   struct pipe_buffer **cbuf = &shader->cbuf;
    VGint param_bytes = paint_constant_buffer_size(shader->paint);
    float temp_buf[MAX_CONSTANTS];
 
    assert(param_bytes <= sizeof(temp_buf));
    paint_fill_constant_buffer(shader->paint, temp_buf);
 
-   if (cbuf->buffer == NULL ||
+   if (*cbuf == NULL ||
        memcmp(temp_buf, shader->constants, param_bytes) != 0)
    {
-      pipe_buffer_reference(&cbuf->buffer, NULL);
+      pipe_buffer_reference(cbuf, NULL);
 
       memcpy(shader->constants, temp_buf, param_bytes);
-      cbuf->buffer = pipe_user_buffer_create(pipe->screen,
-                                             &shader->constants,
-                                             sizeof(shader->constants));
+      *cbuf = pipe_user_buffer_create(pipe->screen,
+                                      &shader->constants,
+                                      sizeof(shader->constants));
    }
 
-   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf);
+   ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_FRAGMENT, 0, *cbuf);
 }
 
 static VGint blend_bind_samplers(struct vg_context *ctx,
diff --git a/src/gallium/state_trackers/vega/vg_context.c b/src/gallium/state_trackers/vega/vg_context.c
index 00d23f5c227..c16ac036e3b 100644
--- a/src/gallium/state_trackers/vega/vg_context.c
+++ b/src/gallium/state_trackers/vega/vg_context.c
@@ -122,8 +122,8 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
 
 void vg_destroy_context(struct vg_context *ctx)
 {
-   struct pipe_constant_buffer *cbuf = &ctx->mask.cbuf;
-   struct pipe_constant_buffer *vsbuf = &ctx->vs_const_buffer;
+   struct pipe_buffer **cbuf = &ctx->mask.cbuf;
+   struct pipe_buffer **vsbuf = &ctx->vs_const_buffer;
 
    util_destroy_blit(ctx->blit);
    renderer_destroy(ctx->renderer);
@@ -131,11 +131,11 @@ void vg_destroy_context(struct vg_context *ctx)
    shader_destroy(ctx->shader);
    paint_destroy(ctx->default_paint);
 
-   if (cbuf && cbuf->buffer)
-      pipe_buffer_reference(&cbuf->buffer, NULL);
+   if (*cbuf)
+      pipe_buffer_reference(cbuf, NULL);
 
-   if (vsbuf && vsbuf->buffer)
-      pipe_buffer_reference(&vsbuf->buffer, NULL);
+   if (*vsbuf)
+      pipe_buffer_reference(vsbuf, NULL);
 
    if (ctx->clear.fs) {
       cso_delete_fragment_shader(ctx->cso_context, ctx->clear.fs);
@@ -371,20 +371,20 @@ void vg_validate_state(struct vg_context *ctx)
          2.f/fb->width, 2.f/fb->height, 1, 1,
          -1, -1, 0, 0
       };
-      struct pipe_constant_buffer *cbuf = &ctx->vs_const_buffer;
+      struct pipe_buffer **cbuf = &ctx->vs_const_buffer;
 
       vg_set_viewport(ctx, VEGA_Y0_BOTTOM);
 
-      pipe_buffer_reference(&cbuf->buffer, NULL);
-      cbuf->buffer = pipe_buffer_create(ctx->pipe->screen, 16,
+      pipe_buffer_reference(cbuf, NULL);
+      *cbuf = pipe_buffer_create(ctx->pipe->screen, 16,
                                         PIPE_BUFFER_USAGE_CONSTANT,
                                         param_bytes);
 
-      if (cbuf->buffer) {
-         st_no_flush_pipe_buffer_write(ctx, cbuf->buffer,
+      if (*cbuf) {
+         st_no_flush_pipe_buffer_write(ctx, *cbuf,
                                        0, param_bytes, vs_consts);
       }
-      ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_VERTEX, 0, cbuf);
+      ctx->pipe->set_constant_buffer(ctx->pipe, PIPE_SHADER_VERTEX, 0, *cbuf);
    }
    if ((ctx->state.dirty & VS_DIRTY)) {
       cso_set_vertex_shader_handle(ctx->cso_context,
diff --git a/src/gallium/state_trackers/vega/vg_context.h b/src/gallium/state_trackers/vega/vg_context.h
index ccc8889c8c5..ff8c1a04217 100644
--- a/src/gallium/state_trackers/vega/vg_context.h
+++ b/src/gallium/state_trackers/vega/vg_context.h
@@ -113,7 +113,7 @@ struct vg_context
    } clear;
 
    struct {
-      struct pipe_constant_buffer cbuf;
+      struct pipe_buffer *cbuf;
       struct pipe_sampler_state sampler;
 
       struct vg_shader *union_fs;
@@ -135,7 +135,7 @@ struct vg_context
 
    struct pipe_sampler_state blend_sampler;
    struct {
-      struct pipe_constant_buffer buffer;
+      struct pipe_buffer *buffer;
       void *color_matrix_fs;
    } filter;
    struct vg_paint *default_paint;
@@ -145,7 +145,7 @@ struct vg_context
    struct vg_shader *plain_vs;
    struct vg_shader *clear_vs;
    struct vg_shader *texture_vs;
-   struct pipe_constant_buffer vs_const_buffer;
+   struct pipe_buffer *vs_const_buffer;
 };
 
 struct vg_object {
diff --git a/src/gallium/state_trackers/vega/vg_tracker.c b/src/gallium/state_trackers/vega/vg_tracker.c
index e5039132758..ff80aab03a3 100644
--- a/src/gallium/state_trackers/vega/vg_tracker.c
+++ b/src/gallium/state_trackers/vega/vg_tracker.c
@@ -36,6 +36,9 @@
 #include "util/u_math.h"
 #include "util/u_rect.h"
 
+/* advertise OpenVG support */
+PUBLIC const int st_api_OpenVG = 1;
+
 static struct pipe_texture *
 create_texture(struct pipe_context *pipe, enum pipe_format format,
                VGint width, VGint height)
@@ -368,14 +371,15 @@ void st_unreference_framebuffer(struct st_framebuffer *stfb)
    /* FIXME */
 }
 
-void st_make_current(struct vg_context *st,
-                     struct st_framebuffer *draw,
-                     struct st_framebuffer *read)
+boolean st_make_current(struct vg_context *st,
+                        struct st_framebuffer *draw,
+                        struct st_framebuffer *read)
 {
    vg_set_current_context(st);
    if (st) {
       st->draw_buffer = draw;
    }
+   return VG_TRUE;
 }
 
 struct vg_context *st_get_current(void)
@@ -425,3 +429,8 @@ int st_unbind_texture_surface(struct pipe_surface *ps, int target, int level)
 {
    return 0;
 }
+
+st_proc st_get_proc_address(const char *procname)
+{
+   return NULL;
+}
diff --git a/src/gallium/state_trackers/vega/vg_tracker.h b/src/gallium/state_trackers/vega/vg_tracker.h
index 5457631106f..c1196954a76 100644
--- a/src/gallium/state_trackers/vega/vg_tracker.h
+++ b/src/gallium/state_trackers/vega/vg_tracker.h
@@ -45,15 +45,19 @@ struct pipe_fence_handle;
 struct pipe_surface;
 
 
+PUBLIC
 struct vg_context *st_create_context(struct pipe_context *pipe,
                                      const void *visual,
                                      struct vg_context *share);
 
+PUBLIC
 void st_destroy_context( struct vg_context *st );
 
+PUBLIC
 void st_copy_context_state(struct vg_context *dst, struct vg_context *src,
                            uint mask);
 
+PUBLIC
 struct st_framebuffer *st_create_framebuffer(const void *visual,
                                              enum pipe_format colorFormat,
                                              enum pipe_format depthFormat,
@@ -61,47 +65,63 @@ struct st_framebuffer *st_create_framebuffer(const void *visual,
                                              uint width, uint height,
                                              void *privateData);
 
+PUBLIC
 void st_resize_framebuffer(struct st_framebuffer *stfb,
                            uint width, uint height);
 
+PUBLIC
 void st_set_framebuffer_surface(struct st_framebuffer *stfb,
                                 uint surfIndex, struct pipe_surface *surf);
 
+PUBLIC
 void st_get_framebuffer_dimensions( struct st_framebuffer *stfb,
 				    uint *width, uint *height);
 
+PUBLIC
 int st_bind_texture_surface(struct pipe_surface *ps, int target, int level,
                             enum pipe_format format);
 
+PUBLIC
 int st_unbind_texture_surface(struct pipe_surface *ps, int target, int level);
 
+PUBLIC
 int st_get_framebuffer_surface(struct st_framebuffer *stfb,
                                uint surfIndex, struct pipe_surface **surf);
 
+PUBLIC
 int st_get_framebuffer_texture(struct st_framebuffer *stfb,
                                uint surfIndex, struct pipe_texture **tex);
 
+PUBLIC
 void *st_framebuffer_private(struct st_framebuffer *stfb);
 
+PUBLIC
 void st_unreference_framebuffer(struct st_framebuffer *stfb);
 
-void st_make_current(struct vg_context *st,
-                     struct st_framebuffer *draw,
-                     struct st_framebuffer *read);
+PUBLIC
+boolean st_make_current(struct vg_context *st,
+                        struct st_framebuffer *draw,
+                        struct st_framebuffer *read);
 
+PUBLIC
 struct vg_context *st_get_current(void);
 
+PUBLIC
 void st_flush(struct vg_context *st, uint pipeFlushFlags,
                struct pipe_fence_handle **fence);
+PUBLIC
 void st_finish(struct vg_context *st);
 
+PUBLIC
 void st_notify_swapbuffers(struct st_framebuffer *stfb);
+PUBLIC
 void st_notify_swapbuffers_complete(struct st_framebuffer *stfb);
 
 
 /** Generic function type */
 typedef void (*st_proc)();
 
+PUBLIC
 st_proc st_get_proc_address(const char *procname);
 
 #endif
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.c b/src/gallium/state_trackers/xorg/xorg_renderer.c
index d80f341e6c2..8f729b565b1 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.c
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.c
@@ -379,14 +379,14 @@ struct xorg_renderer * renderer_create(struct pipe_context *pipe)
 
 void renderer_destroy(struct xorg_renderer *r)
 {
-   struct pipe_constant_buffer *vsbuf = &r->vs_const_buffer;
-   struct pipe_constant_buffer *fsbuf = &r->fs_const_buffer;
+   struct pipe_buffer **vsbuf = &r->vs_const_buffer;
+   struct pipe_buffer **fsbuf = &r->fs_const_buffer;
 
-   if (vsbuf && vsbuf->buffer)
-      pipe_buffer_reference(&vsbuf->buffer, NULL);
+   if (*vsbuf)
+      pipe_buffer_reference(vsbuf, NULL);
 
-   if (fsbuf && fsbuf->buffer)
-      pipe_buffer_reference(&fsbuf->buffer, NULL);
+   if (*fsbuf)
+      pipe_buffer_reference(fsbuf, NULL);
 
    if (r->shaders) {
       xorg_shaders_destroy(r->shaders);
@@ -409,20 +409,20 @@ void renderer_set_constants(struct xorg_renderer *r,
                             const float *params,
                             int param_bytes)
 {
-   struct pipe_constant_buffer *cbuf =
+   struct pipe_buffer **cbuf =
       (shader_type == PIPE_SHADER_VERTEX) ? &r->vs_const_buffer :
       &r->fs_const_buffer;
 
-   pipe_buffer_reference(&cbuf->buffer, NULL);
-   cbuf->buffer = pipe_buffer_create(r->pipe->screen, 16,
-                                     PIPE_BUFFER_USAGE_CONSTANT,
-                                     param_bytes);
+   pipe_buffer_reference(cbuf, NULL);
+   *cbuf = pipe_buffer_create(r->pipe->screen, 16,
+                              PIPE_BUFFER_USAGE_CONSTANT,
+                              param_bytes);
 
-   if (cbuf->buffer) {
-      pipe_buffer_write(r->pipe->screen, cbuf->buffer,
+   if (*cbuf) {
+      pipe_buffer_write(r->pipe->screen, *cbuf,
                         0, param_bytes, params);
    }
-   r->pipe->set_constant_buffer(r->pipe, shader_type, 0, cbuf);
+   r->pipe->set_constant_buffer(r->pipe, shader_type, 0, *cbuf);
 }
 
 
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.h b/src/gallium/state_trackers/xorg/xorg_renderer.h
index 5272cde2b3f..af6aa0567d6 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.h
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.h
@@ -23,8 +23,8 @@ struct xorg_renderer {
 
    int fb_width;
    int fb_height;
-   struct pipe_constant_buffer vs_const_buffer;
-   struct pipe_constant_buffer fs_const_buffer;
+   struct pipe_buffer *vs_const_buffer;
+   struct pipe_buffer *fs_const_buffer;
 
    float buffer[BUF_SIZE];
    int buffer_size;
diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c
index 6b5a41a3727..5bf0e94b627 100644
--- a/src/gallium/state_trackers/xorg/xorg_xv.c
+++ b/src/gallium/state_trackers/xorg/xorg_xv.c
@@ -486,8 +486,11 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
    int dxo, dyo;
    Bool hdtv;
    int x, y, w, h;
-   struct exa_pixmap_priv *dst = exaGetPixmapDriverPrivate(pPixmap);
-   struct pipe_surface *dst_surf = xorg_gpu_surface(pPriv->r->pipe->screen, dst);
+   struct exa_pixmap_priv *dst;
+   struct pipe_surface *dst_surf = NULL;
+
+   exaMoveInPixmap(pPixmap);
+   dst = exaGetPixmapDriverPrivate(pPixmap);
 
    if (dst && !dst->tex) {
 	xorg_exa_set_shared_usage(pPixmap);
@@ -497,6 +500,7 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
    if (!dst || !dst->tex)
       XORG_FALLBACK("Xv destination %s", !dst ? "!dst" : "!dst->tex");
 
+   dst_surf = xorg_gpu_surface(pPriv->r->pipe->screen, dst);
    hdtv = ((src_w >= RES_720P_X) && (src_h >= RES_720P_Y));
 
    REGION_TRANSLATE(pScrn->pScreen, dstRegion, -pPixmap->screen_x,
@@ -516,7 +520,6 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
    bind_samplers(pPriv);
    setup_fs_video_constants(pPriv->r, hdtv);
 
-   exaMoveInPixmap(pPixmap);
    DamageDamageRegion(&pPixmap->drawable, dstRegion);
 
    while (nbox--) {
diff --git a/src/gallium/winsys/drm/Makefile.egl_g3d b/src/gallium/winsys/drm/Makefile.egl_g3d
new file mode 100644
index 00000000000..3ce27258529
--- /dev/null
+++ b/src/gallium/winsys/drm/Makefile.egl_g3d
@@ -0,0 +1,64 @@
+# src/gallium/winsys/drm/Makefile.egl_g3d
+
+# The driver Makefile should define
+#
+#   EGL_DRIVER_NAME, the name of the driver
+#   EGL_DRIVER_SOURCES, the sources of the driver
+#   EGL_DRIVER_LIBS, extra libraries needed by the driver
+#   EGL_DRIVER_PIPES, the pipe drivers of the driver
+#
+# before including this file.
+
+EGL_DRIVER_OBJECTS = $(EGL_DRIVER_SOURCES:.c=.o)
+
+common_LIBS = -ldrm -lm -ldl
+
+x11_ST = $(TOP)/src/gallium/state_trackers/egl_g3d/libeglx11.a
+x11_LIBS = $(common_LIBS) -lX11 -lXext -lXfixes
+
+kms_ST = $(TOP)/src/gallium/state_trackers/egl_g3d/libeglkms.a
+kms_LIBS = $(common_LIBS)
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
+
+
+##### TARGETS #####
+
+EGL_DISPLAY_DRIVERS = $(foreach dpy, $(EGL_DISPLAYS), egl_$(dpy)_$(EGL_DRIVER_NAME).so)
+
+LIB_GALLIUM_DIR = $(TOP)/$(LIB_DIR)/gallium
+EGL_DISPLAY_LIBS = $(foreach drv, $(EGL_DISPLAY_DRIVERS), $(LIB_GALLIUM_DIR)/$(drv))
+
+default: $(EGL_DISPLAY_LIBS)
+
+$(EGL_DISPLAY_LIBS): $(LIB_GALLIUM_DIR)/%.so: %.so
+	@mkdir -p $(LIB_GALLIUM_DIR)
+	$(INSTALL) $^ $(LIB_GALLIUM_DIR)
+
+define mklib-egl
+$(MKLIB) -noprefix -o $@ $(EGL_DRIVER_OBJECTS) \
+	-Wl,--whole-archive $($(1)_ST) -Wl,--no-whole-archive \
+	$(EGL_DRIVER_PIPES) $(GALLIUM_AUXILIARIES) $($(1)_LIBS) $(EGL_DRIVER_LIBS)
+endef
+
+egl_x11_$(EGL_DRIVER_NAME).so: $(EGL_DRIVER_OBJECTS) $(x11_ST) $(EGL_DRIVER_PIPES) $(GALLIUM_AUXILIARIES) Makefile
+	$(call mklib-egl,x11)
+
+egl_kms_$(EGL_DRIVER_NAME).so: $(EGL_DRIVER_OBJECTS) $(kms_ST) $(EGL_DRIVER_PIPES) $(GALLIUM_AUXILIARIES) Makefile
+	$(call mklib-egl,kms)
+
+clean:
+	-rm -f $(EGL_DRIVER_OBJECTS)
+	-rm -f $(EGL_DISPLAY_DRIVERS)
+
+install: $(EGL_DISPLAY_LIBS)
+	@$(INSTALL) -d $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR)
+	@echo "Install $(EGL_DISPLAY_DRIVERS)"
+	@for lib in "$(EGL_DISPLAY_LIBS)"; do \
+		$(MINSTALL) -m 755 "$$lib" $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR); \
+	done
+
+depend:
diff --git a/src/gallium/winsys/drm/i965/egl_g3d/Makefile b/src/gallium/winsys/drm/i965/egl_g3d/Makefile
new file mode 100644
index 00000000000..dd2efe24855
--- /dev/null
+++ b/src/gallium/winsys/drm/i965/egl_g3d/Makefile
@@ -0,0 +1,14 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+EGL_DRIVER_NAME = i965
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_intel
+
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/i965/gem/libi965drm.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/i965/libi965.a
+
+include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/i965/egl_g3d/dummy.c b/src/gallium/winsys/drm/i965/egl_g3d/dummy.c
new file mode 100644
index 00000000000..4a1bc28b0b6
--- /dev/null
+++ b/src/gallium/winsys/drm/i965/egl_g3d/dummy.c
@@ -0,0 +1 @@
+/* mklib expects at least one object file */
diff --git a/src/gallium/winsys/drm/i965/gem/i965_drm_api.c b/src/gallium/winsys/drm/i965/gem/i965_drm_api.c
index fc9678d2b62..9a2203fc6d7 100644
--- a/src/gallium/winsys/drm/i965/gem/i965_drm_api.c
+++ b/src/gallium/winsys/drm/i965/gem/i965_drm_api.c
@@ -228,6 +228,7 @@ destroy(struct drm_api *api)
 
 struct drm_api i965_libdrm_api =
 {
+   .name = "i965",
    .create_context = i965_libdrm_create_context,
    .create_screen = i965_libdrm_create_screen,
    .texture_from_shared_handle = i965_libdrm_texture_from_shared_handle,
diff --git a/src/gallium/winsys/drm/i965/xorg/Makefile b/src/gallium/winsys/drm/i965/xorg/Makefile
index d91d0006efd..65d521fc1ce 100644
--- a/src/gallium/winsys/drm/i965/xorg/Makefile
+++ b/src/gallium/winsys/drm/i965/xorg/Makefile
@@ -1,19 +1,25 @@
-TARGET     = modesetting_drv.so
-CFILES     = $(wildcard ./*.c)
-OBJECTS    = $(patsubst ./%.c,./%.o,$(CFILES))
 TOP        = ../../../../../..
 
-include $(TOP)/configs/current
 
-INCLUDES = \
-	$(shell pkg-config --cflags-only-I pixman-1 xorg-server libdrm xproto) \
-	-I../gem \
-	-I$(TOP)/src/gallium/include \
-	-I$(TOP)/src/gallium/drivers \
-	-I$(TOP)/src/gallium/auxiliary \
-	-I$(TOP)/src/mesa \
-	-I$(TOP)/include \
-	-I$(TOP)/src/egl/main
+GALLIUMDIR = $(TOP)/src/gallium
+
+TARGET     = i965g_drv.so
+
+CFILES     = $(wildcard ./*.c)
+
+include ${TOP}/configs/current
+
+OBJECTS    = $(patsubst ./%.c,./%.o,$(CFILES))
+
+CFLAGS = -DHAVE_CONFIG_H \
+         -g -Wall -Wimplicit-function-declaration -fPIC \
+         $(shell pkg-config --cflags pixman-1 xorg-server libdrm xproto) \
+         -I${GALLIUMDIR}/include \
+         -I${GALLIUMDIR}/drivers \
+         -I${GALLIUMDIR}/auxiliary \
+         -I${TOP}/src/mesa \
+         -I$(TOP)/include \
+         -I$(TOP)/src/egl/main
 
 LIBS = \
 	$(TOP)/src/gallium/state_trackers/xorg/libxorgtracker.a \
@@ -23,19 +29,20 @@ LIBS = \
 	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(GALLIUM_AUXILIARIES)
 
-DRIVER_DEFINES = \
-	-DHAVE_CONFIG_H
-
-
+TARGET_STAGING = $(TOP)/$(LIB_DIR)/gallium/$(TARGET)
 #############################################
 
+all default: $(TARGET) $(TARGET_STAGING)
 
+$(TARGET): $(OBJECTS) Makefile $(GALLIUMDIR)/state_trackers/xorg/libxorgtracker.a $(LIBS)
+	$(TOP)/bin/mklib -noprefix -o $@ \
+	$(OBJECTS) $(LIBS) $(shell pkg-config --libs libdrm) -ldrm_radeon
 
-all default: $(TARGET)
+$(TOP)/$(LIB_DIR)/gallium:
+	mkdir -p $@
 
-$(TARGET): $(OBJECTS) Makefile $(TOP)/src/gallium/state_trackers/xorg/libxorgtracker.a $(LIBS)
-	$(TOP)/bin/mklib -noprefix -o $@ \
-	$(OBJECTS) $(LIBS) $(shell pkg-config --libs libdrm) -ldrm_intel
+$(TARGET_STAGING): $(TARGET) $(TOP)/$(LIB_DIR)/gallium
+	$(INSTALL) $(TARGET) $(TOP)/$(LIB_DIR)/gallium
 
 clean:
 	rm -rf $(OBJECTS) $(TARGET)
@@ -44,14 +51,4 @@ install:
 	$(INSTALL) -d $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR)
 	$(MINSTALL) -m 755 $(TARGET) $(DESTDIR)/$(XORG_DRIVER_INSTALL_DIR)
 
-
-##############################################
-
-
-.c.o:
-	$(CC) -c $(CFLAGS) $(INCLUDES) $(DRIVER_DEFINES) $< -o $@
-
-
-##############################################
-
 .PHONY	= all clean install
diff --git a/src/gallium/winsys/drm/intel/egl_g3d/Makefile b/src/gallium/winsys/drm/intel/egl_g3d/Makefile
new file mode 100644
index 00000000000..cdbb680773c
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/egl_g3d/Makefile
@@ -0,0 +1,14 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+EGL_DRIVER_NAME = i915
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_intel
+
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/intel/gem/libinteldrm.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/i915/libi915.a
+
+include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/intel/egl_g3d/dummy.c b/src/gallium/winsys/drm/intel/egl_g3d/dummy.c
new file mode 100644
index 00000000000..4a1bc28b0b6
--- /dev/null
+++ b/src/gallium/winsys/drm/intel/egl_g3d/dummy.c
@@ -0,0 +1 @@
+/* mklib expects at least one object file */
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
index 5ed2a10af1c..450ae09b345 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
@@ -195,6 +195,7 @@ destroy(struct drm_api *api)
 
 struct drm_api intel_drm_api =
 {
+   .name = "i915",
    .create_context = intel_drm_create_context,
    .create_screen = intel_drm_create_screen,
    .texture_from_shared_handle = intel_drm_texture_from_shared_handle,
diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
index 7106a06492d..4b2c6a1025e 100644
--- a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
+++ b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
@@ -87,6 +87,7 @@ nouveau_drm_create_screen(struct drm_api *api, int fd,
 	case 0x60:
 		init = nv40_screen_create;
 		break;
+	case 0x50:
 	case 0x80:
 	case 0x90:
 	case 0xa0:
@@ -164,6 +165,7 @@ nouveau_drm_create_context(struct drm_api *api, struct pipe_screen *pscreen)
 	case 0x60:
 		init = nv40_create;
 		break;
+	case 0x50:
 	case 0x80:
 	case 0x90:
 	case 0xa0:
@@ -252,6 +254,7 @@ nouveau_drm_handle_from_pt(struct drm_api *api, struct pipe_screen *pscreen,
 }
 
 struct drm_api drm_api_hooks = {
+	.name = "nouveau",
 	.create_screen = nouveau_drm_create_screen,
 	.create_context = nouveau_drm_create_context,
 	.texture_from_shared_handle = nouveau_drm_pt_from_name,
diff --git a/src/gallium/winsys/drm/nouveau/egl_g3d/Makefile b/src/gallium/winsys/drm/nouveau/egl_g3d/Makefile
new file mode 100644
index 00000000000..865a5d56a97
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/egl_g3d/Makefile
@@ -0,0 +1,19 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+EGL_DRIVER_NAME = nouveau
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_nouveau
+
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/nouveau/drm/libnouveaudrm.a \
+	$(TOP)/src/gallium/drivers/nv04/libnv04.a \
+	$(TOP)/src/gallium/drivers/nv10/libnv10.a \
+	$(TOP)/src/gallium/drivers/nv20/libnv20.a \
+	$(TOP)/src/gallium/drivers/nv30/libnv30.a \
+	$(TOP)/src/gallium/drivers/nv40/libnv40.a \
+	$(TOP)/src/gallium/drivers/nv50/libnv50.a \
+	$(TOP)/src/gallium/drivers/nouveau/libnouveau.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a
+
+include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/nouveau/egl_g3d/dummy.c b/src/gallium/winsys/drm/nouveau/egl_g3d/dummy.c
new file mode 100644
index 00000000000..4a1bc28b0b6
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/egl_g3d/dummy.c
@@ -0,0 +1 @@
+/* mklib expects at least one object file */
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
index d2367b245a2..385fa857b56 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
@@ -146,16 +146,17 @@ static void *radeon_buffer_map(struct pipe_winsys *ws,
         (struct radeon_pipe_buffer*)buffer;
     int write = 0;
 
