r600: improve state emission

Slowing migrating to atom based system like
the other radeon drivers.

6 files changed, 182 insertions, 30 deletions

diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h
index 17ac0889011..0cf2eb2daa5 100644
--- a/src/mesa/drivers/dri/r600/r600_context.h
+++ b/src/mesa/drivers/dri/r600/r600_context.h
@@ -137,6 +137,19 @@ extern GLboolean r600CreateContext(const __GLcontextModes * glVisual,
 
 #define R700_CONTEXT_STATES(context) ((R700_CHIP_CONTEXT *)(&context->hw))
 
+#define R600_NEWPRIM( rmesa )			\
+do {						\
+	if ( rmesa->radeon.dma.flush )			\
+		rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	\
+} while (0)
+
+#define R600_STATECHANGE(r600, atom)			\
+do {							\
+	R600_NEWPRIM(r600);					\
+	(atom) = 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);
diff --git a/src/mesa/drivers/dri/r600/r700_chip.c b/src/mesa/drivers/dri/r600/r700_chip.c
index e67e544d538..7f46b0c4688 100644
--- a/src/mesa/drivers/dri/r600/r700_chip.c
+++ b/src/mesa/drivers/dri/r600/r700_chip.c
@@ -258,6 +258,8 @@ GLboolean r700SendDepthTargetState(context_t *context)
 	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;
 }
 
@@ -310,6 +312,8 @@ GLboolean r700SendRenderTargetState(context_t *context, int id)
 	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;
 }
 
@@ -343,6 +347,8 @@ GLboolean r700SendPSState(context_t *context)
 
 	COMMIT_BATCH();
 
+	r700->ps.dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -375,6 +381,8 @@ GLboolean r700SendVSState(context_t *context)
 
 	COMMIT_BATCH();
 
+	r700->vs.dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -416,6 +424,8 @@ GLboolean r700SendFSState(context_t *context)
 
 	COMMIT_BATCH();
 
+	r700->fs.dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -448,6 +458,8 @@ GLboolean r700SendViewportState(context_t *context, int id)
 
 	COMMIT_BATCH();
 
+	r700->viewport[id].dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -485,6 +497,8 @@ GLboolean r700SendSQConfig(context_t *context)
 
 	COMMIT_BATCH();
 
+	r700->sq_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -495,7 +509,7 @@ GLboolean r700SendUCPState(context_t *context)
 	int i;
 
 	for (i = 0; i < R700_MAX_UCP; i++) {
-		if (r700->ucp[i].enabled) {
+		if (r700->ucp[i].enabled && r700->ucp[i].dirty) {
 			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);
@@ -504,6 +518,7 @@ 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;
 		}
 	}
 
@@ -582,6 +597,8 @@ GLboolean r700SendSPIState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->spi_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -628,6 +645,8 @@ GLboolean r700SendVGTState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->vgt_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -643,6 +662,8 @@ GLboolean r700SendSXState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->sx_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -675,6 +696,8 @@ GLboolean r700SendDBState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->db_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -734,6 +757,8 @@ GLboolean r700SendCBState(context_t *context)
 
 	COMMIT_BATCH();
 
+	r700->cb_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -764,6 +789,8 @@ GLboolean r700SendSUState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->su_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -787,6 +814,8 @@ GLboolean r700SendCLState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->cl_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
 
@@ -832,5 +861,7 @@ GLboolean r700SendSCState(context_t *context)
 	END_BATCH();
 	COMMIT_BATCH();
 
+	r700->sc_dirty = GL_FALSE;
+
 	return GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/r600/r700_chip.h b/src/mesa/drivers/dri/r600/r700_chip.h
index 545e4a188a2..c0ec4b0dd59 100644
--- a/src/mesa/drivers/dri/r600/r700_chip.h
+++ b/src/mesa/drivers/dri/r600/r700_chip.h
@@ -275,6 +275,7 @@ typedef struct _R700_CHIP_CONTEXT
 	union UINT_FLOAT             	DB_DEPTH_VIEW             ;  /* 0xA001 */
 	union UINT_FLOAT             	DB_DEPTH_BASE             ;  /* 0xA003 */
 	union UINT_FLOAT             	DB_DEPTH_INFO             ;  /* 0xA004 */
+	GLboolean                       db_target_dirty;
 	union UINT_FLOAT                DB_HTILE_DATA_BASE        ;  /* 0xA005 */
 	union UINT_FLOAT          	DB_STENCIL_CLEAR          ;  /* 0xA00A */
 	union UINT_FLOAT            	DB_DEPTH_CLEAR            ;  /* 0xA00B */
diff --git a/src/mesa/drivers/dri/r600/r700_clear.c b/src/mesa/drivers/dri/r600/r700_clear.c
index 05d4af331e0..6d4ea90ccc6 100644
--- a/src/mesa/drivers/dri/r600/r700_clear.c
+++ b/src/mesa/drivers/dri/r600/r700_clear.c
@@ -46,12 +46,6 @@ static GLboolean r700ClearFast(context_t *context, GLbitfield mask)
     return GL_FALSE;
 }
 
