swr/rast: Support dynamically sized vertex layout

Each shader stage state (VS, TS, GS, SO, BE/CLIP) now has a
vertexAttribOffset to specify the offset to the start of the
general attribute section of the incoming verts for that stage.
It is up to the driver to set this up correctly based on the
active stages. All the shader stages use this value instead of
VERTEX_ATTRIB_START_SLOT to offset to the incoming attributes.

Only the vertex shader stage supports dynamic layout output
currently. The other stages continue to expect the output to be
the fixed layout slots as before. Will be enabling GS next.

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

6 files changed, 49 insertions, 21 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/core/binner.cpp b/src/gallium/drivers/swr/rasterizer/core/binner.cpp
index 036d8b1e7d..19eef9bb15 100644
--- a/src/gallium/drivers/swr/rasterizer/core/binner.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/binner.cpp
@@ -80,12 +80,12 @@ INLINE void ProcessAttributes(
         if (IsSwizzledT::value)
         {
             SWR_ATTRIB_SWIZZLE attribSwizzle = backendState.swizzleMap[i];
-            inputSlot = VERTEX_ATTRIB_START_SLOT + attribSwizzle.sourceAttrib;
+            inputSlot = backendState.vertexAttribOffset + attribSwizzle.sourceAttrib;
 
         }
         else
         {
-            inputSlot = VERTEX_ATTRIB_START_SLOT + i;
+            inputSlot = backendState.vertexAttribOffset + i;
         }
 
         __m128 attrib[3];    // triangle attribs (always 4 wide)
diff --git a/src/gallium/drivers/swr/rasterizer/core/clip.h b/src/gallium/drivers/swr/rasterizer/core/clip.h
index 12b52c5847..4f940d931c 100644
--- a/src/gallium/drivers/swr/rasterizer/core/clip.h
+++ b/src/gallium/drivers/swr/rasterizer/core/clip.h
@@ -489,7 +489,7 @@ public:
             // Compute absolute attrib slot in vertex array
             uint32_t mapSlot = backendState.swizzleEnable ? backendState.swizzleMap[slot].sourceAttrib : slot;
             maxSlot = std::max<int32_t>(maxSlot, mapSlot);
-            uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + mapSlot;
+            uint32_t inputSlot = backendState.vertexAttribOffset + mapSlot;
 
             pa.Assemble(inputSlot, tmpVector);
 
@@ -625,10 +625,10 @@ public:
             }
 
             // transpose attribs
-            pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_ATTRIB_START_SLOT]) + sizeof(float) * inputPrim;
+            pBase = (uint8_t*)(&vertices[0].attrib[backendState.vertexAttribOffset]) + sizeof(float) * inputPrim;
             for (uint32_t attrib = 0; attrib < numAttribs; ++attrib)
             {
-                uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + attrib;
+                uint32_t attribSlot = backendState.vertexAttribOffset + attrib;
                 for (uint32_t c = 0; c < 4; ++c)
                 {
 #if USE_SIMD16_FRONTEND
@@ -746,7 +746,7 @@ public:
             // Compute absolute attrib slot in vertex array
             uint32_t mapSlot = backendState.swizzleEnable ? backendState.swizzleMap[slot].sourceAttrib : slot;
             maxSlot = std::max<int32_t>(maxSlot, mapSlot);
-            uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + mapSlot;
+            uint32_t inputSlot = backendState.vertexAttribOffset + mapSlot;
 
             pa.Assemble_simd16(inputSlot, tmpVector);
 
