cuda: Add more formats for texture read/write (#7012)

* WiP: Add more formats for texture reads

* fix test

* format code

* add  float2/float4 versions for 1D and 3D as well

* fixed review comment

* fix review comments

---------

Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
Co-authored-by: Ellie Hermaszewska <ellieh@nvidia.com>

5 files changed, 231 insertions, 44 deletions

diff --git a/tests/compute/rw-texture-simple.slang b/tests/compute/rw-texture-simple.slang
index f5befcd57..d0d90acb9 100644
--- a/tests/compute/rw-texture-simple.slang
+++ b/tests/compute/rw-texture-simple.slang
@@ -14,6 +14,18 @@ RWTexture1D<float> rwt1D;
 RWTexture2D<float> rwt2D;
 //TEST_INPUT: RWTexture3D(format=R32Float, size=4, content = one, mipMaps = 1):name rwt3D
 RWTexture3D<float> rwt3D;
+//TEST_INPUT: RWTexture1D(format=RG32Float, size=4, content = one, mipMaps = 1):name rwt1D_float2
+RWTexture1D<float2> rwt1D_float2;
+//TEST_INPUT: RWTexture1D(format=RGBA32Float, size=4, content = one, mipMaps = 1):name rwt1D_float4
+RWTexture1D<float4> rwt1D_float4;
+//TEST_INPUT: RWTexture2D(format=RG32Float, size=4, content = one, mipMaps = 1):name rwt2D_float2
+RWTexture2D<float2> rwt2D_float2;
+//TEST_INPUT: RWTexture2D(format=RGBA32Float, size=4, content = one, mipMaps = 1):name rwt2D_float4
+RWTexture2D<float4> rwt2D_float4;
+//TEST_INPUT: RWTexture3D(format=RG32Float, size=4, content = one, mipMaps = 1):name rwt3D_float2
+RWTexture3D<float2> rwt3D_float2;
+//TEST_INPUT: RWTexture3D(format=RGBA32Float, size=4, content = one, mipMaps = 1):name rwt3D_float4
+RWTexture3D<float4> rwt3D_float4;
 
 //TEST_INPUT: ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<float> outputBuffer;
@@ -25,6 +37,7 @@ void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
     
     float val = 0.0f;
  
+    // float texture operations
     val += rwt1D[idx];
     val += rwt2D[uint2(idx, idx)];
     val += rwt3D[uint3(idx, idx, idx)];
@@ -36,6 +49,54 @@ void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
     val += rwt1D[idx];
     val += rwt2D[uint2(idx, idx)];
     val += rwt3D[uint3(idx, idx, idx)];
- 
+    
+    // float2 texture operations for 1D
+    float2 val2_1d = rwt1D_float2[idx];
+    rwt1D_float2[idx] = float2(idx, idx);
+    val2_1d = rwt1D_float2[idx];
+    val += val2_1d.x;
+    val += val2_1d.y;
+    
+    // float2 texture operations for 2D
+    float2 val2 = rwt2D_float2[uint2(idx, idx)];
+    rwt2D_float2[uint2(idx, idx)] = float2(idx, idx);
+    val2 = rwt2D_float2[uint2(idx, idx)];
+    val += val2.x;
+    val += val2.y;
+    
+    // float2 texture operations for 3D
+    float2 val2_3d = rwt3D_float2[uint3(idx, idx, idx)];
+    rwt3D_float2[uint3(idx, idx, idx)] = float2(idx, idx);
+    val2_3d = rwt3D_float2[uint3(idx, idx, idx)];
+    val += val2_3d.x;
+    val += val2_3d.y;
+    
+    // float4 texture operations for 1D
+    float4 val4_1d = rwt1D_float4[idx];
+    rwt1D_float4[idx] = float4(idx, idx, idx, idx);
+    val4_1d = rwt1D_float4[idx];
+    val += val4_1d.x;
+    val += val4_1d.y;
+    val += val4_1d.z;
+    val += val4_1d.w;
+    
+    // float4 texture operations for 2D
+    float4 val4 = rwt2D_float4[uint2(idx, idx)];
+    rwt2D_float4[uint2(idx, idx)] = float4(idx, idx, idx, idx);
+    val4 = rwt2D_float4[uint2(idx, idx)];
+    val += val4.x;
+    val += val4.y;
+    val += val4.z;
+    val += val4.w;
+    
+    // float4 texture operations for 3D
+    float4 val4_3d = rwt3D_float4[uint3(idx, idx, idx)];
+    rwt3D_float4[uint3(idx, idx, idx)] = float4(idx, idx, idx, idx);
+    val4_3d = rwt3D_float4[uint3(idx, idx, idx)];
+    val += val4_3d.x;
+    val += val4_3d.y;
+    val += val4_3d.z;
+    val += val4_3d.w;
+    
     outputBuffer[idx] = val;
 }
diff --git a/tests/compute/rw-texture-simple.slang.expected.txt b/tests/compute/rw-texture-simple.slang.expected.txt
index 1d9023742..3c49364c4 100644
--- a/tests/compute/rw-texture-simple.slang.expected.txt
+++ b/tests/compute/rw-texture-simple.slang.expected.txt
@@ -1,5 +1,5 @@
 type: float
 3.000000
-6.000000
-9.000000
-12.000000
+24.000000
+45.000000
+66.000000
diff --git a/tests/compute/texture-subscript-cuda.slang b/tests/compute/texture-subscript-cuda.slang
index e64f42b19..7c4a2cc78 100644
--- a/tests/compute/texture-subscript-cuda.slang
+++ b/tests/compute/texture-subscript-cuda.slang
@@ -5,14 +5,24 @@
 Texture1D<float> cudaT1D;
 //TEST_INPUT: Texture2D(size=8, content = one):name cudaT2D
 Texture2D<float> cudaT2D;
+//TEST_INPUT: Texture2D(size=8, content = one):name cudaT2D_f2
+Texture2D<float2> cudaT2D_f2;
+//TEST_INPUT: Texture2D(size=8, content = one):name cudaT2D_f4
+Texture2D<float4> cudaT2D_f4;
 //TEST_INPUT: Texture3D(size=8, content = one):name cudaT3D
 Texture3D<float> cudaT3D;
+//TEST_INPUT: Texture3D(size=8, content = one):name cudaT3D_f2
+Texture3D<float2> cudaT3D_f2;
+//TEST_INPUT: Texture3D(size=8, content = one):name cudaT3D_f4
+Texture3D<float4> cudaT3D_f4;
 //TEST_INPUT: TextureCube(size=16, content = one):name cudaTCube
 TextureCube<float> cudaTCube;
 //TEST_INPUT: Texture2D(size=16, content = one, arrayLength=3):name cudaT2DArray
 Texture2DArray<float> cudaT2DArray;
-//TEST_INPUT: TextureCube(size=16, content = one, arrayLength=1):name cudaTCubeArray
-TextureCubeArray<float> cudaTCubeArray;
+//TEST_INPUT: Texture2D(size=16, content = one, arrayLength=3):name cudaT2DArray_f2
+Texture2DArray<float2> cudaT2DArray_f2;
+//TEST_INPUT: Texture2D(size=16, content = one, arrayLength=3):name cudaT2DArray_f4
+Texture2DArray<float4> cudaT2DArray_f4;
 
