r600: convert to using common radeon state atoms

switches more state handling to common code.  We need
should be more fine grained with the state atoms
eventually.

7 files changed, 154 insertions, 226 deletions

diff --git a/src/mesa/drivers/dri/r600/r600_cmdbuf.c b/src/mesa/drivers/dri/r600/r600_cmdbuf.c
index 83687bb7209..38814b6d71e 100644
--- a/src/mesa/drivers/dri/r600/r600_cmdbuf.c
+++ b/src/mesa/drivers/dri/r600/r600_cmdbuf.c
@@ -451,12 +451,10 @@ struct radeon_cs_manager * r600_radeon_cs_manager_legacy_ctor(struct radeon_cont
 
 void r600InitCmdBuf(context_t *r600) /* from rcommonInitCmdBuf */
 {
-    radeonContextPtr rmesa = &r600->radeon;
-    GLuint size;
-    rmesa->hw.max_state_size = 4000; /* rough estimate */
+	radeonContextPtr rmesa = &r600->radeon;
+	GLuint size;
 
-    rmesa->hw.all_dirty = GL_TRUE;
-    rmesa->hw.is_dirty = GL_TRUE;
+	r600InitAtoms(r600);
 
 	/* Initialize command buffer */
 	size = 256 * driQueryOptioni(&rmesa->optionCache,
@@ -480,7 +478,7 @@ void r600InitCmdBuf(context_t *r600) /* from rcommonInitCmdBuf */
 	rmesa->cmdbuf.cs = radeon_cs_create(rmesa->cmdbuf.csm, size);
 	assert(rmesa->cmdbuf.cs != NULL);
 	rmesa->cmdbuf.size = size;
-	
+
 	if (!rmesa->radeonScreen->kernel_mm) {
 		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]);
 		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, rmesa->radeonScreen->gartTextures.size);
diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h
index 0cf2eb2daa5..a9b080baa31 100644
--- a/src/mesa/drivers/dri/r600/r600_context.h
+++ b/src/mesa/drivers/dri/r600/r600_context.h
@@ -55,9 +55,9 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 struct r600_context;
 typedef struct r600_context context_t;
 
-GLboolean r700SendPSState(context_t *context);
-GLboolean r700SendVSState(context_t *context);
-GLboolean r700SendSQConfig(context_t *context);
+extern GLboolean r700SendPSState(context_t *context);
+extern GLboolean r700SendVSState(context_t *context);
+extern GLboolean r700SendFSState(context_t *context);
 
 #include "main/mm.h"
 
@@ -112,6 +112,22 @@ enum
     RIGHT_SHIFT = 2,
 };
 
+struct r600_hw_state {
+	struct radeon_state_atom sq;
+	struct radeon_state_atom db;
+	struct radeon_state_atom db_target;
+	struct radeon_state_atom sc;
+	struct radeon_state_atom cl;
+	struct radeon_state_atom ucp;
+	struct radeon_state_atom su;
+	struct radeon_state_atom cb;
+	struct radeon_state_atom cb_target;
+	struct radeon_state_atom sx;
+	struct radeon_state_atom vgt;
+	struct radeon_state_atom spi;
+	struct radeon_state_atom vpt;
+};
+
 /**
  * \brief R600 context structure.
  */
@@ -121,6 +137,8 @@ struct r600_context {
 	/* ------ */
 	R700_CHIP_CONTEXT hw;
 
+	struct r600_hw_state atoms;
+
 	/* Vertex buffers
 	 */
 	GLvector4f dummy_attrib[_TNL_ATTRIB_MAX];
@@ -143,28 +161,14 @@ do {						\
 		rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	\
 } while (0)
 
-#define R600_STATECHANGE(r600, atom)			\
+#define R600_STATECHANGE(r600, ATOM)			\
 do {							\
 	R600_NEWPRIM(r600);					\
-	(atom) = GL_TRUE;					\
+	r600->atoms.ATOM.dirty = GL_TRUE;					\
 	r600->radeon.hw.is_dirty = GL_TRUE;			\
 } while(0)
 