@@ -877,10 +877,10 @@ public:
             }
 
             // transpose attribs
-            pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_ATTRIB_START_SLOT]) + sizeof(float) * inputPrim;
+            pBase = (uint8_t*)(&vertices[0].attrib[backendState.vertexAttribOffset]) + sizeof(float) * inputPrim;
             for (uint32_t attrib = 0; attrib < numAttribs; ++attrib)
             {
-                uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + attrib;
+                uint32_t attribSlot = backendState.vertexAttribOffset + attrib;
                 for (uint32_t c = 0; c < 4; ++c)
                 {
                     simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
@@ -1230,6 +1230,8 @@ private:
         uint32_t numInAttribs,          // number of attributes per vertex.
         float *pOutVerts)               // array of output positions. We'll write our new intersection point at i*4.
     {
+        uint32_t vertexAttribOffset = this->state.backendState.vertexAttribOffset;
+
         // compute interpolation factor
         simdscalar t;
         switch (ClippingPlane)
@@ -1263,7 +1265,7 @@ private:
         // interpolate attributes and store
         for (uint32_t a = 0; a < numInAttribs; ++a)
         {
-            uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+            uint32_t attribSlot = vertexAttribOffset + a;
             for (uint32_t c = 0; c < 4; ++c)
             {
                 simdscalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
@@ -1312,6 +1314,8 @@ private:
         uint32_t numInAttribs,          // number of attributes per vertex.
         float *pOutVerts)               // array of output positions. We'll write our new intersection point at i*4.
     {
+        uint32_t vertexAttribOffset = this->state.backendState.vertexAttribOffset;
+
         // compute interpolation factor
         simd16scalar t;
         switch (ClippingPlane)
@@ -1345,7 +1349,7 @@ private:
         // interpolate attributes and store
         for (uint32_t a = 0; a < numInAttribs; ++a)
         {
-            uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+            uint32_t attribSlot = vertexAttribOffset + a;
             for (uint32_t c = 0; c < 4; ++c)
             {
                 simd16scalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
@@ -1421,6 +1425,8 @@ private:
     template<SWR_CLIPCODES ClippingPlane>
     simdscalari ClipTriToPlane(const float* pInVerts, const simdscalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
     {
+        uint32_t vertexAttribOffset = this->state.backendState.vertexAttribOffset;
+
         simdscalari vCurIndex = _simd_setzero_si();
         simdscalari vOutIndex = _simd_setzero_si();
         simdscalar vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts));
@@ -1461,7 +1467,7 @@ private:
                 // store attribs
                 for (uint32_t a = 0; a < numInAttribs; ++a)
                 {
-                    uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+                    uint32_t attribSlot = vertexAttribOffset + a;
                     for (uint32_t c = 0; c < 4; ++c)
                     {
                         simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
@@ -1515,6 +1521,8 @@ private:
     template<SWR_CLIPCODES ClippingPlane>
     simd16scalari ClipTriToPlane(const float* pInVerts, const simd16scalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
     {
+        uint32_t vertexAttribOffset = this->state.backendState.vertexAttribOffset;
+
         simd16scalari vCurIndex = _simd16_setzero_si();
         simd16scalari vOutIndex = _simd16_setzero_si();
         simd16scalar vActiveMask = _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex, vNumInPts));
@@ -1555,7 +1563,7 @@ private:
                 // store attribs
                 for (uint32_t a = 0; a < numInAttribs; ++a)
                 {
-                    uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+                    uint32_t attribSlot = vertexAttribOffset + a;
                     for (uint32_t c = 0; c < 4; ++c)
                     {
                         simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
@@ -1609,6 +1617,8 @@ private:
     template<SWR_CLIPCODES ClippingPlane>
     simdscalari ClipLineToPlane(const float* pInVerts, const simdscalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
     {
+        uint32_t vertexAttribOffset = this->state.backendState.vertexAttribOffset;
+
         simdscalari vCurIndex = _simd_setzero_si();
         simdscalari vOutIndex = _simd_setzero_si();
         simdscalar vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts));
@@ -1646,7 +1656,7 @@ private:
                 // interpolate attributes and store
                 for (uint32_t a = 0; a < numInAttribs; ++a)
                 {
-                    uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+                    uint32_t attribSlot = vertexAttribOffset + a;
                     for (uint32_t c = 0; c < 4; ++c)
                     {
                         simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
@@ -1679,7 +1689,7 @@ private:
                 // interpolate attributes and store
                 for (uint32_t a = 0; a < numInAttribs; ++a)
                 {
-                    uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+                    uint32_t attribSlot = vertexAttribOffset + a;
                     for (uint32_t c = 0; c < 4; ++c)
                     {
                         simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, p_in, p, c);
@@ -1699,6 +1709,8 @@ private:
     template<SWR_CLIPCODES ClippingPlane>
     simd16scalari ClipLineToPlane(const float* pInVerts, const simd16scalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
     {
+        uint32_t vertexAttribOffset = this->state.backendState.vertexAttribOffset;
+
         simd16scalari vCurIndex = _simd16_setzero_si();
         simd16scalari vOutIndex = _simd16_setzero_si();
         simd16scalar vActiveMask = _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex, vNumInPts));
@@ -1736,7 +1748,7 @@ private:
                 // interpolate attributes and store
                 for (uint32_t a = 0; a < numInAttribs; ++a)
                 {
-                    uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+                    uint32_t attribSlot = vertexAttribOffset + a;
                     for (uint32_t c = 0; c < 4; ++c)
                     {
                         simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
@@ -1769,7 +1781,7 @@ private:
                 // interpolate attributes and store
                 for (uint32_t a = 0; a < numInAttribs; ++a)
                 {
-                    uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
+                    uint32_t attribSlot = vertexAttribOffset + a;
                     for (uint32_t c = 0; c < 4; ++c)
                     {
                         simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, p_in, p, c);
diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
index 1cd166d83f..9e2f35725c 100644
--- a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
@@ -528,7 +528,7 @@ static void StreamOut(
         while (_BitScanForward(&slot, soMask))
         {
             __m128 attrib[MAX_NUM_VERTS_PER_PRIM];    // prim attribs (always 4 wide)
-            uint32_t paSlot = slot + VERTEX_ATTRIB_START_SLOT;
+            uint32_t paSlot = slot + soState.vertexAttribOffset[streamIndex];
             pa.AssembleSingle(paSlot, primIndex, attrib);
 
