Fix#8080: Batch-4: Enable cuda tests (#8261)

10 files changed, 17 insertions, 12 deletions

diff --git a/tests/hlsl-intrinsic/image-swizzle-write.slang b/tests/hlsl-intrinsic/image-swizzle-write.slang
index 3a4f233bc..6429e7e41 100644
--- a/tests/hlsl-intrinsic/image-swizzle-write.slang
+++ b/tests/hlsl-intrinsic/image-swizzle-write.slang
@@ -1,5 +1,6 @@
 //TEST:SIMPLE(filecheck=CHECK): -entry computeMain -stage compute -target spirv -emit-spirv-directly
 //TEST:SIMPLE(filecheck=CHECK): -entry computeMain -stage compute -target spirv
+//TEST:SIMPLE(filecheck=CHECK-CUDA): -entry computeMain -stage compute -target cuda
 
 layout(rgba8)
 RWTexture2D<float4> texture;
@@ -9,7 +10,8 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
 {
     half h = 1.0h;
     // CHECK: OpImageRead
-    // CEHCK: OpCompositeInsert
+    // CHECK: OpCompositeInsert
     // CHECK: OpImageWrite
+    // CHECK-CUDA: CUsurfObject texture_0
     texture[dispatchThreadID.xy].a = h;
 }
diff --git a/tests/hlsl-intrinsic/texture/texture-sample-count.slang b/tests/hlsl-intrinsic/texture/texture-sample-count.slang
index 0af7b92ba..0457d41b7 100644
--- a/tests/hlsl-intrinsic/texture/texture-sample-count.slang
+++ b/tests/hlsl-intrinsic/texture/texture-sample-count.slang
@@ -1,5 +1,6 @@
 //TEST:SIMPLE(filecheck=CHECK): -target hlsl -profile sm_6_6 -entry main -stage compute
 //TEST:SIMPLE(filecheck=DXIL): -target dxil -profile sm_6_6 -entry main -stage compute
+//TEST:SIMPLE(filecheck=DXIL): -target cuda -profile sm_6_6 -entry main -stage compute
 
 // Test that Texture2DMS accepts an optional sampleCount argument
 // and the argument correctly appears in the output hlsl.
diff --git a/tests/hlsl-intrinsic/wave-active-product.slang b/tests/hlsl-intrinsic/wave-active-product.slang
index a15cbfc6d..1450f1b36 100644
--- a/tests/hlsl-intrinsic/wave-active-product.slang
+++ b/tests/hlsl-intrinsic/wave-active-product.slang
@@ -3,7 +3,7 @@
 //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj -render-feature hardware-device
 //TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -render-feature hardware-device
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-metal -compute -shaderobj
 
diff --git a/tests/hlsl-intrinsic/wave-diverge.slang b/tests/hlsl-intrinsic/wave-diverge.slang
index a18e99f58..a2cf76f63 100644
--- a/tests/hlsl-intrinsic/wave-diverge.slang
+++ b/tests/hlsl-intrinsic/wave-diverge.slang
@@ -3,7 +3,7 @@
 //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj
 //TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-metal -compute -shaderobj
 
diff --git a/tests/hlsl-intrinsic/wave-equality.slang b/tests/hlsl-intrinsic/wave-equality.slang
index 4271c3c11..b513e41d3 100644
--- a/tests/hlsl-intrinsic/wave-equality.slang
+++ b/tests/hlsl-intrinsic/wave-equality.slang
@@ -3,7 +3,7 @@
 //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj
 //TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
@@ -29,4 +29,4 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     value |= WaveActiveAllEqual(v1)           ? 0x20 : 0;     // false
     
     outputBuffer[idx] = value;
-}
\ No newline at end of file
+}
diff --git a/tests/hlsl-intrinsic/wave-matrix.slang b/tests/hlsl-intrinsic/wave-matrix.slang
index c4fe635be..c366db948 100644
--- a/tests/hlsl-intrinsic/wave-matrix.slang
+++ b/tests/hlsl-intrinsic/wave-matrix.slang
@@ -3,7 +3,7 @@
 //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj -render-feature hardware-device
 //DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
@@ -35,4 +35,4 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     matrix<int, 2, 2> r = r0 + matrix<int, 2, 2>(r1) + r6;
    
