swr/rast: more flexible max attribute slots

Ability to allocate space for an arbitrary number (at compile time)
of positions in the vertex layout.

Removes KNOB_NUM_ATTRIBUTES from knobs.h, replaces the VTX slot
number #defines with the SWR_VTX_SLOTS enum (which contains
replacement for NUM_ATTRIBUTES: SWR_VTX_NUM_SLOTS)

Reviewed-by: Bruce Cherniak <[email protected]>

7 files changed, 30 insertions, 27 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
index 8cf234cd67a..08a4fd3b61d 100644
--- a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
@@ -510,7 +510,7 @@ static void StreamOut(
     uint32_t soVertsPerPrim = NumVertsPerPrim(pa.binTopology, false);
 
     // The pPrimData buffer is sparse in that we allocate memory for all 32 attributes for each vertex.
-    uint32_t primDataDwordVertexStride = (KNOB_NUM_ATTRIBUTES * sizeof(float) * 4) / sizeof(uint32_t);
+    uint32_t primDataDwordVertexStride = (SWR_VTX_NUM_SLOTS * sizeof(float) * 4) / sizeof(uint32_t);
 
     SWR_STREAMOUT_CONTEXT soContext = { 0 };
 
@@ -618,13 +618,13 @@ INLINE static T RoundDownEven(T value)
 ///
 /// attribCount will limit the vector copies to those attribs specified
 ///
-/// note: the stride between vertexes is determinded by KNOB_NUM_ATTRIBUTES
+/// note: the stride between vertexes is determinded by SWR_VTX_NUM_SLOTS
 ///
 void PackPairsOfSimdVertexIntoSimd16Vertex(simd16vertex *vertex_simd16, const simdvertex *vertex, uint32_t vertexCount, uint32_t attribCount)
 {
     SWR_ASSERT(vertex);
     SWR_ASSERT(vertex_simd16);
-    SWR_ASSERT(attribCount <= KNOB_NUM_ATTRIBUTES);
+    SWR_ASSERT(attribCount <= SWR_VTX_NUM_SLOTS);
 
     simd16vertex temp;
 
@@ -709,7 +709,7 @@ void ProcessStreamIdBuffer(uint32_t stream, uint8_t* pStreamIdBase, uint32_t num
             }
             curInputByte >>= 2;
         }
-        
+
         *pCutBuffer++ = outByte;
     }
 }
@@ -810,7 +810,7 @@ static void GeometryShaderStage(
             tlsGsContext.vert[i].attrib[attribSlot] = attrib[i];
         }
     }
-    
+
     // assemble position
     pa.Assemble(VERTEX_POSITION_SLOT, attrib);
     for (uint32_t i = 0; i < numVertsPerPrim; ++i)
@@ -890,7 +890,7 @@ static void GeometryShaderStage(
 
             uint8_t* pBase = pInstanceBase + instance * bufferInfo.vertexInstanceStride;
             uint8_t* pCutBase = pCutBufferBase + instance * bufferInfo.cutInstanceStride;
-            
+
             uint32_t numAttribs = state.feNumAttributes;
 
             for (uint32_t stream = 0; stream < MAX_SO_STREAMS; ++stream)
@@ -930,7 +930,7 @@ static void GeometryShaderStage(
                     tempVertex_simd16,
                     reinterpret_cast<const simdvertex *>(pBase),
                     numEmittedVerts,
-                    KNOB_NUM_ATTRIBUTES);
+                    SWR_VTX_NUM_SLOTS);
 
 #endif
 #if USE_SIMD16_FRONTEND
diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.h b/src/gallium/drivers/swr/rasterizer/core/frontend.h
index a9c36b4559b..1ce51bbd5df 100644
--- a/src/gallium/drivers/swr/rasterizer/core/frontend.h
+++ b/src/gallium/drivers/swr/rasterizer/core/frontend.h
@@ -34,7 +34,7 @@
 // TODO: this belongs in state.h alongside the simdvector definition, but there is a llvm codegen issue
 struct simd16vertex
 {
-    simd16vector    attrib[KNOB_NUM_ATTRIBUTES];
+    simd16vector    attrib[SWR_VTX_NUM_SLOTS];
 };
 
 #endif
diff --git a/src/gallium/drivers/swr/rasterizer/core/knobs.h b/src/gallium/drivers/swr/rasterizer/core/knobs.h
index 7928f5d6d76..640b6726ca4 100644
--- a/src/gallium/drivers/swr/rasterizer/core/knobs.h
+++ b/src/gallium/drivers/swr/rasterizer/core/knobs.h
@@ -97,9 +97,6 @@
 // Maximum supported number of active vertex buffer streams
 #define KNOB_NUM_STREAMS                    32
 
-// Maximum supported number of attributes per vertex
-#define KNOB_NUM_ATTRIBUTES                 39
-
 // Maximum supported active viewports and scissors
 #define KNOB_NUM_VIEWPORTS_SCISSORS         16
 
diff --git a/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp b/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp
index af54779653e..4df146e9f99 100644
--- a/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp
@@ -510,7 +510,7 @@ __declspec(thread) volatile uint64_t gToss;
 
 static const uint32_t vertsPerTri = 3, componentsPerAttrib = 4;
 // try to avoid _chkstk insertions; make this thread local
-static THREAD OSALIGNLINE(float) perspAttribsTLS[vertsPerTri * KNOB_NUM_ATTRIBUTES * componentsPerAttrib];
+static THREAD OSALIGNLINE(float) perspAttribsTLS[vertsPerTri * SWR_VTX_NUM_SLOTS * componentsPerAttrib];
 
 INLINE
 void ComputeEdgeData(int32_t a, int32_t b, EDGE& edge)
@@ -1312,7 +1312,7 @@ void RasterizeTriPoint(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile,
     newWorkDesc.pTriBuffer = &newTriBuffer[0];
 