-    if (radeon_bo_is_referenced_by_cs(radeon_buffer->bo, priv->cs)) {
-        priv->flush_cb(priv->flush_data);
-    }
-
     if (flags & PIPE_BUFFER_USAGE_DONTBLOCK) {
         uint32_t domain;
 
         if (radeon_bo_is_busy(radeon_buffer->bo, &domain))
             return NULL;
     }
+
+    if (radeon_bo_is_referenced_by_cs(radeon_buffer->bo, priv->cs)) {
+        priv->flush_cb(priv->flush_data);
+    }
+
     if (flags & PIPE_BUFFER_USAGE_CPU_WRITE) {
         write = 1;
     }
@@ -280,58 +281,3 @@ struct radeon_winsys* radeon_pipe_winsys(int fd)
 
     return radeon_ws;
 }
-#if 0
-static struct pipe_buffer *radeon_buffer_from_handle(struct radeon_screen *radeon_screen,
-                                                  uint32_t handle)
-{
-    struct radeon_pipe_buffer *radeon_buffer;
-    struct radeon_bo *bo = NULL;
-
-    bo = radeon_bo_open(radeon_screen->bom, handle, 0, 0, 0, 0);
-    if (bo == NULL) {
-        return NULL;
-    }
-    radeon_buffer = calloc(1, sizeof(struct radeon_pipe_buffer));
-    if (radeon_buffer == NULL) {
-        radeon_bo_unref(bo);
-        return NULL;
-    }
-    pipe_reference_init(&radeon_buffer->base.reference, 1);
-    radeon_buffer->base.usage = PIPE_BUFFER_USAGE_PIXEL;
-    radeon_buffer->bo = bo;
-    return &radeon_buffer->base;
-}
-
-struct pipe_surface *radeon_surface_from_handle(struct radeon_context *radeon_context,
-                                             uint32_t handle,
-                                             enum pipe_format format,
-                                             int w, int h, int pitch)
-{
-    struct pipe_screen *pipe_screen = radeon_context->pipe_screen;
-    struct pipe_winsys *pipe_winsys = radeon_context->pipe_winsys;
-    struct pipe_texture tmpl;
-    struct pipe_surface *ps;
-    struct pipe_texture *pt;
-    struct pipe_buffer *pb;
-
-    pb = radeon_buffer_from_handle(radeon_context->radeon_screen, handle);
-    if (pb == NULL) {
-        return NULL;
-    }
-    memset(&tmpl, 0, sizeof(tmpl));
-    tmpl.tex_usage = PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
-    tmpl.target = PIPE_TEXTURE_2D;
-    tmpl.width0 = w;
-    tmpl.height0 = h;
-    tmpl.depth0 = 1;
-    tmpl.format = format;
-
-    pt = pipe_screen->texture_blanket(pipe_screen, &tmpl, &pitch, pb);
-    if (pt == NULL) {
-        pipe_buffer_reference(&pb, NULL);
-    }
-    ps = pipe_screen->get_tex_surface(pipe_screen, pt, 0, 0, 0,
-                                      PIPE_BUFFER_USAGE_GPU_WRITE);
-    return ps;
-}
-#endif
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.c b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
index 05194fc52a2..851c2236979 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_drm.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
@@ -252,6 +252,7 @@ static boolean radeon_local_handle_from_texture(struct drm_api *api,
 }
 
 struct drm_api drm_api_hooks = {
+    .name = "radeon",
     .create_screen = radeon_create_screen,
     .create_context = radeon_create_context,
     .texture_from_shared_handle = radeon_texture_from_shared_handle,
diff --git a/src/gallium/winsys/drm/radeon/dri/Makefile b/src/gallium/winsys/drm/radeon/dri/Makefile
index a9889444de8..eaa34180321 100644
--- a/src/gallium/winsys/drm/radeon/dri/Makefile
+++ b/src/gallium/winsys/drm/radeon/dri/Makefile
@@ -2,7 +2,7 @@
 TOP = ../../../../../..
 include $(TOP)/configs/current
 
-LIBNAME = radeon_dri.so
+LIBNAME = radeong_dri.so
 
 MINIGLX_SOURCES =
 
diff --git a/src/gallium/winsys/drm/radeon/egl_g3d/Makefile b/src/gallium/winsys/drm/radeon/egl_g3d/Makefile
new file mode 100644
index 00000000000..9027a5ff46d
--- /dev/null
+++ b/src/gallium/winsys/drm/radeon/egl_g3d/Makefile
@@ -0,0 +1,14 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+EGL_DRIVER_NAME = r300
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS = -ldrm_radeon
+
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/radeon/core/libradeonwinsys.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/r300/libr300.a
+
+include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/radeon/egl_g3d/dummy.c b/src/gallium/winsys/drm/radeon/egl_g3d/dummy.c
new file mode 100644
index 00000000000..4a1bc28b0b6
--- /dev/null
+++ b/src/gallium/winsys/drm/radeon/egl_g3d/dummy.c
@@ -0,0 +1 @@
+/* mklib expects at least one object file */
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
index 5995eee34ba..4f5ccea4677 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
@@ -358,6 +358,7 @@ static struct dri1_api dri1_api_hooks = {
 };
 
 static struct drm_api vmw_drm_api_hooks = {
+   .name = "vmwgfx",
    .create_screen = vmw_drm_create_screen,
    .create_context = vmw_drm_create_context,
    .texture_from_shared_handle = vmw_drm_texture_from_handle,
diff --git a/src/gallium/winsys/drm/vmware/egl_g3d/Makefile b/src/gallium/winsys/drm/vmware/egl_g3d/Makefile
new file mode 100644
index 00000000000..3cf79924e08
--- /dev/null
+++ b/src/gallium/winsys/drm/vmware/egl_g3d/Makefile
@@ -0,0 +1,14 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+EGL_DRIVER_NAME = vmwgfx
+EGL_DRIVER_SOURCES = dummy.c
+EGL_DRIVER_LIBS =
+
+EGL_DRIVER_PIPES = \
+	$(TOP)/src/gallium/winsys/drm/vmware/core/libsvgadrm.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/svga/libsvga.a
+
+include ../../Makefile.egl_g3d
diff --git a/src/gallium/winsys/drm/vmware/egl_g3d/dummy.c b/src/gallium/winsys/drm/vmware/egl_g3d/dummy.c
new file mode 100644
index 00000000000..4a1bc28b0b6
--- /dev/null
+++ b/src/gallium/winsys/drm/vmware/egl_g3d/dummy.c
@@ -0,0 +1 @@
+/* mklib expects at least one object file */
diff --git a/src/gallium/winsys/egl_xlib/egl_xlib.c b/src/gallium/winsys/egl_xlib/egl_xlib.c
index 599973ce127..1d9bac3871c 100644
--- a/src/gallium/winsys/egl_xlib/egl_xlib.c
+++ b/src/gallium/winsys/egl_xlib/egl_xlib.c
@@ -274,7 +274,7 @@ xlib_eglTerminate(_EGLDriver *drv, _EGLDisplay *dpy)
 
 
 static _EGLProc
-xlib_eglGetProcAddress(const char *procname)
+xlib_eglGetProcAddress(_EGLDriver *drv, const char *procname)
 {
    return (_EGLProc) st_get_proc_address(procname);
 }
@@ -751,24 +751,18 @@ xlib_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 static EGLBoolean
 xlib_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw)
 {
-   /* error checking step: */
-   if (!_eglSwapBuffers(drv, dpy, draw))
-      return EGL_FALSE;
-
-   {
-      struct xlib_egl_surface *xsurf = lookup_surface(draw);
-      struct pipe_winsys *pws = xsurf->winsys;
-      struct pipe_surface *psurf;
+   struct xlib_egl_surface *xsurf = lookup_surface(draw);
+   struct pipe_winsys *pws = xsurf->winsys;
+   struct pipe_surface *psurf;
 
-      st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT,
-                                 &psurf);
+   st_get_framebuffer_surface(xsurf->Framebuffer, ST_SURFACE_BACK_LEFT,
+         &psurf);
 
-      st_notify_swapbuffers(xsurf->Framebuffer);
+   st_notify_swapbuffers(xsurf->Framebuffer);
 
-      display_surface(pws, psurf, xsurf);
+   display_surface(pws, psurf, xsurf);
 
-      check_and_update_buffer_size(xsurf);
-   }
+   check_and_update_buffer_size(xsurf);
 
    return EGL_TRUE;
 }
diff --git a/src/gallium/winsys/xlib/xlib.c b/src/gallium/winsys/xlib/xlib.c
index 6dbe05f193e..13117712c23 100644
--- a/src/gallium/winsys/xlib/xlib.c
+++ b/src/gallium/winsys/xlib/xlib.c
@@ -47,6 +47,8 @@ enum mode {
    MODE_SOFTPIPE
 };
 
+/* advertise OpenGL support */
+PUBLIC const int st_api_OpenGL = 1;
 
 static enum mode get_mode()
 {
diff --git a/src/glu/sgi/libnurbs/interface/bezierPatchMesh.h b/src/glu/sgi/libnurbs/interface/bezierPatchMesh.h
index 449329665cc..ba6868a3063 100644
--- a/src/glu/sgi/libnurbs/interface/bezierPatchMesh.h
+++ b/src/glu/sgi/libnurbs/interface/bezierPatchMesh.h
@@ -33,6 +33,7 @@
 #ifndef _BEZIERPATCHMESH_H
 #define _BEZIERPATCHMESH_H
 
+#include <GL/gl.h>
 #include "bezierPatch.h"
 
 typedef struct bezierPatchMesh{
diff --git a/src/glu/sgi/libnurbs/interface/glsurfeval.h b/src/glu/sgi/libnurbs/interface/glsurfeval.h
index 1567c6b098c..621e59391aa 100644
--- a/src/glu/sgi/libnurbs/interface/glsurfeval.h
+++ b/src/glu/sgi/libnurbs/interface/glsurfeval.h
@@ -83,7 +83,7 @@ typedef struct surfEvalMachine{
 
 class StoredVertex {
 public:
-    		StoredVertex() { type = 0; }
+    		StoredVertex() { type = 0; coord[0] = 0; coord[1] = 0; point[0] = 0; point[1] = 0; }
 		~StoredVertex(void) {}
     void	saveEvalCoord(REAL x, REAL y) 
 		    {coord[0] = x; coord[1] = y; type = TYPECOORD; }
diff --git a/src/glu/sgi/libnurbs/internals/mesher.cc b/src/glu/sgi/libnurbs/internals/mesher.cc
index 9cc436adbf3..b2d83f41288 100644
--- a/src/glu/sgi/libnurbs/internals/mesher.cc
+++ b/src/glu/sgi/libnurbs/internals/mesher.cc
@@ -58,6 +58,9 @@ Mesher::Mesher( Backend& b )
 {
     stacksize = 0;
     vdata = 0;
+    last[0] = 0;
+    last[1] = 0;
+    itop = 0;
     lastedge = 0; //needed to prevent purify UMR 
 }
 
diff --git a/src/glu/sgi/libnurbs/internals/reader.cc b/src/glu/sgi/libnurbs/internals/reader.cc
index 6135eef60ee..c59240d26ab 100644
--- a/src/glu/sgi/libnurbs/internals/reader.cc
+++ b/src/glu/sgi/libnurbs/internals/reader.cc
@@ -64,6 +64,7 @@ O_pwlcurve::O_pwlcurve( long _type, long count, INREAL *array, long byte_stride,
     owner = 0;
     pts = trimpts;
     npts = (int) count;
+    save = 0;
     int i;
 
     /* copy user data into internal trimming data structures */
@@ -115,6 +116,7 @@ O_pwlcurve::O_pwlcurve( long _type, long count, INREAL *array, long byte_stride,
     owner = 0;
     pts = trimpts;
     npts = (int) count;
+    save = 0;
 
     /* copy user data into internal trimming data structures */
     switch( _type ) {
diff --git a/src/glu/sgi/libnurbs/internals/renderhints.cc b/src/glu/sgi/libnurbs/internals/renderhints.cc
index a3aa62d42cf..7025f74f5b6 100644
--- a/src/glu/sgi/libnurbs/internals/renderhints.cc
+++ b/src/glu/sgi/libnurbs/internals/renderhints.cc
@@ -54,6 +54,10 @@ Renderhints::Renderhints()
     errorchecking 	= N_MSG;
     subdivisions 	= 6.0;
     tmp1 		= 0.0;
+    displaydomain 	= 0;
+    maxsubdivisions 	= (int) subdivisions;
+    wiretris	 	= 0;
+    wirequads	 	= 0;
 }
 
 void
diff --git a/src/glu/sgi/libnurbs/internals/simplemath.h b/src/glu/sgi/libnurbs/internals/simplemath.h
index 0a060c57ead..d00062dc700 100644
--- a/src/glu/sgi/libnurbs/internals/simplemath.h
+++ b/src/glu/sgi/libnurbs/internals/simplemath.h
@@ -38,6 +38,8 @@
 
 /* simple inline routines */
 
+#include "types.h"
+
 inline int 
 max( int x, int y ) { return ( x < y ) ? y : x; }
 
diff --git a/src/glu/sgi/libnurbs/internals/slicer.cc b/src/glu/sgi/libnurbs/internals/slicer.cc
index 27d2a650d11..1b18d73c176 100644
--- a/src/glu/sgi/libnurbs/internals/slicer.cc
+++ b/src/glu/sgi/libnurbs/internals/slicer.cc
@@ -1181,6 +1181,10 @@ void Slicer::slice(Arc_ptr loop)
 Slicer::Slicer( Backend &b ) 
 	: CoveAndTiler( b ), Mesher( b ), backend( b )
 {
+    oneOverDu = 0;
+    du = 0;
+    dv = 0;
+    isolines = 0;
     ulinear = 0;
     vlinear = 0;
 }
diff --git a/src/glut/glx/win32_menu.c b/src/glut/glx/win32_menu.c
index 5ce264dcdb7..d8e336ceed7 100644
--- a/src/glut/glx/win32_menu.c
+++ b/src/glut/glx/win32_menu.c
@@ -97,7 +97,7 @@ static void
 mapMenu(GLUTmenu * menu, int x, int y)
 {
   TrackPopupMenu((HMENU) menu->win, TPM_LEFTALIGN |
-    (__glutMenuButton == TPM_RIGHTBUTTON) ? TPM_RIGHTBUTTON : TPM_LEFTBUTTON,
+    ((__glutMenuButton == TPM_RIGHTBUTTON) ? TPM_RIGHTBUTTON : TPM_LEFTBUTTON),
     x, y, 0, __glutCurrentWindow->win, NULL);
 }
 
diff --git a/src/glx/x11/dri2.c b/src/glx/x11/dri2.c
index dad04470a00..2cb5d3463af 100644
--- a/src/glx/x11/dri2.c
+++ b/src/glx/x11/dri2.c
@@ -34,12 +34,15 @@
 #ifdef GLX_DIRECT_RENDERING
 
 #define NEED_REPLIES
+#include <stdio.h>
 #include <X11/Xlibint.h>
 #include <X11/extensions/Xext.h>
 #include <X11/extensions/extutil.h>
 #include <X11/extensions/dri2proto.h>
 #include "xf86drm.h"
 #include "dri2.h"
+#include "glxclient.h"
+#include "GL/glxext.h"
 
 /* Allow the build to work with an older versions of dri2proto.h and
  * dri2tokens.h.
@@ -55,6 +58,11 @@ static char dri2ExtensionName[] = DRI2_NAME;
 static XExtensionInfo *dri2Info;
 static XEXT_GENERATE_CLOSE_DISPLAY (DRI2CloseDisplay, dri2Info)
 
+static Bool
+DRI2WireToEvent(Display *dpy, XEvent *event, xEvent *wire);
+static Status
+DRI2EventToWire(Display *dpy, XEvent *event, xEvent *wire);
+
 static /* const */ XExtensionHooks dri2ExtensionHooks = {
   NULL,                   /* create_gc */
   NULL,                   /* copy_gc */
@@ -63,8 +71,8 @@ static /* const */ XExtensionHooks dri2ExtensionHooks = {
   NULL,                   /* create_font */
   NULL,                   /* free_font */
   DRI2CloseDisplay,       /* close_display */
-  NULL,                   /* wire_to_event */
-  NULL,                   /* event_to_wire */
+  DRI2WireToEvent,        /* wire_to_event */
+  DRI2EventToWire,        /* event_to_wire */
   NULL,                   /* error */
   NULL,                   /* error_string */
 };
@@ -75,6 +83,65 @@ static XEXT_GENERATE_FIND_DISPLAY (DRI2FindDisplay,
                                    &dri2ExtensionHooks,
                                    0, NULL)
 
+static Bool
+DRI2WireToEvent(Display *dpy, XEvent *event, xEvent *wire)
+{
+   XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+
+   XextCheckExtension(dpy, info, dri2ExtensionName, False);
+
+   switch ((wire->u.u.type & 0x7f) - info->codes->first_event) {
+   case DRI2_BufferSwapComplete:
+   {
+      GLXBufferSwapComplete *aevent = (GLXBufferSwapComplete *)event;
+      xDRI2BufferSwapComplete *awire = (xDRI2BufferSwapComplete *)wire;
+      switch (awire->type) {
+      case DRI2_EXCHANGE_COMPLETE:
+	 aevent->event_type = GLX_EXCHANGE_COMPLETE;
+	 break;
+      case DRI2_BLIT_COMPLETE:
+	 aevent->event_type = GLX_BLIT_COMPLETE;
+	 break;
+      case DRI2_FLIP_COMPLETE:
+	 aevent->event_type = GLX_FLIP_COMPLETE;
+	 break;
+      default:
+	 /* unknown swap completion type */
+	 return False;
+      }
+      aevent->drawable = awire->drawable;
+      aevent->ust = ((CARD64)awire->ust_hi << 32) | awire->ust_lo;
+      aevent->msc = ((CARD64)awire->msc_hi << 32) | awire->msc_lo;
+      aevent->sbc = ((CARD64)awire->sbc_hi << 32) | awire->sbc_lo;
+      return True;
+   }
+   default:
+      /* client doesn't support server event */
+      break;
+   }
+
+   return False;
+}
+
+/* We don't actually support this.  It doesn't make sense for clients to
+ * send each other DRI2 events.
+ */
+static Status
+DRI2EventToWire(Display *dpy, XEvent *event, xEvent *wire)
+{
+   XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+
+   XextCheckExtension(dpy, info, dri2ExtensionName, False);
+
+   switch (event->type) {
+   default:
+      /* client doesn't support server event */
+      break;
+   }
+
+   return Success;
+}
+
 Bool
 DRI2QueryExtension(Display * dpy, int *eventBase, int *errorBase)
 {
@@ -380,4 +447,177 @@ DRI2CopyRegion(Display * dpy, XID drawable, XserverRegion region,
    SyncHandle();
 }
 
+static void
+load_swap_req(xDRI2SwapBuffersReq *req, CARD64 target, CARD64 divisor,
+	     CARD64 remainder)
+{
+    req->target_msc_hi = target >> 32;
+    req->target_msc_lo = target & 0xffffffff;
+    req->divisor_hi = divisor >> 32;
+    req->divisor_lo = divisor & 0xffffffff;
+    req->remainder_hi = remainder >> 32;
+    req->remainder_lo = remainder & 0xffffffff;
+}
+
+static CARD64
+vals_to_card64(CARD32 lo, CARD32 hi)
+{
+    return (CARD64)hi << 32 | lo;
+}
+
+void DRI2SwapBuffers(Display *dpy, XID drawable, CARD64 target_msc,
+		     CARD64 divisor, CARD64 remainder, CARD64 *count)
+{
+    XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+    xDRI2SwapBuffersReq *req;
+    xDRI2SwapBuffersReply rep;
+
+    XextSimpleCheckExtension (dpy, info, dri2ExtensionName);
+
+    LockDisplay(dpy);
+    GetReq(DRI2SwapBuffers, req);
+    req->reqType = info->codes->major_opcode;
+    req->dri2ReqType = X_DRI2SwapBuffers;
+    req->drawable = drawable;
+    load_swap_req(req, target_msc, divisor, remainder);
+
+    _XReply(dpy, (xReply *)&rep, 0, xFalse);
+
+    *count = vals_to_card64(rep.swap_lo, rep.swap_hi);
+
+    UnlockDisplay(dpy);
+    SyncHandle();
+}
+
+Bool DRI2GetMSC(Display *dpy, XID drawable, CARD64 *ust, CARD64 *msc,
+		CARD64 *sbc)
+{
+    XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+    xDRI2GetMSCReq *req;
+    xDRI2MSCReply rep;
+
+    XextCheckExtension (dpy, info, dri2ExtensionName, False);
+
+    LockDisplay(dpy);
+    GetReq(DRI2GetMSC, req);
+    req->reqType = info->codes->major_opcode;
+    req->dri2ReqType = X_DRI2GetMSC;
+    req->drawable = drawable;
+
+    if (!_XReply(dpy, (xReply *)&rep, 0, xFalse)) {
+	UnlockDisplay(dpy);
+	SyncHandle();
+	return False;
+    }
+
+    *ust = vals_to_card64(rep.ust_lo, rep.ust_hi);
+    *msc = vals_to_card64(rep.msc_lo, rep.msc_hi);
+    *sbc = vals_to_card64(rep.sbc_lo, rep.sbc_hi);
+
+    UnlockDisplay(dpy);
+    SyncHandle();
+
+    return True;
+}
+
+static void
+load_msc_req(xDRI2WaitMSCReq *req, CARD64 target, CARD64 divisor,
+	     CARD64 remainder)
+{
+    req->target_msc_hi = target >> 32;
+    req->target_msc_lo = target & 0xffffffff;
+    req->divisor_hi = divisor >> 32;
+    req->divisor_lo = divisor & 0xffffffff;
+    req->remainder_hi = remainder >> 32;
+    req->remainder_lo = remainder & 0xffffffff;
+}
+
+Bool DRI2WaitMSC(Display *dpy, XID drawable, CARD64 target_msc, CARD64 divisor,
+		 CARD64 remainder, CARD64 *ust, CARD64 *msc, CARD64 *sbc)
+{
+    XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+    xDRI2WaitMSCReq *req;
+    xDRI2MSCReply rep;
+
+    XextCheckExtension (dpy, info, dri2ExtensionName, False);
+
+    LockDisplay(dpy);
+    GetReq(DRI2WaitMSC, req);
+    req->reqType = info->codes->major_opcode;
+    req->dri2ReqType = X_DRI2WaitMSC;
+    req->drawable = drawable;
+    load_msc_req(req, target_msc, divisor, remainder);
+
+    if (!_XReply(dpy, (xReply *)&rep, 0, xFalse)) {
+	UnlockDisplay(dpy);
+	SyncHandle();
+	return False;
+    }
+
+    *ust = ((CARD64)rep.ust_hi << 32) | (CARD64)rep.ust_lo;
+    *msc = ((CARD64)rep.msc_hi << 32) | (CARD64)rep.msc_lo;
+    *sbc = ((CARD64)rep.sbc_hi << 32) | (CARD64)rep.sbc_lo;
+
+    UnlockDisplay(dpy);
+    SyncHandle();
+
+    return True;
+}
+
+static void
+load_sbc_req(xDRI2WaitSBCReq *req, CARD64 target)
+{
+    req->target_sbc_hi = target >> 32;
+    req->target_sbc_lo = target & 0xffffffff;
+}
+
+Bool DRI2WaitSBC(Display *dpy, XID drawable, CARD64 target_sbc, CARD64 *ust,
+		 CARD64 *msc, CARD64 *sbc)
+{
+    XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+    xDRI2WaitSBCReq *req;
+    xDRI2MSCReply rep;
+
+    XextCheckExtension (dpy, info, dri2ExtensionName, False);
+
+    LockDisplay(dpy);
+    GetReq(DRI2WaitSBC, req);
+    req->reqType = info->codes->major_opcode;
+    req->dri2ReqType = X_DRI2WaitSBC;
+    req->drawable = drawable;
+    load_sbc_req(req, target_sbc);
+
+    if (!_XReply(dpy, (xReply *)&rep, 0, xFalse)) {
+	UnlockDisplay(dpy);
+	SyncHandle();
+	return False;
+    }
+
+    *ust = ((CARD64)rep.ust_hi << 32) | rep.ust_lo;
+    *msc = ((CARD64)rep.msc_hi << 32) | rep.msc_lo;
+    *sbc = ((CARD64)rep.sbc_hi << 32) | rep.sbc_lo;
+
+    UnlockDisplay(dpy);
+    SyncHandle();
+
+    return True;
+}
+
+void DRI2SwapInterval(Display *dpy, XID drawable, int interval)
+{
+    XExtDisplayInfo *info = DRI2FindDisplay(dpy);
+    xDRI2SwapIntervalReq *req;
+
+    XextSimpleCheckExtension (dpy, info, dri2ExtensionName);
+
+    LockDisplay(dpy);
+    GetReq(DRI2SwapInterval, req);
+    req->reqType = info->codes->major_opcode;
+    req->dri2ReqType = X_DRI2SwapInterval;
+    req->drawable = drawable;
+    req->interval = interval;
+    UnlockDisplay(dpy);
+    SyncHandle();
+}
+
 #endif /* GLX_DIRECT_RENDERING */
diff --git a/src/glx/x11/dri2.h b/src/glx/x11/dri2.h
index a6fe66e136d..114e9f8f965 100644
--- a/src/glx/x11/dri2.h
+++ b/src/glx/x11/dri2.h
@@ -85,4 +85,22 @@ DRI2CopyRegion(Display * dpy, XID drawable,
                XserverRegion region,
                CARD32 dest, CARD32 src);
 