     outputBuffer[idx] = r[0][0] + r[0][1] + r[1][0] + r[1][1];
-}
\ No newline at end of file
+}
diff --git a/tests/hlsl-intrinsic/wave-prefix-count-bits.slang b/tests/hlsl-intrinsic/wave-prefix-count-bits.slang
index e16a8428b..bb1ad3ab0 100644
--- a/tests/hlsl-intrinsic/wave-prefix-count-bits.slang
+++ b/tests/hlsl-intrinsic/wave-prefix-count-bits.slang
@@ -3,7 +3,7 @@
 //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj -render-feature hardware-device
 //TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -render-feature hardware-device
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
@@ -14,4 +14,4 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     int idx = int(dispatchThreadID.x);
     
     outputBuffer[idx] = int(WavePrefixCountBits(bool(idx & 5)));
-}
\ No newline at end of file
+}
diff --git a/tests/hlsl-intrinsic/wave-prefix-sum-fp16.slang b/tests/hlsl-intrinsic/wave-prefix-sum-fp16.slang
index 03fa39da8..50258f1c3 100644
--- a/tests/hlsl-intrinsic/wave-prefix-sum-fp16.slang
+++ b/tests/hlsl-intrinsic/wave-prefix-sum-fp16.slang
@@ -1,6 +1,7 @@
 //TEST:SIMPLE(filecheck=CHECK_SPV):-target spirv -entry computeMain -stage compute -emit-spirv-directly
 //TEST:SIMPLE(filecheck=CHECK_SPV):-target spirv -entry computeMain -stage compute
 //TEST:SIMPLE(filecheck=CHECK_WGSL):-target wgsl -entry computeMain -stage compute
+//TEST:SIMPLE(filecheck=CHECK_CUDA):-target cuda -entry computeMain -stage compute
 //TEST:SIMPLE(filecheck=CHECK_METAL):-target metal -entry computeMain -stage compute
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer
@@ -16,6 +17,7 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     // CHECK_SPV: OpGroupNonUniformFAdd
     // CHECK_WGSL: subgroupExclusiveAdd
     // CHECK_METAL: simd_prefix_exclusive_sum
+    // CHECK_CUDA: _wavePrefixSumMultiple
     float2 r1 = WavePrefixSum(v1);
     
     outputBuffer[idx] = (int)r1.x;
diff --git a/tests/hlsl-intrinsic/wave-read-lane-at.slang b/tests/hlsl-intrinsic/wave-read-lane-at.slang
index bae063209..391a8900c 100644
--- a/tests/hlsl-intrinsic/wave-read-lane-at.slang
+++ b/tests/hlsl-intrinsic/wave-read-lane-at.slang
@@ -4,7 +4,7 @@
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj -render-feature hardware-device
 // Disabled on VK because glsl can't do WaveReadLaneAt on matrix. 
 //DISABLE_TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
@@ -52,4 +52,4 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     }
     
     outputBuffer[idx] = value;
-}
\ No newline at end of file
+}
diff --git a/tests/hlsl-intrinsic/wave.slang b/tests/hlsl-intrinsic/wave.slang
index f7e52b887..9cbe7ec7a 100644
--- a/tests/hlsl-intrinsic/wave.slang
+++ b/tests/hlsl-intrinsic/wave.slang
@@ -3,7 +3,7 @@
 //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0 -shaderobj
 //TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
-//TEST:COMPARE_COMPUTE_EX:-cuda -compute -render-features cuda_sm_7_0 -shaderobj
+//TEST:COMPARE_COMPUTE_EX:-cuda -compute -capability cuda_sm_7_0 -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj
 //TEST:COMPARE_COMPUTE_EX:-metal -compute -shaderobj