HLSL: Add gather, improve proto generator machine for upcoming 2DMS/Shadow

Test/hlsl.gather.array.dx10.frag

+SamplerState       g_sSamp : register(s0);
+
+Texture1DArray          g_tTex1df4a : register(t1);
+
+uniform Texture1DArray <float4> g_tTex1df4 : register(t0);
+Texture1DArray <int4>   g_tTex1di4a;
+Texture1DArray <uint4>  g_tTex1du4a;
+
+Texture2DArray <float4> g_tTex2df4a;
+Texture2DArray <int4>   g_tTex2di4a;
+Texture2DArray <uint4>  g_tTex2du4a;
+
+TextureCubeArray <float4> g_tTexcdf4a;
+TextureCubeArray <int4>   g_tTexcdi4a;
+TextureCubeArray <uint4>  g_tTexcdu4a;
+
+struct PS_OUTPUT
+{
+    float4 Color : SV_Target0;
+    float  Depth : SV_Depth;
+};
+
+PS_OUTPUT main()
+{
+   PS_OUTPUT psout;
+
+   // No 1D gathers
+
+   float4 txval20 = g_tTex2df4a . Gather(g_sSamp, float3(0.1, 0.2, 0.3));
+   int4   txval21 = g_tTex2di4a . Gather(g_sSamp, float3(0.3, 0.4, 0.5));
+   uint4  txval22 = g_tTex2du4a . Gather(g_sSamp, float3(0.5, 0.6, 0.7));
+
+   // no 3D gathers
+
+   float4 txval40 = g_tTexcdf4a . Gather(g_sSamp, float4(0.1, 0.2, 0.3, 0.4));
+   int4   txval41 = g_tTexcdi4a . Gather(g_sSamp, float4(0.4, 0.5, 0.6, 0.7));
+   uint4  txval42 = g_tTexcdu4a . Gather(g_sSamp, float4(0.7, 0.8, 0.9, 1.0));
+
+   psout.Color = 1.0;
+   psout.Depth = 1.0;
+
+   return psout;
+}

Test/hlsl.gather.basic.dx10.frag

+SamplerState       g_sSamp : register(s0);
+uniform sampler2D          g_sSamp2d;
+
+Texture1D          g_tTex1df4a : register(t1);
+
+uniform Texture1D <float4> g_tTex1df4 : register(t0);
+Texture1D <int4>   g_tTex1di4;
+Texture1D <uint4>  g_tTex1du4;
+
+Texture2D <float4> g_tTex2df4;
+Texture2D <int4>   g_tTex2di4;
+Texture2D <uint4>  g_tTex2du4;
+
+Texture3D <float4> g_tTex3df4;
+Texture3D <int4>   g_tTex3di4;
+Texture3D <uint4>  g_tTex3du4;
+
+TextureCube <float4> g_tTexcdf4;
+TextureCube <int4>   g_tTexcdi4;
+TextureCube <uint4>  g_tTexcdu4;
+
+struct PS_OUTPUT
+{
+    float4 Color : SV_Target0;
+    float  Depth : SV_Depth;
+};
+
+PS_OUTPUT main()
+{
+   PS_OUTPUT psout;
+
+   // no 1D gathers
+
+   float4 txval20 = g_tTex2df4 . Gather(g_sSamp, float2(0.1, 0.2));
+   int4   txval21 = g_tTex2di4 . Gather(g_sSamp, float2(0.3, 0.4));
+   uint4  txval22 = g_tTex2du4 . Gather(g_sSamp, float2(0.5, 0.6));
+
+   // no 3D gathers
+
+   float4 txval40 = g_tTexcdf4 . Gather(g_sSamp, float3(0.1, 0.2, 0.3));
+   int4   txval41 = g_tTexcdi4 . Gather(g_sSamp, float3(0.4, 0.5, 0.6));
+   uint4  txval42 = g_tTexcdu4 . Gather(g_sSamp, float3(0.7, 0.8, 0.9));
+
+   psout.Color = 1.0;
+   psout.Depth = 1.0;
+
+   return psout;
+}