+extern void
+DRI2SwapBuffers(Display *dpy, XID drawable, CARD64 target_msc, CARD64 divisor,
+		CARD64 remainder, CARD64 *count);
+
+extern Bool
+DRI2GetMSC(Display *dpy, XID drawable, CARD64 *ust, CARD64 *msc, CARD64 *sbc);
+
+extern Bool
+DRI2WaitMSC(Display *dpy, XID drawable, CARD64 target_msc, CARD64 divisor,
+	    CARD64 remainder, CARD64 *ust, CARD64 *msc, CARD64 *sbc);
+
+extern Bool
+DRI2WaitSBC(Display *dpy, XID drawable, CARD64 target_sbc, CARD64 *ust,
+	    CARD64 *msc, CARD64 *sbc);
+
+extern void
+DRI2SwapInterval(Display *dpy, XID drawable, int interval);
+
 #endif
diff --git a/src/glx/x11/dri2_glx.c b/src/glx/x11/dri2_glx.c
index 89efe3ab292..7b0c52b50d4 100644
--- a/src/glx/x11/dri2_glx.c
+++ b/src/glx/x11/dri2_glx.c
@@ -35,6 +35,7 @@
 #include <X11/Xlib.h>
 #include <X11/extensions/Xfixes.h>
 #include <X11/extensions/Xdamage.h>
+#include "glapi.h"
 #include "glxclient.h"
 #include "glcontextmodes.h"
 #include "xf86dri.h"
@@ -46,6 +47,7 @@
 #include "xf86drm.h"
 #include "dri2.h"
 #include "dri_common.h"
+#include "../../mesa/drivers/dri/common/dri_util.h"
 
 #undef DRI2_MINOR
 #define DRI2_MINOR 1
@@ -64,6 +66,7 @@ struct __GLXDRIdisplayPrivateRec
    int driMajor;
    int driMinor;
    int driPatch;
+   int swapAvailable;
 };
 
 struct __GLXDRIcontextPrivateRec
@@ -81,6 +84,7 @@ struct __GLXDRIdrawablePrivateRec
    int width, height;
    int have_back;
    int have_fake_front;
+   int swap_interval;
 };
 
 static void dri2WaitX(__GLXDRIdrawable * pdraw);
@@ -197,9 +201,31 @@ dri2CreateDrawable(__GLXscreenConfigs * psc,
    return &pdraw->base;
 }
 
+static int
+dri2DrawableGetMSC(__GLXscreenConfigs *psc, __GLXDRIdrawable *pdraw,
+		   int64_t *ust, int64_t *msc, int64_t *sbc)
+{
+   return DRI2GetMSC(psc->dpy, pdraw->xDrawable, ust, msc, sbc);
+}
+
+static int
+dri2WaitForMSC(__GLXDRIdrawable *pdraw, int64_t target_msc, int64_t divisor,
+	       int64_t remainder, int64_t *ust, int64_t *msc, int64_t *sbc)
+{
+   return DRI2WaitMSC(pdraw->psc->dpy, pdraw->xDrawable, target_msc, divisor,
+		      remainder, ust, msc, sbc);
+}
+
+static int
+dri2WaitForSBC(__GLXDRIdrawable *pdraw, int64_t target_sbc, int64_t *ust,
+	       int64_t *msc, int64_t *sbc)
+{
+   return DRI2WaitSBC(pdraw->psc->dpy, pdraw->xDrawable, target_sbc, ust, msc,
+		      sbc);
+}
+
 static void
-dri2CopySubBuffer(__GLXDRIdrawable * pdraw,
-                  int x, int y, int width, int height)
+dri2CopySubBuffer(__GLXDRIdrawable *pdraw, int x, int y, int width, int height)
 {
    __GLXDRIdrawablePrivate *priv = (__GLXDRIdrawablePrivate *) pdraw;
    XRectangle xrect;
@@ -232,15 +258,7 @@ dri2CopySubBuffer(__GLXDRIdrawable * pdraw,
 }
 
 static void
-dri2SwapBuffers(__GLXDRIdrawable * pdraw)
-{
-   __GLXDRIdrawablePrivate *priv = (__GLXDRIdrawablePrivate *) pdraw;
-
-   dri2CopySubBuffer(pdraw, 0, 0, priv->width, priv->height);
-}
-
-static void
-dri2WaitX(__GLXDRIdrawable * pdraw)
+dri2WaitX(__GLXDRIdrawable *pdraw)
 {
    __GLXDRIdrawablePrivate *priv = (__GLXDRIdrawablePrivate *) pdraw;
    XRectangle xrect;
@@ -342,6 +360,38 @@ process_buffers(__GLXDRIdrawablePrivate * pdraw, DRI2Buffer * buffers,
 
 }
 
+static int64_t
+dri2SwapBuffers(__GLXDRIdrawable *pdraw, int64_t target_msc, int64_t divisor,
+		int64_t remainder)
+{
+    __GLXDRIdrawablePrivate *priv = (__GLXDRIdrawablePrivate *) pdraw;
+    __GLXdisplayPrivate *dpyPriv = __glXInitialize(priv->base.psc->dpy);
+    __GLXDRIdisplayPrivate *pdp =
+	(__GLXDRIdisplayPrivate *)dpyPriv->dri2Display;
+    int64_t ret;
+
+#ifdef __DRI2_FLUSH
+    if (pdraw->psc->f)
+    	(*pdraw->psc->f->flush)(pdraw->driDrawable);
+#endif
+
+    /* Old servers can't handle swapbuffers */
+    if (!pdp->swapAvailable) {
+       dri2CopySubBuffer(pdraw, 0, 0, priv->width, priv->height);
+       return 0;
+    }
+
+    DRI2SwapBuffers(pdraw->psc->dpy, pdraw->xDrawable, target_msc, divisor,
+		    remainder, &ret);
+
+#if __DRI2_FLUSH_VERSION >= 2
+    if (pdraw->psc->f)
+       (*pdraw->psc->f->flushInvalidate)(pdraw->driDrawable);
+#endif
+
+    return ret;
+}
+
 static __DRIbuffer *
 dri2GetBuffers(__DRIdrawable * driDrawable,
                int *width, int *height,
@@ -390,6 +440,23 @@ dri2GetBuffersWithFormat(__DRIdrawable * driDrawable,
    return pdraw->buffers;
 }
 
+static void
+dri2SetSwapInterval(__GLXDRIdrawable *pdraw, int interval)
+{
+   __GLXDRIdrawablePrivate *priv =  (__GLXDRIdrawablePrivate *) pdraw;
+
+   DRI2SwapInterval(priv->base.psc->dpy, pdraw->xDrawable, interval);
+   priv->swap_interval = interval;
+}
+
+static unsigned int
+dri2GetSwapInterval(__GLXDRIdrawable *pdraw)
+{
+   __GLXDRIdrawablePrivate *priv =  (__GLXDRIdrawablePrivate *) pdraw;
+
+  return priv->swap_interval;
+}
+
 static const __DRIdri2LoaderExtension dri2LoaderExtension = {
    {__DRI_DRI2_LOADER, __DRI_DRI2_LOADER_VERSION},
    dri2GetBuffers,
@@ -437,7 +504,7 @@ dri2CreateScreen(__GLXscreenConfigs * psc, int screen,
    psc->ext_list_first_time = GL_TRUE;
 
    if (!DRI2Connect(psc->dpy, RootWindow(psc->dpy, screen),
-                    &driverName, &deviceName))
+		    &driverName, &deviceName))
       return NULL;
 
    psc->driver = driOpenDriver(driverName);
@@ -454,9 +521,9 @@ dri2CreateScreen(__GLXscreenConfigs * psc, int screen,
 
    for (i = 0; extensions[i]; i++) {
       if (strcmp(extensions[i]->name, __DRI_CORE) == 0)
-         psc->core = (__DRIcoreExtension *) extensions[i];
+	 psc->core = (__DRIcoreExtension *) extensions[i];
       if (strcmp(extensions[i]->name, __DRI_DRI2) == 0)
-         psc->dri2 = (__DRIdri2Extension *) extensions[i];
+	 psc->dri2 = (__DRIdri2Extension *) extensions[i];
    }
 
    if (psc->core == NULL || psc->dri2 == NULL) {
@@ -485,16 +552,17 @@ dri2CreateScreen(__GLXscreenConfigs * psc, int screen,
     */
    psc->__driScreen =
       psc->dri2->createNewScreen(screen, psc->fd, ((pdp->driMinor < 1)
-                                                   ? loader_extensions_old
-                                                   : loader_extensions),
-                                 &driver_configs, psc);
+						   ? loader_extensions_old
+						   : loader_extensions),
+				 &driver_configs, psc);
 
    if (psc->__driScreen == NULL) {
       ErrorMessageF("failed to create dri screen\n");
       return NULL;
    }
 
-   driBindExtensions(psc, 1);
+   driBindCommonExtensions(psc);
+   dri2BindExtensions(psc);
 
    psc->configs = driConvertConfigs(psc->core, psc->configs, driver_configs);
    psc->visuals = driConvertConfigs(psc->core, psc->visuals, driver_configs);
@@ -507,6 +575,19 @@ dri2CreateScreen(__GLXscreenConfigs * psc, int screen,
    psp->swapBuffers = dri2SwapBuffers;
    psp->waitGL = dri2WaitGL;
    psp->waitX = dri2WaitX;
+   if (pdp->driMinor >= 2) {
+      psp->getDrawableMSC = dri2DrawableGetMSC;
+      psp->waitForMSC = dri2WaitForMSC;
+      psp->waitForSBC = dri2WaitForSBC;
+      psp->setSwapInterval = dri2SetSwapInterval;
+      psp->getSwapInterval = dri2GetSwapInterval;
+   } else {
+      psp->getDrawableMSC = NULL;
+      psp->waitForMSC = NULL;
+      psp->waitForSBC = NULL;
+      psp->setSwapInterval = NULL;
+      psp->getSwapInterval = NULL;
+   }
 
    /* DRI2 suports SubBuffer through DRI2CopyRegion, so it's always
     * available.*/
@@ -518,7 +599,7 @@ dri2CreateScreen(__GLXscreenConfigs * psc, int screen,
 
    return psp;
 
- handle_error:
+handle_error:
    Xfree(driverName);
    Xfree(deviceName);
 
@@ -559,6 +640,9 @@ dri2CreateDisplay(Display * dpy)
    }
 
    pdp->driPatch = 0;
+   pdp->swapAvailable = 0;
+   if (pdp->driMinor >= 2)
+      pdp->swapAvailable = 1;
 
    pdp->base.destroyDisplay = dri2DestroyDisplay;
    pdp->base.createScreen = dri2CreateScreen;
diff --git a/src/glx/x11/dri_common.c b/src/glx/x11/dri_common.c
index 9c825ad74ab..e4034161bb3 100644
--- a/src/glx/x11/dri_common.c
+++ b/src/glx/x11/dri_common.c
@@ -336,8 +336,9 @@ driConvertConfigs(const __DRIcoreExtension * core,
    return head.next;
 }
 
+/* Bind DRI1 specific extensions */
 _X_HIDDEN void
-driBindExtensions(__GLXscreenConfigs * psc, int dri2)
+driBindExtensions(__GLXscreenConfigs *psc)
 {
    const __DRIextension **extensions;
    int i;
@@ -345,35 +346,13 @@ driBindExtensions(__GLXscreenConfigs * psc, int dri2)
    extensions = psc->core->getExtensions(psc->__driScreen);
 
    for (i = 0; extensions[i]; i++) {
-#ifdef __DRI_COPY_SUB_BUFFER
-      if (strcmp(extensions[i]->name, __DRI_COPY_SUB_BUFFER) == 0) {
-         psc->driCopySubBuffer =
-            (__DRIcopySubBufferExtension *) extensions[i];
-         __glXEnableDirectExtension(psc, "GLX_MESA_copy_sub_buffer");
-      }
-#endif
-
 #ifdef __DRI_SWAP_CONTROL
       /* No DRI2 support for swap_control at the moment, since SwapBuffers
        * is done by the X server */
-      if (strcmp(extensions[i]->name, __DRI_SWAP_CONTROL) == 0 && !dri2) {
-         psc->swapControl = (__DRIswapControlExtension *) extensions[i];
-         __glXEnableDirectExtension(psc, "GLX_SGI_swap_control");
-         __glXEnableDirectExtension(psc, "GLX_MESA_swap_control");
-      }
-#endif
-
-#ifdef __DRI_ALLOCATE
-      if (strcmp(extensions[i]->name, __DRI_ALLOCATE) == 0) {
-         psc->allocate = (__DRIallocateExtension *) extensions[i];
-         __glXEnableDirectExtension(psc, "GLX_MESA_allocate_memory");
-      }
-#endif
-
-#ifdef __DRI_FRAME_TRACKING
-      if (strcmp(extensions[i]->name, __DRI_FRAME_TRACKING) == 0) {
-         psc->frameTracking = (__DRIframeTrackingExtension *) extensions[i];
-         __glXEnableDirectExtension(psc, "GLX_MESA_swap_frame_usage");
+      if (strcmp(extensions[i]->name, __DRI_SWAP_CONTROL) == 0) {
+	 psc->swapControl = (__DRIswapControlExtension *) extensions[i];
+	 __glXEnableDirectExtension(psc, "GLX_SGI_swap_control");
+	 __glXEnableDirectExtension(psc, "GLX_MESA_swap_control");
       }
 #endif
 
@@ -390,23 +369,77 @@ driBindExtensions(__GLXscreenConfigs * psc, int dri2)
        * GLX_OML_sync_control if implemented. */
 #endif
 
-#ifdef __DRI_READ_DRAWABLE
-      if (strcmp(extensions[i]->name, __DRI_READ_DRAWABLE) == 0) {
-         __glXEnableDirectExtension(psc, "GLX_SGI_make_current_read");
-      }
-#endif
+      /* Ignore unknown extensions */
+   }
+}
 
+/* Bind DRI2 specific extensions */
+_X_HIDDEN void
+dri2BindExtensions(__GLXscreenConfigs *psc)
+{
+   const __DRIextension **extensions;
+   int i;
+
+   extensions = psc->core->getExtensions(psc->__driScreen);
+
+   for (i = 0; extensions[i]; i++) {
 #ifdef __DRI_TEX_BUFFER
-      if ((strcmp(extensions[i]->name, __DRI_TEX_BUFFER) == 0) && dri2) {
-         psc->texBuffer = (__DRItexBufferExtension *) extensions[i];
-         __glXEnableDirectExtension(psc, "GLX_EXT_texture_from_pixmap");
+      if ((strcmp(extensions[i]->name, __DRI_TEX_BUFFER) == 0)) {
+	 psc->texBuffer = (__DRItexBufferExtension *) extensions[i];
+	 __glXEnableDirectExtension(psc, "GLX_EXT_texture_from_pixmap");
       }
 #endif
 
+      __glXEnableDirectExtension(psc, "GLX_SGI_video_sync");
+      __glXEnableDirectExtension(psc, "GLX_SGI_swap_control");
+      __glXEnableDirectExtension(psc, "GLX_MESA_swap_control");
+
+      /* FIXME: if DRI2 version supports it... */
+      __glXEnableDirectExtension(psc, "INTEL_swap_event");
+
 #ifdef __DRI2_FLUSH
-      if ((strcmp(extensions[i]->name, __DRI2_FLUSH) == 0) && dri2) {
-         psc->f = (__DRI2flushExtension *) extensions[i];
-         /* internal driver extension, no GL extension exposed */
+      if ((strcmp(extensions[i]->name, __DRI2_FLUSH) == 0)) {
+	 psc->f = (__DRI2flushExtension *) extensions[i];
+	 /* internal driver extension, no GL extension exposed */
+      }
+#endif
+   }
+}
+
+/* Bind extensions common to DRI1 and DRI2 */
+_X_HIDDEN void
+driBindCommonExtensions(__GLXscreenConfigs *psc)
+{
+   const __DRIextension **extensions;
+   int i;
+
+   extensions = psc->core->getExtensions(psc->__driScreen);
+
+   for (i = 0; extensions[i]; i++) {
+#ifdef __DRI_COPY_SUB_BUFFER
+      if (strcmp(extensions[i]->name, __DRI_COPY_SUB_BUFFER) == 0) {
+	 psc->driCopySubBuffer = (__DRIcopySubBufferExtension *) extensions[i];
+	 __glXEnableDirectExtension(psc, "GLX_MESA_copy_sub_buffer");
+      }
+#endif
+
+#ifdef __DRI_ALLOCATE
+      if (strcmp(extensions[i]->name, __DRI_ALLOCATE) == 0) {
+	 psc->allocate = (__DRIallocateExtension *) extensions[i];
+	 __glXEnableDirectExtension(psc, "GLX_MESA_allocate_memory");
+      }
+#endif
+
+#ifdef __DRI_FRAME_TRACKING
+      if (strcmp(extensions[i]->name, __DRI_FRAME_TRACKING) == 0) {
+	 psc->frameTracking = (__DRIframeTrackingExtension *) extensions[i];
+	 __glXEnableDirectExtension(psc, "GLX_MESA_swap_frame_usage");
+      }
+#endif
+
+#ifdef __DRI_READ_DRAWABLE
+      if (strcmp(extensions[i]->name, __DRI_READ_DRAWABLE) == 0) {
+	 __glXEnableDirectExtension(psc, "GLX_SGI_make_current_read");
       }
 #endif
 
diff --git a/src/glx/x11/dri_common.h b/src/glx/x11/dri_common.h
index 61ac9c64160..bb178db7875 100644
--- a/src/glx/x11/dri_common.h
+++ b/src/glx/x11/dri_common.h
@@ -56,6 +56,8 @@ extern void ErrorMessageF(const char *f, ...);
 
 extern void *driOpenDriver(const char *driverName);
 
-extern void driBindExtensions(__GLXscreenConfigs * psc, int dri2);
+extern void driBindExtensions(__GLXscreenConfigs * psc);
+extern void dri2BindExtensions(__GLXscreenConfigs * psc);
+extern void driBindCommonExtensions(__GLXscreenConfigs * psc);
 
 #endif /* _DRI_COMMON_H */
diff --git a/src/glx/x11/dri_glx.c b/src/glx/x11/dri_glx.c
index a890feeab20..e658644eca5 100644
--- a/src/glx/x11/dri_glx.c
+++ b/src/glx/x11/dri_glx.c
@@ -607,10 +607,12 @@ driCreateDrawable(__GLXscreenConfigs * psc,
    return pdraw;
 }
 
-static void
-driSwapBuffers(__GLXDRIdrawable * pdraw)
+static int64_t
+driSwapBuffers(__GLXDRIdrawable * pdraw, int64_t unused1, int64_t unused2,
+	       int64_t unused3)
 {
    (*pdraw->psc->core->swapBuffers) (pdraw->driDrawable);
+   return 0;
 }
 
 static void
@@ -670,9 +672,9 @@ driCreateScreen(__GLXscreenConfigs * psc, int screen,
 
    for (i = 0; extensions[i]; i++) {
       if (strcmp(extensions[i]->name, __DRI_CORE) == 0)
-         psc->core = (__DRIcoreExtension *) extensions[i];
+	 psc->core = (__DRIcoreExtension *) extensions[i];
       if (strcmp(extensions[i]->name, __DRI_LEGACY) == 0)
-         psc->legacy = (__DRIlegacyExtension *) extensions[i];
+	 psc->legacy = (__DRIlegacyExtension *) extensions[i];
    }
 
    if (psc->core == NULL || psc->legacy == NULL) {
@@ -688,7 +690,9 @@ driCreateScreen(__GLXscreenConfigs * psc, int screen,
       return NULL;
    }
 
-   driBindExtensions(psc, 0);
+   driBindExtensions(psc);
+   driBindCommonExtensions(psc);
+
    if (psc->driCopySubBuffer)
       psp->copySubBuffer = driCopySubBuffer;
 
diff --git a/src/glx/x11/drisw_glx.c b/src/glx/x11/drisw_glx.c
index 1866b2cc870..6a51d748afe 100644
--- a/src/glx/x11/drisw_glx.c
+++ b/src/glx/x11/drisw_glx.c
@@ -398,7 +398,8 @@ driCreateScreen(__GLXscreenConfigs * psc, int screen,
       goto handle_error;
    }
 
-   driBindExtensions(psc, 0);
+   driBindExtensions(psc);
+   driBindCommonExtensions(psc);
 
    psc->configs = driConvertConfigs(psc->core, psc->configs, driver_configs);
    psc->visuals = driConvertConfigs(psc->core, psc->visuals, driver_configs);
diff --git a/src/glx/x11/glxclient.h b/src/glx/x11/glxclient.h
index 00ee14fb05b..ded4f5a434a 100644
--- a/src/glx/x11/glxclient.h
+++ b/src/glx/x11/glxclient.h
@@ -121,26 +121,35 @@ struct __GLXDRIdisplayRec
                                     __GLXdisplayPrivate * priv);
 };
 
-struct __GLXDRIscreenRec
-{
-
-   void (*destroyScreen) (__GLXscreenConfigs * psc);
-
-   __GLXDRIcontext *(*createContext) (__GLXscreenConfigs * psc,
-                                      const __GLcontextModes * mode,
-                                      GLXContext gc,
-                                      GLXContext shareList, int renderType);
-
-   __GLXDRIdrawable *(*createDrawable) (__GLXscreenConfigs * psc,
-                                        XID drawable,
-                                        GLXDrawable glxDrawable,
-                                        const __GLcontextModes * modes);
-
-   void (*swapBuffers) (__GLXDRIdrawable * pdraw);
-   void (*copySubBuffer) (__GLXDRIdrawable * pdraw,
-                          int x, int y, int width, int height);
-   void (*waitX) (__GLXDRIdrawable * pdraw);
-   void (*waitGL) (__GLXDRIdrawable * pdraw);
+struct __GLXDRIscreenRec {
+
+   void (*destroyScreen)(__GLXscreenConfigs *psc);
+
+   __GLXDRIcontext *(*createContext)(__GLXscreenConfigs *psc,
+				     const __GLcontextModes *mode,
+				     GLXContext gc,
+				     GLXContext shareList, int renderType);
+
+   __GLXDRIdrawable *(*createDrawable)(__GLXscreenConfigs *psc,
+				       XID drawable,
+				       GLXDrawable glxDrawable,
+				       const __GLcontextModes *modes);
+
+   int64_t (*swapBuffers)(__GLXDRIdrawable *pdraw, int64_t target_msc,
+			  int64_t divisor, int64_t remainder);
+   void (*copySubBuffer)(__GLXDRIdrawable *pdraw,
+			 int x, int y, int width, int height);
+   void (*waitX)(__GLXDRIdrawable *pdraw);
+   void (*waitGL)(__GLXDRIdrawable *pdraw);
+   int (*getDrawableMSC)(__GLXscreenConfigs *psc, __GLXDRIdrawable *pdraw,
+			 int64_t *ust, int64_t *msc, int64_t *sbc);
+   int (*waitForMSC)(__GLXDRIdrawable *pdraw, int64_t target_msc,
+		     int64_t divisor, int64_t remainder, int64_t *ust,
+		     int64_t *msc, int64_t *sbc);
+   int (*waitForSBC)(__GLXDRIdrawable *pdraw, int64_t target_sbc, int64_t *ust,
+		     int64_t *msc, int64_t *sbc);
+   void (*setSwapInterval)(__GLXDRIdrawable *pdraw, int interval);
+   int (*getSwapInterval)(__GLXDRIdrawable *pdraw);
 };
 
 struct __GLXDRIcontextRec
diff --git a/src/glx/x11/glxcmds.c b/src/glx/x11/glxcmds.c
index d1c68dd02e1..c3be974ea91 100644
--- a/src/glx/x11/glxcmds.c
+++ b/src/glx/x11/glxcmds.c
@@ -982,7 +982,7 @@ glXSwapBuffers(Display * dpy, GLXDrawable drawable)
 
    if (pdraw != NULL) {
       glFlush();
-      (*pdraw->psc->driScreen->swapBuffers) (pdraw);
+      (*pdraw->psc->driScreen->swapBuffers)(pdraw, 0, 0, 0);
       return;
    }
 #endif
@@ -1884,6 +1884,7 @@ __glXSwapIntervalSGI(int interval)
 {
    xGLXVendorPrivateReq *req;
    GLXContext gc = __glXGetCurrentContext();
+   __GLXscreenConfigs *psc;
    Display *dpy;
    CARD32 *interval_ptr;
    CARD8 opcode;
@@ -1898,20 +1899,30 @@ __glXSwapIntervalSGI(int interval)
 
