r600: rework DB render setup

- consolidate DB render setup
- only enable perfect ZPASS counts and cull disable
when OQ is active
- enable early Z

4 files changed, 73 insertions, 42 deletions

diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.c b/src/mesa/drivers/dri/r600/r700_fragprog.c
index 0f549ead9c..ccafd433bf 100644
--- a/src/mesa/drivers/dri/r600/r700_fragprog.c
+++ b/src/mesa/drivers/dri/r600/r700_fragprog.c
@@ -393,26 +393,6 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx)
     SETfield(r700->ps.SQ_PGM_EXPORTS_PS.u32All, fp->r700Shader.exportMode,
              EXPORT_MODE_shift, EXPORT_MODE_mask);
 
-    R600_STATECHANGE(context, db);
-
-    if(fp->r700Shader.killIsUsed)
-    {
-	    SETbit(r700->DB_SHADER_CONTROL.u32All, KILL_ENABLE_bit);
-    }
-    else
-    {
-        CLEARbit(r700->DB_SHADER_CONTROL.u32All, KILL_ENABLE_bit);
-    }
-
-    if(fp->r700Shader.depthIsExported)
-    {
-	    SETbit(r700->DB_SHADER_CONTROL.u32All, Z_EXPORT_ENABLE_bit);
-    }
-    else
-    {
-        CLEARbit(r700->DB_SHADER_CONTROL.u32All, Z_EXPORT_ENABLE_bit);
-    }
-
     // emit ps input map
     unBit = 1 << FRAG_ATTRIB_WPOS;
     if(mesa_fp->Base.InputsRead & unBit)
@@ -479,9 +459,12 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx)
 	    }
     }
 
-    R600_STATECHANGE(context, cb);
     exportCount = (r700->ps.SQ_PGM_EXPORTS_PS.u32All & EXPORT_MODE_mask) / (1 << EXPORT_MODE_shift);
-    r700->CB_SHADER_CONTROL.u32All = (1 << exportCount) - 1;
+    if (r700->CB_SHADER_CONTROL.u32All != ((1 << exportCount) - 1))
+    {
+	    R600_STATECHANGE(context, cb);
+	    r700->CB_SHADER_CONTROL.u32All = (1 << exportCount) - 1;
+    }
 
     /* sent out shader constants. */
     paramList = fp->mesa_program.Base.Parameters;
diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c
index c345b9d8ac..47f89c91f8 100644
--- a/src/mesa/drivers/dri/r600/r700_render.c
+++ b/src/mesa/drivers/dri/r600/r700_render.c
@@ -59,9 +59,7 @@
 
 void r700WaitForIdle(context_t *context);
 void r700WaitForIdleClean(context_t *context);
-GLboolean r700SendTextureState(context_t *context);
 static unsigned int r700PrimitiveType(int prim);
-void r600UpdateTextureState(GLcontext * ctx);
 GLboolean r700SyncSurf(context_t *context,
 		       struct radeon_bo *pbo,
 		       uint32_t read_domain,
@@ -891,7 +889,7 @@ static GLboolean r700TryDrawPrims(GLcontext *ctx,
     r700SetScissor(context);
     r700SetupVertexProgram(ctx);
     r700SetupFragmentProgram(ctx);
-    r600UpdateTextureState(ctx);
+    r700UpdateShaderStates(ctx);
 
     GLuint emit_end = r700PredictRenderSize(ctx, prim, ib, nr_prims)
                     + context->radeon.cmdbuf.cs->cdw;
diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c
index b95fc87d30..41000dc8ce 100644
--- a/src/mesa/drivers/dri/r600/r700_state.c
+++ b/src/mesa/drivers/dri/r600/r700_state.c
@@ -54,7 +54,7 @@
 #include "r700_fragprog.h"
 #include "r700_vertprog.h"
 
-
+void r600UpdateTextureState(GLcontext * ctx);
 static void r700SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state);
 static void r700UpdatePolygonMode(GLcontext * ctx);
 static void r700SetPolygonOffsetState(GLcontext * ctx, GLboolean state);
@@ -191,6 +191,70 @@ static void r700InvalidateState(GLcontext * ctx, GLuint new_state) //-----------
     context->radeon.NewGLState |= new_state;
 }
 