-extern GLboolean r700SendSPIState(context_t *context);
-extern GLboolean r700SendVGTState(context_t *context);
-extern GLboolean r700SendSXState(context_t *context);
-extern GLboolean r700SendDBState(context_t *context);
-extern GLboolean r700SendCBState(context_t *context);
-extern GLboolean r700SendSUState(context_t *context);
-extern GLboolean r700SendCLState(context_t *context);
-extern GLboolean r700SendSCState(context_t *context);
-extern GLboolean r700SendViewportState(context_t *context, int id);
-extern GLboolean r700SendRenderTargetState(context_t *context, int id);
 extern GLboolean r700SendTextureState(context_t *context);
-extern GLboolean r700SendDepthTargetState(context_t *context);
-extern GLboolean r700SendUCPState(context_t *context);
-extern GLboolean r700SendFSState(context_t *context);
-extern void r700EmitState(GLcontext * ctx);
 
 extern GLboolean r700SyncSurf(context_t *context,
 			      struct radeon_bo *pbo,
@@ -180,6 +184,8 @@ extern void      r700SetupVTXConstants(GLcontext  * ctx,
 				       unsigned int stride,
 				       unsigned int Count);    /* number of vectors in stream */
 
+extern void r600InitAtoms(context_t *context);
+
 #define RADEON_D_CAPTURE 0
 #define RADEON_D_PLAYBACK 1
 #define RADEON_D_PLAYBACK_RAW 2
diff --git a/src/mesa/drivers/dri/r600/r600_emit.c b/src/mesa/drivers/dri/r600/r600_emit.c
index b695ed95838..b0c7294682a 100644
--- a/src/mesa/drivers/dri/r600/r600_emit.c
+++ b/src/mesa/drivers/dri/r600/r600_emit.c
@@ -49,7 +49,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 void r600EmitCacheFlush(context_t *rmesa)
 {
-	BATCH_LOCALS(&rmesa->radeon);
 }
 
 GLboolean r600EmitShader(GLcontext * ctx, 
diff --git a/src/mesa/drivers/dri/r600/r700_chip.c b/src/mesa/drivers/dri/r600/r700_chip.c
index 7f46b0c4688..2d68f021dfc 100644
--- a/src/mesa/drivers/dri/r600/r700_chip.c
+++ b/src/mesa/drivers/dri/r600/r700_chip.c
@@ -27,6 +27,7 @@
 
 #include "main/imports.h"
 #include "main/glheader.h"
+#include "main/simple_list.h"
 
 #include "r600_context.h"
 #include "r600_cmdbuf.h"
@@ -220,8 +221,9 @@ int r700SetupStreams(GLcontext * ctx)
     return R600_FALLBACK_NONE;
 }
 
-GLboolean r700SendDepthTargetState(context_t *context)
+static void r700SendDepthTargetState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	struct radeon_renderbuffer *rrb;
 	BATCH_LOCALS(&context->radeon);
@@ -229,7 +231,7 @@ GLboolean r700SendDepthTargetState(context_t *context)
 	rrb = radeon_get_depthbuffer(&context->radeon);
 	if (!rrb || !rrb->bo) {
 		fprintf(stderr, "no rrb\n");
-		return GL_FALSE;
+		return;
 	}
 
         BEGIN_BATCH_NO_AUTOSTATE(8 + 2);
@@ -255,31 +257,27 @@ GLboolean r700SendDepthTargetState(context_t *context)
 
 	COMMIT_BATCH();
 
-	r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
-		     DB_ACTION_ENA_bit | DB_DEST_BASE_ENA_bit);
-
-	r700->db_target_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendRenderTargetState(context_t *context, int id)
+static void r700SendRenderTargetState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	struct radeon_renderbuffer *rrb;
 	BATCH_LOCALS(&context->radeon);
+	int id = 0;
 
 	rrb = radeon_get_colorbuffer(&context->radeon);
 	if (!rrb || !rrb->bo) {
 		fprintf(stderr, "no rrb\n");
-		return GL_FALSE;
+		return;
 	}
 
 	if (id > R700_MAX_RENDER_TARGETS)
-		return GL_FALSE;
+		return;
 
 	if (!r700->render_target[id].enabled)
-		return GL_FALSE;
+		return;
 