 //TEST_INPUT: ubuffer(data=[0 0 0 0 0 0 0], stride=4):out,name cudaOutputBuffer
 RWStructuredBuffer<float> cudaOutputBuffer;
@@ -23,39 +33,69 @@ void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
 {
     int idx = dispatchThreadID.x;
     
-	switch (idx)
-	{
-	case 1:
-		{
-			int var = 0;
-			float result = cudaT1D[0];
-			// This is not supported in PTX.
-			//cudaOutputBuffer[idx] = result;
-		}	
-	break;
-	
-	case 2:
-		{
-			int2 var = int2(1, 2);
-			float result = cudaT2D[var];
-			cudaOutputBuffer[idx] = result;
-		}
-	break;
-
-	case 3:
-		{
-			int3 var = int3(1, 1, 1);
-			float result = cudaT3D[var];
-			cudaOutputBuffer[idx] = result;
-		}
-	break;
-
-	case 4:
-		{
-			int3 var = int3(0, 0, 1);
-			float result = cudaT2DArray[var];
-			cudaOutputBuffer[idx] = result;
-		}
-	break;
-	}
+    switch (idx)
+    {
+    case 1:
+        {
+            int var = 0;
+            float result = cudaT1D[0];
+            // This is not supported in PTX.
+            //cudaOutputBuffer[idx] = result;
+        }    
+    break;
+    
+    case 2:
+        {
+            int2 var = int2(1, 2);
+            float result = cudaT2D[var];
+            cudaOutputBuffer[idx] = result;
+            
+            float2 result2 = cudaT2D_f2[var];
+            cudaOutputBuffer[idx] += result2.x;
+            cudaOutputBuffer[idx] += result2.y;
+
+            float4 result4 = cudaT2D_f4[var];
+            cudaOutputBuffer[idx] += result4.x;
+            cudaOutputBuffer[idx] += result4.y;
+            cudaOutputBuffer[idx] += result4.z;
+            cudaOutputBuffer[idx] += result4.w;
+        }
+    break;
+
+    case 3:
+        {
+            int3 var = int3(1, 1, 1);
+            float result = cudaT3D[var];
+            cudaOutputBuffer[idx] = result;
+            
+            float2 result2 = cudaT3D_f2[var];
+            cudaOutputBuffer[idx] += result2.x;
+            cudaOutputBuffer[idx] += result2.y;
+
+            float4 result4 = cudaT3D_f4[var];
+            cudaOutputBuffer[idx] += result4.x;
+            cudaOutputBuffer[idx] += result4.y;
+            cudaOutputBuffer[idx] += result4.z;
+            cudaOutputBuffer[idx] += result4.w;
+        }
+    break;
+
+    case 4:
+        {
+            int3 var = int3(0, 0, 1);
+            float result = cudaT2DArray[var];
+            cudaOutputBuffer[idx] = result;
+            
+            float2 result2 = cudaT2DArray_f2[var];
+            cudaOutputBuffer[idx] += result2.x;
+            cudaOutputBuffer[idx] += result2.y;
+
+            float4 result4 = cudaT2DArray_f4[var];
+            cudaOutputBuffer[idx] += result4.x;
+            cudaOutputBuffer[idx] += result4.y;
+            cudaOutputBuffer[idx] += result4.z;
+            cudaOutputBuffer[idx] += result4.w;
+        }
+    break;
+    }
 } 
\ No newline at end of file
diff --git a/tests/compute/texture-subscript-cuda.slang.expected.txt b/tests/compute/texture-subscript-cuda.slang.expected.txt
index 133a47e56..698717361 100644
--- a/tests/compute/texture-subscript-cuda.slang.expected.txt
+++ b/tests/compute/texture-subscript-cuda.slang.expected.txt
@@ -1,7 +1,7 @@
 0
 0
-3F800000
-3F800000