+static void r700SetDBRenderState(GLcontext * ctx)
+{
+	context_t *context = R700_CONTEXT(ctx);
+	R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
+	struct r700_fragment_program *fp = (struct r700_fragment_program *)
+		(ctx->FragmentProgram._Current);
+
+	R600_STATECHANGE(context, db);
+
+	SETbit(r700->DB_SHADER_CONTROL.u32All, DUAL_EXPORT_ENABLE_bit);
+	SETfield(r700->DB_SHADER_CONTROL.u32All, EARLY_Z_THEN_LATE_Z, Z_ORDER_shift, Z_ORDER_mask);
+	/* XXX not sure if this is required */
+	if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)
+		SETbit(r700->DB_RENDER_OVERRIDE.u32All, FORCE_SHADER_Z_ORDER_bit);
+	/* XXX need to enable htile for hiz/s */
+	SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIZ_ENABLE_shift, FORCE_HIZ_ENABLE_mask);
+	SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE0_shift, FORCE_HIS_ENABLE0_mask);
+	SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE1_shift, FORCE_HIS_ENABLE1_mask);
+
+	if (context->radeon.query.current)
+	{
+		SETbit(r700->DB_RENDER_OVERRIDE.u32All, NOOP_CULL_DISABLE_bit);
+		if (context->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV770)
+		{
+			SETbit(r700->DB_RENDER_CONTROL.u32All, PERFECT_ZPASS_COUNTS_bit);
+		}
+	}
+	else
+	{
+		CLEARbit(r700->DB_RENDER_OVERRIDE.u32All, NOOP_CULL_DISABLE_bit);
+		if (context->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV770)
+		{
+			CLEARbit(r700->DB_RENDER_CONTROL.u32All, PERFECT_ZPASS_COUNTS_bit);
+		}
+	}
+
+	if (fp)
+	{
+		if (fp->r700Shader.killIsUsed)
+		{
+			SETbit(r700->DB_SHADER_CONTROL.u32All, KILL_ENABLE_bit);
+		}
+		else
+		{
+			CLEARbit(r700->DB_SHADER_CONTROL.u32All, KILL_ENABLE_bit);
+		}
+
+		if (fp->r700Shader.depthIsExported)
+		{
+			SETbit(r700->DB_SHADER_CONTROL.u32All, Z_EXPORT_ENABLE_bit);
+		}
+		else
+		{
+			CLEARbit(r700->DB_SHADER_CONTROL.u32All, Z_EXPORT_ENABLE_bit);
+		}
+	}
+}
+
+void r700UpdateShaderStates(GLcontext * ctx)
+{
+	r700SetDBRenderState(ctx);
+	r600UpdateTextureState(ctx);
+}
+
 static void r700SetDepthState(GLcontext * ctx)
 {
 	context_t *context = R700_CONTEXT(ctx);
@@ -1672,24 +1736,10 @@ void r700InitState(GLcontext * ctx) //-------------------
     r700Enable(ctx, GL_DEPTH_TEST, ctx->Depth.Test);
     r700DepthMask(ctx, ctx->Depth.Mask);
     r700DepthFunc(ctx, ctx->Depth.Func);
-    SETbit(r700->DB_SHADER_CONTROL.u32All, DUAL_EXPORT_ENABLE_bit);
-
     r700->DB_DEPTH_CLEAR.u32All     = 0x3F800000;
-
-    r700->DB_RENDER_CONTROL.u32All  = 0;
     SETbit(r700->DB_RENDER_CONTROL.u32All, STENCIL_COMPRESS_DISABLE_bit);
     SETbit(r700->DB_RENDER_CONTROL.u32All, DEPTH_COMPRESS_DISABLE_bit);
-    r700->DB_RENDER_OVERRIDE.u32All = 0;
-    if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)
-	    SETbit(r700->DB_RENDER_OVERRIDE.u32All, FORCE_SHADER_Z_ORDER_bit);
-    SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIZ_ENABLE_shift, FORCE_HIZ_ENABLE_mask);
-    SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE0_shift, FORCE_HIS_ENABLE0_mask);
-    SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE1_shift, FORCE_HIS_ENABLE1_mask);
-    SETbit(r700->DB_RENDER_OVERRIDE.u32All, NOOP_CULL_DISABLE_bit);
-    if (context->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV770)
-    {
-	    CLEARbit(r700->DB_RENDER_CONTROL.u32All, PERFECT_ZPASS_COUNTS_bit);
-    }
+    r700SetDBRenderState(ctx);
 
     r700->DB_ALPHA_TO_MASK.u32All = 0;
     SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET0_shift, ALPHA_TO_MASK_OFFSET0_mask);
diff --git a/src/mesa/drivers/dri/r600/r700_state.h b/src/mesa/drivers/dri/r600/r700_state.h
index 209189d8d7..60c6a7f23c 100644
--- a/src/mesa/drivers/dri/r600/r700_state.h
+++ b/src/mesa/drivers/dri/r600/r700_state.h
@@ -35,7 +35,7 @@
 
 extern void r700UpdateStateParameters(GLcontext * ctx, GLuint new_state);
 extern void r700UpdateShaders (GLcontext * ctx);
-extern void r700UpdateShaders2(GLcontext * ctx);
+extern void r700UpdateShaderStates(GLcontext * ctx);
 
 extern void r700UpdateViewportOffset(GLcontext * ctx);