         BEGIN_BATCH_NO_AUTOSTATE(3 + 2);
 	R600_OUT_BATCH_REGSEQ(CB_COLOR0_BASE + (4 * id), 1);
@@ -309,12 +307,6 @@ GLboolean r700SendRenderTargetState(context_t *context, int id)
 
 	COMMIT_BATCH();
 
-	r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
-		     CB_ACTION_ENA_bit | (1 << (id + 6)));
-
-	r700->render_target[id].dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
 GLboolean r700SendPSState(context_t *context)
@@ -429,16 +421,18 @@ GLboolean r700SendFSState(context_t *context)
 	return GL_TRUE;
 }
 
-GLboolean r700SendViewportState(context_t *context, int id)
+static void r700SendViewportState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
+	int id = 0;
 
 	if (id > R700_MAX_VIEWPORTS)
-		return GL_FALSE;
+		return;
 
 	if (!r700->viewport[id].enabled)
-		return GL_FALSE;
+		return;
 
         BEGIN_BATCH_NO_AUTOSTATE(16);
 	R600_OUT_BATCH_REGSEQ(PA_SC_VPORT_SCISSOR_0_TL + (8 * id), 2);
@@ -458,13 +452,11 @@ GLboolean r700SendViewportState(context_t *context, int id)
 
 	COMMIT_BATCH();
 
-	r700->viewport[id].dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendSQConfig(context_t *context)
+static void r700SendSQConfig(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -496,20 +488,17 @@ GLboolean r700SendSQConfig(context_t *context)
         END_BATCH();
 
 	COMMIT_BATCH();
-
-	r700->sq_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendUCPState(context_t *context)
+static void r700SendUCPState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 	int i;
 
 	for (i = 0; i < R700_MAX_UCP; i++) {
-		if (r700->ucp[i].enabled && r700->ucp[i].dirty) {
+		if (r700->ucp[i].enabled) {
 			BEGIN_BATCH_NO_AUTOSTATE(6);
 			R600_OUT_BATCH_REGSEQ(PA_CL_UCP_0_X + (16 * i), 4);
 			R600_OUT_BATCH(r700->ucp[i].PA_CL_UCP_0_X.u32All);
@@ -518,15 +507,13 @@ GLboolean r700SendUCPState(context_t *context)
 			R600_OUT_BATCH(r700->ucp[i].PA_CL_UCP_0_W.u32All);
 			END_BATCH();
 			COMMIT_BATCH();
-			r700->ucp[i].dirty = GL_FALSE;
 		}
 	}
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendSPIState(context_t *context)
+static void r700SendSPIState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 	unsigned int ui;
@@ -596,14 +583,11 @@ GLboolean r700SendSPIState(context_t *context)
 
 	END_BATCH();
 	COMMIT_BATCH();
-
-	r700->spi_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendVGTState(context_t *context)
+static void r700SendVGTState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -644,14 +628,11 @@ GLboolean r700SendVGTState(context_t *context)
 
 	END_BATCH();
 	COMMIT_BATCH();
-
-	r700->vgt_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendSXState(context_t *context)
+static void r700SendSXState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -661,14 +642,11 @@ GLboolean r700SendSXState(context_t *context)
 	R600_OUT_BATCH_REGVAL(SX_ALPHA_REF, r700->SX_ALPHA_REF.u32All);
 	END_BATCH();
 	COMMIT_BATCH();
-
-	r700->sx_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendDBState(context_t *context)
+static void r700SendDBState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -695,14 +673,11 @@ GLboolean r700SendDBState(context_t *context)
 
 	END_BATCH();
 	COMMIT_BATCH();
-
-	r700->db_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendCBState(context_t *context)
+static void r700SendCBState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 	unsigned int ui;
@@ -757,13 +732,11 @@ GLboolean r700SendCBState(context_t *context)
 
 	COMMIT_BATCH();
 
-	r700->cb_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendSUState(context_t *context)
+static void r700SendSUState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -789,13 +762,11 @@ GLboolean r700SendSUState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
-	r700->su_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
-GLboolean r700SendCLState(context_t *context)
+static void r700SendCLState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -813,15 +784,12 @@ GLboolean r700SendCLState(context_t *context)
 