-#define R600_NEWPRIM( rmesa )			\
-  do {						\
-  if ( rmesa->radeon.dma.flush )			\
-    rmesa->radeon.dma.flush( rmesa->radeon.glCtx );	\
-  } while (0)
-
 void r700Clear(GLcontext * ctx, GLbitfield mask)
 {
     context_t *context = R700_CONTEXT(ctx);
diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c
index 1cc886d5e9c..52f8f90dad3 100644
--- a/src/mesa/drivers/dri/r600/r700_render.c
+++ b/src/mesa/drivers/dri/r600/r700_render.c
@@ -332,30 +332,70 @@ 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__);
 
-	r700SendSQConfig(context);
-	r700SendUCPState(context);
-	r700SendSCState(context);
-	r700SendSUState(context);
-	r700SendCLState(context);
-	r700SendCBState(context);
-	r700SendDBState(context);
-	r700SendSXState(context);
-	r700SendVGTState(context);
-	r700SendSPIState(context);
-	r700SendViewportState(context, 0);
-	r700SendRenderTargetState(context, 0);
-	r700SendDepthTargetState(context);
+	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;
 
 }
 
diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c
index 835b5e18c2c..ada394ec61b 100644
--- a/src/mesa/drivers/dri/r600/r700_state.c
+++ b/src/mesa/drivers/dri/r600/r700_state.c
@@ -148,8 +148,15 @@ void r700UpdateViewportOffset(GLcontext * ctx) //------------------
 	GLfloat tx = v[MAT_TX] + xoffset;
 	GLfloat ty = (-v[MAT_TY]) + yoffset;
 
-	r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All = tx;
-	r700->viewport[id].PA_CL_VPORT_YOFFSET.f32All = ty;
+	if (r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All != tx ||
+	    r700->viewport[id].PA_CL_VPORT_YOFFSET.f32All != ty) {
+		/* Note: this should also modify whatever data the context reset
+		 * code uses...
+		 */
+		R600_STATECHANGE(context, r700->viewport[id].dirty);
+		r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All = tx;
+		r700->viewport[id].PA_CL_VPORT_YOFFSET.f32All = ty;
+	}
 
 	radeonUpdateScissor(ctx);
 }
@@ -161,6 +168,10 @@ void r700UpdateViewportOffset(GLcontext * ctx) //------------------
 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);
 
 	r700SetRenderTarget(context, 0);
 	r700SetDepthTarget(context);
@@ -233,6 +244,9 @@ static void r700InvalidateState(GLcontext * ctx, GLuint new_state) //-----------
 
 	r700UpdateStateParameters(ctx, new_state);
 
+    R600_STATECHANGE(context, r700->cl_dirty);
+    R600_STATECHANGE(context, r700->spi_dirty);
+
     if(GL_TRUE == r700->bEnablePerspective)
     {
         /* Do scale XY and Z by 1/W0 for perspective correction on pos. For orthogonal case, set both to one. */
@@ -256,14 +270,15 @@ static void r700InvalidateState(GLcontext * ctx, GLuint new_state) //-----------
         SETbit(r700->SPI_PS_IN_CONTROL_0.u32All, LINEAR_GRADIENT_ENA_bit);
     }
 
-	context->radeon.NewGLState |= new_state;
+    context->radeon.NewGLState |= new_state;
 }
 
 static void r700SetDepthState(GLcontext * ctx)
 {
 	context_t *context = R700_CONTEXT(ctx);
+	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 
-    R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
+	R600_STATECHANGE(context, r700->db_dirty);
 
     if (ctx->Depth.Test)
     {
@@ -331,6 +346,8 @@ static void r700SetAlphaState(GLcontext * ctx)
 	uint32_t alpha_func = REF_ALWAYS;
 	GLboolean really_enabled = ctx->Color.AlphaEnabled;
 
+	R600_STATECHANGE(context, r700->sx_dirty);
+
 	switch (ctx->Color.AlphaFunc) {
 	case GL_NEVER:
 		alpha_func = REF_NEVER;
@@ -383,6 +400,8 @@ 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);
+
 	r700->CB_BLEND_RED.f32All = cf[0];
 	r700->CB_BLEND_GREEN.f32All = cf[1];
 	r700->CB_BLEND_BLUE.f32All = cf[2];
@@ -451,6 +470,8 @@ static void r700SetBlendState(GLcontext * ctx)
 	int id = 0;
 	uint32_t blend_reg = 0, eqn, eqnA;
 