 #ifdef __DRI_SWAP_CONTROL
    if (gc->driContext) {
-      __GLXscreenConfigs *const psc = GetGLXScreenConfigs(gc->currentDpy,
-                                                          gc->screen);
+      __GLXscreenConfigs * const psc = GetGLXScreenConfigs( gc->currentDpy,
+							    gc->screen );
       __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(gc->currentDpy,
-                                                  gc->currentDrawable,
-                                                  NULL);
+						  gc->currentDrawable,
+						  NULL);
       if (psc->swapControl != NULL && pdraw != NULL) {
-         psc->swapControl->setSwapInterval(pdraw->driDrawable, interval);
-         return 0;
+	 psc->swapControl->setSwapInterval(pdraw->driDrawable, interval);
+	 return 0;
       }
-      else {
-         return GLX_BAD_CONTEXT;
+      else if (pdraw == NULL) {
+	 return GLX_BAD_CONTEXT;
       }
    }
 #endif
+   psc = GetGLXScreenConfigs( gc->currentDpy, gc->screen);
+
+   if (gc->driContext && psc->driScreen && psc->driScreen->setSwapInterval) {
+      __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(gc->currentDpy,
+						  gc->currentDrawable,
+						  NULL);
+      psc->driScreen->setSwapInterval(pdraw, interval);
+      return 0;
+   }
+
    dpy = gc->currentDpy;
    opcode = __glXSetupForCommand(dpy);
    if (!opcode) {
@@ -1943,13 +1954,13 @@ __glXSwapIntervalSGI(int interval)
 static int
 __glXSwapIntervalMESA(unsigned int interval)
 {
-#ifdef __DRI_SWAP_CONTROL
    GLXContext gc = __glXGetCurrentContext();
 
    if (interval < 0) {
       return GLX_BAD_VALUE;
    }
 
+#ifdef __DRI_SWAP_CONTROL
    if (gc != NULL && gc->driContext) {
       __GLXscreenConfigs *const psc = GetGLXScreenConfigs(gc->currentDpy,
                                                           gc->screen);
@@ -1963,10 +1974,20 @@ __glXSwapIntervalMESA(unsigned int interval)
          }
       }
    }
-#else
-   (void) interval;
 #endif
 
+   if (gc != NULL && gc->driContext) {
+      __GLXscreenConfigs *psc;
+
+      psc = GetGLXScreenConfigs( gc->currentDpy, gc->screen);
+      if (psc->driScreen && psc->driScreen->setSwapInterval) {
+         __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(gc->currentDpy,
+						     gc->currentDrawable, NULL);
+	 psc->driScreen->setSwapInterval(pdraw, interval);
+	 return 0;
+      }
+   }
+
    return GLX_BAD_CONTEXT;
 }
 
@@ -1990,6 +2011,16 @@ __glXGetSwapIntervalMESA(void)
       }
    }
 #endif
+   if (gc != NULL && gc->driContext) {
+      __GLXscreenConfigs *psc;
+
+      psc = GetGLXScreenConfigs( gc->currentDpy, gc->screen);
+      if (psc->driScreen && psc->driScreen->getSwapInterval) {
+         __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(gc->currentDpy,
+						     gc->currentDrawable, NULL);
+	 return psc->driScreen->getSwapInterval(pdraw);
+      }
+   }
 
    return 0;
 }
@@ -2102,64 +2133,73 @@ __glXQueryFrameTrackingMESA(Display * dpy, GLXDrawable drawable,
 static int
 __glXGetVideoSyncSGI(unsigned int *count)
 {
+   int64_t ust, msc, sbc;
+   int ret;
+   GLXContext gc = __glXGetCurrentContext();
+   __GLXscreenConfigs *psc;
+   __GLXDRIdrawable *pdraw;
+
+   if (!gc || !gc->driContext)
+      return GLX_BAD_CONTEXT;
+
+   psc = GetGLXScreenConfigs(gc->currentDpy, gc->screen);
+   pdraw = GetGLXDRIDrawable(gc->currentDpy, gc->currentDrawable, NULL);
+
    /* FIXME: Looking at the GLX_SGI_video_sync spec in the extension registry,
     * FIXME: there should be a GLX encoding for this call.  I can find no
     * FIXME: documentation for the GLX encoding.
     */
 #ifdef __DRI_MEDIA_STREAM_COUNTER
-   GLXContext gc = __glXGetCurrentContext();
-
-
-   if (gc != NULL && gc->driContext) {
-      __GLXscreenConfigs *const psc = GetGLXScreenConfigs(gc->currentDpy,
-                                                          gc->screen);
-      if (psc->msc && psc->driScreen) {
-         __GLXDRIdrawable *pdraw =
-            GetGLXDRIDrawable(gc->currentDpy, gc->currentDrawable, NULL);
-         int64_t temp;
-         int ret;
+   if ( psc->msc && psc->driScreen ) {
+      ret = (*psc->msc->getDrawableMSC)(psc->__driScreen,
+					pdraw->driDrawable, &msc);
+      *count = (unsigned) msc;
 
-         ret = (*psc->msc->getDrawableMSC) (psc->__driScreen,
-                                            pdraw->driDrawable, &temp);
-         *count = (unsigned) temp;
-
-         return (ret == 0) ? 0 : GLX_BAD_CONTEXT;
-      }
+      return (ret == 0) ? 0 : GLX_BAD_CONTEXT;
    }
-#else
-   (void) count;
 #endif
+   if (psc->driScreen && psc->driScreen->getDrawableMSC) {
+      ret = psc->driScreen->getDrawableMSC(psc, pdraw, &ust, &msc, &sbc);
+      *count = (unsigned) msc;
+      return (ret == True) ? 0 : GLX_BAD_CONTEXT;
+   }
+
    return GLX_BAD_CONTEXT;
 }
 
 static int
 __glXWaitVideoSyncSGI(int divisor, int remainder, unsigned int *count)
 {
-#ifdef __DRI_MEDIA_STREAM_COUNTER
    GLXContext gc = __glXGetCurrentContext();
+   __GLXscreenConfigs *psc;
+   __GLXDRIdrawable *pdraw;
+   int64_t ust, msc, sbc;
+   int ret;
 
    if (divisor <= 0 || remainder < 0)
       return GLX_BAD_VALUE;
 
-   if (gc != NULL && gc->driContext) {
-      __GLXscreenConfigs *const psc = GetGLXScreenConfigs(gc->currentDpy,
-                                                          gc->screen);
-      if (psc->msc != NULL && psc->driScreen) {
-         __GLXDRIdrawable *pdraw =
-            GetGLXDRIDrawable(gc->currentDpy, gc->currentDrawable, NULL);
-         int ret;
-         int64_t msc;
-         int64_t sbc;
-
-         ret = (*psc->msc->waitForMSC) (pdraw->driDrawable, 0,
-                                        divisor, remainder, &msc, &sbc);
-         *count = (unsigned) msc;
-         return (ret == 0) ? 0 : GLX_BAD_CONTEXT;
-      }
+   if (!gc || !gc->driContext)
+      return GLX_BAD_CONTEXT;
+
+   psc = GetGLXScreenConfigs( gc->currentDpy, gc->screen);
+   pdraw = GetGLXDRIDrawable(gc->currentDpy, gc->currentDrawable, NULL);
+
+#ifdef __DRI_MEDIA_STREAM_COUNTER
+   if (psc->msc != NULL && psc->driScreen ) {
+      ret = (*psc->msc->waitForMSC)(pdraw->driDrawable, 0,
+				    divisor, remainder, &msc, &sbc);
+      *count = (unsigned) msc;
+      return (ret == 0) ? 0 : GLX_BAD_CONTEXT;
    }
-#else
-   (void) count;
 #endif
+   if (psc->driScreen && psc->driScreen->waitForMSC) {
+      ret = psc->driScreen->waitForMSC(pdraw, 0, divisor, remainder, &ust, &msc,
+				       &sbc);
+      *count = (unsigned) msc;
+      return (ret == True) ? 0 : GLX_BAD_CONTEXT;
+   }
+
    return GLX_BAD_CONTEXT;
 }
 
@@ -2312,27 +2352,29 @@ static Bool
 __glXGetSyncValuesOML(Display * dpy, GLXDrawable drawable,
                       int64_t * ust, int64_t * msc, int64_t * sbc)
 {
-#if defined(__DRI_SWAP_BUFFER_COUNTER) && defined(__DRI_MEDIA_STREAM_COUNTER)
-   __GLXdisplayPrivate *const priv = __glXInitialize(dpy);
+   __GLXdisplayPrivate * const priv = __glXInitialize(dpy);
+   int i, ret;
+   __GLXDRIdrawable *pdraw;
+   __GLXscreenConfigs *psc;
 
-   if (priv != NULL) {
-      int i;
-      __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(dpy, drawable, &i);
-      __GLXscreenConfigs *const psc = &priv->screenConfigs[i];
+   if (!priv)
+      return False;
 
-      assert((pdraw == NULL) || (i != -1));
-      return ((pdraw && psc->sbc && psc->msc)
-              && ((*psc->msc->getMSC) (psc->driScreen, msc) == 0)
-              && ((*psc->sbc->getSBC) (pdraw->driDrawable, sbc) == 0)
-              && (__glXGetUST(ust) == 0));
-   }
-#else
-   (void) dpy;
-   (void) drawable;
-   (void) ust;
-   (void) msc;
-   (void) sbc;
+   pdraw = GetGLXDRIDrawable(dpy, drawable, &i);
+   psc = &priv->screenConfigs[i];
+
+#if defined(__DRI_SWAP_BUFFER_COUNTER) && defined(__DRI_MEDIA_STREAM_COUNTER)
+   if (pdraw && psc->sbc && psc->sbc)
+      return ( (pdraw && psc->sbc && psc->msc)
+	       && ((*psc->msc->getMSC)(psc->driScreen, msc) == 0)
+	       && ((*psc->sbc->getSBC)(pdraw->driDrawable, sbc) == 0)
+	       && (__glXGetUST(ust) == 0) );
 #endif
+   if (pdraw && psc && psc->driScreen && psc->driScreen->getDrawableMSC) {
+      ret = psc->driScreen->getDrawableMSC(psc, pdraw, ust, msc, sbc);
+      return ret;
+   }
+
    return False;
 }
 
@@ -2440,11 +2482,14 @@ static int64_t
 __glXSwapBuffersMscOML(Display * dpy, GLXDrawable drawable,
                        int64_t target_msc, int64_t divisor, int64_t remainder)
 {
-#ifdef __DRI_SWAP_BUFFER_COUNTER
+   GLXContext gc = __glXGetCurrentContext();
    int screen;
    __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(dpy, drawable, &screen);
    __GLXscreenConfigs *const psc = GetGLXScreenConfigs(dpy, screen);
 
+   if (!pdraw || !gc->driContext) /* no GLX for this */
+      return -1;
+
    /* The OML_sync_control spec says these should "generate a GLX_BAD_VALUE
     * error", but it also says "It [glXSwapBuffersMscOML] will return a value
     * of -1 if the function failed because of errors detected in the input
@@ -2455,18 +2500,19 @@ __glXSwapBuffersMscOML(Display * dpy, GLXDrawable drawable,
    if (divisor > 0 && remainder >= divisor)
       return -1;
 
-   if (pdraw != NULL && psc->counters != NULL)
-      return (*psc->sbc->swapBuffersMSC) (pdraw->driDrawable, target_msc,
-                                          divisor, remainder);
+#ifdef __DRI_SWAP_BUFFER_COUNTER
+   if (psc->counters != NULL)
+      return (*psc->sbc->swapBuffersMSC)(pdraw->driDrawable, target_msc,
+					 divisor, remainder);
+#endif
 
-#else
-   (void) dpy;
-   (void) drawable;
-   (void) target_msc;
-   (void) divisor;
-   (void) remainder;
+#ifdef GLX_DIRECT_RENDERING
+   if (psc->driScreen && psc->driScreen->swapBuffers)
+      return (*psc->driScreen->swapBuffers)(pdraw, target_msc, divisor,
+					    remainder);
 #endif
-   return 0;
+
+   return -1;
 }
 
 
@@ -2476,12 +2522,14 @@ __glXWaitForMscOML(Display * dpy, GLXDrawable drawable,
                    int64_t remainder, int64_t * ust,
                    int64_t * msc, int64_t * sbc)
 {
-#ifdef __DRI_MEDIA_STREAM_COUNTER
-   int screen = 0;
+   int screen;
    __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(dpy, drawable, &screen);
-   __GLXscreenConfigs *const psc = GetGLXScreenConfigs(dpy, screen);
+   __GLXscreenConfigs * const psc = GetGLXScreenConfigs( dpy, screen );
    int ret;
 
+   fprintf(stderr, "waitmsc: %lld, %lld, %lld\n", target_msc, divisor,
+	   remainder);
+
    /* The OML_sync_control spec says these should "generate a GLX_BAD_VALUE
     * error", but the return type in the spec is Bool.
     */
@@ -2490,7 +2538,9 @@ __glXWaitForMscOML(Display * dpy, GLXDrawable drawable,
    if (divisor > 0 && remainder >= divisor)
       return False;
 
+#ifdef __DRI_MEDIA_STREAM_COUNTER
    if (pdraw != NULL && psc->msc != NULL) {
+      fprintf(stderr, "dri1 msc\n");
       ret = (*psc->msc->waitForMSC) (pdraw->driDrawable, target_msc,
                                      divisor, remainder, msc, sbc);
 
@@ -2499,16 +2549,14 @@ __glXWaitForMscOML(Display * dpy, GLXDrawable drawable,
        */
       return ((ret == 0) && (__glXGetUST(ust) == 0));
    }
-#else
-   (void) dpy;
-   (void) drawable;
-   (void) target_msc;
-   (void) divisor;
-   (void) remainder;
-   (void) ust;
-   (void) msc;
-   (void) sbc;
 #endif
+   if (pdraw && psc->driScreen && psc->driScreen->waitForMSC) {
+      ret = psc->driScreen->waitForMSC(pdraw, target_msc, divisor, remainder,
+				       ust, msc, sbc);
+      return ret;
+   }
+
+   fprintf(stderr, "no drawable??\n");
    return False;
 }
 
@@ -2518,7 +2566,6 @@ __glXWaitForSbcOML(Display * dpy, GLXDrawable drawable,
                    int64_t target_sbc, int64_t * ust,
                    int64_t * msc, int64_t * sbc)
 {
-#ifdef __DRI_SWAP_BUFFER_COUNTER
    int screen;
    __GLXDRIdrawable *pdraw = GetGLXDRIDrawable(dpy, drawable, &screen);
    __GLXscreenConfigs *const psc = GetGLXScreenConfigs(dpy, screen);
@@ -2529,7 +2576,7 @@ __glXWaitForSbcOML(Display * dpy, GLXDrawable drawable,
     */
    if (target_sbc < 0)
       return False;
-
+#ifdef __DRI_SWAP_BUFFER_COUNTER
    if (pdraw != NULL && psc->sbc != NULL) {
       ret =
          (*psc->sbc->waitForSBC) (pdraw->driDrawable, target_sbc, msc, sbc);
@@ -2539,14 +2586,11 @@ __glXWaitForSbcOML(Display * dpy, GLXDrawable drawable,
        */
       return ((ret == 0) && (__glXGetUST(ust) == 0));
    }
-#else
-   (void) dpy;
-   (void) drawable;
-   (void) target_sbc;
-   (void) ust;
-   (void) msc;
-   (void) sbc;
 #endif
+   if (pdraw && psc->driScreen && psc->driScreen->waitForMSC) {
+      ret = psc->driScreen->waitForSBC(pdraw, target_sbc, ust, msc, sbc);
+      return ret;
+   }
    return False;
 }
 
diff --git a/src/glx/x11/glxext.c b/src/glx/x11/glxext.c
index 5633a3e4a29..fe65216c411 100644
--- a/src/glx/x11/glxext.c
+++ b/src/glx/x11/glxext.c
@@ -101,6 +101,10 @@ __glXCloseDisplay(Display * dpy, XExtCodes * codes)
 static
 XEXT_GENERATE_ERROR_STRING(__glXErrorString, __glXExtensionName,
                            __GLX_NUMBER_ERRORS, error_list)
+static Bool
+__glXWireToEvent(Display *dpy, XEvent *event, xEvent *wire);
+static Status
+__glXEventToWire(Display *dpy, XEvent *event, xEvent *wire);
 
 static /* const */ XExtensionHooks __glXExtensionHooks = {
   NULL,                   /* create_gc */
@@ -110,8 +114,8 @@ static /* const */ XExtensionHooks __glXExtensionHooks = {
   NULL,                   /* create_font */
   NULL,                   /* free_font */
   __glXCloseDisplay,      /* close_display */
-  NULL,                   /* wire_to_event */
-  NULL,                   /* event_to_wire */
+  __glXWireToEvent,       /* wire_to_event */
+  __glXEventToWire,       /* event_to_wire */
   NULL,                   /* error */
   __glXErrorString,       /* error_string */
 };
@@ -121,6 +125,89 @@ XEXT_GENERATE_FIND_DISPLAY(__glXFindDisplay, __glXExtensionInfo,
                            __glXExtensionName, &__glXExtensionHooks,
                            __GLX_NUMBER_EVENTS, NULL)
 
+/*
+ * GLX events are a bit funky.  We don't stuff the X event code into
+ * our user exposed (via XNextEvent) structure.  Instead we use the GLX
+ * private event code namespace (and hope it doesn't conflict).  Clients
+ * have to know that bit 15 in the event type field means they're getting
+ * a GLX event, and then handle the various sub-event types there, rather
+ * than simply checking the event code and handling it directly.
+ */
+
+static Bool
+__glXWireToEvent(Display *dpy, XEvent *event, xEvent *wire)
+{
+   XExtDisplayInfo *info = __glXFindDisplay(dpy);
+
+   XextCheckExtension(dpy, info, __glXExtensionName, False);
+
+   switch ((wire->u.u.type & 0x7f) - info->codes->first_event) {
+   case GLX_PbufferClobber:
+   {
+      GLXPbufferClobberEvent *aevent = (GLXPbufferClobberEvent *)event;
+      xGLXPbufferClobberEvent *awire = (xGLXPbufferClobberEvent *)wire;
+      aevent->event_type = awire->type;
+      aevent->serial = awire->sequenceNumber;
+      aevent->event_type = awire->event_type;
+      aevent->draw_type = awire->draw_type;
+      aevent->drawable = awire->drawable;
+      aevent->buffer_mask = awire->buffer_mask;
+      aevent->aux_buffer = awire->aux_buffer;
+      aevent->x = awire->x;
+      aevent->y = awire->y;
+      aevent->width = awire->width;
+      aevent->height = awire->height;
+      aevent->count = awire->count;
+      return True;
+   }
+   case GLX_BufferSwapComplete:
+   {
+      GLXBufferSwapComplete *aevent = (GLXBufferSwapComplete *)event;
+      xGLXBufferSwapComplete *awire = (xGLXBufferSwapComplete *)wire;
+      aevent->event_type = awire->event_type;
+      aevent->drawable = awire->drawable;
+      aevent->ust = ((CARD64)awire->ust_hi << 32) | awire->ust_lo;
+      aevent->msc = ((CARD64)awire->msc_hi << 32) | awire->msc_lo;
+      aevent->sbc = ((CARD64)awire->sbc_hi << 32) | awire->sbc_lo;
+      return True;
+   }
+   default:
+      /* client doesn't support server event */
+      break;
+   }
+
+   return False;
+}
+
+/* We don't actually support this.  It doesn't make sense for clients to
+ * send each other GLX events.
+ */
+static Status
+__glXEventToWire(Display *dpy, XEvent *event, xEvent *wire)
+{
+   XExtDisplayInfo *info = __glXFindDisplay(dpy);
+
+   XextCheckExtension(dpy, info, __glXExtensionName, False);
+
+   switch (event->type) {
+   case GLX_DAMAGED:
+      break;
+   case GLX_SAVED:
+      break;
+   case GLX_EXCHANGE_COMPLETE:
+      break;
+   case GLX_BLIT_COMPLETE:
+      break;
+   case GLX_FLIP_COMPLETE:
+      break;
+   default:
+      /* client doesn't support server event */
+      break;
+   }
+
+   return Success;
+}
+
 /************************************************************************/
 /*
 ** Free the per screen configs data as well as the array of
diff --git a/src/glx/x11/glxextensions.c b/src/glx/x11/glxextensions.c
index 6852128e2ae..25a5c492930 100644
--- a/src/glx/x11/glxextensions.c
+++ b/src/glx/x11/glxextensions.c
@@ -104,6 +104,7 @@ static const struct extension_info known_glx_extensions[] = {
    { GLX(SGIX_swap_group),             VER(0,0), N, N, N, N },
    { GLX(SGIX_visual_select_group),    VER(0,0), Y, Y, N, N },
    { GLX(EXT_texture_from_pixmap),     VER(0,0), Y, N, N, N },
+   { GLX(INTEL_swap_event),            VER(1,4), Y, Y, N, N },
    { NULL }
 };
 
diff --git a/src/glx/x11/glxextensions.h b/src/glx/x11/glxextensions.h
index 652c5db1c81..f556b1239c9 100644
--- a/src/glx/x11/glxextensions.h
+++ b/src/glx/x11/glxextensions.h
@@ -65,7 +65,8 @@ enum
    SGIX_swap_barrier_bit,
    SGIX_swap_group_bit,
    SGIX_visual_select_group_bit,
-   EXT_texture_from_pixmap_bit
+   EXT_texture_from_pixmap_bit,
+   INTEL_swap_event_bit,
 };
 
 enum
diff --git a/src/mesa/Makefile b/src/mesa/Makefile
index a815f46b4a1..f845d93fbd3 100644
--- a/src/mesa/Makefile
+++ b/src/mesa/Makefile
@@ -42,7 +42,7 @@ libglapi.a: $(GLAPI_OBJECTS)
 
 ######################################################################
 # Device drivers
-driver_subdirs: libmesa.a libglapi.a
+driver_subdirs: libmesa.a libglapi.a libmesagallium.a
 	@ (cd drivers && $(MAKE))
 
 
diff --git a/src/mesa/drivers/common/meta.c b/src/mesa/drivers/common/meta.c
index c4dbfa6d7d9..e500359bb78 100644
--- a/src/mesa/drivers/common/meta.c
+++ b/src/mesa/drivers/common/meta.c
@@ -510,9 +510,9 @@ _mesa_meta_begin(GLcontext *ctx, GLbitfield state)
       _mesa_LoadIdentity();
       _mesa_MatrixMode(GL_PROJECTION);
       _mesa_LoadIdentity();
-      _mesa_Ortho(0.0F, ctx->DrawBuffer->Width,
-                  0.0F, ctx->DrawBuffer->Height,
-                  -1.0F, 1.0F);
+      _mesa_Ortho(0.0, ctx->DrawBuffer->Width,
+                  0.0, ctx->DrawBuffer->Height,
+                  -1.0, 1.0);
       save->ClipPlanesEnabled = ctx->Transform.ClipPlanesEnabled;
       if (ctx->Transform.ClipPlanesEnabled) {
          GLuint i;
@@ -2321,6 +2321,26 @@ _mesa_meta_GenerateMipmap(GLcontext *ctx, GLenum target,
 
    _mesa_set_enable(ctx, target, GL_TRUE);
 
+   /* setup vertex positions */
+   {
+      verts[0].x = 0.0F;
+      verts[0].y = 0.0F;
+      verts[1].x = 1.0F;
+      verts[1].y = 0.0F;
+      verts[2].x = 1.0F;
+      verts[2].y = 1.0F;
+      verts[3].x = 0.0F;
+      verts[3].y = 1.0F;
+      
+      /* upload new vertex data */
+      _mesa_BufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, sizeof(verts), verts);
+   }
+
+   /* setup projection matrix */
+   _mesa_MatrixMode(GL_PROJECTION);
+   _mesa_LoadIdentity();
+   _mesa_Ortho(0.0, 1.0, 0.0, 1.0, -1.0, 1.0);
+
    /* texture is already locked, unlock now */
    _mesa_unlock_texture(ctx, texObj);
 
@@ -2387,21 +2407,6 @@ _mesa_meta_GenerateMipmap(GLcontext *ctx, GLenum target,
          }
       }
 
-      /* setup vertex positions */
-      {
-         verts[0].x = 0.0F;
-         verts[0].y = 0.0F;
-         verts[1].x = (GLfloat) dstWidth;
-         verts[1].y = 0.0F;
-         verts[2].x = (GLfloat) dstWidth;
-         verts[2].y = (GLfloat) dstHeight;
-         verts[3].x = 0.0F;
-         verts[3].y = (GLfloat) dstHeight;
-
-         /* upload new vertex data */
-         _mesa_BufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, sizeof(verts), verts);
-      }
-
       /* limit sampling to src level */
       _mesa_TexParameteri(target, GL_TEXTURE_BASE_LEVEL, srcLevel);
       _mesa_TexParameteri(target, GL_TEXTURE_MAX_LEVEL, srcLevel);
@@ -2440,6 +2445,12 @@ _mesa_meta_GenerateMipmap(GLcontext *ctx, GLenum target,
          break;
       }
 
