swr/rast: Remove explicit primitive id slot in the vertex layout

- Remove any special casing in the PS stage when primitive ID is input.
  Treat as a normal attribute that must be set up properly in the FE linkage.
- Remove primitive id from the PS_CONTEXT and TRI_FLAGS

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

7 files changed, 33 insertions, 58 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/core/backend.h b/src/gallium/drivers/swr/rasterizer/core/backend.h
index dba5041140f..035948652bc 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend.h
+++ b/src/gallium/drivers/swr/rasterizer/core/backend.h
@@ -498,7 +498,6 @@ void SetupPixelShaderContext(SWR_PS_CONTEXT *psContext, const SWR_MULTISAMPLE_PO
     psContext->pAttribs = work.pAttribs;
     psContext->pPerspAttribs = work.pPerspAttribs;
     psContext->frontFace = work.triFlags.frontFacing;
-    psContext->primID = work.triFlags.primID;
     psContext->renderTargetArrayIndex = work.triFlags.renderTargetArrayIndex;
 
     // save Ia/Ib/Ic and Ja/Jb/Jc if we need to reevaluate i/j/k in the shader because of pull attribs
diff --git a/src/gallium/drivers/swr/rasterizer/core/binner.cpp b/src/gallium/drivers/swr/rasterizer/core/binner.cpp
index daadd5f3d33..06078cdcfa1 100644
--- a/src/gallium/drivers/swr/rasterizer/core/binner.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/binner.cpp
@@ -809,7 +809,6 @@ endBinTriangles:
         TRIANGLE_WORK_DESC &desc = work.desc.tri;
 
         desc.triFlags.frontFacing = state.forceFront ? 1 : ((frontFaceMask >> triIndex) & 1);
-        desc.triFlags.primID = pPrimID[triIndex];
         desc.triFlags.renderTargetArrayIndex = aRTAI[triIndex];
         desc.triFlags.viewportIndex = pViewportIndex[triIndex];
 
@@ -1257,7 +1256,6 @@ endBinTriangles:
         TRIANGLE_WORK_DESC &desc = work.desc.tri;
 
         desc.triFlags.frontFacing = state.forceFront ? 1 : ((frontFaceMask >> triIndex) & 1);
-        desc.triFlags.primID = pPrimID[triIndex];
         desc.triFlags.renderTargetArrayIndex = aRTAI[triIndex];
         desc.triFlags.viewportIndex = pViewportIndex[triIndex];
 
@@ -1451,7 +1449,6 @@ void BinPostSetupPoints(
 
             // points are always front facing
             desc.triFlags.frontFacing = 1;
-            desc.triFlags.primID = pPrimID[primIndex];
             desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex];
             desc.triFlags.viewportIndex = pViewportIndex[primIndex];
 
@@ -1601,7 +1598,6 @@ void BinPostSetupPoints(
             TRIANGLE_WORK_DESC &desc = work.desc.tri;
 
             desc.triFlags.frontFacing = 1;
-            desc.triFlags.primID = pPrimID[primIndex];
             desc.triFlags.pointSize = aPointSize[primIndex];
             desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex];
             desc.triFlags.viewportIndex = pViewportIndex[primIndex];
@@ -1823,7 +1819,6 @@ void BinPostSetupPoints_simd16(
 
             // points are always front facing
             desc.triFlags.frontFacing = 1;
-            desc.triFlags.primID = pPrimID[primIndex];
             desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex];
             desc.triFlags.viewportIndex = pViewportIndex[primIndex];
 
@@ -1978,7 +1973,6 @@ void BinPostSetupPoints_simd16(
             TRIANGLE_WORK_DESC &desc = work.desc.tri;
 
             desc.triFlags.frontFacing = 1;
-            desc.triFlags.primID = pPrimID[primIndex];
             desc.triFlags.pointSize = aPointSize[primIndex];
             desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex];
             desc.triFlags.viewportIndex = pViewportIndex[primIndex];
@@ -2246,7 +2240,6 @@ void BinPostSetupLines(
         TRIANGLE_WORK_DESC &desc = work.desc.tri;
 
         desc.triFlags.frontFacing = 1;
-        desc.triFlags.primID = pPrimID[primIndex];
         desc.triFlags.yMajor = (yMajorMask >> primIndex) & 1;
         desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex];
         desc.triFlags.viewportIndex = pViewportIndex[primIndex];
@@ -2464,7 +2457,6 @@ void BinPostSetupLines_simd16(
         TRIANGLE_WORK_DESC &desc = work.desc.tri;
 