Test/hlsl.gather.basic.dx10.vert

+SamplerState       g_sSamp : register(s0);
+uniform sampler2D          g_sSamp2d;
+
+Texture1D          g_tTex1df4a : register(t1);
+
+uniform Texture1D <float4> g_tTex1df4 : register(t0);
+Texture1D <int4>   g_tTex1di4;
+Texture1D <uint4>  g_tTex1du4;
+
+Texture2D <float4> g_tTex2df4;
+Texture2D <int4>   g_tTex2di4;
+Texture2D <uint4>  g_tTex2du4;
+
+Texture3D <float4> g_tTex3df4;
+Texture3D <int4>   g_tTex3di4;
+Texture3D <uint4>  g_tTex3du4;
+
+TextureCube <float4> g_tTexcdf4;
+TextureCube <int4>   g_tTexcdi4;
+TextureCube <uint4>  g_tTexcdu4;
+
+struct VS_OUTPUT
+{
+    float4 Pos : SV_Position;
+};
+
+VS_OUTPUT main()
+{
+   VS_OUTPUT vsout;
+
+   // no 1D gathers
+
+   float4 txval20 = g_tTex2df4 . Gather(g_sSamp, float2(0.1, 0.2));
+   int4   txval21 = g_tTex2di4 . Gather(g_sSamp, float2(0.3, 0.4));
+   uint4  txval22 = g_tTex2du4 . Gather(g_sSamp, float2(0.5, 0.6));
+
+   // no 3D gathers
+
+   float4 txval40 = g_tTexcdf4 . Gather(g_sSamp, float3(0.1, 0.2, 0.3));
+   int4   txval41 = g_tTexcdi4 . Gather(g_sSamp, float3(0.4, 0.5, 0.6));
+   uint4  txval42 = g_tTexcdu4 . Gather(g_sSamp, float3(0.7, 0.8, 0.9));
+
+   vsout.Pos = float4(0,0,0,0);
+
+   return vsout;
+}

Test/hlsl.gather.offset.dx10.frag

+SamplerState       g_sSamp : register(s0);
+
+Texture1D          g_tTex1df4a : register(t1);
+
+uniform Texture1D <float4> g_tTex1df4 : register(t0);
+Texture1D <int4>   g_tTex1di4;
+Texture1D <uint4>  g_tTex1du4;
+
+Texture2D <float4> g_tTex2df4;
+Texture2D <int4>   g_tTex2di4;
+Texture2D <uint4>  g_tTex2du4;
+
+Texture3D <float4> g_tTex3df4;
+Texture3D <int4>   g_tTex3di4;
+Texture3D <uint4>  g_tTex3du4;
+
+TextureCube <float4> g_tTexcdf4;
+TextureCube <int4>   g_tTexcdi4;
+TextureCube <uint4>  g_tTexcdu4;
+
+struct PS_OUTPUT
+{
+    float4 Color : SV_Target0;
+    float  Depth : SV_Depth;
+};
+
+PS_OUTPUT main()
+{
+   PS_OUTPUT psout;
+
+   // no 1D Gathers
+
+   float4 txval20 = g_tTex2df4 . Gather(g_sSamp, float2(0.1, 0.2), int2(1,0));
+   int4   txval21 = g_tTex2di4 . Gather(g_sSamp, float2(0.3, 0.4), int2(1,1));
+   uint4  txval22 = g_tTex2du4 . Gather(g_sSamp, float2(0.5, 0.6), int2(1,-1));
+
+   // no 3D gathers
+   // No Cube offset gathers
+
+   psout.Color = 1.0;
+   psout.Depth = 1.0;
+
+   return psout;
+}