 	END_BATCH();
 	COMMIT_BATCH();
-
-	r700->cl_dirty = GL_FALSE;
-
-	return GL_TRUE;
 }
 
 // XXX need to split this up
-GLboolean r700SendSCState(context_t *context)
+static void r700SendSCState(GLcontext *ctx, struct radeon_state_atom *atom)
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
 	BATCH_LOCALS(&context->radeon);
 
@@ -860,8 +828,47 @@ GLboolean r700SendSCState(context_t *context)
 
 	END_BATCH();
 	COMMIT_BATCH();
+}
 
-	r700->sc_dirty = GL_FALSE;
+static int check_always(GLcontext *ctx, struct radeon_state_atom *atom)
+{
+	return atom->cmd_size;
+}
 
-	return GL_TRUE;
+#define ALLOC_STATE( ATOM, SZ, EMIT )					\
+do {									\
+      context->atoms.ATOM.cmd_size = (SZ);					\
+      context->atoms.ATOM.cmd = NULL;					\
+      context->atoms.ATOM.name = #ATOM;					\
+      context->atoms.ATOM.idx = 0;						\
+      context->atoms.ATOM.check = check_always;				\
+      context->atoms.ATOM.dirty = GL_FALSE;				\
+      context->atoms.ATOM.emit = (EMIT);					\
+      context->radeon.hw.max_state_size += (SZ);			\
+      insert_at_tail(&context->radeon.hw.atomlist, &context->atoms.ATOM); \
+} while (0)
+
+void r600InitAtoms(context_t *context)
+{
+
+	/* Setup the atom linked list */
+	make_empty_list(&context->radeon.hw.atomlist);
+	context->radeon.hw.atomlist.name = "atom-list";
+
+	ALLOC_STATE(sq, 34, r700SendSQConfig);
+	ALLOC_STATE(db, 27, r700SendDBState);
+	ALLOC_STATE(db_target, 19, r700SendDepthTargetState);
+	ALLOC_STATE(sc, 47, r700SendSCState);
+	ALLOC_STATE(cl, 18, r700SendCLState);
+	ALLOC_STATE(ucp, 36, r700SendUCPState);
+	ALLOC_STATE(su, 19, r700SendSUState);
+	ALLOC_STATE(cb, 39, r700SendCBState);
+	ALLOC_STATE(cb_target, 32, r700SendRenderTargetState);
+	ALLOC_STATE(sx, 9, r700SendSXState);
+	ALLOC_STATE(vgt, 41, r700SendVGTState);
+	ALLOC_STATE(spi, (59 + R700_MAX_SHADER_EXPORTS), r700SendSPIState);
+	ALLOC_STATE(vpt, 16, r700SendViewportState);
+
+	context->radeon.hw.is_dirty = GL_TRUE;
+	context->radeon.hw.all_dirty = GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/r600/r700_ioctl.c b/src/mesa/drivers/dri/r600/r700_ioctl.c
index e0e506d665d..72a89789767 100644
--- a/src/mesa/drivers/dri/r600/r700_ioctl.c
+++ b/src/mesa/drivers/dri/r600/r700_ioctl.c
@@ -41,33 +41,10 @@
 #include "r700_ioctl.h"
 #include "r700_clear.h"
 
-static void r700Flush(GLcontext *ctx)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s %d\n", __FUNCTION__, radeon->cmdbuf.cs->cdw);
-
-	/* okay if we have no cmds in the buffer &&
-	   we have no DMA flush &&
-	   we have no DMA buffer allocated.
-	   then no point flushing anything at all.
-	*/
-	if (!radeon->dma.flush && !radeon->cmdbuf.cs->cdw && is_empty_list(&radeon->dma.reserved))
-		return;
-
-	if (radeon->dma.flush)
-		radeon->dma.flush( ctx );
-
-	r700EmitState(ctx);
-
-	if (radeon->cmdbuf.cs->cdw)
-		rcommonFlushCmdBuf(radeon, __FUNCTION__);
-}
 
 void r700InitIoctlFuncs(struct dd_function_table *functions)
 {
 	functions->Clear = r700Clear;
 	functions->Finish = radeonFinish;
-	functions->Flush = r700Flush;
+	functions->Flush = radeonFlush;
 }
diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c
index 52f8f90dad3..f0cd357c765 100644
--- a/src/mesa/drivers/dri/r600/r700_render.c
+++ b/src/mesa/drivers/dri/r600/r700_render.c
@@ -332,96 +332,28 @@ static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim
 