         desc.triFlags.frontFacing = 1;
-        desc.triFlags.primID = pPrimID[primIndex];
         desc.triFlags.yMajor = (yMajorMask >> primIndex) & 1;
         desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex];
         desc.triFlags.viewportIndex = pViewportIndex[primIndex];
diff --git a/src/gallium/drivers/swr/rasterizer/core/context.h b/src/gallium/drivers/swr/rasterizer/core/context.h
index 62332db173b..fa308feade2 100644
--- a/src/gallium/drivers/swr/rasterizer/core/context.h
+++ b/src/gallium/drivers/swr/rasterizer/core/context.h
@@ -62,7 +62,6 @@ struct TRI_FLAGS
     uint32_t coverageMask : (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
     uint32_t reserved : 32 - 1 - 1 - (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
     float pointSize;
-    uint32_t primID;
     uint32_t renderTargetArrayIndex;
     uint32_t viewportIndex;
 };
diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
index dfbbc580b63..e2660c1cc90 100644
--- a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp
@@ -948,18 +948,7 @@ static void GeometryShaderStage(
                             if (HasRastT::value && state.soState.streamToRasterizer == stream)
                             {
 #if USE_SIMD16_FRONTEND
-                                simd16scalari vPrimId;
-                                // pull primitiveID from the GS output if available
-                                if (state.gsState.emitsPrimitiveID)
-                                {
-                                    simd16vector primIdAttrib[3];
-                                    gsPa.Assemble_simd16(VERTEX_PRIMID_SLOT, primIdAttrib);
-                                    vPrimId = _simd16_castps_si(primIdAttrib[state.frontendState.topologyProvokingVertex].x);
-                                }
-                                else
-                                {
-                                    vPrimId = _simd16_set1_epi32(pPrimitiveId[inputPrim]);
-                                }
+                                simd16scalari vPrimId = _simd16_set1_epi32(pPrimitiveId[inputPrim]);
 
                                 // use viewport array index if GS declares it as an output attribute. Otherwise use index 0.
                                 simd16scalari vViewPortIdx;
@@ -983,18 +972,7 @@ static void GeometryShaderStage(
                                 gsPa.useAlternateOffset = false;
                                 pfnClipFunc(pDC, gsPa, workerId, attrib_simd16, GenMask(gsPa.NumPrims()), vPrimId, vViewPortIdx);
 #else
-                                simdscalari vPrimId;
-                                // pull primitiveID from the GS output if available
-                                if (state.gsState.emitsPrimitiveID)
-                                {
-                                    simdvector primIdAttrib[3];
-                                    gsPa.Assemble(VERTEX_PRIMID_SLOT, primIdAttrib);
-                                    vPrimId = _simd_castps_si(primIdAttrib[state.frontendState.topologyProvokingVertex].x);
-                                }
-                                else
-                                {
-                                    vPrimId = _simd_set1_epi32(pPrimitiveId[inputPrim]);
-                                }
+                                simdscalari vPrimId = _simd_set1_epi32(pPrimitiveId[inputPrim]);
 
                                 // use viewport array index if GS declares it as an output attribute. Otherwise use index 0.
                                 simdscalari vViewPortIdx;
diff --git a/src/gallium/drivers/swr/rasterizer/core/state.h b/src/gallium/drivers/swr/rasterizer/core/state.h
index 3db637036cd..f7030511474 100644
--- a/src/gallium/drivers/swr/rasterizer/core/state.h
+++ b/src/gallium/drivers/swr/rasterizer/core/state.h
@@ -186,11 +186,10 @@ enum SWR_VTX_SLOTS
     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),
+    VERTEX_CLIPCULL_DIST_LO_SLOT     = (34 + VERTEX_POSITION_END_SLOT), // VS writes lower 4 clip/cull dist
+    VERTEX_CLIPCULL_DIST_HI_SLOT     = (35 + VERTEX_POSITION_END_SLOT), // VS writes upper 4 clip/cull dist
+    VERTEX_POINT_SIZE_SLOT           = (36 + VERTEX_POSITION_END_SLOT), // VS writes point size here
+    VERTEX_VIEWPORT_ARRAY_INDEX_SLOT = (37 + VERTEX_POSITION_END_SLOT),
     SWR_VTX_NUM_SLOTS                 = VERTEX_VIEWPORT_ARRAY_INDEX_SLOT,
 };
 
@@ -343,7 +342,6 @@ struct SWR_PS_CONTEXT
                                 // OUT: result color per rendertarget
 
     uint32_t frontFace;                 // IN: front- 1, back- 0
-    uint32_t primID;                    // IN: primitive ID
     uint32_t sampleIndex;               // IN: sampleIndex
     uint32_t renderTargetArrayIndex;    // IN: render target array index from GS
     uint32_t rasterizerSampleCount;     // IN: sample count used by the rasterizer
@@ -716,9 +714,6 @@ struct SWR_GS_STATE
     // geometry shader emits renderTargetArrayIndex
     bool emitsRenderTargetArrayIndex;
 
