swr/rast: Fix read-back of render target array index

The last FE stage can emit render target array index. Currently we only
check to see if GS is emitting it. Moved the state to BACKEND_STATE and
plumbed the driver to set it.

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

src/gallium/drivers/swr/rasterizer/core/api.cpp

@@ -680,7 +680,7 @@ void SwrSetBlendFunc(
 // update guardband multipliers for the viewport
 void updateGuardbands(API_STATE *pState)
 {
-    uint32_t numGbs = pState->gsState.emitsRenderTargetArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
+    uint32_t numGbs = pState->backendState.readRenderTargetArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
 
     for(uint32_t i = 0; i < numGbs; ++i)
     {

src/gallium/drivers/swr/rasterizer/core/binner.cpp

@@ -763,7 +763,7 @@ endBinTriangles:
 
     // store render target array index
     OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH];
-    if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+    if (state.backendState.readRenderTargetArrayIndex)
     {
         simdvector vRtai[3];
         pa.Assemble(VERTEX_SGV_SLOT, vRtai);
@@ -1213,7 +1213,7 @@ endBinTriangles:
 
     // store render target array index
     OSALIGNSIMD16(uint32_t) aRTAI[KNOB_SIMD16_WIDTH];
-    if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+    if (state.backendState.readRenderTargetArrayIndex)
     {
         simd16vector vRtai[3];
         pa.Assemble_simd16(VERTEX_SGV_SLOT, vRtai);
@@ -1423,7 +1423,7 @@ void BinPostSetupPoints(
 
         // store render target array index
         OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH];
-        if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+        if (state.backendState.readRenderTargetArrayIndex)
         {
             simdvector vRtai;
             pa.Assemble(VERTEX_SGV_SLOT, &vRtai);
@@ -1563,7 +1563,7 @@ void BinPostSetupPoints(
 
         // store render target array index
         OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH];
-        if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+        if (state.backendState.readRenderTargetArrayIndex)
         {
             simdvector vRtai[2];
             pa.Assemble(VERTEX_SGV_SLOT, vRtai);
@@ -1793,7 +1793,7 @@ void BinPostSetupPoints_simd16(
 
         // store render target array index
         OSALIGNSIMD16(uint32_t) aRTAI[KNOB_SIMD16_WIDTH];
-        if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+        if (state.backendState.readRenderTargetArrayIndex)
         {
             simd16vector vRtai;
             pa.Assemble_simd16(VERTEX_SGV_SLOT, &vRtai);
@@ -1938,7 +1938,7 @@ void BinPostSetupPoints_simd16(
 
         // store render target array index
         OSALIGNSIMD16(uint32_t) aRTAI[KNOB_SIMD16_WIDTH];
-        if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+        if (state.backendState.readRenderTargetArrayIndex)
         {
             simd16vector vRtai[2];
             pa.Assemble_simd16(VERTEX_SGV_SLOT, vRtai);
@@ -2219,7 +2219,7 @@ void BinPostSetupLines(
 
     // store render target array index
     OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH];
-    if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+    if (state.backendState.readRenderTargetArrayIndex)
     {
         simdvector vRtai[2];
         pa.Assemble(VERTEX_SGV_SLOT, vRtai);
@@ -2436,7 +2436,7 @@ void BinPostSetupLines_simd16(
 
     // store render target array index
     OSALIGNSIMD16(uint32_t) aRTAI[KNOB_SIMD16_WIDTH];
-    if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex)
+    if (state.backendState.readRenderTargetArrayIndex)
     {
         simd16vector vRtai[2];
         pa.Assemble_simd16(VERTEX_SGV_SLOT, vRtai);

src/gallium/drivers/swr/rasterizer/core/state.h

@@ -710,9 +710,6 @@ struct SWR_GS_STATE
     // instance count
     uint32_t instanceCount;
 
-    // geometry shader emits renderTargetArrayIndex
-    bool emitsRenderTargetArrayIndex;
-
     // geometry shader emits ViewportArrayIndex
     bool emitsViewportArrayIndex;
 
@@ -1074,6 +1071,8 @@ struct SWR_BACKEND_STATE
                                         // setting up attributes for the backend, otherwise
                                         // all attributes up to numAttributes will be sent
     SWR_ATTRIB_SWIZZLE swizzleMap[32];
+
+    bool readRenderTargetArrayIndex;    // Forward render target array index from last FE stage to the backend
 };
 
 

src/gallium/drivers/swr/swr_shader.cpp

@@ -547,7 +547,6 @@ BuilderSWR::CompileGS(struct swr_context *ctx, swr_jit_gs_key &key)
    pGS->maxNumVerts = info->properties[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES];
    pGS->instanceCount = info->properties[TGSI_PROPERTY_GS_INVOCATIONS];
 
-   pGS->emitsRenderTargetArrayIndex = info->writes_layer;
    pGS->emitsViewportArrayIndex = info->writes_viewport_index;
 
    // XXX: single stream for now...

src/gallium/drivers/swr/swr_state.cpp

@@ -1755,6 +1755,13 @@ swr_update_derived(struct pipe_context *pipe,
       (ctx->rasterizer->flatshade ? ctx->fs->flatConstantMask : 0);
    backendState.pointSpriteTexCoordMask = ctx->fs->pointSpriteMask;
 
+   if (ctx->gs)
+      backendState.readRenderTargetArrayIndex =
+         ctx->gs->info.base.writes_layer;
+   else
+      backendState.readRenderTargetArrayIndex =
+         ctx->vs->info.base.writes_layer;
+
    SwrSetBackendState(ctx->swrContext, &backendState);
 
    /* Ensure that any in-progress attachment change StoreTiles finish */