+      assert(dstWidth == ctx->DrawBuffer->Width);
+      assert(dstHeight == ctx->DrawBuffer->Height);
+
+      /* setup viewport */
+      _mesa_set_viewport(ctx, 0, 0, dstWidth, dstHeight);
+
       _mesa_DrawArrays(GL_TRIANGLE_FAN, 0, 4);
    }
 
diff --git a/src/mesa/drivers/dri/common/dri_util.c b/src/mesa/drivers/dri/common/dri_util.c
index 0e01d742656..3649c296662 100644
--- a/src/mesa/drivers/dri/common/dri_util.c
+++ b/src/mesa/drivers/dri/common/dri_util.c
@@ -454,6 +454,7 @@ driCreateNewDrawable(__DRIscreen *psp, const __DRIconfig *config,
 
     pdp->driScreenPriv = psp;
     pdp->driContextPriv = &psp->dummyContextPriv;
+    pdp->validBuffers = GL_FALSE;
 
     if (!(*psp->DriverAPI.CreateBuffer)(psp, pdp, &config->modes,
 					renderType == GLX_PIXMAP_BIT)) {
diff --git a/src/mesa/drivers/dri/common/dri_util.h b/src/mesa/drivers/dri/common/dri_util.h
index 35d8b8ff932..95df702f1ae 100644
--- a/src/mesa/drivers/dri/common/dri_util.h
+++ b/src/mesa/drivers/dri/common/dri_util.h
@@ -376,6 +376,8 @@ struct __DRIdrawableRec {
      * GLX_MESA_swap_control.
      */
     unsigned int swap_interval;
+
+    GLboolean validBuffers;
 };
 
 /**
diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c
index cc1052f757c..f316e0cda49 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c
@@ -692,7 +692,7 @@ void emit_xpd(struct brw_compile *p,
 {
    GLuint i;
 
-   assert(!(mask & WRITEMASK_W) == WRITEMASK_X);
+   assert((mask & WRITEMASK_W) != WRITEMASK_W);
    
    for (i = 0 ; i < 3; i++) {
       if (mask & (1<<i)) {
diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c
index 13a28f0419d..3f6634c65a7 100644
--- a/src/mesa/drivers/dri/intel/intel_context.c
+++ b/src/mesa/drivers/dri/intel/intel_context.c
@@ -70,8 +70,6 @@ int INTEL_DEBUG = (0);
 #define DRIVER_DATE_GEM                 "GEM " DRIVER_DATE
 
 
-static void intel_flush(GLcontext *ctx, GLboolean needs_mi_flush);
-
 static const GLubyte *
 intelGetString(GLcontext * ctx, GLenum name)
 {
@@ -378,6 +376,7 @@ intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable)
        }
    }
 
+   drawable->validBuffers = GL_TRUE;
    driUpdateFramebufferSize(&intel->ctx, drawable);
 }
 
@@ -461,7 +460,7 @@ intelInvalidateState(GLcontext * ctx, GLuint new_state)
       intel->vtbl.invalidate_state( intel, new_state );
 }
 
-static void
+void
 intel_flush(GLcontext *ctx, GLboolean needs_mi_flush)
 {
    struct intel_context *intel = intel_context(ctx);
diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h
index c7b72358362..07207bfbec9 100644
--- a/src/mesa/drivers/dri/intel/intel_context.h
+++ b/src/mesa/drivers/dri/intel/intel_context.h
@@ -412,6 +412,7 @@ extern GLboolean intelInitContext(struct intel_context *intel,
 
 extern void intelFinish(GLcontext * ctx);
 extern void intelFlush(GLcontext * ctx);
+extern void intel_flush(GLcontext * ctx, GLboolean needs_mi_flush);
 
 extern void intelInitDriverFunctions(struct dd_function_table *functions);
 
diff --git a/src/mesa/drivers/dri/intel/intel_extensions.h b/src/mesa/drivers/dri/intel/intel_extensions.h
index 1d1c97a4a9b..e78e07356e7 100644
--- a/src/mesa/drivers/dri/intel/intel_extensions.h
+++ b/src/mesa/drivers/dri/intel/intel_extensions.h
@@ -32,5 +32,8 @@
 extern void
 intelInitExtensions(GLcontext *ctx);
 
+extern void
+intelFlushDrawable(__DRIdrawable *drawable);
+
 
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c
index 5165716e092..e240957197d 100644
--- a/src/mesa/drivers/dri/intel/intel_screen.c
+++ b/src/mesa/drivers/dri/intel/intel_screen.c
@@ -113,10 +113,36 @@ static const __DRItexBufferExtension intelTexBufferExtension = {
    intelSetTexBuffer2,
 };
 
+static void
+intelDRI2Flush(__DRIdrawable *drawable)
+{
+   struct intel_context *intel = drawable->driContextPriv->driverPrivate;
+
+   if (intel->gen < 4)
+      INTEL_FIREVERTICES(intel);
+
+   if (intel->batch->map != intel->batch->ptr)
+      intel_batchbuffer_flush(intel->batch);
+}
+
+static void
+intelDRI2FlushInvalidate(__DRIdrawable *drawable)
+{
+   intelDRI2Flush(drawable);
+   drawable->validBuffers = GL_FALSE;
+}
+
+static const struct __DRI2flushExtensionRec intelFlushExtension = {
+    { __DRI2_FLUSH, __DRI2_FLUSH_VERSION },
+    intelDRI2Flush,
+    intelDRI2FlushInvalidate,
+};
+
 static const __DRIextension *intelScreenExtensions[] = {
     &driReadDrawableExtension,
     &intelTexOffsetExtension.base,
     &intelTexBufferExtension.base,
+    &intelFlushExtension.base,
     NULL
 };
 
@@ -311,11 +337,10 @@ __DRIconfig **intelInitScreen2(__DRIscreen *psp)
    intelScreenPrivate *intelScreen;
    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
+       GLX_NONE, GLX_SWAP_UNDEFINED_OML,
+       GLX_SWAP_EXCHANGE_OML, GLX_SWAP_COPY_OML
    };
    uint8_t depth_bits[4], stencil_bits[4], msaa_samples_array[1];
    int color;
diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c
index 307669f87ec..6f41eafd0ef 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_image.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_image.c
@@ -742,7 +742,8 @@ intelSetTexBuffer2(__DRIcontext *pDRICtx, GLint target,
    if (!intelObj)
       return;
 
-   intel_update_renderbuffers(pDRICtx, dPriv);
+   if (!dPriv->validBuffers)
+      intel_update_renderbuffers(pDRICtx, dPriv);
 
    rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
    /* If the region isn't set, then intel_update_renderbuffers was unable
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.c b/src/mesa/drivers/dri/r300/compiler/radeon_program.c
index 0dbc5380bb5..a3c41d7bd44 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.c
@@ -94,10 +94,11 @@ unsigned int rc_find_free_temporary(struct radeon_compiler * c)
 {
 	char used[RC_REGISTER_MAX_INDEX];
 	unsigned int i;
+	struct rc_instruction * rcinst;
 
 	memset(used, 0, sizeof(used));
 
-	for (struct rc_instruction * rcinst = c->Program.Instructions.Next; rcinst != &c->Program.Instructions; rcinst = rcinst->Next) {
+	for (rcinst = c->Program.Instructions.Next; rcinst != &c->Program.Instructions; rcinst = rcinst->Next) {
 		const struct rc_sub_instruction *inst = &rcinst->U.I;
 		const struct rc_opcode_info *opcode = rc_get_opcode_info(inst->Opcode);
 		unsigned int k;
@@ -168,8 +169,9 @@ void rc_remove_instruction(struct rc_instruction * inst)
 unsigned int rc_recompute_ips(struct radeon_compiler * c)
 {
 	unsigned int ip = 0;
+	struct rc_instruction * inst;
 
-	for(struct rc_instruction * inst = c->Program.Instructions.Next;
+	for(inst = c->Program.Instructions.Next;
 	    inst != &c->Program.Instructions;
 	    inst = inst->Next) {
 		inst->IP = ip++;
diff --git a/src/mesa/drivers/dri/r300/r300_blit.c b/src/mesa/drivers/dri/r300/r300_blit.c
index ea626d942dc..2eec27e900f 100644
--- a/src/mesa/drivers/dri/r300/r300_blit.c
+++ b/src/mesa/drivers/dri/r300/r300_blit.c
@@ -181,8 +181,6 @@ static uint32_t mesa_format_to_us_format(gl_format mesa_format)
 {
     switch(mesa_format)
     {
-        case MESA_FORMAT_S8_Z24:
-        case MESA_FORMAT_X8_Z24:
         case MESA_FORMAT_RGBA8888: // x
             return EASY_US_FORMAT(R500_OUT_FMT_C4_8, A, B, G, R, 0);
         case MESA_FORMAT_RGB565: // x
@@ -216,7 +214,8 @@ static uint32_t mesa_format_to_us_format(gl_format mesa_format)
             return EASY_US_FORMAT(R500_OUT_FMT_C4_16, R, G, B, A, 0xf);
 
         default:
-            assert(!"Invalid format for US output\n");
+            fprintf(stderr, "Unsupported format %s for US output\n", _mesa_get_format_name(mesa_format));
+            assert(0);
             return 0;
     }
 }
@@ -541,6 +540,19 @@ GLboolean r300_blit(struct r300_context *r300,
                     unsigned reg_height,
                     unsigned flip_y)
 {
+    if (_mesa_get_format_bits(src_mesaformat, GL_DEPTH_BITS) > 0)
+        return GL_FALSE;
+
+    /* Make sure that colorbuffer has even width - hw limitation */
+    if (dst_pitch % 2 > 0)
+        ++dst_pitch;
+
+    /* Rendering to small buffer doesn't work.
+     * Looks like a hw limitation.
+     */
+    if (dst_pitch < 32)
+        return GL_FALSE;
+
     /* Need to clamp the region size to make sure
      * we don't read outside of the source buffer
      * or write outside of the destination buffer.
@@ -562,7 +574,7 @@ GLboolean r300_blit(struct r300_context *r300,
         fprintf(stderr, "src: size [%d x %d], pitch %d, "
                 "offset [%d x %d], format %s, bo %p\n",
                 src_width, src_height, src_pitch,
-                src_offset, src_y_offset,
+                src_x_offset, src_y_offset,
                 _mesa_get_format_name(src_mesaformat),
                 src_bo);
         fprintf(stderr, "dst: pitch %d, offset[%d x %d], format %s, bo %p\n",
@@ -571,6 +583,9 @@ GLboolean r300_blit(struct r300_context *r300,
         fprintf(stderr, "region: %d x %d\n", reg_width, reg_height);
     }
 
+    /* Flush is needed to make sure that source buffer has correct data */
+    radeonFlush(r300->radeon.glCtx);
+
     if (!validate_buffers(r300, src_bo, dst_bo))
         return GL_FALSE;
 