 }
 
-static void r700EmitAtoms(GLcontext * ctx, GLboolean dirty)
-{
-	context_t *context = R700_CONTEXT(ctx);
-	radeonContextPtr radeon = &context->radeon;
-	R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context);
-
-	if ((r700->sq_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendSQConfig(context);
-	r700SendUCPState(context);
-	if ((r700->sc_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendSCState(context);
-	if ((r700->su_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendSUState(context);
-	if ((r700->cl_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendCLState(context);
-	if ((r700->cb_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendCBState(context);
-	if ((r700->db_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendDBState(context);
-	if ((r700->sx_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendSXState(context);
-	if ((r700->vgt_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendVGTState(context);
-	if ((r700->spi_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendSPIState(context);
-	if ((r700->viewport[0].dirty || radeon->hw.all_dirty) == dirty)
-		r700SendViewportState(context, 0);
-	if ((r700->render_target[0].dirty || radeon->hw.all_dirty) == dirty)
-		r700SendRenderTargetState(context, 0);
-	if ((r700->db_target_dirty || radeon->hw.all_dirty) == dirty)
-		r700SendDepthTargetState(context);
-
-}
-
-void r700EmitState(GLcontext * ctx)
-{
-	context_t *context = R700_CONTEXT(ctx);
-	radeonContextPtr radeon = &context->radeon;
-
-	if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	if (radeon->vtbl.pre_emit_state)
-		radeon->vtbl.pre_emit_state(radeon);
-
-	if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty)
-		return;
-
-	rcommonEnsureCmdBufSpace(&context->radeon,
-				 652, __FUNCTION__);
-
-	if (!radeon->cmdbuf.cs->cdw) {
-		if (RADEON_DEBUG & DEBUG_STATE)
-			fprintf(stderr, "Begin reemit state\n");
-
-		r700EmitAtoms(ctx, GL_FALSE);
-	}
-
-	if (RADEON_DEBUG & DEBUG_STATE)
-		fprintf(stderr, "Begin dirty state\n");
-
-	r700EmitAtoms(ctx, GL_TRUE);
-	radeon->hw.is_dirty = GL_FALSE;
-	radeon->hw.all_dirty = GL_FALSE;
-
-}
-
 static GLboolean r700RunRender(GLcontext * ctx,
 			                   struct tnl_pipeline_stage *stage)
 {
     context_t *context = R700_CONTEXT(ctx);
     radeonContextPtr radeon = &context->radeon;
-    unsigned int i, ind_count = 0;
+    unsigned int i, ind_count = 0, id = 0;
     TNLcontext *tnl = TNL_CONTEXT(ctx);
     struct vertex_buffer *vb = &tnl->vb;
+    struct radeon_renderbuffer *rrb;
 
     for (i = 0; i < vb->PrimitiveCount; i++)
 	    ind_count += vb->Primitive[i].count + 10;
 
     /* just an estimate, need to properly calculate this */
     rcommonEnsureCmdBufSpace(&context->radeon,
-			     radeon->hw.max_state_size + ind_count, __FUNCTION__);
+			     radeon->hw.max_state_size + ind_count + 1000, __FUNCTION__);
 
     r700Start3D(context);
-
     r700UpdateShaders(ctx);
     r700SetScissor(context);
     r700SetupShaders(ctx);
-
-    r700EmitState(ctx);
+    radeonEmitState(radeon);
 
     /* richard test code */
     for (i = 0; i < vb->PrimitiveCount; i++) {
@@ -434,6 +366,16 @@ static GLboolean r700RunRender(GLcontext * ctx,
     /* Flush render op cached for last several quads. */
     r700WaitForIdleClean(context);
 