Test/hlsl.gather.offsetarray.dx10.frag

+SamplerState       g_sSamp : register(s0);
+
+Texture1DArray          g_tTex1df4a : register(t1);
+
+uniform Texture1DArray <float4> g_tTex1df4 : register(t0);
+Texture1DArray <int4>   g_tTex1di4;
+Texture1DArray <uint4>  g_tTex1du4;
+
+Texture2DArray <float4> g_tTex2df4;
+Texture2DArray <int4>   g_tTex2di4;
+Texture2DArray <uint4>  g_tTex2du4;
+
+struct PS_OUTPUT
+{
+    float4 Color : SV_Target0;
+    float  Depth : SV_Depth;
+};
+
+PS_OUTPUT main()
+{
+   PS_OUTPUT psout;
+
+   // No 1D gathers
+
+   float4 txval20 = g_tTex2df4 . Gather(g_sSamp, float3(0.1, 0.2, 0.3), int2(1,0));
+   int4   txval21 = g_tTex2di4 . Gather(g_sSamp, float3(0.3, 0.4, 0.4), int2(1,1));
+   uint4  txval22 = g_tTex2du4 . Gather(g_sSamp, float3(0.5, 0.6, 0.7), int2(1,-1));
+
+   // No 3D gathers
+   // No Cube offset gathers
+
+   psout.Color = 1.0;
+   psout.Depth = 1.0;
+
+   return psout;
+}

Test/hlsl.samplelevel.array.dx10.frag

@@ -3,16 +3,16 @@ SamplerState       g_sSamp : register(s0);
 Texture1DArray          g_tTex1df4a : register(t1);
 
 uniform Texture1DArray <float4> g_tTex1df4 : register(t0);
-Texture1DArray <int4>   g_tTex1di4;
-Texture1DArray <uint4>  g_tTex1du4;
+Texture1DArray <int4>   g_tTex1di4a;
+Texture1DArray <uint4>  g_tTex1du4a;
 
-Texture2DArray <float4> g_tTex2df4;
-Texture2DArray <int4>   g_tTex2di4;
-Texture2DArray <uint4>  g_tTex2du4;
+Texture2DArray <float4> g_tTex2df4a;
+Texture2DArray <int4>   g_tTex2di4a;
+Texture2DArray <uint4>  g_tTex2du4a;
 
-TextureCubeArray <float4> g_tTexcdf4;
-TextureCubeArray <int4>   g_tTexcdi4;
-TextureCubeArray <uint4>  g_tTexcdu4;
+TextureCubeArray <float4> g_tTexcdf4a;
+TextureCubeArray <int4>   g_tTexcdi4a;
+TextureCubeArray <uint4>  g_tTexcdu4a;
 
 struct PS_OUTPUT
 {
@@ -24,17 +24,17 @@ PS_OUTPUT main()
 {
    PS_OUTPUT psout;
 
-   float4 txval10 = g_tTex1df4 . SampleLevel(g_sSamp, float2(0.1, 0.2), 0.75);
-   int4   txval11 = g_tTex1di4 . SampleLevel(g_sSamp, float2(0.2, 0.3), 0.75);
-   uint4  txval12 = g_tTex1du4 . SampleLevel(g_sSamp, float2(0.3, 0.4), 0.75);
+   float4 txval10 = g_tTex1df4a . SampleLevel(g_sSamp, float2(0.1, 0.2), 0.75);
+   int4   txval11 = g_tTex1di4a . SampleLevel(g_sSamp, float2(0.2, 0.3), 0.75);
+   uint4  txval12 = g_tTex1du4a . SampleLevel(g_sSamp, float2(0.3, 0.4), 0.75);
 