diff --git a/src/mesa/drivers/dri/r600/Makefile b/src/mesa/drivers/dri/r600/Makefile
index 26f47b72687..b90efce17a1 100644
--- a/src/mesa/drivers/dri/r600/Makefile
+++ b/src/mesa/drivers/dri/r600/Makefile
@@ -59,6 +59,8 @@ DRIVER_SOURCES = \
 		 r700_render.c   \
 		 r600_tex.c      \
 		 r600_texstate.c      \
+		 r600_blit.c     \
+		 r600_texcopy.c  \
 		 r700_debug.c    \
 		 $(RADEON_COMMON_SOURCES) \
 		 $(EGL_SOURCES) \
diff --git a/src/mesa/drivers/dri/r600/r600_blit.c b/src/mesa/drivers/dri/r600/r600_blit.c
new file mode 100644
index 00000000000..de3a58706e6
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/r600_blit.c
@@ -0,0 +1,1584 @@
+/*
+ * Copyright (C) 2009 Advanced Micro Devices, 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, 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 "r600_context.h"
+
+#include "r600_blit.h"
+#include "r600_blit_shaders.h"
+#include "r600_cmdbuf.h"
+
+static inline void
+set_render_target(context_t *context, struct radeon_bo *bo, gl_format mesa_format,
+                  int pitch, int w, int h, intptr_t dst_offset)
+{
+    uint32_t cb_color0_base, cb_color0_size = 0, cb_color0_info = 0, cb_color0_view = 0;
+    int nPitchInPixel, id = 0;
+    uint32_t comp_swap, format, bpp = _mesa_get_format_bytes(mesa_format);
+    BATCH_LOCALS(&context->radeon);
+
+    cb_color0_base = dst_offset / 256;
+
+    nPitchInPixel = pitch/bpp;
+    SETfield(cb_color0_size, (nPitchInPixel / 8) - 1,
+             PITCH_TILE_MAX_shift, PITCH_TILE_MAX_mask);
+    SETfield(cb_color0_size, ((nPitchInPixel * h) / 64) - 1,
+             SLICE_TILE_MAX_shift, SLICE_TILE_MAX_mask);
+
+    SETfield(cb_color0_info, ENDIAN_NONE, ENDIAN_shift, ENDIAN_mask);
+    SETfield(cb_color0_info, ARRAY_LINEAR_GENERAL,
+             CB_COLOR0_INFO__ARRAY_MODE_shift, CB_COLOR0_INFO__ARRAY_MODE_mask);
+
+    SETbit(cb_color0_info, BLEND_BYPASS_bit);
+
+    switch(mesa_format) {
+    case MESA_FORMAT_RGBA8888:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_SIGNED_RGBA8888:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_SNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_RGBA8888_REV:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_STD;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_SIGNED_RGBA8888_REV:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_STD;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_SNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ARGB8888:
+    case MESA_FORMAT_XRGB8888:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_ALT;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ARGB8888_REV:
+    case MESA_FORMAT_XRGB8888_REV:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_RGB565:
+            format = COLOR_5_6_5;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_RGB565_REV:
+            format = COLOR_5_6_5;
+            comp_swap = SWAP_STD;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ARGB4444:
+            format = COLOR_4_4_4_4;
+            comp_swap = SWAP_ALT;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ARGB4444_REV:
+            format = COLOR_4_4_4_4;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ARGB1555:
+            format = COLOR_1_5_5_5;
+            comp_swap = SWAP_ALT;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ARGB1555_REV:
+            format = COLOR_1_5_5_5;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_AL88:
+            format = COLOR_8_8;
+            comp_swap = SWAP_STD;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_AL88_REV:
+            format = COLOR_8_8;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_RGB332:
+            format = COLOR_3_3_2;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_A8:
+            format = COLOR_8;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_I8:
+    case MESA_FORMAT_CI8:
+            format = COLOR_8;
+            comp_swap = SWAP_STD;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_L8:
+            format = COLOR_8;
+            comp_swap = SWAP_ALT;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_RGBA_FLOAT32:
+            format = COLOR_32_32_32_32_FLOAT;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, BLEND_FLOAT32_bit);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_RGBA_FLOAT16:
+            format = COLOR_16_16_16_16_FLOAT;
+            comp_swap = SWAP_STD_REV;
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ALPHA_FLOAT32:
+            format = COLOR_32_FLOAT;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, BLEND_FLOAT32_bit);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_ALPHA_FLOAT16:
+            format = COLOR_16_FLOAT;
+            comp_swap = SWAP_ALT_REV;
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_LUMINANCE_FLOAT32:
+            format = COLOR_32_FLOAT;
+            comp_swap = SWAP_ALT;
+	    SETbit(cb_color0_info, BLEND_FLOAT32_bit);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_LUMINANCE_FLOAT16:
+            format = COLOR_16_FLOAT;
+            comp_swap = SWAP_ALT;
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32:
+            format = COLOR_32_32_FLOAT;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, BLEND_FLOAT32_bit);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16:
+            format = COLOR_16_16_FLOAT;
+            comp_swap = SWAP_ALT_REV;
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_INTENSITY_FLOAT32: /* X, X, X, X */
+            format = COLOR_32_FLOAT;
+            comp_swap = SWAP_STD;
+	    SETbit(cb_color0_info, BLEND_FLOAT32_bit);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_INTENSITY_FLOAT16: /* X, X, X, X */
+            format = COLOR_16_FLOAT;
+            comp_swap = SWAP_STD;
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_FLOAT, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_X8_Z24:
+    case MESA_FORMAT_S8_Z24:
+            format = COLOR_8_24;
+            comp_swap = SWAP_STD;
+	    SETfield(cb_color0_info, ARRAY_1D_TILED_THIN1,
+		     CB_COLOR0_INFO__ARRAY_MODE_shift, CB_COLOR0_INFO__ARRAY_MODE_mask);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_Z24_S8:
+            format = COLOR_24_8;
+            comp_swap = SWAP_STD;
+	    SETfield(cb_color0_info, ARRAY_1D_TILED_THIN1,
+		     CB_COLOR0_INFO__ARRAY_MODE_shift, CB_COLOR0_INFO__ARRAY_MODE_mask);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_Z16:
+            format = COLOR_16;
+            comp_swap = SWAP_STD;
+	    SETfield(cb_color0_info, ARRAY_1D_TILED_THIN1,
+		     CB_COLOR0_INFO__ARRAY_MODE_shift, CB_COLOR0_INFO__ARRAY_MODE_mask);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_Z32:
+            format = COLOR_32;
+            comp_swap = SWAP_STD;
+	    SETfield(cb_color0_info, ARRAY_1D_TILED_THIN1,
+		     CB_COLOR0_INFO__ARRAY_MODE_shift, CB_COLOR0_INFO__ARRAY_MODE_mask);
+	    CLEARbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_SRGBA8:
+            format = COLOR_8_8_8_8;
+            comp_swap = SWAP_STD_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_SRGB, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_SLA8:
+            format = COLOR_8_8;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_SRGB, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    case MESA_FORMAT_SL8:
+            format = COLOR_8;
+            comp_swap = SWAP_ALT_REV;
+	    SETbit(cb_color0_info, SOURCE_FORMAT_bit);
+	    SETfield(cb_color0_info, NUMBER_SRGB, NUMBER_TYPE_shift, NUMBER_TYPE_mask);
+            break;
+    default:
+            fprintf(stderr,"Invalid format for copy %s\n",_mesa_get_format_name(mesa_format));
+            assert("Invalid format for US output\n");
+            return;
+    }
+
+    SETfield(cb_color0_info, format, CB_COLOR0_INFO__FORMAT_shift,
+             CB_COLOR0_INFO__FORMAT_mask);
+    SETfield(cb_color0_info, comp_swap, COMP_SWAP_shift, COMP_SWAP_mask);
+
+    BEGIN_BATCH_NO_AUTOSTATE(3 + 2);
+    R600_OUT_BATCH_REGSEQ(CB_COLOR0_BASE + (4 * id), 1);
+    R600_OUT_BATCH(cb_color0_base);
+    R600_OUT_BATCH_RELOC(0,
+			 bo,
+			 0,
+			 0, RADEON_GEM_DOMAIN_VRAM | RADEON_GEM_DOMAIN_GTT, 0);
+    END_BATCH();
+
+    if ((context->radeon.radeonScreen->chip_family > CHIP_FAMILY_R600) &&
+	(context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)) {
+	    BEGIN_BATCH_NO_AUTOSTATE(2);
+	    R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_BASE_UPDATE, 0));
+	    R600_OUT_BATCH((2 << id));
+	    END_BATCH();
+    }
+
+    BEGIN_BATCH_NO_AUTOSTATE(18);
+    R600_OUT_BATCH_REGVAL(CB_COLOR0_SIZE + (4 * id), cb_color0_size);
+    R600_OUT_BATCH_REGVAL(CB_COLOR0_VIEW + (4 * id), cb_color0_view);
+    R600_OUT_BATCH_REGVAL(CB_COLOR0_INFO + (4 * id), cb_color0_info);
+    R600_OUT_BATCH_REGVAL(CB_COLOR0_TILE + (4 * id), 0);
+    R600_OUT_BATCH_REGVAL(CB_COLOR0_FRAG + (4 * id), 0);
+    R600_OUT_BATCH_REGVAL(CB_COLOR0_MASK + (4 * id), 0);
+    END_BATCH();
+
+    COMMIT_BATCH();
+
+}
+
+static inline void load_shaders(GLcontext * ctx)
+{
+
+    radeonContextPtr radeonctx = RADEON_CONTEXT(ctx);
+    context_t *context = R700_CONTEXT(ctx);
+    int i, size;
+    uint32_t *shader;
+
+    if (context->blit_bo_loaded == 1)
+        return;
+
+    size = 4096;
+    context->blit_bo = radeon_bo_open(radeonctx->radeonScreen->bom, 0,
+                                      size, 256, RADEON_GEM_DOMAIN_GTT, 0);
+    radeon_bo_map(context->blit_bo, 1);
+    shader = context->blit_bo->ptr;
+
+    for(i=0; i<sizeof(r6xx_vs)/4; i++) {
+        shader[128+i] = r6xx_vs[i];
+    }
+    for(i=0; i<sizeof(r6xx_ps)/4; i++) {
+        shader[256+i] = r6xx_ps[i];
+    }
+
+    radeon_bo_unmap(context->blit_bo);
+    context->blit_bo_loaded = 1;
+
+}
+
+static inline void
+set_shaders(context_t *context)
+{
+    struct radeon_bo * pbo = context->blit_bo;
+    BATCH_LOCALS(&context->radeon);
+
+    uint32_t sq_pgm_start_fs = (512 >> 8);
+    uint32_t sq_pgm_resources_fs = 0;
+    uint32_t sq_pgm_cf_offset_fs = 0;
+
+    uint32_t sq_pgm_start_vs = (512 >> 8);
+    uint32_t sq_pgm_resources_vs = (1 << NUM_GPRS_shift);
+    uint32_t sq_pgm_cf_offset_vs = 0;
+
+    uint32_t sq_pgm_start_ps = (1024 >> 8);
+    uint32_t sq_pgm_resources_ps = (1 << NUM_GPRS_shift);
+    uint32_t sq_pgm_cf_offset_ps = 0;
+    uint32_t sq_pgm_exports_ps = (1 << 1);
+
+    r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit);
+
+    /* FS */
+    BEGIN_BATCH_NO_AUTOSTATE(3 + 2);
+    R600_OUT_BATCH_REGSEQ(SQ_PGM_START_FS, 1);
+    R600_OUT_BATCH(sq_pgm_start_fs);
+    R600_OUT_BATCH_RELOC(sq_pgm_start_fs,
+			 pbo,
+			 sq_pgm_start_fs,
+			 RADEON_GEM_DOMAIN_GTT, 0, 0);
+    END_BATCH();
+
+    BEGIN_BATCH_NO_AUTOSTATE(6);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_FS, sq_pgm_resources_fs);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_FS, sq_pgm_cf_offset_fs);
+    END_BATCH();
+
+    /* VS */
+    BEGIN_BATCH_NO_AUTOSTATE(3 + 2);
+    R600_OUT_BATCH_REGSEQ(SQ_PGM_START_VS, 1);
+    R600_OUT_BATCH(sq_pgm_start_vs);
+    R600_OUT_BATCH_RELOC(sq_pgm_start_vs,
+		         pbo,
+		         sq_pgm_start_vs,
+		         RADEON_GEM_DOMAIN_GTT, 0, 0);
+    END_BATCH();
+
+    BEGIN_BATCH_NO_AUTOSTATE(6);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_VS, sq_pgm_resources_vs);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_VS, sq_pgm_cf_offset_vs);
+    END_BATCH();
+
+    /* PS */
+    BEGIN_BATCH_NO_AUTOSTATE(3 + 2);
+    R600_OUT_BATCH_REGSEQ(SQ_PGM_START_PS, 1);
+    R600_OUT_BATCH(sq_pgm_start_ps);
+    R600_OUT_BATCH_RELOC(sq_pgm_start_ps,
+		         pbo,
+		         sq_pgm_start_ps,
+		         RADEON_GEM_DOMAIN_GTT, 0, 0);
+    END_BATCH();
+
+    BEGIN_BATCH_NO_AUTOSTATE(9);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_RESOURCES_PS, sq_pgm_resources_ps);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_EXPORTS_PS, sq_pgm_exports_ps);
+    R600_OUT_BATCH_REGVAL(SQ_PGM_CF_OFFSET_PS, sq_pgm_cf_offset_ps);
+    END_BATCH();
+
+    BEGIN_BATCH_NO_AUTOSTATE(18);
+    R600_OUT_BATCH_REGVAL(SPI_VS_OUT_CONFIG, 0); //EXPORT_COUNT is - 1
+    R600_OUT_BATCH_REGVAL(SPI_VS_OUT_ID_0, 0);
+    R600_OUT_BATCH_REGVAL(SPI_PS_INPUT_CNTL_0, SEL_CENTROID_bit);
+    R600_OUT_BATCH_REGVAL(SPI_PS_IN_CONTROL_0, (1 << NUM_INTERP_shift));
+    R600_OUT_BATCH_REGVAL(SPI_PS_IN_CONTROL_1, 0);
+    R600_OUT_BATCH_REGVAL(SPI_INTERP_CONTROL_0, 0);
+    END_BATCH();
+
+    COMMIT_BATCH();
+
+}
+
+static inline void
+set_vtx_resource(context_t *context)
+{
+    struct radeon_bo *bo = context->blit_bo;
+    BATCH_LOCALS(&context->radeon);
+
+    BEGIN_BATCH_NO_AUTOSTATE(6);
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CTL_CONST, 1));
+    R600_OUT_BATCH(mmSQ_VTX_BASE_VTX_LOC - ASIC_CTL_CONST_BASE_INDEX);
+    R600_OUT_BATCH(0);
+
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CTL_CONST, 1));
+    R600_OUT_BATCH(mmSQ_VTX_START_INST_LOC - ASIC_CTL_CONST_BASE_INDEX);
+    R600_OUT_BATCH(0);
+    END_BATCH();
+    COMMIT_BATCH();
+
+    if ((context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV610) ||
+	(context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV620) ||
+	(context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS780) ||
+	(context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS880) ||
+	(context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV710))
+	    r700SyncSurf(context, bo, RADEON_GEM_DOMAIN_GTT, 0, TC_ACTION_ENA_bit);
+    else
+	    r700SyncSurf(context, bo, RADEON_GEM_DOMAIN_GTT, 0, VC_ACTION_ENA_bit);
+
+    BEGIN_BATCH_NO_AUTOSTATE(9 + 2);
+
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_RESOURCE, 7));
+    R600_OUT_BATCH(SQ_FETCH_RESOURCE_VS_OFFSET * FETCH_RESOURCE_STRIDE);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(48 - 1);
+    R600_OUT_BATCH(16 << SQ_VTX_CONSTANT_WORD2_0__STRIDE_shift);
+    R600_OUT_BATCH(1 << MEM_REQUEST_SIZE_shift);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(SQ_TEX_VTX_VALID_BUFFER << SQ_TEX_RESOURCE_WORD6_0__TYPE_shift);
+    R600_OUT_BATCH_RELOC(SQ_VTX_CONSTANT_WORD0_0,
+                         bo,
+                         SQ_VTX_CONSTANT_WORD0_0,
+                         RADEON_GEM_DOMAIN_GTT, 0, 0);
+    END_BATCH();
+    COMMIT_BATCH();
+
+}
+
+static inline void
+set_tex_resource(context_t * context,
+		 gl_format mesa_format, struct radeon_bo *bo, int w, int h,
+		 int pitch, intptr_t src_offset)
+{
+    uint32_t sq_tex_resource0, sq_tex_resource1, sq_tex_resource2, sq_tex_resource4, sq_tex_resource6;
+    int bpp = _mesa_get_format_bytes(mesa_format);
+    int TexelPitch = pitch/bpp;
+
+    sq_tex_resource0 = sq_tex_resource1 = sq_tex_resource2 = sq_tex_resource4 = sq_tex_resource6 = 0;
+    BATCH_LOCALS(&context->radeon);
+
+    SETfield(sq_tex_resource0, SQ_TEX_DIM_2D, DIM_shift, DIM_mask);
+    SETfield(sq_tex_resource0, ARRAY_LINEAR_GENERAL,
+                 SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+                 SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+
+    switch (mesa_format) {
+    case MESA_FORMAT_RGBA8888:
+    case MESA_FORMAT_SIGNED_RGBA8888:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    if (mesa_format == MESA_FORMAT_SIGNED_RGBA8888) {
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_X_shift, FORMAT_COMP_X_mask);
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_Y_shift, FORMAT_COMP_Y_mask);
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_Z_shift, FORMAT_COMP_Z_mask);
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_W_shift, FORMAT_COMP_W_mask);
+	    }
+	    break;
+    case MESA_FORMAT_RGBA8888_REV:
+    case MESA_FORMAT_SIGNED_RGBA8888_REV:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    if (mesa_format == MESA_FORMAT_SIGNED_RGBA8888_REV) {
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_X_shift, FORMAT_COMP_X_mask);
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_Y_shift, FORMAT_COMP_Y_mask);
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_Z_shift, FORMAT_COMP_Z_mask);
+		    SETfield(sq_tex_resource4, SQ_FORMAT_COMP_SIGNED,
+			     FORMAT_COMP_W_shift, FORMAT_COMP_W_mask);
+	    }
+	    break;
+    case MESA_FORMAT_ARGB8888:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_XRGB8888:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ARGB8888_REV:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_XRGB8888_REV:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_RGB565:
+	    SETfield(sq_tex_resource1, FMT_5_6_5,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_RGB565_REV:
+	    SETfield(sq_tex_resource1, FMT_5_6_5,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ARGB4444:
+	    SETfield(sq_tex_resource1, FMT_4_4_4_4,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ARGB4444_REV:
+	    SETfield(sq_tex_resource1, FMT_4_4_4_4,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ARGB1555:
+	    SETfield(sq_tex_resource1, FMT_1_5_5_5,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ARGB1555_REV:
+	    SETfield(sq_tex_resource1, FMT_1_5_5_5,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_AL88:
+    case MESA_FORMAT_AL88_REV: /* TODO : Check this. */
+	    SETfield(sq_tex_resource1, FMT_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_RGB332:
+	    SETfield(sq_tex_resource1, FMT_3_3_2,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_A8: /* ZERO, ZERO, ZERO, X */
+	    SETfield(sq_tex_resource1, FMT_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_L8: /* X, X, X, ONE */
+	    SETfield(sq_tex_resource1, FMT_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_I8: /* X, X, X, X */
+    case MESA_FORMAT_CI8:
+	    SETfield(sq_tex_resource1, FMT_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_RGBA_FLOAT32:
+	    SETfield(sq_tex_resource1, FMT_32_32_32_32_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_RGBA_FLOAT16:
+	    SETfield(sq_tex_resource1, FMT_16_16_16_16_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ALPHA_FLOAT32: /* ZERO, ZERO, ZERO, X */
+	    SETfield(sq_tex_resource1, FMT_32_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_ALPHA_FLOAT16: /* ZERO, ZERO, ZERO, X */
+	    SETfield(sq_tex_resource1, FMT_16_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_LUMINANCE_FLOAT32: /* X, X, X, ONE */
+	    SETfield(sq_tex_resource1, FMT_32_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_LUMINANCE_FLOAT16: /* X, X, X, ONE */
+	    SETfield(sq_tex_resource1, FMT_16_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32:
+	    SETfield(sq_tex_resource1, FMT_32_32_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16:
+	    SETfield(sq_tex_resource1, FMT_16_16_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_INTENSITY_FLOAT32: /* X, X, X, X */
+	    SETfield(sq_tex_resource1, FMT_32_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_INTENSITY_FLOAT16: /* X, X, X, X */
+	    SETfield(sq_tex_resource1, FMT_16_FLOAT,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_Z16:
+	    SETbit(sq_tex_resource0, TILE_TYPE_bit);
+	    SETfield(sq_tex_resource0, ARRAY_1D_TILED_THIN1,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+	    SETfield(sq_tex_resource1, FMT_16,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_X8_Z24:
+	    SETbit(sq_tex_resource0, TILE_TYPE_bit);
+	    SETfield(sq_tex_resource0, ARRAY_1D_TILED_THIN1,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+	    SETfield(sq_tex_resource1, FMT_8_24,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_S8_Z24:
+	    SETbit(sq_tex_resource0, TILE_TYPE_bit);
+	    SETfield(sq_tex_resource0, ARRAY_1D_TILED_THIN1,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+	    SETfield(sq_tex_resource1, FMT_8_24,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_Z24_S8:
+	    SETbit(sq_tex_resource0, TILE_TYPE_bit);
+	    SETfield(sq_tex_resource0, ARRAY_1D_TILED_THIN1,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+	    SETfield(sq_tex_resource1, FMT_24_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_0,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_Z32:
+	    SETbit(sq_tex_resource0, TILE_TYPE_bit);
+	    SETfield(sq_tex_resource0, ARRAY_1D_TILED_THIN1,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+	    SETfield(sq_tex_resource1, FMT_32,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_S8:
+	    SETbit(sq_tex_resource0, TILE_TYPE_bit);
+	    SETfield(sq_tex_resource0, ARRAY_1D_TILED_THIN1,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_shift,
+		     SQ_TEX_RESOURCE_WORD0_0__TILE_MODE_mask);
+	    SETfield(sq_tex_resource1, FMT_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    break;
+    case MESA_FORMAT_SRGBA8:
+	    SETfield(sq_tex_resource1, FMT_8_8_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_W,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Z,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    SETbit(sq_tex_resource4, SQ_TEX_RESOURCE_WORD4_0__FORCE_DEGAMMA_bit);
+	    break;
+    case MESA_FORMAT_SLA8:
+	    SETfield(sq_tex_resource1, FMT_8_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_Y,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    SETbit(sq_tex_resource4, SQ_TEX_RESOURCE_WORD4_0__FORCE_DEGAMMA_bit);
+	    break;
+    case MESA_FORMAT_SL8: /* X, X, X, ONE */
+	    SETfield(sq_tex_resource1, FMT_8,
+		     SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
+
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_X,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+	    SETfield(sq_tex_resource4, SQ_SEL_1,
+		     SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
+	    SETbit(sq_tex_resource4, SQ_TEX_RESOURCE_WORD4_0__FORCE_DEGAMMA_bit);
+	    break;
+    default:
+            fprintf(stderr,"Invalid format for copy %s\n",_mesa_get_format_name(mesa_format));
+            assert("Invalid format for US output\n");
+            return;
+    };
+
+    SETfield(sq_tex_resource0, (TexelPitch/8)-1, PITCH_shift, PITCH_mask);
+    SETfield(sq_tex_resource0, w - 1, TEX_WIDTH_shift, TEX_WIDTH_mask);
+    SETfield(sq_tex_resource1, h - 1, TEX_HEIGHT_shift, TEX_HEIGHT_mask);
+
+    sq_tex_resource2 = src_offset / 256;
+
+    SETfield(sq_tex_resource6, SQ_TEX_VTX_VALID_TEXTURE,
+             SQ_TEX_RESOURCE_WORD6_0__TYPE_shift,
+             SQ_TEX_RESOURCE_WORD6_0__TYPE_mask);
+
+    r700SyncSurf(context, bo,
+                 RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM,
+		 0, TC_ACTION_ENA_bit);
+
+    BEGIN_BATCH_NO_AUTOSTATE(9 + 4);
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_RESOURCE, 7));
+    R600_OUT_BATCH(0 * 7);
+
+    R600_OUT_BATCH(sq_tex_resource0);
+    R600_OUT_BATCH(sq_tex_resource1);
+    R600_OUT_BATCH(sq_tex_resource2);
+    R600_OUT_BATCH(0); //SQ_TEX_RESOURCE3
+    R600_OUT_BATCH(sq_tex_resource4);
+    R600_OUT_BATCH(0); //SQ_TEX_RESOURCE5
+    R600_OUT_BATCH(sq_tex_resource6);
+    R600_OUT_BATCH_RELOC(0,
+		     bo,
+		     0,
+		     RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+    R600_OUT_BATCH_RELOC(0,
+		     bo,
+		     0,
+		     RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
+    END_BATCH();
+    COMMIT_BATCH();
+}
+
+static inline void
+set_tex_sampler(context_t * context)
+{
+    uint32_t sq_tex_sampler_word0 = 0, sq_tex_sampler_word1 = 0, sq_tex_sampler_word2 = 0;
+    int i = 0;
+
+    SETbit(sq_tex_sampler_word2, SQ_TEX_SAMPLER_WORD2_0__TYPE_bit);
+
+    BATCH_LOCALS(&context->radeon);
+
+    BEGIN_BATCH_NO_AUTOSTATE(5);
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_SAMPLER, 3));
+    R600_OUT_BATCH(i * 3);
+    R600_OUT_BATCH(sq_tex_sampler_word0);
+    R600_OUT_BATCH(sq_tex_sampler_word1);
+    R600_OUT_BATCH(sq_tex_sampler_word2);
+    END_BATCH();
+
+}
+
+static inline void
+set_scissors(context_t *context, int x1, int y1, int x2, int y2)
+{
+    BATCH_LOCALS(&context->radeon);
+
+    BEGIN_BATCH_NO_AUTOSTATE(17);
+    R600_OUT_BATCH_REGSEQ(PA_SC_SCREEN_SCISSOR_TL, 2);
+    R600_OUT_BATCH((x1 << 0) | (y1 << 16));
+    R600_OUT_BATCH((x2 << 0) | (y2 << 16));
+
+    R600_OUT_BATCH_REGSEQ(PA_SC_WINDOW_OFFSET, 3);
+    R600_OUT_BATCH(0); //PA_SC_WINDOW_OFFSET
+    R600_OUT_BATCH((x1 << 0) | (y1 << 16) | (WINDOW_OFFSET_DISABLE_bit)); //PA_SC_WINDOW_SCISSOR_TL
+    R600_OUT_BATCH((x2 << 0) | (y2 << 16));
+
+    R600_OUT_BATCH_REGSEQ(PA_SC_GENERIC_SCISSOR_TL, 2);
+    R600_OUT_BATCH((x1 << 0) | (y1 << 16) | (WINDOW_OFFSET_DISABLE_bit));
+    R600_OUT_BATCH((x2 << 0) | (y2 << 16));
+
+    /* XXX 16 of these PA_SC_VPORT_SCISSOR_0_TL_num ... */
+    R600_OUT_BATCH_REGSEQ(PA_SC_VPORT_SCISSOR_0_TL, 2 );
+    R600_OUT_BATCH((x1 << 0) | (y1 << 16) | (WINDOW_OFFSET_DISABLE_bit));
+    R600_OUT_BATCH((x2 << 0) | (y2 << 16));
+    END_BATCH();
+
+    COMMIT_BATCH();
+
+}
+
+static inline void
+set_vb_data(context_t * context, int src_x, int src_y, int dst_x, int dst_y,
+            int w, int h, int src_h, unsigned flip_y)
+{
+    float *vb;
+    radeon_bo_map(context->blit_bo, 1);
+    vb = context->blit_bo->ptr;
+
+    vb[0] = (float)(dst_x);
+    vb[1] = (float)(dst_y);
+    vb[2] = (float)(src_x);
+    vb[3] = (flip_y) ? (float)(src_h - src_y) : (float)src_y;
+
+    vb[4] = (float)(dst_x);
+    vb[5] = (float)(dst_y + h);
+    vb[6] = (float)(src_x);
+    vb[7] = (flip_y) ? (float)(src_h - (src_y + h)) : (float)(src_y + h);
+
+    vb[8] = (float)(dst_x + w);
+    vb[9] = (float)(dst_y + h);
+    vb[10] = (float)(src_x + w);
+    vb[11] = (flip_y) ? (float)(src_h - (src_y + h)) : (float)(src_y + h);
+
+    radeon_bo_unmap(context->blit_bo);
+
+}
+
+static inline void
+draw_auto(context_t *context)
+{
+    BATCH_LOCALS(&context->radeon);
+    uint32_t vgt_primitive_type = 0, vgt_index_type = 0, vgt_draw_initiator = 0, vgt_num_indices;
+
+    SETfield(vgt_primitive_type, DI_PT_RECTLIST,
+             VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift,
+             VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask);
+    SETfield(vgt_index_type, DI_INDEX_SIZE_16_BIT, INDEX_TYPE_shift,
+             INDEX_TYPE_mask);
+    SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift,
+             MAJOR_MODE_mask);
+    SETfield(vgt_draw_initiator, DI_SRC_SEL_AUTO_INDEX, SOURCE_SELECT_shift,
+             SOURCE_SELECT_mask);
+
+    vgt_num_indices = 3;
+
+    BEGIN_BATCH_NO_AUTOSTATE(10);
+    // prim
+    R600_OUT_BATCH_REGSEQ(VGT_PRIMITIVE_TYPE, 1);
+    R600_OUT_BATCH(vgt_primitive_type);
+    // index type
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0));
+    R600_OUT_BATCH(vgt_index_type);
+    // num instances
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0));
+    R600_OUT_BATCH(1);
+    //
+    R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_AUTO, 1));
+    R600_OUT_BATCH(vgt_num_indices);
+    R600_OUT_BATCH(vgt_draw_initiator);
+
+    END_BATCH();
+    COMMIT_BATCH();
+}
+
+static inline void
+set_default_state(context_t *context)
+{
+    int ps_prio = 0;
+    int vs_prio = 1;
+    int gs_prio = 2;
+    int es_prio = 3;
+    int num_ps_gprs;
+    int num_vs_gprs;
+    int num_gs_gprs;
+    int num_es_gprs;
+    int num_temp_gprs;
+    int num_ps_threads;
+    int num_vs_threads;
+    int num_gs_threads;
+    int num_es_threads;
+    int num_ps_stack_entries;
+    int num_vs_stack_entries;
+    int num_gs_stack_entries;
+    int num_es_stack_entries;
+    uint32_t sq_config, sq_gpr_resource_mgmt_1, sq_gpr_resource_mgmt_2;
+    uint32_t sq_thread_resource_mgmt, sq_stack_resource_mgmt_1, sq_stack_resource_mgmt_2;
+    uint32_t ta_cntl_aux, db_watermarks, sq_dyn_gpr_cntl_ps_flush_req, db_debug;
+    BATCH_LOCALS(&context->radeon);
+
+    switch (context->radeon.radeonScreen->chip_family) {
+    case CHIP_FAMILY_R600:
+	    num_ps_gprs = 192;
+	    num_vs_gprs = 56;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 136;
+	    num_vs_threads = 48;
+	    num_gs_threads = 4;
+	    num_es_threads = 4;
+	    num_ps_stack_entries = 128;
+	    num_vs_stack_entries = 128;
+	    num_gs_stack_entries = 0;
+	    num_es_stack_entries = 0;
+	    break;
+    case CHIP_FAMILY_RV630:
+    case CHIP_FAMILY_RV635:
+	    num_ps_gprs = 84;
+	    num_vs_gprs = 36;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 144;
+	    num_vs_threads = 40;
+	    num_gs_threads = 4;
+	    num_es_threads = 4;
+	    num_ps_stack_entries = 40;
+	    num_vs_stack_entries = 40;
+	    num_gs_stack_entries = 32;
+	    num_es_stack_entries = 16;
+	    break;
+    case CHIP_FAMILY_RV610:
+    case CHIP_FAMILY_RV620:
+    case CHIP_FAMILY_RS780:
+    case CHIP_FAMILY_RS880:
+    default:
+	    num_ps_gprs = 84;
+	    num_vs_gprs = 36;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 136;
+	    num_vs_threads = 48;
+	    num_gs_threads = 4;
+	    num_es_threads = 4;
+	    num_ps_stack_entries = 40;
+	    num_vs_stack_entries = 40;
+	    num_gs_stack_entries = 32;
+	    num_es_stack_entries = 16;
+	    break;
+    case CHIP_FAMILY_RV670:
+	    num_ps_gprs = 144;
+	    num_vs_gprs = 40;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 136;
+	    num_vs_threads = 48;
+	    num_gs_threads = 4;
+	    num_es_threads = 4;
+	    num_ps_stack_entries = 40;
+	    num_vs_stack_entries = 40;
+	    num_gs_stack_entries = 32;
+	    num_es_stack_entries = 16;
+	    break;
+    case CHIP_FAMILY_RV770:
+	    num_ps_gprs = 192;
+	    num_vs_gprs = 56;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 188;
+	    num_vs_threads = 60;
+	    num_gs_threads = 0;
+	    num_es_threads = 0;
+	    num_ps_stack_entries = 256;
+	    num_vs_stack_entries = 256;
+	    num_gs_stack_entries = 0;
+	    num_es_stack_entries = 0;
+	    break;
+    case CHIP_FAMILY_RV730:
+    case CHIP_FAMILY_RV740:
+	    num_ps_gprs = 84;
+	    num_vs_gprs = 36;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 188;
+	    num_vs_threads = 60;
+	    num_gs_threads = 0;
+	    num_es_threads = 0;
+	    num_ps_stack_entries = 128;
+	    num_vs_stack_entries = 128;
+	    num_gs_stack_entries = 0;
+	    num_es_stack_entries = 0;
+	    break;
+    case CHIP_FAMILY_RV710:
+	    num_ps_gprs = 192;
+	    num_vs_gprs = 56;
+	    num_temp_gprs = 4;
+	    num_gs_gprs = 0;
+	    num_es_gprs = 0;
+	    num_ps_threads = 144;
+	    num_vs_threads = 48;
+	    num_gs_threads = 0;
+	    num_es_threads = 0;
+	    num_ps_stack_entries = 128;
+	    num_vs_stack_entries = 128;
+	    num_gs_stack_entries = 0;
+	    num_es_stack_entries = 0;
+	    break;
+    }
+
+    sq_config = 0;
+    if ((context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV610) ||
+        (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV620) ||
+	(context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS780) ||
+	(context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS880) ||
+        (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV710))
+	    CLEARbit(sq_config, VC_ENABLE_bit);
+    else
+	    SETbit(sq_config, VC_ENABLE_bit);
+    SETbit(sq_config, DX9_CONSTS_bit);
+    SETbit(sq_config, ALU_INST_PREFER_VECTOR_bit);
+    SETfield(sq_config, ps_prio, PS_PRIO_shift, PS_PRIO_mask);
+    SETfield(sq_config, vs_prio, VS_PRIO_shift, VS_PRIO_mask);
+    SETfield(sq_config, gs_prio, GS_PRIO_shift, GS_PRIO_mask);
+    SETfield(sq_config, es_prio, ES_PRIO_shift, ES_PRIO_mask);
+
+    sq_gpr_resource_mgmt_1 = 0;
+    SETfield(sq_gpr_resource_mgmt_1, num_ps_gprs, NUM_PS_GPRS_shift, NUM_PS_GPRS_mask);
+    SETfield(sq_gpr_resource_mgmt_1, num_vs_gprs, NUM_VS_GPRS_shift, NUM_VS_GPRS_mask);
+    SETfield(sq_gpr_resource_mgmt_1, num_temp_gprs,
+	     NUM_CLAUSE_TEMP_GPRS_shift, NUM_CLAUSE_TEMP_GPRS_mask);
+
+    sq_gpr_resource_mgmt_2 = 0;
+    SETfield(sq_gpr_resource_mgmt_2, num_gs_gprs, NUM_GS_GPRS_shift, NUM_GS_GPRS_mask);
+    SETfield(sq_gpr_resource_mgmt_2, num_es_gprs, NUM_ES_GPRS_shift, NUM_ES_GPRS_mask);
+
+    sq_thread_resource_mgmt = 0;
+    SETfield(sq_thread_resource_mgmt, num_ps_threads,
+	     NUM_PS_THREADS_shift, NUM_PS_THREADS_mask);
+    SETfield(sq_thread_resource_mgmt, num_vs_threads,
+	     NUM_VS_THREADS_shift, NUM_VS_THREADS_mask);
+    SETfield(sq_thread_resource_mgmt, num_gs_threads,
+	     NUM_GS_THREADS_shift, NUM_GS_THREADS_mask);
+    SETfield(sq_thread_resource_mgmt, num_es_threads,
+	     NUM_ES_THREADS_shift, NUM_ES_THREADS_mask);
+
+    sq_stack_resource_mgmt_1 = 0;
+    SETfield(sq_stack_resource_mgmt_1, num_ps_stack_entries,
+	     NUM_PS_STACK_ENTRIES_shift, NUM_PS_STACK_ENTRIES_mask);
+    SETfield(sq_stack_resource_mgmt_1, num_vs_stack_entries,
+	     NUM_VS_STACK_ENTRIES_shift, NUM_VS_STACK_ENTRIES_mask);
+
+    sq_stack_resource_mgmt_2 = 0;
+    SETfield(sq_stack_resource_mgmt_2, num_gs_stack_entries,
+	     NUM_GS_STACK_ENTRIES_shift, NUM_GS_STACK_ENTRIES_mask);
+    SETfield(sq_stack_resource_mgmt_2, num_es_stack_entries,
+	     NUM_ES_STACK_ENTRIES_shift, NUM_ES_STACK_ENTRIES_mask);
+
+    ta_cntl_aux = 0;
+    SETfield(ta_cntl_aux, 28, TD_FIFO_CREDIT_shift, TD_FIFO_CREDIT_mask);
+    db_watermarks = 0;
+    SETfield(db_watermarks, 4, DEPTH_FREE_shift, DEPTH_FREE_mask);
+    SETfield(db_watermarks, 16, DEPTH_FLUSH_shift, DEPTH_FLUSH_mask);
+    SETfield(db_watermarks, 0, FORCE_SUMMARIZE_shift, FORCE_SUMMARIZE_mask);
+    SETfield(db_watermarks, 4, DEPTH_PENDING_FREE_shift, DEPTH_PENDING_FREE_mask);
+    sq_dyn_gpr_cntl_ps_flush_req = 0;
+    db_debug = 0;
+    if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) {
+	    SETfield(ta_cntl_aux, 3, GRADIENT_CREDIT_shift, GRADIENT_CREDIT_mask);
+	    db_debug = 0x82000000;
+	    SETfield(db_watermarks, 16, DEPTH_CACHELINE_FREE_shift, DEPTH_CACHELINE_FREE_mask);
+    } else {
+	    SETfield(ta_cntl_aux, 2, GRADIENT_CREDIT_shift, GRADIENT_CREDIT_mask);
+	    SETfield(db_watermarks, 4, DEPTH_CACHELINE_FREE_shift, DEPTH_CACHELINE_FREE_mask);
+	    SETbit(sq_dyn_gpr_cntl_ps_flush_req, VS_PC_LIMIT_ENABLE_bit);
+    }
+
+    BEGIN_BATCH_NO_AUTOSTATE(117);
+    R600_OUT_BATCH_REGSEQ(SQ_CONFIG, 6);
+    R600_OUT_BATCH(sq_config);
+    R600_OUT_BATCH(sq_gpr_resource_mgmt_1);
+    R600_OUT_BATCH(sq_gpr_resource_mgmt_2);
+    R600_OUT_BATCH(sq_thread_resource_mgmt);
+    R600_OUT_BATCH(sq_stack_resource_mgmt_1);
+    R600_OUT_BATCH(sq_stack_resource_mgmt_2);
+
+    R600_OUT_BATCH_REGVAL(TA_CNTL_AUX, ta_cntl_aux);
+    R600_OUT_BATCH_REGVAL(VC_ENHANCE, 0);
+    R600_OUT_BATCH_REGVAL(R7xx_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, sq_dyn_gpr_cntl_ps_flush_req);
+    R600_OUT_BATCH_REGVAL(DB_DEBUG, db_debug);
+    R600_OUT_BATCH_REGVAL(DB_WATERMARKS, db_watermarks);
+
+    R600_OUT_BATCH_REGSEQ(SQ_ESGS_RING_ITEMSIZE, 9);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+
+    R600_OUT_BATCH_REGVAL(CB_CLRCMP_CONTROL,
+                         (CLRCMP_SEL_SRC << CLRCMP_FCN_SEL_shift));
+    R600_OUT_BATCH_REGVAL(SQ_VTX_BASE_VTX_LOC, 0);
+    R600_OUT_BATCH_REGVAL(SQ_VTX_START_INST_LOC, 0);
+    R600_OUT_BATCH_REGVAL(DB_DEPTH_INFO, 0);
+    R600_OUT_BATCH_REGVAL(DB_DEPTH_CONTROL, 0);
+    R600_OUT_BATCH_REGVAL(CB_SHADER_MASK, (OUTPUT0_ENABLE_mask));
+    R600_OUT_BATCH_REGVAL(CB_TARGET_MASK, (TARGET0_ENABLE_mask));
+    R600_OUT_BATCH_REGVAL(R7xx_CB_SHADER_CONTROL, (RT0_ENABLE_bit));
+    R600_OUT_BATCH_REGVAL(CB_COLOR_CONTROL, (0xcc << ROP3_shift));
+
+    R600_OUT_BATCH_REGVAL(PA_CL_VTE_CNTL, VTX_XY_FMT_bit);
+    R600_OUT_BATCH_REGVAL(PA_CL_VS_OUT_CNTL, 0);
+    R600_OUT_BATCH_REGVAL(PA_CL_CLIP_CNTL, CLIP_DISABLE_bit);
+    R600_OUT_BATCH_REGVAL(PA_SU_SC_MODE_CNTL, (FACE_bit) |
+        (POLYMODE_PTYPE__TRIANGLES << POLYMODE_FRONT_PTYPE_shift) |
+        (POLYMODE_PTYPE__TRIANGLES << POLYMODE_BACK_PTYPE_shift));
+    R600_OUT_BATCH_REGVAL(PA_SU_VTX_CNTL, (PIX_CENTER_bit) |
+        (X_ROUND_TO_EVEN << PA_SU_VTX_CNTL__ROUND_MODE_shift) |
+        (X_1_256TH << QUANT_MODE_shift));
+
+    R600_OUT_BATCH_REGSEQ(VGT_MAX_VTX_INDX, 4);
+    R600_OUT_BATCH(2048);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+
+    R600_OUT_BATCH_REGSEQ(VGT_OUTPUT_PATH_CNTL, 13);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+
+    R600_OUT_BATCH_REGVAL(VGT_PRIMITIVEID_EN, 0);
+    R600_OUT_BATCH_REGVAL(VGT_MULTI_PRIM_IB_RESET_EN, 0);
+    R600_OUT_BATCH_REGVAL(VGT_INSTANCE_STEP_RATE_0, 0);
+    R600_OUT_BATCH_REGVAL(VGT_INSTANCE_STEP_RATE_1, 0);
+
+    R600_OUT_BATCH_REGSEQ(VGT_STRMOUT_EN, 3);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+    R600_OUT_BATCH(0);
+
+    R600_OUT_BATCH_REGVAL(VGT_STRMOUT_BUFFER_EN, 0);
+
+    END_BATCH();
+    COMMIT_BATCH();
+}
+
+static GLboolean validate_buffers(context_t *rmesa,
+                                  struct radeon_bo *src_bo,
+                                  struct radeon_bo *dst_bo)
+{
+    int ret;
+    radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
+                                      src_bo, RADEON_GEM_DOMAIN_VRAM, 0);
+
+    radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
+                                      dst_bo, 0, RADEON_GEM_DOMAIN_VRAM);
+
+    radeon_cs_space_add_persistent_bo(rmesa->radeon.cmdbuf.cs,
+                                      rmesa->blit_bo, RADEON_GEM_DOMAIN_GTT, 0);
+
+    ret = radeon_cs_space_check_with_bo(rmesa->radeon.cmdbuf.cs,
+					rmesa->blit_bo,
+					RADEON_GEM_DOMAIN_GTT, 0);
+    if (ret)
+        return GL_FALSE;
+
+    ret = radeon_cs_space_check_with_bo(rmesa->radeon.cmdbuf.cs,
+                                        first_elem(&rmesa->radeon.dma.reserved)->bo,
+                                        RADEON_GEM_DOMAIN_GTT, 0);
+    if (ret)
+        return GL_FALSE;
+
+    return GL_TRUE;
+}
+
+GLboolean r600_blit(context_t *context,
+                    struct radeon_bo *src_bo,
+                    intptr_t src_offset,
+                    gl_format src_mesaformat,
+                    unsigned src_pitch,
+                    unsigned src_width,
+                    unsigned src_height,
+                    unsigned src_x,
+                    unsigned src_y,
+                    struct radeon_bo *dst_bo,
+                    intptr_t dst_offset,
+                    gl_format dst_mesaformat,
+                    unsigned dst_pitch,
+                    unsigned dst_width,
+                    unsigned dst_height,
+                    unsigned dst_x,
+                    unsigned dst_y,
+                    unsigned w,
+                    unsigned h,
+                    unsigned flip_y)
+{
+    int id = 0;
+
+    /* not sure blit to depth works or not yet */
+    if (_mesa_get_format_bits(src_mesaformat, GL_DEPTH_BITS) > 0)
+	    return GL_FALSE;
+
+    if (src_bo == dst_bo) {
+        return GL_FALSE;
+    }
+
+    if (0) {
+        fprintf(stderr, "src: width %d, height %d, pitch %d vs %d, format %s\n",
+                src_width, src_height, src_pitch,
+                _mesa_format_row_stride(src_mesaformat, src_width),
+                _mesa_get_format_name(src_mesaformat));
+        fprintf(stderr, "dst: width %d, height %d, pitch %d, format %s\n",
+                dst_width, dst_height,
+                _mesa_format_row_stride(dst_mesaformat, dst_width),
+                _mesa_get_format_name(dst_mesaformat));
+    }
+
+    /* Flush is needed to make sure that source buffer has correct data */
+    radeonFlush(context->radeon.glCtx);
+
+    rcommonEnsureCmdBufSpace(&context->radeon, 302, __FUNCTION__);
+
+    /* load shaders */
+    load_shaders(context->radeon.glCtx);
+
+    if (!validate_buffers(context, src_bo, dst_bo))
+        return GL_FALSE;
+
+    /* set clear state */
+    /* 117 */
+    set_default_state(context);
+
+    /* shaders */
+    /* 72 */
+    set_shaders(context);
+
+    /* src */
+    /* 20 */
+    set_tex_resource(context, src_mesaformat, src_bo,
+		     src_width, src_height, src_pitch, src_offset);
+
+    /* 5 */
+    set_tex_sampler(context);
+
+    /* dst */
+    /* 25 */
+    set_render_target(context, dst_bo, dst_mesaformat,
+		      dst_pitch, dst_width, dst_height, dst_offset);
+    /* scissors */
+    /* 17 */
+    set_scissors(context, dst_x, dst_y, dst_x + dst_width, dst_y + dst_height);
+
+    set_vb_data(context, src_x, src_y, dst_x, dst_y, w, h, src_height, flip_y);
+    /* Vertex buffer setup */
+    /* 24 */
+    set_vtx_resource(context);
+
+    /* draw */
+    /* 10 */
+    draw_auto(context);
+
+    /* 7 */
+    r700SyncSurf(context, dst_bo, 0,
+                 RADEON_GEM_DOMAIN_VRAM|RADEON_GEM_DOMAIN_GTT,
+		 CB_ACTION_ENA_bit | (1 << (id + 6)));
+
+    /* 5 */
+    r700WaitForIdleClean(context);
+
+    radeonFlush(context->radeon.glCtx);
+
+    return GL_TRUE;
+}
diff --git a/src/mesa/drivers/dri/r600/r600_blit.h b/src/mesa/drivers/dri/r600/r600_blit.h
new file mode 100644
index 00000000000..f30c13c288d
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/r600_blit.h
@@ -0,0 +1,21 @@
+GLboolean r600_blit(context_t *context,
+                    struct radeon_bo *src_bo,
+                    intptr_t src_offset,
+                    gl_format src_mesaformat,
+                    unsigned src_pitch,
+                    unsigned src_width,
+                    unsigned src_height,
+                    unsigned src_x_offset,
+                    unsigned src_y_offset,
+                    struct radeon_bo *dst_bo,
+                    intptr_t dst_offset,
+                    gl_format dst_mesaformat,
+                    unsigned dst_pitch,
+                    unsigned dst_width,
+                    unsigned dst_height,
+                    unsigned dst_x_offset,
+                    unsigned dst_y_offset,
+                    unsigned w,
+                    unsigned h,
+                    unsigned flip_y);
+
diff --git a/src/mesa/drivers/dri/r600/r600_blit_shaders.h b/src/mesa/drivers/dri/r600/r600_blit_shaders.h
new file mode 100644
index 00000000000..492dde96368
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/r600_blit_shaders.h
@@ -0,0 +1,28 @@
+const uint32_t r6xx_vs[] =
+{
+        0x00000004, // CF_DWORD0(ADDR(4))
+        0x81000000, // SQ_CF_INST_VTX COUNT(1)
+        0x0000203c, // CF_EXP_IMP CF_POS0 SQ_EXPORT_POS RW_GPR(0) ELEM_SIZE(0)
+        0x94000b08, // SQ_CF_INST_EXPORT_DONE SWZ XY01 BARRIER(1)
+        0x00004000, // CF_EXP_IMP 0 SQ_EXPORT_PARAM RW_GPR(0) ELEM_SIZE(0)
+        0x14200b1a, // SQ_CF_INST_EXPORT_DONE SWZ ZW01 EOP(1) BARRIER(0)
+        0x00000000,
+        0x00000000,
+        0x3c000000, // SQ_VTX_INST_FETCH BUFFER_ID(0) MEGA_FETCH_COUNT(16)
+        0x68cd1000, // DST_GPR(0) DST_SWZ: XYZW DATA_FORMAT(35) SQ_NUM_FORMAT_SCALED SQ_FORMAT_COMP_SIGNED
+        0x00080000, // ENDIAN_SWAP(SQ_ENDIAN_NONE) MEGA_FETCH(1)
+        0x00000000, // VTX_DWORD_PAD
+};
+
+const uint32_t r6xx_ps[] =
+{
+        0x00000002, // CF_DWORD0 AADR(2)
+        0x80800000, // SQ_CF_INST_TEX COUNT(1)
+        0x00000000, // CF_ALLOC_IMP_EXP0 SQ_EXPORT_PIXEL RW_GPR(0) ELEM_SIZE(0)
+        0x94200688, // SQ_CF_INST_EXPORT_DONE EOP(1) BARRIER(1) SWZ: XYZW
+        0x00000010, // SQ_TEX_INST_SAMPLE SRC_GPR(0) RESOURCE_ID(0)
+        0x000d1000, // DST_GPR(0) SWZ: XYZW TEX_UNNORMALIZED
+        0xb0800000, // SAMPLER_ID(0) SRC_SWZ XYZW
+        0x00000000, // TEX_DWORD_PAD
+};
+
diff --git a/src/mesa/drivers/dri/r600/r600_context.c b/src/mesa/drivers/dri/r600/r600_context.c
index cb549497f54..55680fad32d 100644
--- a/src/mesa/drivers/dri/r600/r600_context.c
+++ b/src/mesa/drivers/dri/r600/r600_context.c
@@ -384,6 +384,10 @@ GLboolean r600CreateContext(const __GLcontextModes * glVisual,
 	r700InitIoctlFuncs(&functions);
 	radeonInitBufferObjectFuncs(&functions);
 