+    rrb = radeon_get_colorbuffer(&context->radeon);
+    if (!rrb || !rrb->bo)
+	    r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
+			 CB_ACTION_ENA_bit | (1 << (id + 6)));
+
+    rrb = radeon_get_depthbuffer(&context->radeon);
+    if (!rrb || !rrb->bo)
+	    r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM,
+			 DB_ACTION_ENA_bit | DB_DEST_BASE_ENA_bit);
+
     radeonReleaseArrays(ctx, ~0);
 
     return GL_FALSE;
diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c
index ada394ec61b..6b44cc0ceb6 100644
--- a/src/mesa/drivers/dri/r600/r700_state.c
+++ b/src/mesa/drivers/dri/r600/r700_state.c
@@ -153,7 +153,7 @@ void r700UpdateViewportOffset(GLcontext * ctx) //------------------
 		/* Note: this should also modify whatever data the context reset
 		 * code uses...
 		 */
-		R600_STATECHANGE(context, r700->viewport[id].dirty);
+		R600_STATECHANGE(context, vpt);
 		r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All = tx;
 		r700->viewport[id].PA_CL_VPORT_YOFFSET.f32All = ty;
 	}
@@ -170,8 +170,8 @@ void r700UpdateDrawBuffer(GLcontext * ctx) /* TODO */ //---------------------
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->render_target[0].dirty);
-	R600_STATECHANGE(context, r700->db_target_dirty);
+	R600_STATECHANGE(context, cb_target);
+	R600_STATECHANGE(context, db_target);
 
 	r700SetRenderTarget(context, 0);
 	r700SetDepthTarget(context);
@@ -244,8 +244,8 @@ static void r700InvalidateState(GLcontext * ctx, GLuint new_state) //-----------
 
 	r700UpdateStateParameters(ctx, new_state);
 
-    R600_STATECHANGE(context, r700->cl_dirty);
-    R600_STATECHANGE(context, r700->spi_dirty);
+    R600_STATECHANGE(context, cl);
+    R600_STATECHANGE(context, spi);
 
     if(GL_TRUE == r700->bEnablePerspective)
     {
@@ -278,7 +278,7 @@ static void r700SetDepthState(GLcontext * ctx)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->db_dirty);
+	R600_STATECHANGE(context, db);
 
     if (ctx->Depth.Test)
     {
@@ -346,7 +346,7 @@ static void r700SetAlphaState(GLcontext * ctx)
 	uint32_t alpha_func = REF_ALWAYS;
 	GLboolean really_enabled = ctx->Color.AlphaEnabled;
 
-	R600_STATECHANGE(context, r700->sx_dirty);
+	R600_STATECHANGE(context, sx);
 
 	switch (ctx->Color.AlphaFunc) {
 	case GL_NEVER:
@@ -400,7 +400,7 @@ static void r700BlendColor(GLcontext * ctx, const GLfloat cf[4]) //-------------
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->cb_dirty);
+	R600_STATECHANGE(context, cb);
 
 	r700->CB_BLEND_RED.f32All = cf[0];
 	r700->CB_BLEND_GREEN.f32All = cf[1];
@@ -470,7 +470,7 @@ static void r700SetBlendState(GLcontext * ctx)
 	int id = 0;
 	uint32_t blend_reg = 0, eqn, eqnA;
 
-	R600_STATECHANGE(context, r700->cb_dirty);
+	R600_STATECHANGE(context, cb);
 
 	if (RGBA_LOGICOP_ENABLED(ctx) || !ctx->Color.BlendEnabled) {
 		SETfield(blend_reg,
@@ -661,7 +661,7 @@ static void r700SetLogicOpState(GLcontext *ctx)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
 
-	R600_STATECHANGE(context, r700->cb_dirty);
+	R600_STATECHANGE(context, cb);
 
 	if (RGBA_LOGICOP_ENABLED(ctx))
 		SETfield(r700->CB_COLOR_CONTROL.u32All,
@@ -685,7 +685,7 @@ static void r700UpdateCulling(GLcontext * ctx)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
 
-    R600_STATECHANGE(context, r700->su_dirty);
+    R600_STATECHANGE(context, su);
 
     CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, FACE_bit);
     CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, CULL_FRONT_bit);
@@ -733,7 +733,7 @@ static void r700UpdateLineStipple(GLcontext * ctx)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
 