-   float4 txval20 = g_tTex2df4 . SampleLevel(g_sSamp, float3(0.1, 0.2, 0.3), 0.75);
-   int4   txval21 = g_tTex2di4 . SampleLevel(g_sSamp, float3(0.3, 0.4, 0.5), 0.75);
-   uint4  txval22 = g_tTex2du4 . SampleLevel(g_sSamp, float3(0.5, 0.6, 0.7), 0.75);
+   float4 txval20 = g_tTex2df4a . SampleLevel(g_sSamp, float3(0.1, 0.2, 0.3), 0.75);
+   int4   txval21 = g_tTex2di4a . SampleLevel(g_sSamp, float3(0.3, 0.4, 0.5), 0.75);
+   uint4  txval22 = g_tTex2du4a . SampleLevel(g_sSamp, float3(0.5, 0.6, 0.7), 0.75);
 
-   float4 txval40 = g_tTexcdf4 . SampleLevel(g_sSamp, float4(0.1, 0.2, 0.3, 0.4), 0.75);
-   int4   txval41 = g_tTexcdi4 . SampleLevel(g_sSamp, float4(0.4, 0.5, 0.6, 0.7), 0.75);
-   uint4  txval42 = g_tTexcdu4 . SampleLevel(g_sSamp, float4(0.7, 0.8, 0.9, 1.0), 0.75);
+   float4 txval40 = g_tTexcdf4a . SampleLevel(g_sSamp, float4(0.1, 0.2, 0.3, 0.4), 0.75);
+   int4   txval41 = g_tTexcdi4a . SampleLevel(g_sSamp, float4(0.4, 0.5, 0.6, 0.7), 0.75);
+   uint4  txval42 = g_tTexcdu4a . SampleLevel(g_sSamp, float4(0.7, 0.8, 0.9, 1.0), 0.75);
 
    psout.Color = 1.0;
    psout.Depth = 1.0;

gtests/Hlsl.FromFile.cpp

@@ -81,6 +81,11 @@ INSTANTIATE_TEST_CASE_P(
         {"hlsl.float1.frag", "PixelShaderFunction"},
         {"hlsl.float4.frag", "PixelShaderFunction"},
         {"hlsl.forLoop.frag", "PixelShaderFunction"},
+        {"hlsl.gather.array.dx10.frag", "main"},
+        {"hlsl.gather.basic.dx10.frag", "main"},
+        {"hlsl.gather.basic.dx10.vert", "main"},
+        {"hlsl.gather.offset.dx10.frag", "main"},
+        {"hlsl.gather.offsetarray.dx10.frag", "main"},
         {"hlsl.getdimensions.dx10.frag", "main"},
         {"hlsl.getdimensions.dx10.vert", "main"},
         {"hlsl.if.frag", "PixelShaderFunction"},

hlsl/hlslParseHelper.cpp

@@ -1221,6 +1221,35 @@ void HlslParseContext::decomposeSampleMethods(const TSourceLoc& loc, TIntermType
             break;
         }
 
+    case EOpMethodGather:
+        {
+            TIntermTyped* argTex    = argAggregate->getSequence()[0]->getAsTyped();
+            TIntermTyped* argSamp   = argAggregate->getSequence()[1]->getAsTyped();
+            TIntermTyped* argCoord  = argAggregate->getSequence()[2]->getAsTyped();
+            TIntermTyped* argOffset = nullptr;
+
+            // Offset is optional
+            if (argAggregate->getSequence().size() == 4)
+                argOffset = argAggregate->getSequence()[3]->getAsTyped();
+
+            const TOperator textureOp = (argOffset == nullptr ? EOpTextureGather : EOpTextureGatherOffset);
+            TIntermAggregate* txgather = new TIntermAggregate(textureOp);
+
+            TIntermAggregate* txcombine = handleSamplerTextureCombine(loc, argTex, argSamp);
+
+            txgather->getSequence().push_back(txcombine);
+            txgather->getSequence().push_back(argCoord);
+
+            if (argOffset != nullptr)
+                txgather->getSequence().push_back(argOffset);
+
+            txgather->setType(node->getType());
+            txgather->setLoc(loc);
+            node = txgather;
+
+            break;
+        }
+        
     default:
         break; // most pass through unchanged
     }