     // create a copy of the attrib buffer to write our adjusted attribs to
-    OSALIGNSIMD(float) newAttribBuffer[4 * 3 * KNOB_NUM_ATTRIBUTES];
+    OSALIGNSIMD(float) newAttribBuffer[4 * 3 * SWR_VTX_NUM_SLOTS];
     newWorkDesc.pAttribs = &newAttribBuffer[0];
 
     newWorkDesc.pUserClipBuffer = workDesc.pUserClipBuffer;
@@ -1597,7 +1597,7 @@ void RasterizeLine(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, voi
     newWorkDesc.pTriBuffer = &newTriBuffer[0];
 
     // create a copy of the attrib buffer to write our adjusted attribs to
-    OSALIGNSIMD(float) newAttribBuffer[4 * 3 * KNOB_NUM_ATTRIBUTES];
+    OSALIGNSIMD(float) newAttribBuffer[4 * 3 * SWR_VTX_NUM_SLOTS];
     newWorkDesc.pAttribs = &newAttribBuffer[0];
 
     const __m128 vBloat0 = _mm_set_ps(0.5f, -0.5f, -0.5f, 0.5f);
diff --git a/src/gallium/drivers/swr/rasterizer/core/state.h b/src/gallium/drivers/swr/rasterizer/core/state.h
index 535b85e8593..3d0b4ff951f 100644
--- a/src/gallium/drivers/swr/rasterizer/core/state.h
+++ b/src/gallium/drivers/swr/rasterizer/core/state.h
@@ -179,19 +179,25 @@ enum SWR_OUTER_TESSFACTOR_ID
 ///        space for up to 32 attributes, as well as any SGV values generated
 ///        by the pipeline
 /////////////////////////////////////////////////////////////////////////
-#define VERTEX_POSITION_SLOT 0
-#define VERTEX_ATTRIB_START_SLOT 1
-#define VERTEX_ATTRIB_END_SLOT 32
-#define VERTEX_RTAI_SLOT 33         // GS writes RenderTargetArrayIndex here
-#define VERTEX_PRIMID_SLOT 34       // GS writes PrimId here
-#define VERTEX_CLIPCULL_DIST_LO_SLOT 35 // VS writes lower 4 clip/cull dist
-#define VERTEX_CLIPCULL_DIST_HI_SLOT 36 // VS writes upper 4 clip/cull dist
-#define VERTEX_POINT_SIZE_SLOT 37       // VS writes point size here
-#define VERTEX_VIEWPORT_ARRAY_INDEX_SLOT 38
+enum SWR_VTX_SLOTS
+{
+    VERTEX_POSITION_SLOT             = 0,
+    VERTEX_POSITION_END_SLOT         = 0,
+    VERTEX_ATTRIB_START_SLOT         = ( 1 + VERTEX_POSITION_END_SLOT),
+    VERTEX_ATTRIB_END_SLOT           = (32 + VERTEX_POSITION_END_SLOT),
+    VERTEX_RTAI_SLOT                 = (33 + VERTEX_POSITION_END_SLOT), // GS writes RenderTargetArrayIndex here
+    VERTEX_PRIMID_SLOT               = (34 + VERTEX_POSITION_END_SLOT), // GS writes PrimId here
+    VERTEX_CLIPCULL_DIST_LO_SLOT     = (35 + VERTEX_POSITION_END_SLOT), // VS writes lower 4 clip/cull dist
+    VERTEX_CLIPCULL_DIST_HI_SLOT     = (36 + VERTEX_POSITION_END_SLOT), // VS writes upper 4 clip/cull dist
+    VERTEX_POINT_SIZE_SLOT           = (37 + VERTEX_POSITION_END_SLOT), // VS writes point size here
+    VERTEX_VIEWPORT_ARRAY_INDEX_SLOT = (38 + VERTEX_POSITION_END_SLOT),
+    SWR_VTX_NUM_SLOTS                 = VERTEX_VIEWPORT_ARRAY_INDEX_SLOT,
+};
+
 // SoAoSoA
 struct simdvertex
 {
-    simdvector    attrib[KNOB_NUM_ATTRIBUTES];
+    simdvector    attrib[SWR_VTX_NUM_SLOTS];
 };
 
 //////////////////////////////////////////////////////////////////////////
@@ -226,7 +232,7 @@ struct ScalarAttrib
 
 struct ScalarCPoint
 {
-    ScalarAttrib attrib[KNOB_NUM_ATTRIBUTES];
+    ScalarAttrib attrib[SWR_VTX_NUM_SLOTS];
 };
 
 //////////////////////////////////////////////////////////////////////////
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h
index d5cec70c2e8..4f456afffce 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h
@@ -95,7 +95,7 @@ enum ComponentControl
 struct FETCH_COMPILE_STATE
 {
     uint32_t numAttribs{ 0 };
-    INPUT_ELEMENT_DESC layout[KNOB_NUM_ATTRIBUTES];
+    INPUT_ELEMENT_DESC layout[SWR_VTX_NUM_SLOTS];
     SWR_FORMAT indexType;
     uint32_t cutIndex{ 0xffffffff };
 
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp
index 2c19321fb3f..dbceb36c213 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp
@@ -241,7 +241,7 @@ struct StreamOutJit : public Builder
 
             // increment stream and output buffer pointers
             // stream verts are always 32*4 dwords apart
-            pStreamData = GEP(pStreamData, C(KNOB_NUM_ATTRIBUTES * 4));
+            pStreamData = GEP(pStreamData, C(SWR_VTX_NUM_SLOTS * 4));
 
             // output buffers offset using pitch in buffer state
             for (uint32_t b : activeSOBuffers)