swr/rast: Add some archrast stats

Add stats for degenerate and backfacing primitive counts

Wire archrast stats for alpha blend and alpha test.
pass value to jitter, upon return have archrast event increment a value

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

9 files changed, 105 insertions, 11 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/archrast/archrast.cpp b/src/gallium/drivers/swr/rasterizer/archrast/archrast.cpp
index 1f87dbabcd..12dfc0e337 100644
--- a/src/gallium/drivers/swr/rasterizer/archrast/archrast.cpp
+++ b/src/gallium/drivers/swr/rasterizer/archrast/archrast.cpp
@@ -73,6 +73,18 @@ namespace ArchRast
         uint32_t rasterTiles = 0;
     };
 
+    struct CullStats
+    {
+        uint32_t degeneratePrimCount = 0;
+        uint32_t backfacePrimCount = 0;
+    };
+
+    struct AlphaStats
+    {
+        uint32_t alphaTestCount = 0;
+        uint32_t alphaBlendCount = 0;
+    };
+
     //////////////////////////////////////////////////////////////////////////
     /// @brief Event handler that handles API thread events. This is shared
     ///        between the API and its caller (e.g. driver shim) but typically
@@ -280,7 +292,12 @@ namespace ArchRast
             // Rasterized Subspans
             EventHandlerFile::Handle(RasterTiles(drawId, rastStats.rasterTiles));
 
-            //Reset Internal Counters
+            // Alpha Subspans
+            EventHandlerFile::Handle(AlphaEvent(drawId, mAlphaStats.alphaTestCount, mAlphaStats.alphaBlendCount));
+
+            // Primitive Culling
+            EventHandlerFile::Handle(CullEvent(drawId, mCullStats.backfacePrimCount, mCullStats.degeneratePrimCount));
+            
             mDSSingleSample = {};
             mDSSampleRate = {};
             mDSCombined = {};
@@ -288,6 +305,8 @@ namespace ArchRast
             mDSNullPS = {};
 
             rastStats = {};
+            mCullStats = {};
+            mAlphaStats = {};
             mNeedFlush = false;
         }
 
@@ -327,6 +346,18 @@ namespace ArchRast
             rastStats.rasterTiles += event.data.rasterTiles;
         }
 
+        virtual void Handle(const CullInfoEvent& event)
+        {
+            mCullStats.degeneratePrimCount += _mm_popcnt_u32(event.data.validMask ^ (event.data.validMask & ~event.data.degeneratePrimMask));
+            mCullStats.backfacePrimCount   += _mm_popcnt_u32(event.data.validMask ^ (event.data.validMask & ~event.data.backfacePrimMask));
+        }
+
+        virtual void Handle(const AlphaInfoEvent& event)
+        {
+            mAlphaStats.alphaTestCount  += event.data.alphaTestEnable;
+            mAlphaStats.alphaBlendCount += event.data.alphaBlendEnable;
+        }
+
     protected:
         bool mNeedFlush;
         // Per draw stats
@@ -340,6 +371,8 @@ namespace ArchRast
         TEStats mTS = {};
         GSStats mGS = {};
         RastStats rastStats = {};
+        CullStats mCullStats = {};
+        AlphaStats mAlphaStats = {};
 
     };
 
diff --git a/src/gallium/drivers/swr/rasterizer/archrast/events.proto b/src/gallium/drivers/swr/rasterizer/archrast/events.proto
index 7d9a68d502..deb0373bf5 100644
--- a/src/gallium/drivers/swr/rasterizer/archrast/events.proto
+++ b/src/gallium/drivers/swr/rasterizer/archrast/events.proto
@@ -180,6 +180,7 @@ event LateStencilSampleRate
     uint64_t failCount;
 };
 
+// Total Early-Z counts, SingleSample and SampleRate
 event EarlyZ
 {
     uint32_t drawId;
@@ -187,6 +188,7 @@ event EarlyZ
     uint64_t failCount;
 };	
 
+// Total LateZ counts, SingleSample and SampleRate
 event LateZ
 {
     uint32_t drawId;
@@ -194,6 +196,7 @@ event LateZ
     uint64_t failCount;
 };
 
+// Total EarlyStencil counts, SingleSample and SampleRate
 event EarlyStencil
 {
     uint32_t drawId; 
@@ -201,6 +204,7 @@ event EarlyStencil
     uint64_t failCount;
 };
 