-    R600_STATECHANGE(context, r700->sc_dirty);
+    R600_STATECHANGE(context, sc);
 
     if (ctx->Line.StippleFlag)
     {
@@ -817,7 +817,7 @@ static void r700ColorMask(GLcontext * ctx,
 			     (a ? 8 : 0));
 
 	if (mask != r700->CB_SHADER_MASK.u32All) {
-		R600_STATECHANGE(context, r700->cb_dirty);
+		R600_STATECHANGE(context, cb);
 		SETfield(r700->CB_SHADER_MASK.u32All, mask, OUTPUT0_ENABLE_shift, OUTPUT0_ENABLE_mask);
 	}
 }
@@ -875,7 +875,7 @@ static void r700ShadeModel(GLcontext * ctx, GLenum mode) //--------------------
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->spi_dirty);
+	R600_STATECHANGE(context, spi);
 
 	/* also need to set/clear FLAT_SHADE bit per param in SPI_PS_INPUT_CNTL_[0-31] */
 	switch (mode) {
@@ -898,7 +898,7 @@ static void r700PointSize(GLcontext * ctx, GLfloat size)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->su_dirty);
+	R600_STATECHANGE(context, su);
 
 	/* We need to clamp to user defined range here, because
 	 * the HW clamping happens only for per vertex point size. */
@@ -920,7 +920,7 @@ static void r700PointParameter(GLcontext * ctx, GLenum pname, const GLfloat * pa
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->su_dirty);
+	R600_STATECHANGE(context, su);
 
 	/* format is 12.4 fixed point */
 	switch (pname) {
@@ -1006,7 +1006,7 @@ static void r700SetStencilState(GLcontext * ctx, GLboolean state)
 	}
 
 	if (hw_stencil) {
-		R600_STATECHANGE(context, r700->db_dirty);
+		R600_STATECHANGE(context, db);
 		if (state)
 			SETbit(r700->DB_DEPTH_CONTROL.u32All, STENCIL_ENABLE_bit);
 		else
@@ -1024,7 +1024,7 @@ static void r700StencilFuncSeparate(GLcontext * ctx, GLenum face,
 	//fixme
 	//r300CatchStencilFallback(ctx);
 
-	R600_STATECHANGE(context, r700->db_dirty);
+	R600_STATECHANGE(context, db);
 
 	//front
 	SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.Ref[0],
@@ -1055,7 +1055,7 @@ static void r700StencilMaskSeparate(GLcontext * ctx, GLenum face, GLuint mask) /
 	//fixme
 	//r300CatchStencilFallback(ctx);
 
-	R600_STATECHANGE(context, r700->db_dirty);
+	R600_STATECHANGE(context, db);
 
 	// front
 	SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.WriteMask[0],
@@ -1077,7 +1077,7 @@ static void r700StencilOpSeparate(GLcontext * ctx, GLenum face,
 	//fixme
 	//r300CatchStencilFallback(ctx);
 
-	R600_STATECHANGE(context, r700->db_dirty);
+	R600_STATECHANGE(context, db);
 
 	SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.FailFunc[0]),
 		 STENCILFAIL_shift, STENCILFAIL_mask);
@@ -1121,7 +1121,7 @@ static void r700UpdateWindow(GLcontext * ctx, int id) //--------------------
 	GLfloat sz = v[MAT_SZ] * depthScale;
 	GLfloat tz = v[MAT_TZ] * depthScale;
 
-	R600_STATECHANGE(context, r700->viewport[id].dirty);
+	R600_STATECHANGE(context, vpt);
 
 	r700->viewport[id].PA_CL_VPORT_XSCALE.f32All  = sx;
 	r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All = tx;
@@ -1160,7 +1160,7 @@ static void r700LineWidth(GLcontext * ctx, GLfloat widthf) //---------------
     R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
     uint32_t lineWidth = (uint32_t)((widthf * 0.5) * (1 << 4));
 
-    R600_STATECHANGE(context, r700->su_dirty);
+    R600_STATECHANGE(context, su);
 
     if (lineWidth > 0xFFFF)
 	    lineWidth = 0xFFFF;
@@ -1173,7 +1173,7 @@ static void r700LineStipple(GLcontext *ctx, GLint factor, GLushort pattern)
     context_t *context = R700_CONTEXT(ctx);
     R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-    R600_STATECHANGE(context, r700->sc_dirty);
+    R600_STATECHANGE(context, sc);
 