+	R600_STATECHANGE(context, r700->cb_dirty);
+
 	if (RGBA_LOGICOP_ENABLED(ctx) || !ctx->Color.BlendEnabled) {
 		SETfield(blend_reg,
 			 BLEND_ONE, COLOR_SRCBLEND_shift, COLOR_SRCBLEND_mask);
@@ -637,8 +658,11 @@ static GLuint translate_logicop(GLenum logicop)
  */
 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);
+
 	if (RGBA_LOGICOP_ENABLED(ctx))
 		SETfield(r700->CB_COLOR_CONTROL.u32All,
 			 translate_logicop(ctx->Color.LogicOp), ROP3_shift, ROP3_mask);
@@ -658,7 +682,10 @@ static void r700LogicOpcode(GLcontext *ctx, GLenum logicop)
 
 static void r700UpdateCulling(GLcontext * ctx)
 {
-    R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
+	context_t *context = R700_CONTEXT(ctx);
+	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
+
+    R600_STATECHANGE(context, r700->su_dirty);
 
     CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, FACE_bit);
     CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, CULL_FRONT_bit);
@@ -703,7 +730,11 @@ static void r700UpdateCulling(GLcontext * ctx)
 
 static void r700UpdateLineStipple(GLcontext * ctx)
 {
-    R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
+	context_t *context = R700_CONTEXT(ctx);
+	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
+
+    R600_STATECHANGE(context, r700->sc_dirty);
+
     if (ctx->Line.StippleFlag)
     {
 	SETbit(r700->PA_SC_MODE_CNTL.u32All, LINE_STIPPLE_ENABLE_bit);
@@ -778,14 +809,17 @@ static void r700Enable(GLcontext * ctx, GLenum cap, GLboolean state) //---------
 static void r700ColorMask(GLcontext * ctx,
 			  GLboolean r, GLboolean g, GLboolean b, GLboolean a) //------------------
 {
+	context_t *context = R700_CONTEXT(ctx);
 	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw);
 	unsigned int mask = ((r ? 1 : 0) |
 			     (g ? 2 : 0) |
 			     (b ? 4 : 0) |
 			     (a ? 8 : 0));
 
-	if (mask != r700->CB_SHADER_MASK.u32All)
+	if (mask != r700->CB_SHADER_MASK.u32All) {
+		R600_STATECHANGE(context, r700->cb_dirty);
 		SETfield(r700->CB_SHADER_MASK.u32All, mask, OUTPUT0_ENABLE_shift, OUTPUT0_ENABLE_mask);
+	}
 }
 
 /**
@@ -841,6 +875,8 @@ 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);
+
 	/* also need to set/clear FLAT_SHADE bit per param in SPI_PS_INPUT_CNTL_[0-31] */
 	switch (mode) {
 	case GL_FLAT:
@@ -862,6 +898,8 @@ 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);
+
 	/* We need to clamp to user defined range here, because
 	 * the HW clamping happens only for per vertex point size. */
 	size = CLAMP(size, ctx->Point.MinSize, ctx->Point.MaxSize);
@@ -882,6 +920,8 @@ 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);
+
 	/* format is 12.4 fixed point */
 	switch (pname) {
 	case GL_POINT_SIZE_MIN:
@@ -966,6 +1006,7 @@ static void r700SetStencilState(GLcontext * ctx, GLboolean state)
 	}
 
 	if (hw_stencil) {
+		R600_STATECHANGE(context, r700->db_dirty);
 		if (state)
 			SETbit(r700->DB_DEPTH_CONTROL.u32All, STENCIL_ENABLE_bit);
 		else
@@ -983,6 +1024,8 @@ static void r700StencilFuncSeparate(GLcontext * ctx, GLenum face,
 	//fixme
 	//r300CatchStencilFallback(ctx);
 
+	R600_STATECHANGE(context, r700->db_dirty);
+
 	//front
 	SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.Ref[0],
 		 STENCILREF_shift, STENCILREF_mask);
@@ -1012,6 +1055,8 @@ static void r700StencilMaskSeparate(GLcontext * ctx, GLenum face, GLuint mask) /
 	//fixme
 	//r300CatchStencilFallback(ctx);
 
+	R600_STATECHANGE(context, r700->db_dirty);
+
 	// front
 	SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.WriteMask[0],
 		 STENCILWRITEMASK_shift, STENCILWRITEMASK_mask);
@@ -1032,6 +1077,8 @@ static void r700StencilOpSeparate(GLcontext * ctx, GLenum face,
 	//fixme
 	//r300CatchStencilFallback(ctx);
 
+	R600_STATECHANGE(context, r700->db_dirty);
+
 	SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.FailFunc[0]),
 		 STENCILFAIL_shift, STENCILFAIL_mask);
 	SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.ZFailFunc[0]),
@@ -1074,7 +1121,7 @@ static void r700UpdateWindow(GLcontext * ctx, int id) //--------------------
 	GLfloat sz = v[MAT_SZ] * depthScale;
 	GLfloat tz = v[MAT_TZ] * depthScale;
 