+// Total LateStencil counts, SingleSample and SampleRate
 event LateStencil
 {
     uint32_t drawId; 
@@ -302,3 +306,18 @@ event ClipperEvent
     uint32_t trivialAcceptCount;
     uint32_t mustClipCount;
 };
+
+event CullEvent
+{
+    uint32_t drawId;
+    uint64_t backfacePrimCount;
+    uint64_t degeneratePrimCount;
+};
+
+event AlphaEvent
+{
+    uint32_t drawId;
+    uint32_t alphaTestCount;
+    uint32_t alphaBlendCount;
+};
+
diff --git a/src/gallium/drivers/swr/rasterizer/archrast/events_private.proto b/src/gallium/drivers/swr/rasterizer/archrast/events_private.proto
index f0a93107e5..37593befbc 100644
--- a/src/gallium/drivers/swr/rasterizer/archrast/events_private.proto
+++ b/src/gallium/drivers/swr/rasterizer/archrast/events_private.proto
@@ -120,6 +120,21 @@ event ClipInfoEvent
     uint32_t clipMask;
 };
 
+event CullInfoEvent
+{
+    uint32_t drawId;
+    uint64_t degeneratePrimMask;
+    uint64_t backfacePrimMask;
+    uint32_t validMask;
+};
+
+event AlphaInfoEvent
+{
+    uint32_t drawId;
+    uint32_t alphaTestEnable;
+    uint32_t alphaBlendEnable;
+};
+
 event DrawInstancedEvent
 {
     uint32_t drawId;
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_impl.h b/src/gallium/drivers/swr/rasterizer/core/backend_impl.h
index 8c539e31dc..dd349a1d3c 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_impl.h
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_impl.h
@@ -709,8 +709,8 @@ static INLINE void CalcSampleBarycentrics(const BarycentricCoeffs& coeffs, SWR_P
 }
 
 // Merge Output to 4x2 SIMD Tile Format
-INLINE void OutputMerger4x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState,
-    const PFN_BLEND_JIT_FUNC (&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar const &depthPassMask, uint32_t renderTargetMask)
+INLINE void OutputMerger4x2(DRAW_CONTEXT *pDC, SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState,
+    const PFN_BLEND_JIT_FUNC (&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar const &depthPassMask, uint32_t renderTargetMask, uint32_t workerId)
 {
     // type safety guaranteed from template instantiation in BEChooser<>::GetFunc
     const uint32_t rasterTileColorOffset = RasterTileColorOffset(sample);
@@ -747,6 +747,9 @@ INLINE void OutputMerger4x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
             }
         }
 
+        // Track alpha events
+        AR_EVENT(AlphaInfoEvent(pDC->drawId, blendContext.isAlphaTested, blendContext.isAlphaBlended));
+
         // final write mask 
         simdscalari outputMask = _simd_castps_si(_simd_and_ps(coverageMask, depthPassMask));
 
@@ -777,8 +780,8 @@ INLINE void OutputMerger4x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
 
 #if USE_8x2_TILE_BACKEND
 // Merge Output to 8x2 SIMD16 Tile Format
-INLINE void OutputMerger8x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState,
-    const PFN_BLEND_JIT_FUNC(&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar const &depthPassMask, uint32_t renderTargetMask, bool useAlternateOffset)
+INLINE void OutputMerger8x2(DRAW_CONTEXT *pDC, SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState,
+    const PFN_BLEND_JIT_FUNC(&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar const &depthPassMask, uint32_t renderTargetMask, bool useAlternateOffset, uint32_t workerId)
 {
     // type safety guaranteed from template instantiation in BEChooser<>::GetFunc
     uint32_t rasterTileColorOffset = RasterTileColorOffset(sample);
@@ -836,6 +839,9 @@ INLINE void OutputMerger8x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
             }
         }
 
+        // Track alpha events
+        AR_EVENT(AlphaInfoEvent(pDC->drawId, blendContext.isAlphaTested, blendContext.isAlphaBlended));
+
         // final write mask 
         simdscalari outputMask = _simd_castps_si(_simd_and_ps(coverageMask, depthPassMask));
 
@@ -1003,9 +1009,9 @@ void BackendPixelRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t
                 
                 // broadcast the results of the PS to all passing pixels
 #if USE_8x2_TILE_BACKEND
-                OutputMerger8x2(psContext, psContext.pColorBuffer, sample, &state.blendState,state.pfnBlendFunc, coverageMask, depthMask, state.psState.renderTargetMask, useAlternateOffset);
+                OutputMerger8x2(pDC, psContext, psContext.pColorBuffer, sample, &state.blendState,state.pfnBlendFunc, coverageMask, depthMask, state.psState.renderTargetMask, useAlternateOffset, workerId);
 #else // USE_8x2_TILE_BACKEND
-                OutputMerger4x2(psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, coverageMask, depthMask, state.psState.renderTargetMask);
+                OutputMerger4x2(pDC, psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, coverageMask, depthMask, state.psState.renderTargetMask, workerId);
 #endif // USE_8x2_TILE_BACKEND
 
                 if(!state.psState.forceEarlyZ && !T::bForcedSampleCount)
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp b/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp
index 16418f7dc8..4982025a78 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp
@@ -196,9 +196,9 @@ void BackendSampleRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_
                     // output merger
                     RDTSC_BEGIN(BEOutputMerger, pDC->drawId);
 #if USE_8x2_TILE_BACKEND
-                    OutputMerger8x2(psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask, useAlternateOffset);
+                    OutputMerger8x2(pDC, psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask, useAlternateOffset, workerId);
 #else
-                    OutputMerger4x2(psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask);
+                    OutputMerger4x2(pDC, psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask, workerId);
 #endif
 