-3F800000
+40E00000
+40E00000
+40E00000
 0
 0
\ No newline at end of file
diff --git a/tests/compute/texture-subscript-uint-cuda.slang b/tests/compute/texture-subscript-uint-cuda.slang
new file mode 100644
index 000000000..d2b6cec59
--- /dev/null
+++ b/tests/compute/texture-subscript-uint-cuda.slang
@@ -0,0 +1,86 @@
+// Test for verifying subscript operator support with uint types in cuda.
+
+//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-cuda -compute -shaderobj
+//TEST_INPUT: Texture1D(size=4, content = one):name cudaT1D_u
+Texture1D<uint> cudaT1D_u;
+//TEST_INPUT: Texture2D(size=8, content = one):name cudaT2D_u
+Texture2D<uint> cudaT2D_u;
+//TEST_INPUT: Texture2D(size=8, content = one):name cudaT2D_u2
+Texture2D<uint2> cudaT2D_u2;
+//TEST_INPUT: Texture2D(size=8, content = one):name cudaT2D_u4
+Texture2D<uint4> cudaT2D_u4;
+//TEST_INPUT: Texture3D(size=8, content = one):name cudaT3D_u
+Texture3D<uint> cudaT3D_u;
+//TEST_INPUT: Texture3D(size=8, content = one):name cudaT3D_u2
+Texture3D<uint2> cudaT3D_u2;
+//TEST_INPUT: Texture3D(size=8, content = one):name cudaT3D_u4
+Texture3D<uint4> cudaT3D_u4;
+//TEST_INPUT: TextureCube(size=16, content = one):name cudaTCube_u
+TextureCube<uint> cudaTCube_u;
+//TEST_INPUT: Texture2D(size=16, content = one, arrayLength=3):name cudaT2DArray_u
+Texture2DArray<uint> cudaT2DArray_u;
+//TEST_INPUT: Texture2D(size=16, content = one, arrayLength=3):name cudaT2DArray_u2
+Texture2DArray<uint2> cudaT2DArray_u2;
+//TEST_INPUT: Texture2D(size=16, content = one, arrayLength=3):name cudaT2DArray_u4
+Texture2DArray<uint4> cudaT2DArray_u4;
+
+//TEST_INPUT: ubuffer(data=[0 0 0], stride=4):out,name cudaOutputBuffer
+RWStructuredBuffer<uint> cudaOutputBuffer;
+
+[numthreads(3, 1, 1)]
+[shader("compute")]
+void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int idx = dispatchThreadID.x;
+    
+    switch (idx)
+    {   
+    case 0:
+    {
+        int2 var = int2(1, 2);
+        uint result = cudaT2D_u[var];
+        cudaOutputBuffer[idx] = result;
+            
+        uint2 result2 = cudaT2D_u2[var];
+        cudaOutputBuffer[idx] += result2.x;
+
+        uint4 result4 = cudaT2D_u4[var];
+        cudaOutputBuffer[idx] += result4.x;
+        cudaOutputBuffer[idx] += result4.y;
+    }
+    break;
+
+    case 1:
+    {
+        int3 var = int3(1, 1, 1);
+        uint result = cudaT3D_u[var];
+        cudaOutputBuffer[idx] = result;
+            
+        uint2 result2 = cudaT3D_u2[var];
+        cudaOutputBuffer[idx] += result2.x;
+
+        uint4 result4 = cudaT3D_u4[var];
+        cudaOutputBuffer[idx] += result4.x;
+        cudaOutputBuffer[idx] += result4.w;
+    }
+    break;
+
+    case 2:
+    {
+        int3 var = int3(0, 0, 1);
+        uint result = cudaT2DArray_u[var];
+        cudaOutputBuffer[idx] = result;
+            
+        uint2 result2 = cudaT2DArray_u2[var];
+        cudaOutputBuffer[idx] += result2.y;
+
+        uint4 result4 = cudaT2DArray_u4[var];
+        cudaOutputBuffer[idx] += result4.x;
+        cudaOutputBuffer[idx] += result4.y;
+    }
+    break;
+    }
+} 
+// CHECK:FE000000
+// CHECK:FE000000
+// CHECK:FE000000
\ No newline at end of file