-	/* TODO : Need DMA flush as well. */
+	R600_STATECHANGE(context, r700->viewport[id].dirty);
 
 	r700->viewport[id].PA_CL_VPORT_XSCALE.f32All  = sx;
 	r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All = tx;
@@ -1112,10 +1159,13 @@ static void r700LineWidth(GLcontext * ctx, GLfloat widthf) //---------------
     context_t *context = R700_CONTEXT(ctx);
     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);
+
     if (lineWidth > 0xFFFF)
-	lineWidth = 0xFFFF;
+	    lineWidth = 0xFFFF;
     SETfield(r700->PA_SU_LINE_CNTL.u32All,(uint16_t)lineWidth,
-	PA_SU_LINE_CNTL__WIDTH_shift, PA_SU_LINE_CNTL__WIDTH_mask);
+	     PA_SU_LINE_CNTL__WIDTH_shift, PA_SU_LINE_CNTL__WIDTH_mask);
 }
 
 static void r700LineStipple(GLcontext *ctx, GLint factor, GLushort pattern)
@@ -1123,6 +1173,8 @@ 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);
+
     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);
     SETfield(r700->PA_SC_LINE_STIPPLE.u32All, 1, AUTO_RESET_CNTL_shift, AUTO_RESET_CNTL_mask);
@@ -1133,6 +1185,8 @@ 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);
+
 	if (state) {
 		SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_FRONT_ENABLE_bit);
 		SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_BACK_ENABLE_bit);
@@ -1161,6 +1215,8 @@ static void r700PolygonOffset(GLcontext * ctx, GLfloat factor, GLfloat units) //
 
 	factor *= 12.0;
 
+	R600_STATECHANGE(context, r700->su_dirty);
+
 	r700->PA_SU_POLY_OFFSET_FRONT_SCALE.f32All = factor;
 	r700->PA_SU_POLY_OFFSET_FRONT_OFFSET.f32All = constant;
 	r700->PA_SU_POLY_OFFSET_BACK_SCALE.f32All = factor;
@@ -1172,6 +1228,8 @@ 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);
+
 	SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DISABLE_POLY_MODE, POLY_MODE_shift, POLY_MODE_mask);
 
 	/* Only do something if a polygon mode is wanted, default is GL_FILL */
@@ -1247,6 +1305,8 @@ 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);
+
 	r700->ucp[p].PA_CL_UCP_0_X.u32All = ip[0];
 	r700->ucp[p].PA_CL_UCP_0_Y.u32All = ip[1];
 	r700->ucp[p].PA_CL_UCP_0_Z.u32All = ip[2];
@@ -1260,6 +1320,9 @@ static void r700SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state)
 	GLuint p;
 
 	p = cap - GL_CLIP_PLANE0;
+
+	R600_STATECHANGE(context, r700->cl_dirty);
+
 	if (state) {
 		r700->PA_CL_CLIP_CNTL.u32All |= (UCP_ENA_0_bit << p);
 		r700->ucp[p].enabled = GL_TRUE;
@@ -1293,6 +1356,8 @@ void r700SetScissor(context_t *context) //---------------
 		y2 = rrb->dPriv->y + rrb->dPriv->h;
 	}
 
+	R600_STATECHANGE(context, r700->sc_dirty);
+
 	/* window */
 	SETbit(r700->PA_SC_WINDOW_SCISSOR_TL.u32All, WINDOW_OFFSET_DISABLE_bit);
 	SETfield(r700->PA_SC_WINDOW_SCISSOR_TL.u32All, x1,
@@ -1361,6 +1426,9 @@ static void r700SetRenderTarget(context_t *context, int id)
 	    return;
     }
 
+    R600_STATECHANGE(context, r700->render_target[id].dirty);
+    R600_STATECHANGE(context, r700->cb_dirty);
+
     /* screen/window/view */
     SETfield(r700->CB_TARGET_MASK.u32All, 0xF, (4 * id), TARGET0_ENABLE_mask);
 
@@ -1407,6 +1475,9 @@ static void r700SetDepthTarget(context_t *context)
     if (!rrb)
 	    return;
 
+    R600_STATECHANGE(context, r700->db_target_dirty);
+    R600_STATECHANGE(context, r700->db_dirty);
+
     /* depth buf */
     r700->DB_DEPTH_SIZE.u32All = 0;
     r700->DB_DEPTH_BASE.u32All = 0;
@@ -1467,6 +1538,8 @@ static void r700InitSQConfig(GLcontext * ctx)
     int num_gs_stack_entries;
     int num_es_stack_entries;
 
+    R600_STATECHANGE(context, r700->sq_dirty);
+
     // SQ
     ps_prio = 0;
     vs_prio = 1;