                     // do final depth write after all pixel kills
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp b/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp
index 4cc1ed5266..452fba13ed 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp
@@ -181,9 +181,9 @@ void BackendSingleSample(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint3
                 // output merger
                 RDTSC_BEGIN(BEOutputMerger, pDC->drawId);
 #if USE_8x2_TILE_BACKEND
-                OutputMerger8x2(psContext, psContext.pColorBuffer, 0, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask, useAlternateOffset);
+                OutputMerger8x2(pDC, psContext, psContext.pColorBuffer, 0, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask, useAlternateOffset, workerId);
 #else
-                OutputMerger4x2(psContext, psContext.pColorBuffer, 0, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask);
+                OutputMerger4x2(pDC, psContext, psContext.pColorBuffer, 0, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.renderTargetMask, workerId, workerId);
 #endif
 
                 // do final depth write after all pixel kills
diff --git a/src/gallium/drivers/swr/rasterizer/core/binner.cpp b/src/gallium/drivers/swr/rasterizer/core/binner.cpp
index c9a37cb17a..d31fd37095 100644
--- a/src/gallium/drivers/swr/rasterizer/core/binner.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/binner.cpp
@@ -773,6 +773,8 @@ void SIMDCALL BinTrianglesImpl(
         RDTSC_EVENT(FECullZeroAreaAndBackface, _mm_popcnt_u32(origTriMask ^ triMask), 0);
     }
 
+    AR_EVENT(CullInfoEvent(pDC->drawId, cullZeroAreaMask, cullTris, origTriMask));
+
     /// Note: these variable initializations must stay above any 'goto endBenTriangles'
     // compute per tri backface
     uint32_t frontFaceMask = frontWindingTris;
diff --git a/src/gallium/drivers/swr/rasterizer/core/state.h b/src/gallium/drivers/swr/rasterizer/core/state.h
index 8c26ec60a2..22acbe054a 100644
--- a/src/gallium/drivers/swr/rasterizer/core/state.h
+++ b/src/gallium/drivers/swr/rasterizer/core/state.h
@@ -887,6 +887,8 @@ struct SWR_BLEND_CONTEXT
     simdvector*             result;
     simdscalari*            oMask;
     simdscalari*            pMask;
+    uint32_t                isAlphaTested;
+    uint32_t                isAlphaBlended;
 };
 
 //////////////////////////////////////////////////////////////////////////
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp
index 6b7efbfb6d..912a88fd00 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp
@@ -557,6 +557,8 @@ struct BlendJit : public Builder
         ppoMask->setName("ppoMask");
         Value* ppMask = LOAD(pBlendContext, { 0, SWR_BLEND_CONTEXT_pMask });
         ppMask->setName("pMask");
+        Value* AlphaTest1 = LOAD(pBlendContext, { 0, SWR_BLEND_CONTEXT_isAlphaBlended });
+        ppMask->setName("AlphaTest1");
 
         static_assert(KNOB_COLOR_HOT_TILE_FORMAT == R32G32B32A32_FLOAT, "Unsupported hot tile format");
         Value* dst[4];
@@ -590,12 +592,22 @@ struct BlendJit : public Builder
         // alpha test
         if (state.desc.alphaTestEnable)
         {
+            // Gather for archrast stats
+            STORE(C(1), pBlendContext, { 0, SWR_BLEND_CONTEXT_isAlphaTested });
             AlphaTest(state, pBlendState, pSrc0Alpha, ppMask);
         }
+        else
+        {
+            // Gather for archrast stats
+            STORE(C(0), pBlendContext, { 0, SWR_BLEND_CONTEXT_isAlphaTested });
+        }
 
         // color blend
         if (state.blendState.blendEnable)
         {
+            // Gather for archrast stats
+            STORE(C(1), pBlendContext, { 0, SWR_BLEND_CONTEXT_isAlphaBlended });
+
             // clamp sources
             Clamp(state.format, src);
             Clamp(state.format, src1);
@@ -647,6 +659,11 @@ struct BlendJit : public Builder
                 STORE(result[i], pResult, { 0, i });
             }
         }
+        else
+        {
+            // Gather for archrast stats
+            STORE(C(0), pBlendContext, { 0, SWR_BLEND_CONTEXT_isAlphaBlended });
+        }
         
         if(state.blendState.logicOpEnable)
         {