             // Attribute offset is relative offset from start of vertex.
@@ -792,12 +792,12 @@ static void GeometryShaderStage(
     // assemble all attributes for the input primitive
     for (uint32_t slot = 0; slot < pState->numInputAttribs; ++slot)
     {
-        uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + slot;
+        uint32_t attribSlot = pState->vertexAttribOffset + slot;
         pa.Assemble(attribSlot, attrib);
 
         for (uint32_t i = 0; i < numVertsPerPrim; ++i)
         {
-            tlsGsContext.vert[i].attrib[attribSlot] = attrib[i];
+            tlsGsContext.vert[i].attrib[VERTEX_ATTRIB_START_SLOT + slot] = attrib[i];
         }
     }
 
@@ -1131,12 +1131,12 @@ static void TessellationStages(
     // assemble all attributes for the input primitives
     for (uint32_t slot = 0; slot < tsState.numHsInputAttribs; ++slot)
     {
-        uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + slot;
+        uint32_t attribSlot = tsState.vertexAttribOffset + slot;
         pa.Assemble(attribSlot, simdattrib);
 
         for (uint32_t i = 0; i < numVertsPerPrim; ++i)
         {
-            hsContext.vert[i].attrib[attribSlot] = simdattrib[i];
+            hsContext.vert[i].attrib[VERTEX_ATTRIB_START_SLOT + slot] = simdattrib[i];
         }
     }
 
diff --git a/src/gallium/drivers/swr/rasterizer/core/state.h b/src/gallium/drivers/swr/rasterizer/core/state.h
index 2440d44572..4fbd74ddc4 100644
--- a/src/gallium/drivers/swr/rasterizer/core/state.h
+++ b/src/gallium/drivers/swr/rasterizer/core/state.h
@@ -673,6 +673,9 @@ struct SWR_STREAMOUT_STATE
     // Number of attributes, including position, per vertex that are streamed out.
     // This should match number of bits in stream mask.
     uint32_t streamNumEntries[MAX_SO_STREAMS];
+
+    // Offset to the start of the attributes of the input vertices, in simdvector units
+    uint32_t vertexAttribOffset[MAX_SO_STREAMS];
 };
 
 //////////////////////////////////////////////////////////////////////////
@@ -718,6 +721,9 @@ struct SWR_GS_STATE
     // when single stream is enabled, singleStreamID dictates which stream is being output.
     // field ignored if isSingleStream is false
     uint32_t singleStreamID;
+
+    // Offset to the start of the attributes of the input vertices, in simdvector units
+    uint32_t vertexAttribOffset;
 };
 
 
@@ -773,6 +779,9 @@ struct SWR_TS_STATE
     uint32_t                numHsInputAttribs;
     uint32_t                numHsOutputAttribs;
     uint32_t                numDsOutputAttribs;
+
+    // Offset to the start of the attributes of the input vertices, in simdvector units
+    uint32_t vertexAttribOffset;
 };
 
 // output merger state
@@ -1047,6 +1056,9 @@ struct SWR_BACKEND_STATE
 
     bool readRenderTargetArrayIndex;    // Forward render target array index from last FE stage to the backend
     bool readViewportArrayIndex;        // Read viewport array index from last FE stage during binning
+    
+	// Offset to the start of the attributes of the input vertices, in simdvector units
+    uint32_t vertexAttribOffset;
 };
 
 
diff --git a/src/gallium/drivers/swr/swr_shader.cpp b/src/gallium/drivers/swr/swr_shader.cpp
index dfc54fa712..83b49c4763 100644
--- a/src/gallium/drivers/swr/swr_shader.cpp
+++ b/src/gallium/drivers/swr/swr_shader.cpp
@@ -551,6 +551,8 @@ BuilderSWR::CompileGS(struct swr_context *ctx, swr_jit_gs_key &key)
    pGS->isSingleStream = true;
    pGS->singleStreamID = 0;
 
+   pGS->vertexAttribOffset = VERTEX_ATTRIB_START_SLOT; // TODO: optimize
+
    struct swr_geometry_shader *gs = ctx->gs;
 
    LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][TGSI_NUM_CHANNELS];
diff --git a/src/gallium/drivers/swr/swr_state.cpp b/src/gallium/drivers/swr/swr_state.cpp
index 7a8786d96f..897ce998ff 100644
--- a/src/gallium/drivers/swr/swr_state.cpp
+++ b/src/gallium/drivers/swr/swr_state.cpp
@@ -351,6 +351,7 @@ swr_create_vs_state(struct pipe_context *pipe,
       for (uint32_t i = 0; i < MAX_SO_STREAMS; i++) {
         swr_vs->soState.streamNumEntries[i] =
              _mm_popcnt_u32(swr_vs->soState.streamMasks[i]);
+        swr_vs->soState.vertexAttribOffset[i] = VERTEX_ATTRIB_START_SLOT; // TODO: optimize
        }
    }
 
@@ -1747,6 +1748,7 @@ swr_update_derived(struct pipe_context *pipe,
       &ctx->vs->info.base;
    backendState.readRenderTargetArrayIndex = pLastFE->writes_layer;
    backendState.readViewportArrayIndex = pLastFE->writes_viewport_index;
+   backendState.vertexAttribOffset = VERTEX_ATTRIB_START_SLOT; // TODO: optimize
 
    SwrSetBackendState(ctx->swrContext, &backendState);