-    // geometry shader emits PrimitiveID
-    bool emitsPrimitiveID;
-
     // geometry shader emits ViewportArrayIndex
     bool emitsViewportArrayIndex;
 
diff --git a/src/gallium/drivers/swr/swr_shader.cpp b/src/gallium/drivers/swr/swr_shader.cpp
index 2f495f59c23..44988892691 100644
--- a/src/gallium/drivers/swr/swr_shader.cpp
+++ b/src/gallium/drivers/swr/swr_shader.cpp
@@ -406,8 +406,6 @@ BuilderSWR::swr_gs_llvm_emit_vertex(const struct lp_build_tgsi_gs_iface *gs_base
        uint32_t attribSlot = attrib;
        if (iface->info->output_semantic_name[attrib] == TGSI_SEMANTIC_PSIZE)
           attribSlot = VERTEX_POINT_SIZE_SLOT;
-       else if (iface->info->output_semantic_name[attrib] == TGSI_SEMANTIC_PRIMID)
-          attribSlot = VERTEX_PRIMID_SLOT;
        else if (iface->info->output_semantic_name[attrib] == TGSI_SEMANTIC_LAYER)
           attribSlot = VERTEX_RTAI_SLOT;
 
@@ -534,7 +532,6 @@ BuilderSWR::CompileGS(struct swr_context *ctx, swr_jit_gs_key &key)
    pGS->instanceCount = info->properties[TGSI_PROPERTY_GS_INVOCATIONS];
 
    pGS->emitsRenderTargetArrayIndex = info->writes_layer;
-   pGS->emitsPrimitiveID = info->writes_primid;
    pGS->emitsViewportArrayIndex = info->writes_viewport_index;
 
    // XXX: single stream for now...
@@ -1043,23 +1040,23 @@ BuilderSWR::CompileFS(struct swr_context *ctx, swr_jit_fs_key &key)
          inputs[attrib][3] =
             wrap(LOAD(pPS, {0, SWR_PS_CONTEXT_vOneOverW, PixelPositions_center}, "vOneOverW"));
          continue;
-      } else if (semantic_name == TGSI_SEMANTIC_PRIMID) {
-         Value *primID = LOAD(pPS, {0, SWR_PS_CONTEXT_primID}, "primID");
-         inputs[attrib][0] = wrap(VECTOR_SPLAT(JM()->mVWidth, primID));
-         inputs[attrib][1] = wrap(VIMMED1(0));
-         inputs[attrib][2] = wrap(VIMMED1(0));
-         inputs[attrib][3] = wrap(VIMMED1(0));
-         continue;
       }
 
       unsigned linkedAttrib =
          locate_linkage(semantic_name, semantic_idx, pPrevShader);
 
-      if (semantic_name == TGSI_SEMANTIC_GENERIC &&
+      uint32_t extraAttribs = 0;
+      if (semantic_name == TGSI_SEMANTIC_PRIMID && !ctx->gs) {
+         /* non-gs generated primID - need to grab from swizzleMap override */
+         linkedAttrib = pPrevShader->num_outputs - 1;
+         swr_fs->constantMask |= 1 << linkedAttrib;
+         extraAttribs++;
+      } else if (semantic_name == TGSI_SEMANTIC_GENERIC &&
           key.sprite_coord_enable & (1 << semantic_idx)) {
          /* we add an extra attrib to the backendState in swr_update_derived. */
-         linkedAttrib = pPrevShader->num_outputs - 1;
+         linkedAttrib = pPrevShader->num_outputs + extraAttribs - 1;
          swr_fs->pointSpriteMask |= (1 << linkedAttrib);
+         extraAttribs++;
       } else if (linkedAttrib == 0xFFFFFFFF) {
          inputs[attrib][0] = wrap(VIMMED1(0.0f));
          inputs[attrib][1] = wrap(VIMMED1(0.0f));
diff --git a/src/gallium/drivers/swr/swr_state.cpp b/src/gallium/drivers/swr/swr_state.cpp
index 2d036f91abf..abdcef7b2f3 100644
--- a/src/gallium/drivers/swr/swr_state.cpp
+++ b/src/gallium/drivers/swr/swr_state.cpp
@@ -1729,9 +1729,24 @@ swr_update_derived(struct pipe_context *pipe,
 
    // set up backend state
    SWR_BACKEND_STATE backendState = {0};
-   backendState.numAttributes =
-      ((ctx->gs ? ctx->gs->info.base.num_outputs : ctx->vs->info.base.num_outputs) - 1) +
-      (ctx->rasterizer->sprite_coord_enable ? 1 : 0);
+   if (ctx->gs) {
+      backendState.numAttributes = ctx->gs->info.base.num_outputs - 1;
+   } else {
+      backendState.numAttributes = ctx->vs->info.base.num_outputs - 1;
+      if (ctx->fs->info.base.uses_primid) {
+         backendState.numAttributes++;
+         backendState.swizzleEnable = true;
+         for (unsigned i = 0; i < sizeof(backendState.numComponents); i++) {
+            backendState.swizzleMap[i].sourceAttrib = i;
+         }
+         backendState.swizzleMap[ctx->vs->info.base.num_outputs - 1].constantSource =
+            SWR_CONSTANT_SOURCE_PRIM_ID;
+         backendState.swizzleMap[ctx->vs->info.base.num_outputs - 1].componentOverrideMask = 1;
+      }
+   }
+   if (ctx->rasterizer->sprite_coord_enable)
+      backendState.numAttributes++;
+
    backendState.numAttributes = std::min((size_t)backendState.numAttributes,
                                          sizeof(backendState.numComponents));
    for (unsigned i = 0; i < backendState.numAttributes; i++)