+	if (r600->radeon.radeonScreen->kernel_mm) {
+                r600_init_texcopy_functions(&functions);
+        }
+	
 	if (!radeonInitContext(&r600->radeon, &functions,
 			       glVisual, driContextPriv,
 			       sharedContextPrivate)) {
diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h
index a1b4af715e2..1f7cd4096bf 100644
--- a/src/mesa/drivers/dri/r600/r600_context.h
+++ b/src/mesa/drivers/dri/r600/r600_context.h
@@ -148,6 +148,8 @@ struct r600_context {
 	GLint      nNumActiveAos;
 	StreamDesc stream_desc[VERT_ATTRIB_MAX];
     struct r700_index_buffer ind_buf;
+	struct radeon_bo *blit_bo;
+	GLboolean blit_bo_loaded;
 };
 
 #define R700_CONTEXT(ctx)		((context_t *)(ctx->DriverCtx))
@@ -178,9 +180,12 @@ extern GLboolean r700SyncSurf(context_t *context,
 			      uint32_t write_domain,
 			      uint32_t sync_type);
 
+extern void r700WaitForIdleClean(context_t *context);
+
 extern void r700Start3D(context_t *context);
 extern void r600InitAtoms(context_t *context);
 extern void r700InitDraw(GLcontext *ctx);
+extern void r600_init_texcopy_functions(struct dd_function_table *table);
 
 #define RADEON_D_CAPTURE 0
 #define RADEON_D_PLAYBACK 1
diff --git a/src/mesa/drivers/dri/r600/r600_texcopy.c b/src/mesa/drivers/dri/r600/r600_texcopy.c
new file mode 100644
index 00000000000..287a82ea8f3
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/r600_texcopy.c
@@ -0,0 +1,169 @@
+/*
+ * 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.
+ *
+ */
+
+#include "radeon_common.h"
+#include "r600_context.h"
+
+#include "main/image.h"
+#include "main/teximage.h"
+#include "main/texstate.h"
+#include "drivers/common/meta.h"
+
+#include "radeon_mipmap_tree.h"
+#include "r600_blit.h"
+#include <main/debug.h>
+
+// TODO:
+// need to pass correct pitch for small dst textures!
+static GLboolean
+do_copy_texsubimage(GLcontext *ctx,
+                    GLenum target, GLint level,
+                    struct radeon_tex_obj *tobj,
+                    radeon_texture_image *timg,
+                    GLint dstx, GLint dsty,
+                    GLint x, GLint y,
+                    GLsizei width, GLsizei height)
+{
+    context_t *context = R700_CONTEXT(ctx);
+    struct radeon_renderbuffer *rrb;
+
+    if (_mesa_get_format_bits(timg->base.TexFormat, GL_DEPTH_BITS) > 0) {
+        rrb = radeon_get_depthbuffer(&context->radeon);
+    } else {
+        rrb = radeon_get_colorbuffer(&context->radeon);
+    }
+
+    if (!timg->mt) {
+        radeon_validate_texture_miptree(ctx, &tobj->base);
+    }
+
+    assert(rrb && rrb->bo);
+    assert(timg->mt->bo);
+    assert(timg->base.Width >= dstx + width);
+    assert(timg->base.Height >= dsty + height);
+
+    intptr_t src_offset = rrb->draw_offset;
+    intptr_t dst_offset = radeon_miptree_image_offset(timg->mt, _mesa_tex_target_to_face(target), level);
+
+    if (src_offset % 256 || dst_offset % 256) {
+        return GL_FALSE;
+    }
+
+    if (0) {
+        fprintf(stderr, "%s: copying to face %d, level %d\n",
+                __FUNCTION__, _mesa_tex_target_to_face(target), level);
+        fprintf(stderr, "to: x %d, y %d, offset %d\n", dstx, dsty, (uint32_t) dst_offset);
+        fprintf(stderr, "from (%dx%d) width %d, height %d, offset %d, pitch %d\n",
+                x, y, rrb->base.Width, rrb->base.Height, (uint32_t) src_offset, rrb->pitch/rrb->cpp);
+        fprintf(stderr, "src size %d, dst size %d\n", rrb->bo->size, timg->mt->bo->size);
+
+    }
+
+
+    /* blit from src buffer to texture */
+    return r600_blit(context, rrb->bo, src_offset, rrb->base.Format, rrb->pitch,
+                     rrb->base.Width, rrb->base.Height, x, y,
+                     timg->mt->bo, dst_offset, timg->base.TexFormat,
+                     timg->mt->levels[level].rowstride, timg->base.Width, timg->base.Height,
+                     dstx, dsty, width, height, 1);
+}
+
+static void
+r600CopyTexImage2D(GLcontext *ctx, GLenum target, GLint level,
+                   GLenum internalFormat,
+                   GLint x, GLint y, GLsizei width, GLsizei height,
+                   GLint border)
+{
+    struct gl_texture_unit *texUnit = _mesa_get_current_tex_unit(ctx);
+    struct gl_texture_object *texObj =
+        _mesa_select_tex_object(ctx, texUnit, target);
+    struct gl_texture_image *texImage =
+        _mesa_select_tex_image(ctx, texObj, target, level);
+    int srcx, srcy, dstx, dsty;
+
+    if (border)
+        goto fail;
+
+    /* Setup or redefine the texture object, mipmap tree and texture
+     * image.  Don't populate yet.
+     */
+    ctx->Driver.TexImage2D(ctx, target, level, internalFormat,
+                           width, height, border,
+                           GL_RGBA, GL_UNSIGNED_BYTE, NULL,
+                           &ctx->DefaultPacking, texObj, texImage);
+
+    srcx = x;
+    srcy = y;
+    dstx = 0;
+    dsty = 0;
+    if (!_mesa_clip_copytexsubimage(ctx,
+                                    &dstx, &dsty,
+                                    &srcx, &srcy,
+                                    &width, &height)) {
+        return;
+    }
+
+    if (!do_copy_texsubimage(ctx, target, level,
+                             radeon_tex_obj(texObj), (radeon_texture_image *)texImage,
+                             0, 0, x, y, width, height)) {
+        goto fail;
+    }
+
+    return;
+
+fail:
+    _mesa_meta_CopyTexImage2D(ctx, target, level, internalFormat, x, y,
+                              width, height, border);
+}
+
+static void
+r600CopyTexSubImage2D(GLcontext *ctx, GLenum target, GLint level,
+                      GLint xoffset, GLint yoffset,
+                      GLint x, GLint y,
+                      GLsizei width, GLsizei height)
+{
+    struct gl_texture_unit *texUnit = _mesa_get_current_tex_unit(ctx);
+    struct gl_texture_object *texObj = _mesa_select_tex_object(ctx, texUnit, target);
+    struct gl_texture_image *texImage = _mesa_select_tex_image(ctx, texObj, target, level);
+
+    if (!do_copy_texsubimage(ctx, target, level,
+                             radeon_tex_obj(texObj), (radeon_texture_image *)texImage,
+                             xoffset, yoffset, x, y, width, height)) {
+
+       //DEBUG_FALLBACKS
+
+        _mesa_meta_CopyTexSubImage2D(ctx, target, level,
+                                     xoffset, yoffset, x, y, width, height);
+    }
+}
+
+
+void r600_init_texcopy_functions(struct dd_function_table *table)
+{
+    table->CopyTexImage2D = r600CopyTexImage2D;
+    table->CopyTexSubImage2D = r600CopyTexSubImage2D;
+}
diff --git a/src/mesa/drivers/dri/r600/r600_texstate.c b/src/mesa/drivers/dri/r600/r600_texstate.c
index ae252c995be..b8466bdd75e 100644
--- a/src/mesa/drivers/dri/r600/r600_texstate.c
+++ b/src/mesa/drivers/dri/r600/r600_texstate.c
@@ -357,37 +357,37 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, gl_format mesa
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_32_32_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
 
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
-			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
+			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
 		break;
 	case MESA_FORMAT_RGBA_FLOAT16:
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_16_16_16_16_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
 
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
-			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
+			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
 		break;
 	case MESA_FORMAT_RGB_FLOAT32: /* X, Y, Z, ONE */
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_32_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
 
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
@@ -396,11 +396,11 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, gl_format mesa
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_16_16_16_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
 
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
@@ -461,26 +461,26 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, gl_format mesa
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
 
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
-			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
+			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
 		break;
 	case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16:
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_16_16_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
 
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask);
-		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
-			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
 		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X,
+			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask);
+		SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y,
 			 SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask);
 		break;
 	case MESA_FORMAT_INTENSITY_FLOAT32: /* X, X, X, X */
diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
index 45b608a1b98..bf46eb8aaba 100644
--- a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
+++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
@@ -182,7 +182,7 @@ static int cs_begin(struct radeon_cs_int *cs,
         uint32_t tmp, *ptr;
 	int num = (ndw > 0x3FF) ? ndw : 0x3FF;
 
-        tmp = (cs->cdw + 1 + num) & (~num);
+        tmp = (cs->cdw + ndw + 0x3ff) & (~0x3ff);
         ptr = (uint32_t*)realloc(cs->packets, 4 * tmp);
         if (ptr == NULL) {
             return -ENOMEM;
diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c
index b334ea05e5b..cd02bfbcf5d 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c
@@ -412,6 +412,7 @@ static GLuint radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs )
 	space_required += vbuf;
       else
 	space_required += index + elts;
+      space_required += VB->Primitive[i].count * 3;
       space_required += AOS_BUFSZ(nr_aos);
     }
     space_required += SCISSOR_BUFSZ;
diff --git a/src/mesa/drivers/dri/tdfx/tdfx_dd.c b/src/mesa/drivers/dri/tdfx/tdfx_dd.c
index 8472df607ae..ed8a3315492 100644
--- a/src/mesa/drivers/dri/tdfx/tdfx_dd.c
+++ b/src/mesa/drivers/dri/tdfx/tdfx_dd.c
@@ -91,7 +91,7 @@ static const GLubyte *tdfxDDGetString( GLcontext *ctx, GLenum name )
       else {
 	 /* unexpected result: replace spaces with hyphens */
 	 int i;
-	 for (i = 0; hardware[i] && (i < sizeof(hardware)); i++) {
+	 for (i = 0; i < sizeof(hardware) && hardware[i]; i++) {
 	    if (hardware[i] == ' ' || hardware[i] == '\t') {
 	       hardware[i] = '-';
 	    }
diff --git a/src/mesa/drivers/x11/glxapi.c b/src/mesa/drivers/x11/glxapi.c
index 02eea25a712..a17c2c3ffc6 100644
--- a/src/mesa/drivers/x11/glxapi.c
+++ b/src/mesa/drivers/x11/glxapi.c
@@ -1173,6 +1173,9 @@ _glxapi_get_extensions(void)
 #ifdef GLX_EXT_texture_from_pixmap
       "GLX_EXT_texture_from_pixmap",
 #endif
+#ifdef GLX_INTEL_swap_event
+      "GLX_INTEL_swap_event",
+#endif
       NULL
    };
    return extensions;
diff --git a/src/mesa/shader/nvfragparse.c b/src/mesa/shader/nvfragparse.c
index b739a6aa07c..8ee7c930626 100644
--- a/src/mesa/shader/nvfragparse.c
+++ b/src/mesa/shader/nvfragparse.c
@@ -642,7 +642,7 @@ Parse_SwizzleSuffix(const GLubyte *token, GLuint swizzle[4])
    else {
       /* 4-component swizzle (vector) */
       GLint k;
-      for (k = 0; token[k] && k < 4; k++) {
+      for (k = 0; k < 4 && token[k]; k++) {
          if (token[k] == 'x')
             swizzle[k] = 0;
          else if (token[k] == 'y')
diff --git a/src/mesa/shader/prog_optimize.c b/src/mesa/shader/prog_optimize.c
index 83d119e031e..ce411731b22 100644
--- a/src/mesa/shader/prog_optimize.c
+++ b/src/mesa/shader/prog_optimize.c
@@ -577,7 +577,7 @@ _mesa_remove_extra_moves(struct gl_program *prog)
 
             /* get pointer to previous instruction */
             prevI = i - 1;
-            while (removeInst[prevI] && prevI > 0)
+            while (prevI > 0 && removeInst[prevI])
                prevI--;
             prevInst = prog->Instructions + prevI;
 
diff --git a/src/mesa/sources.mak b/src/mesa/sources.mak
index c42f61af5ea..a4d09c75bb6 100644
--- a/src/mesa/sources.mak
+++ b/src/mesa/sources.mak
@@ -204,7 +204,6 @@ STATETRACKER_SOURCES = \
 	state_tracker/st_cb_strings.c \
 	state_tracker/st_cb_texture.c \
 	state_tracker/st_cb_viewport.c \
-	state_tracker/st_api.c \
 	state_tracker/st_context.c \
 	state_tracker/st_debug.c \
 	state_tracker/st_draw.c \
diff --git a/src/mesa/state_tracker/st_atom_constbuf.c b/src/mesa/state_tracker/st_atom_constbuf.c
index 77153889b68..d70cf877e84 100644
--- a/src/mesa/state_tracker/st_atom_constbuf.c
+++ b/src/mesa/state_tracker/st_atom_constbuf.c
@@ -57,7 +57,7 @@ void st_upload_constants( struct st_context *st,
                           unsigned shader_type)
 {
    struct pipe_context *pipe = st->pipe;
-   struct pipe_constant_buffer *cbuf = &st->state.constants[shader_type];
+   struct pipe_buffer **cbuf = &st->state.constants[shader_type];
 
    assert(shader_type == PIPE_SHADER_VERTEX ||
           shader_type == PIPE_SHADER_FRAGMENT);
@@ -71,8 +71,8 @@ void st_upload_constants( struct st_context *st,
       /* We always need to get a new buffer, to keep the drivers simple and
        * avoid gratuitous rendering synchronization.
        */
-      pipe_buffer_reference(&cbuf->buffer, NULL );
-      cbuf->buffer = pipe_buffer_create(pipe->screen, 16,
+      pipe_buffer_reference(cbuf, NULL );
+      *cbuf = pipe_buffer_create(pipe->screen, 16,
                                         PIPE_BUFFER_USAGE_CONSTANT,
 					paramBytes );
 
@@ -84,12 +84,12 @@ void st_upload_constants( struct st_context *st,
       }
 
       /* load Mesa constants into the constant buffer */
-      if (cbuf->buffer)
-         st_no_flush_pipe_buffer_write(st, cbuf->buffer,
+      if (cbuf)
+         st_no_flush_pipe_buffer_write(st, *cbuf,
 				       0, paramBytes,
 				       params->ParameterValues);
 