     SETfield(r700->PA_SC_LINE_STIPPLE.u32All, pattern, LINE_PATTERN_shift, LINE_PATTERN_mask);
     SETfield(r700->PA_SC_LINE_STIPPLE.u32All, (factor-1), REPEAT_COUNT_shift, REPEAT_COUNT_mask);
@@ -1185,7 +1185,7 @@ static void r700SetPolygonOffsetState(GLcontext * ctx, GLboolean state)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->su_dirty);
+	R600_STATECHANGE(context, su);
 
 	if (state) {
 		SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_FRONT_ENABLE_bit);
@@ -1215,7 +1215,7 @@ static void r700PolygonOffset(GLcontext * ctx, GLfloat factor, GLfloat units) //
 
 	factor *= 12.0;
 
-	R600_STATECHANGE(context, r700->su_dirty);
+	R600_STATECHANGE(context, su);
 
 	r700->PA_SU_POLY_OFFSET_FRONT_SCALE.f32All = factor;
 	r700->PA_SU_POLY_OFFSET_FRONT_OFFSET.f32All = constant;
@@ -1228,7 +1228,7 @@ static void r700UpdatePolygonMode(GLcontext * ctx)
 	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-	R600_STATECHANGE(context, r700->su_dirty);
+	R600_STATECHANGE(context, su);
 
 	SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DISABLE_POLY_MODE, POLY_MODE_shift, POLY_MODE_mask);
 
@@ -1305,7 +1305,7 @@ static void r700ClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq )
 	p = (GLint) plane - (GLint) GL_CLIP_PLANE0;
 	ip = (GLint *)ctx->Transform._ClipUserPlane[p];
 
-	R600_STATECHANGE(context, r700->ucp[p].dirty);
+	R600_STATECHANGE(context, ucp);
 
 	r700->ucp[p].PA_CL_UCP_0_X.u32All = ip[0];
 	r700->ucp[p].PA_CL_UCP_0_Y.u32All = ip[1];
@@ -1321,7 +1321,7 @@ static void r700SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state)
 
 	p = cap - GL_CLIP_PLANE0;
 
-	R600_STATECHANGE(context, r700->cl_dirty);
+	R600_STATECHANGE(context, cl);
 
 	if (state) {
 		r700->PA_CL_CLIP_CNTL.u32All |= (UCP_ENA_0_bit << p);
@@ -1356,7 +1356,7 @@ void r700SetScissor(context_t *context) //---------------
 		y2 = rrb->dPriv->y + rrb->dPriv->h;
 	}
 
-	R600_STATECHANGE(context, r700->sc_dirty);
+	R600_STATECHANGE(context, sc);
 
 	/* window */
 	SETbit(r700->PA_SC_WINDOW_SCISSOR_TL.u32All, WINDOW_OFFSET_DISABLE_bit);
@@ -1426,8 +1426,8 @@ static void r700SetRenderTarget(context_t *context, int id)
 	    return;
     }
 
-    R600_STATECHANGE(context, r700->render_target[id].dirty);
-    R600_STATECHANGE(context, r700->cb_dirty);
+    R600_STATECHANGE(context, cb_target);
+    R600_STATECHANGE(context, cb);
 
     /* screen/window/view */
     SETfield(r700->CB_TARGET_MASK.u32All, 0xF, (4 * id), TARGET0_ENABLE_mask);
@@ -1475,8 +1475,7 @@ static void r700SetDepthTarget(context_t *context)
     if (!rrb)
 	    return;
 
-    R600_STATECHANGE(context, r700->db_target_dirty);
-    R600_STATECHANGE(context, r700->db_dirty);
+    R600_STATECHANGE(context, db_target);
 
     /* depth buf */
     r700->DB_DEPTH_SIZE.u32All = 0;
@@ -1538,7 +1537,7 @@ static void r700InitSQConfig(GLcontext * ctx)
     int num_gs_stack_entries;
     int num_es_stack_entries;
 
-    R600_STATECHANGE(context, r700->sq_dirty);
+    R600_STATECHANGE(context, sq);
 
     // SQ
     ps_prio = 0;