-      st->pipe->set_constant_buffer(st->pipe, shader_type, 0, cbuf);
+      st->pipe->set_constant_buffer(st->pipe, shader_type, 0, *cbuf);
    }
    else {
       st->constants.tracked_state[shader_type].dirty.mesa = 0x0;
diff --git a/src/mesa/state_tracker/st_atom_shader.c b/src/mesa/state_tracker/st_atom_shader.c
index 46c8cbb3098..176f3ea68d7 100644
--- a/src/mesa/state_tracker/st_atom_shader.c
+++ b/src/mesa/state_tracker/st_atom_shader.c
@@ -35,8 +35,6 @@
  *   Brian Paul
  */
 
-
-
 #include "main/imports.h"
 #include "main/mtypes.h"
 #include "main/macros.h"
@@ -57,9 +55,7 @@
 
 
 
-
-
-/*
+/**
  * Translate fragment program if needed.
  */
 static void
@@ -155,8 +151,10 @@ find_translated_vp(struct st_context *st,
 }
 
 
-
-
+/**
+ * Return pointer to a pass-through fragment shader.
+ * This shader is used when a texture is missing/incomplete.
+ */
 static void *
 get_passthrough_fs(struct st_context *st)
 {
@@ -168,6 +166,11 @@ get_passthrough_fs(struct st_context *st)
    return st->passthrough_fs;
 }
 
+
+/**
+ * Update fragment program state/atom.  This involves translating the
+ * Mesa fragment program into a gallium fragment program and binding it.
+ */
 static void
 update_fp( struct st_context *st )
 {
@@ -191,6 +194,7 @@ update_fp( struct st_context *st )
    }
 }
 
+
 const struct st_tracked_state st_update_fp = {
    "st_update_fp",					/* name */
    {							/* dirty */
@@ -202,7 +206,10 @@ const struct st_tracked_state st_update_fp = {
 
 
 
-
+/**
+ * Update vertex program state/atom.  This involves translating the
+ * Mesa vertex program into a gallium fragment program and binding it.
+ */
 static void
 update_vp( struct st_context *st )
 {
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index 0cec23f2b3c..f01053cdacc 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -680,7 +680,22 @@ st_TexImage(GLcontext * ctx,
     * conversion and copy:
     */
    if (compressed_src) {
-      memcpy(texImage->Data, pixels, imageSize);
+      const GLuint srcImageStride = _mesa_format_row_stride(texImage->TexFormat, width);
+      if(dstRowStride == srcImageStride)
+         memcpy(texImage->Data, pixels, imageSize);
+      else
+      {
+         char *dst = texImage->Data;
+         const char *src = pixels;
+         int i;
+
+         for(i = 0; i < height; ++i)
+         {
+            memcpy(dst, src, srcImageStride);
+            dst += dstRowStride;
+            src += srcImageStride;
+         }
+      }
    }
    else {
       const GLuint srcImageStride =
diff --git a/src/mesa/state_tracker/st_cb_viewport.c b/src/mesa/state_tracker/st_cb_viewport.c
index 75b0a219ce3..ab11c5b4fe0 100644
--- a/src/mesa/state_tracker/st_cb_viewport.c
+++ b/src/mesa/state_tracker/st_cb_viewport.c
@@ -42,8 +42,8 @@ static void st_viewport(GLcontext * ctx, GLint x, GLint y,
 {
    struct st_context *st = ctx->st;
 
-   if (st->pipe->winsys && st->pipe->winsys->update_buffer)
-      st->pipe->winsys->update_buffer( st->pipe->winsys,
+   if (st->pipe->screen && st->pipe->screen->update_buffer)
+      st->pipe->screen->update_buffer( st->pipe->screen,
                                        st->pipe->priv );
 }
 
diff --git a/src/mesa/state_tracker/st_context.c b/src/mesa/state_tracker/st_context.c
index e4f18c842ca..45bb6b3cb7e 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -218,8 +218,8 @@ static void st_destroy_context_priv( struct st_context *st )
    }
 
    for (i = 0; i < Elements(st->state.constants); i++) {
-      if (st->state.constants[i].buffer) {
-         pipe_buffer_reference(&st->state.constants[i].buffer, NULL);
+      if (st->state.constants[i]) {
+         pipe_buffer_reference(&st->state.constants[i], NULL);
       }
    }
 
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index 831909a3f8f..2c4943cfb0a 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -92,7 +92,7 @@ struct st_context
       struct pipe_sampler_state             samplers[PIPE_MAX_SAMPLERS];
       struct pipe_sampler_state             *sampler_list[PIPE_MAX_SAMPLERS];
       struct pipe_clip_state clip;
-      struct pipe_constant_buffer constants[2];
+      struct pipe_buffer *constants[2];
       struct pipe_framebuffer_state framebuffer;
       struct pipe_texture *sampler_texture[PIPE_MAX_SAMPLERS];
       struct pipe_scissor_state scissor;
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index e54f21be600..b0d5b993a77 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -365,6 +365,7 @@ setup_interleaved_attribs(GLcontext *ctx,
 
       velements[attr].src_offset =
          (unsigned) (arrays[mesaAttr]->Ptr - offset0);
+      velements[attr].instance_divisor = 0;
       velements[attr].vertex_buffer_index = 0;
       velements[attr].nr_components = arrays[mesaAttr]->Size;
       velements[attr].src_format =
@@ -454,6 +455,7 @@ setup_non_interleaved_attribs(GLcontext *ctx,
       /* common-case setup */
       vbuffer[attr].stride = stride; /* in bytes */
       vbuffer[attr].max_index = max_index;
+      velements[attr].instance_divisor = 0;
       velements[attr].vertex_buffer_index = attr;
       velements[attr].nr_components = arrays[mesaAttr]->Size;
       velements[attr].src_format
@@ -522,7 +524,6 @@ st_draw_vbo(GLcontext *ctx,
    struct pipe_context *pipe = ctx->st->pipe;
    const struct st_vertex_program *vp;
    const struct st_vp_varient *vpv;
-   const struct pipe_shader_state *vs;
    struct pipe_vertex_buffer vbuffer[PIPE_MAX_SHADER_INPUTS];
    GLuint attr;
    struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
@@ -550,7 +551,6 @@ st_draw_vbo(GLcontext *ctx,
    /* must get these after state validation! */
    vp = ctx->st->vp;
    vpv = ctx->st->vp_varient;
-   vs = &vpv->state;
 
 #if 0
    if (MESA_VERBOSE & VERBOSE_GLSL) {
diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c
index cfc0caac983..a05d6dd06bd 100644
--- a/src/mesa/state_tracker/st_draw_feedback.c
+++ b/src/mesa/state_tracker/st_draw_feedback.c
@@ -177,6 +177,7 @@ st_feedback_draw_vbo(GLcontext *ctx,
       /* common-case setup */
       vbuffers[attr].stride = arrays[mesaAttr]->StrideB; /* in bytes */
       vbuffers[attr].max_index = max_index;
+      velements[attr].instance_divisor = 0;
       velements[attr].vertex_buffer_index = attr;
       velements[attr].nr_components = arrays[mesaAttr]->Size;
       velements[attr].src_format = 
@@ -239,11 +240,11 @@ st_feedback_draw_vbo(GLcontext *ctx,
 
    /* map constant buffers */
    mapped_constants = pipe_buffer_map(pipe->screen,
-                                      st->state.constants[PIPE_SHADER_VERTEX].buffer,
+                                      st->state.constants[PIPE_SHADER_VERTEX],
                                       PIPE_BUFFER_USAGE_CPU_READ);
    draw_set_mapped_constant_buffer(st->draw, PIPE_SHADER_VERTEX,
                                    mapped_constants,
-                                   st->state.constants[PIPE_SHADER_VERTEX].buffer->size);
+                                   st->state.constants[PIPE_SHADER_VERTEX]->size);
 
 
    /* draw here */
@@ -253,7 +254,7 @@ st_feedback_draw_vbo(GLcontext *ctx,
 
 
    /* unmap constant buffers */
-   pipe_buffer_unmap(pipe->screen, st->state.constants[PIPE_SHADER_VERTEX].buffer);
+   pipe_buffer_unmap(pipe->screen, st->state.constants[PIPE_SHADER_VERTEX]);
 
    /*
     * unmap vertex/index buffers
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index 6a869fae904..5c87e47ca3d 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -44,7 +44,6 @@
 
 #include "st_debug.h"
 #include "st_context.h"
-#include "st_atom.h"
 #include "st_program.h"
 #include "st_mesa_to_tgsi.h"
 #include "cso_cache/cso_context.h"
diff --git a/src/mesa/state_tracker/st_public.h b/src/mesa/state_tracker/st_public.h
index a5fdac32d1f..98c19817c87 100644
--- a/src/mesa/state_tracker/st_public.h
+++ b/src/mesa/state_tracker/st_public.h
@@ -56,15 +56,19 @@ struct pipe_surface;
 struct pipe_texture;
 
 
+PUBLIC
 struct st_context *st_create_context(struct pipe_context *pipe,
                                      const __GLcontextModes *visual,
                                      struct st_context *share);
 
+PUBLIC
 void st_destroy_context( struct st_context *st );
 
+PUBLIC
 void st_copy_context_state(struct st_context *dst, struct st_context *src,
                            uint mask);
 
+PUBLIC
 struct st_framebuffer *st_create_framebuffer( const __GLcontextModes *visual,
                                               enum pipe_format colorFormat,
                                               enum pipe_format depthFormat,
@@ -72,50 +76,67 @@ struct st_framebuffer *st_create_framebuffer( const __GLcontextModes *visual,
                                               uint width, uint height,
                                               void *privateData);
 
+PUBLIC
 void st_resize_framebuffer( struct st_framebuffer *stfb,
                             uint width, uint height );
 
+PUBLIC
 void st_set_framebuffer_surface(struct st_framebuffer *stfb,
                                 uint surfIndex, struct pipe_surface *surf);
 
+PUBLIC
 void st_get_framebuffer_dimensions( struct st_framebuffer *stfb,
 				    uint *width, uint *height);
 
+PUBLIC
 int st_get_framebuffer_surface(struct st_framebuffer *stfb,
                                uint surfIndex, struct pipe_surface **surface);
 
+PUBLIC
 int st_get_framebuffer_texture(struct st_framebuffer *stfb,
                                uint surfIndex, struct pipe_texture **texture);
 
+PUBLIC
 void *st_framebuffer_private( struct st_framebuffer *stfb );
 
+PUBLIC
 void st_unreference_framebuffer( struct st_framebuffer *stfb );
 
+PUBLIC
 GLboolean st_make_current(struct st_context *st,
                           struct st_framebuffer *draw,
                           struct st_framebuffer *read);
 
+PUBLIC
 struct st_context *st_get_current(void);
 
+PUBLIC
 void st_flush( struct st_context *st, uint pipeFlushFlags,
                struct pipe_fence_handle **fence );
+PUBLIC
 void st_finish( struct st_context *st );
 
+PUBLIC
 void st_notify_swapbuffers(struct st_framebuffer *stfb);
 
+PUBLIC
 void st_swapbuffers(struct st_framebuffer *stfb,
                     struct pipe_surface **front_left,
                     struct pipe_surface **front_right);
 
+PUBLIC
 int st_bind_texture_surface(struct pipe_surface *ps, int target, int level,
                             enum pipe_format format);
+PUBLIC
 int st_unbind_texture_surface(struct pipe_surface *ps, int target, int level);
 
 /** Redirect rendering into stfb's surface to a texture image */
+PUBLIC
 int st_bind_teximage(struct st_framebuffer *stfb, uint surfIndex,
                      int target, int format, int level);
 
 /** Undo surface-to-texture binding */
+PUBLIC
 int st_release_teximage(struct st_framebuffer *stfb, uint surfIndex,
                         int target, int format, int level);
 
@@ -123,6 +144,7 @@ int st_release_teximage(struct st_framebuffer *stfb, uint surfIndex,
 /** Generic function type */
 typedef void (*st_proc)();
 
+PUBLIC
 st_proc st_get_proc_address(const char *procname);
 
 
diff --git a/src/mesa/state_tracker/st_texture.c b/src/mesa/state_tracker/st_texture.c
index 8a3e4cd3ac3..b210ac91873 100644
--- a/src/mesa/state_tracker/st_texture.c
+++ b/src/mesa/state_tracker/st_texture.c
@@ -35,7 +35,6 @@
 #include "main/texfetch.h"
 #include "main/teximage.h"
 #include "main/texobj.h"
-#include "main/texstore.h"
 
 #undef Elements  /* fix re-defined macro warning */
 
diff --git a/src/mesa/swrast/s_accum.c b/src/mesa/swrast/s_accum.c
index 0e0876efcba..cf53f01b7c1 100644
--- a/src/mesa/swrast/s_accum.c
+++ b/src/mesa/swrast/s_accum.c
@@ -27,7 +27,6 @@
 #include "main/context.h"
 #include "main/macros.h"
 #include "main/imports.h"
-#include "main/fbobject.h"
 
 #include "s_accum.h"
 #include "s_context.h"
diff --git a/src/mesa/swrast/s_atifragshader.c b/src/mesa/swrast/s_atifragshader.c
index e88ff191239..353e9999d61 100644
--- a/src/mesa/swrast/s_atifragshader.c
+++ b/src/mesa/swrast/s_atifragshader.c
@@ -23,7 +23,6 @@
 #include "main/colormac.h"
 #include "main/context.h"
 #include "main/macros.h"
-#include "shader/program.h"
 #include "shader/atifragshader.h"
 #include "swrast/s_atifragshader.h"
 
diff --git a/src/mesa/swrast/s_bitmap.c b/src/mesa/swrast/s_bitmap.c
index 46c63aa6458..59e26e9ea31 100644
--- a/src/mesa/swrast/s_bitmap.c
+++ b/src/mesa/swrast/s_bitmap.c
@@ -33,7 +33,6 @@
 #include "main/condrender.h"
 #include "main/image.h"
 #include "main/macros.h"
-#include "main/pixel.h"
 
 #include "s_context.h"
 #include "s_span.h"
diff --git a/src/mesa/swrast/s_copypix.c b/src/mesa/swrast/s_copypix.c
index 986b6aff4f0..e881d1be30c 100644
--- a/src/mesa/swrast/s_copypix.c
+++ b/src/mesa/swrast/s_copypix.c
@@ -28,11 +28,9 @@
 #include "main/colormac.h"
 #include "main/condrender.h"
 #include "main/convolve.h"
-#include "main/histogram.h"
 #include "main/image.h"
 #include "main/macros.h"
 #include "main/imports.h"
-#include "main/pixel.h"
 
 #include "s_context.h"
 #include "s_depth.h"
diff --git a/src/mesa/swrast/s_depth.c b/src/mesa/swrast/s_depth.c
index c37a54eb3eb..0b6bb7e3ec9 100644
--- a/src/mesa/swrast/s_depth.c
+++ b/src/mesa/swrast/s_depth.c
@@ -28,7 +28,6 @@
 #include "main/formats.h"
 #include "main/macros.h"
 #include "main/imports.h"
-#include "main/fbobject.h"
 
 #include "s_depth.h"
 #include "s_context.h"
diff --git a/src/mesa/swrast/s_drawpix.c b/src/mesa/swrast/s_drawpix.c
index 55a4c4c3c61..248d6cc1c04 100644
--- a/src/mesa/swrast/s_drawpix.c
+++ b/src/mesa/swrast/s_drawpix.c
@@ -31,7 +31,6 @@
 #include "main/image.h"
 #include "main/macros.h"
 #include "main/imports.h"
-#include "main/pixel.h"
 #include "main/state.h"
 
 #include "s_context.h"
diff --git a/src/mesa/swrast/s_feedback.c b/src/mesa/swrast/s_feedback.c
index 47ed25ee100..2e6066983d5 100644
--- a/src/mesa/swrast/s_feedback.c
+++ b/src/mesa/swrast/s_feedback.c
@@ -25,7 +25,6 @@
 #include "main/glheader.h"
 #include "main/colormac.h"
 #include "main/context.h"
-#include "main/enums.h"
 #include "main/feedback.h"
 #include "main/macros.h"
 
diff --git a/src/mesa/swrast/s_fragprog.c b/src/mesa/swrast/s_fragprog.c
index a22d34415d5..9ac33a26a65 100644
--- a/src/mesa/swrast/s_fragprog.c
+++ b/src/mesa/swrast/s_fragprog.c
@@ -25,7 +25,6 @@
 #include "main/glheader.h"
 #include "main/colormac.h"
 #include "main/context.h"
-#include "main/texstate.h"
 #include "shader/prog_instruction.h"
 
 #include "s_fragprog.h"
diff --git a/src/mesa/swrast/s_lines.c b/src/mesa/swrast/s_lines.c
index 23cb9b57eff..5411229d700 100644
--- a/src/mesa/swrast/s_lines.c
+++ b/src/mesa/swrast/s_lines.c
@@ -29,7 +29,6 @@
 #include "main/macros.h"
 #include "s_aaline.h"
 #include "s_context.h"
-#include "s_depth.h"
 #include "s_feedback.h"
 #include "s_lines.h"
 #include "s_span.h"
diff --git a/src/mesa/swrast/s_points.c b/src/mesa/swrast/s_points.c
index 50ec2063a55..6b955429e94 100644
--- a/src/mesa/swrast/s_points.c
+++ b/src/mesa/swrast/s_points.c
@@ -27,7 +27,6 @@
 #include "main/colormac.h"
 #include "main/context.h"
 #include "main/macros.h"
-#include "main/texstate.h"
 #include "s_context.h"
 #include "s_feedback.h"
 #include "s_points.h"
diff --git a/src/mesa/swrast/s_readpix.c b/src/mesa/swrast/s_readpix.c
index 44a11cd6dd2..94fb974eab0 100644
--- a/src/mesa/swrast/s_readpix.c
+++ b/src/mesa/swrast/s_readpix.c
@@ -33,7 +33,6 @@
 #include "main/image.h"
 #include "main/macros.h"
 #include "main/imports.h"
-#include "main/pixel.h"
 #include "main/state.h"
 
 #include "s_context.h"
diff --git a/src/mesa/swrast/s_texcombine.c b/src/mesa/swrast/s_texcombine.c
index 889164b9861..594b71a03c6 100644
--- a/src/mesa/swrast/s_texcombine.c
+++ b/src/mesa/swrast/s_texcombine.c
@@ -29,7 +29,6 @@
 #include "main/colormac.h"
 #include "main/image.h"
 #include "main/imports.h"
-#include "main/pixel.h"
 #include "shader/prog_instruction.h"
 
 #include "s_context.h"
diff --git a/src/mesa/tnl/t_context.c b/src/mesa/tnl/t_context.c
index db21b4589de..5a14e595a04 100644
--- a/src/mesa/tnl/t_context.c
+++ b/src/mesa/tnl/t_context.c
@@ -38,7 +38,6 @@
 #include "tnl.h"
 #include "t_context.h"
 #include "t_pipeline.h"
-#include "t_vp_build.h"
 
 #include "vbo/vbo.h"
 
diff --git a/src/mesa/tnl/t_draw.c b/src/mesa/tnl/t_draw.c
index d31b29b9b40..38757a0e288 100644
--- a/src/mesa/tnl/t_draw.c
+++ b/src/mesa/tnl/t_draw.c
@@ -29,15 +29,11 @@
 #include "main/condrender.h"
 #include "main/context.h"
 #include "main/imports.h"
-#include "main/state.h"
 #include "main/mtypes.h"
 #include "main/macros.h"
 #include "main/enums.h"
 
 #include "t_context.h"
-#include "t_pipeline.h"
-#include "t_vp_build.h"
-#include "t_vertex.h"
 #include "tnl.h"
 
 
diff --git a/src/mesa/tnl/t_pipeline.c b/src/mesa/tnl/t_pipeline.c
index 01b30babb48..946b29e250f 100644
--- a/src/mesa/tnl/t_pipeline.c
+++ b/src/mesa/tnl/t_pipeline.c
@@ -28,7 +28,6 @@
 #include "main/glheader.h"
 #include "main/context.h"
 #include "main/imports.h"
-#include "main/state.h"
 #include "main/mtypes.h"
 
 #include "t_context.h"
diff --git a/src/mesa/tnl/t_rasterpos.c b/src/mesa/tnl/t_rasterpos.c
index 99b67874556..13b84a7d77b 100644
--- a/src/mesa/tnl/t_rasterpos.c
+++ b/src/mesa/tnl/t_rasterpos.c
@@ -29,7 +29,6 @@
 #include "main/feedback.h"
 #include "main/light.h"
 #include "main/macros.h"
-#include "main/rastpos.h"
 #include "main/simple_list.h"
 #include "main/mtypes.h"
 
diff --git a/src/mesa/tnl/t_vb_program.c b/src/mesa/tnl/t_vb_program.c
index c289cdfbaac..5396548666f 100644
--- a/src/mesa/tnl/t_vb_program.c
+++ b/src/mesa/tnl/t_vb_program.c
@@ -40,7 +40,6 @@
 #include "shader/prog_statevars.h"
 #include "shader/prog_execute.h"
 #include "swrast/s_context.h"
-#include "swrast/s_texfilter.h"
 
 #include "tnl/tnl.h"
 #include "tnl/t_context.h"
@@ -390,6 +389,13 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 #endif
          COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
       }
+
+      /* FOGC is a special case.  Fragment shader expects (f,0,0,1) */
+      if (program->Base.OutputsWritten & BITFIELD64_BIT(VERT_RESULT_FOGC)) {
+         store->results[VERT_RESULT_FOGC].data[i][1] = 0.0;
+         store->results[VERT_RESULT_FOGC].data[i][2] = 0.0;
+         store->results[VERT_RESULT_FOGC].data[i][3] = 1.0;
+      }
 #ifdef NAN_CHECK
       ASSERT(machine.Outputs[0][3] != 0.0F);
 #endif
diff --git a/src/mesa/vbo/vbo_exec.c b/src/mesa/vbo/vbo_exec.c
index e168a89ea5e..a057befed0a 100644
--- a/src/mesa/vbo/vbo_exec.c
+++ b/src/mesa/vbo/vbo_exec.c
@@ -28,9 +28,6 @@
 
 #include "main/api_arrayelt.h"
 #include "main/glheader.h"
-#include "main/imports.h"
-#include "main/context.h"
-#include "main/macros.h"
 #include "main/mtypes.h"
 #include "main/vtxfmt.h"
 
diff --git a/src/mesa/vbo/vbo_exec_array.c b/src/mesa/vbo/vbo_exec_array.c
index 6de8f059b79..2c82f7c9c5c 100644
--- a/src/mesa/vbo/vbo_exec_array.c
+++ b/src/mesa/vbo/vbo_exec_array.c
@@ -35,7 +35,6 @@
 #include "main/bufferobj.h"
 #include "main/enums.h"
 #include "main/macros.h"
-#include "glapi/dispatch.h"
 
 #include "vbo_context.h"
 
diff --git a/src/mesa/vbo/vbo_exec_draw.c b/src/mesa/vbo/vbo_exec_draw.c
index 4f43856016c..d7dbbceb1b7 100644
--- a/src/mesa/vbo/vbo_exec_draw.c
+++ b/src/mesa/vbo/vbo_exec_draw.c
@@ -30,7 +30,6 @@
 #include "main/context.h"
 #include "main/enums.h"
 #include "main/state.h"
-#include "main/macros.h"
 
 #include "vbo_context.h"
 
diff --git a/src/mesa/vbo/vbo_save.c b/src/mesa/vbo/vbo_save.c
index 9757c3d9f61..10f705cf844 100644
--- a/src/mesa/vbo/vbo_save.c
+++ b/src/mesa/vbo/vbo_save.c
@@ -28,8 +28,6 @@
 
 #include "main/mtypes.h"
 #include "main/bufferobj.h"
-#include "main/dlist.h"
-#include "main/vtxfmt.h"
 #include "main/imports.h"
 
 #include "vbo_context.h"
diff --git a/src/mesa/vbo/vbo_save_loopback.c b/src/mesa/vbo/vbo_save_loopback.c
index b7a74e4535a..f13a16e3b51 100644
--- a/src/mesa/vbo/vbo_save_loopback.c
+++ b/src/mesa/vbo/vbo_save_loopback.c
@@ -29,7 +29,6 @@
 #include "main/glheader.h"
 #include "main/enums.h"
 #include "main/imports.h"
-#include "main/macros.h"
 #include "main/mtypes.h"
 #include "glapi/dispatch.h"
 #include "glapi/glapi.h"
diff --git a/src/mesa/vbo/vbo_split_copy.c b/src/mesa/vbo/vbo_split_copy.c
index c45190b9dd3..2ca111217ce 100644
--- a/src/mesa/vbo/vbo_split_copy.c
+++ b/src/mesa/vbo/vbo_split_copy.c
@@ -34,7 +34,6 @@
 #include "main/imports.h"
 #include "main/image.h"
 #include "main/macros.h"
-#include "main/enums.h"
 #include "main/mtypes.h"
 
 #include "vbo_split.h"
@@ -221,8 +220,6 @@ begin( struct copy_context *copy, GLenum mode, GLboolean begin_flag )
 {
    struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
 
-/*    _mesa_printf("begin %s (%d)\n", _mesa_lookup_prim_by_nr(mode), begin_flag); */
-		
    prim->mode = mode;
    prim->begin = begin_flag;
 }
diff --git a/src/mesa/x86/x86_xform.c b/src/mesa/x86/x86_xform.c
index 52f6b25d817..c834e2b468b 100644
--- a/src/mesa/x86/x86_xform.c
+++ b/src/mesa/x86/x86_xform.c
@@ -30,7 +30,6 @@
 #include "main/glheader.h"
 #include "main/context.h"
 #include "math/m_xform.h"
-#include "tnl/t_context.h"
 
 #include "x86_xform.h"
 #include "common_x86_asm.h"