diff options
Diffstat (limited to 'tests')
6 files changed, 59 insertions, 62 deletions
diff --git a/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-vote.slang b/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-vote.slang index 23466b742..759bf0e11 100644 --- a/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-vote.slang +++ b/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-vote.slang @@ -3,7 +3,7 @@ //TEST:SIMPLE(filecheck=CHECK_HLSL): -allow-glsl -stage compute -entry computeMain -target hlsl -DTARGET_HLSL // not testing cuda due to missing impl -//DISABLE_TEST:SIMPLE(filecheck=CHECK_CUDA): -allow-glsl -stage compute -entry computeMain -target cuda -DTARGET_CUDA +//DISABLE_TEST:SIMPLE(filecheck=CHECK_CUDA): -allow-glsl -stage compute -entry computeMain -target cuda -DTARGET_CUDA // not testing cpp due to missing impl //DISABLE_TEST:SIMPLE(filecheck=CHECK_CPP): -allow-glsl -stage compute -entry computeMain -target cpp -DTARGET_CPP @@ -23,7 +23,7 @@ buffer MyBlockName } inputBuffer; //TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer -buffer MyBlockName2 +buffer MyBlockName2 { uint data[]; } outputBuffer; @@ -73,7 +73,7 @@ bool testAllEqual() { & testVAllEqual<double, 2>() & testVAllEqual<double, 3>() & testVAllEqual<double, 4>() - & test1AllEqual<half>() + & test1AllEqual<half>() & testVAllEqual<half, 2>() & testVAllEqual<half, 3>() & testVAllEqual<half, 4>() @@ -81,15 +81,15 @@ bool testAllEqual() { & testVAllEqual<int, 2>() & testVAllEqual<int, 3>() & testVAllEqual<int, 4>() - & test1AllEqual<int8_t>() + & test1AllEqual<int8_t>() & testVAllEqual<int8_t, 2>() & testVAllEqual<int8_t, 3>() & testVAllEqual<int8_t, 4>() - & test1AllEqual<int16_t>() + & test1AllEqual<int16_t>() & testVAllEqual<int16_t, 2>() & testVAllEqual<int16_t, 3>() & testVAllEqual<int16_t, 4>() - & test1AllEqual<int64_t>() + & test1AllEqual<int64_t>() & testVAllEqual<int64_t, 2>() & testVAllEqual<int64_t, 3>() & testVAllEqual<int64_t, 4>() @@ -97,15 +97,15 @@ bool testAllEqual() { & testVAllEqual<uint, 2>() & testVAllEqual<uint, 3>() & testVAllEqual<uint, 4>() - & test1AllEqual<uint8_t>() + & test1AllEqual<uint8_t>() & testVAllEqual<uint8_t, 2>() & testVAllEqual<uint8_t, 3>() & testVAllEqual<uint8_t, 4>() - & test1AllEqual<uint16_t>() + & test1AllEqual<uint16_t>() & testVAllEqual<uint16_t, 2>() & testVAllEqual<uint16_t, 3>() & testVAllEqual<uint16_t, 4>() - & test1AllEqual<uint64_t>() + & test1AllEqual<uint64_t>() & testVAllEqual<uint64_t, 2>() & testVAllEqual<uint64_t, 3>() & testVAllEqual<uint64_t, 4>() @@ -117,12 +117,12 @@ bool testAllEqual() { } [[ForceInline]] -void _barrier() +void MyBarrier() { -#if !defined(WGPU) - subgroupBarrier(); -#else +#if defined(WGPU) GroupMemoryBarrier(); +#else + subgroupBarrier(); #endif } @@ -133,53 +133,57 @@ void computeMain() // one is true, rest false, positive outputBuffer.data[0] = 1; - -#if !defined(WGPU) - bool t1 = inputBuffer.data[0] == gl_GlobalInvocationID.x; -#else - // There is no subgroup barrier for WGSL and workgroup barrier requries non uniform control flow. + +#if defined(WGPU) + // There is no subgroup barrier for WGSL. + // And workgroup barrier requries non uniform control flow. bool t1 = true; +#else + bool t1 = inputBuffer.data[0] == gl_GlobalInvocationID.x; #endif - if (subgroupAny(t1)) { - _barrier(); + + if (subgroupAny(t1)) + { outputBuffer.data[0] = 2; } + MyBarrier(); // all false, negative outputBuffer.data[1] = 1; t1 = false; - if (!subgroupAny(t1)) { - _barrier(); + if (!subgroupAny(t1)) + { outputBuffer.data[1] = 2; } + MyBarrier(); // all true, positive outputBuffer.data[2] = 1; t1 = true; - if (subgroupAll(t1)) { - _barrier(); + if (subgroupAll(t1)) + { outputBuffer.data[2] = 2; } + MyBarrier(); // all false, negative outputBuffer.data[3] = 1; t1 = false; - if (!subgroupAll(t1)) { - _barrier(); + if (!subgroupAll(t1)) + { outputBuffer.data[3] = 2; } + MyBarrier(); outputBuffer.data[4] = 1; - // All equal intrinsic is not supported on WGSL as of time of writing. #if !defined(WGPU) && !defined(METAL) - if (testAllEqual()) { - subgroupBarrier(); + if (testAllEqual()) +#endif + { outputBuffer.data[4] = 2; } -#else - outputBuffer.data[4] = 2; -#endif + MyBarrier(); // CHECK_GLSL: void main( // CHECK_SPV: OpEntryPoint diff --git a/tests/hlsl-intrinsic/wave-active-product.slang b/tests/hlsl-intrinsic/wave-active-product.slang index e59e8b6fe..49d87f474 100644 --- a/tests/hlsl-intrinsic/wave-active-product.slang +++ b/tests/hlsl-intrinsic/wave-active-product.slang @@ -4,28 +4,33 @@ //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -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 -capability cuda_sm_7_0 -shaderobj -//TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj //TEST:COMPARE_COMPUTE_EX:-metal -compute -shaderobj -//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer +// WGSL doesn't support wave functions in a dynamic control flow; it works with uniform control flow. +//DISABLE_TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj + +//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer RWStructuredBuffer<int> outputBuffer; [numthreads(8, 1, 1)] void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) { const int idx = int(dispatchThreadID.x); - + #if 1 if (idx < 3) { // Diverge!! - outputBuffer[idx] = -1; + outputBuffer[idx] = -1; // thread 0, 1 and 2 will get 0xFFFFFFFF return; } + + // thread 2, 3, 4, 5, 6, 7 will get the result of 2 * 3 * 4 * 5 * 6 * 7 + // which is 2520, which is 9D8 in hex outputBuffer[idx] = WaveActiveProduct(idx); #else - /// NOTE! Can't say I totally understand WaveActiveProduct. + /// NOTE! Can't say I totally understand WaveActiveProduct. /// The following returns 0x240 on CUDA - which is what I'd expect /// On DX12, it returns 0 diff --git a/tests/hlsl-intrinsic/wave-active-product.slang.expected.txt b/tests/hlsl-intrinsic/wave-active-product.slang.expected.txt index dbe392009..50bb4de5b 100644 --- a/tests/hlsl-intrinsic/wave-active-product.slang.expected.txt +++ b/tests/hlsl-intrinsic/wave-active-product.slang.expected.txt @@ -6,11 +6,3 @@ FFFFFFFF 9D8 9D8 9D8 -0 -0 -0 -0 -0 -0 -0 -0 diff --git a/tests/hlsl-intrinsic/wave-diverge.slang b/tests/hlsl-intrinsic/wave-diverge.slang index 183a23c79..d177c1eb7 100644 --- a/tests/hlsl-intrinsic/wave-diverge.slang +++ b/tests/hlsl-intrinsic/wave-diverge.slang @@ -4,9 +4,11 @@ //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -profile cs_6_0 -shaderobj //TEST(vulkan):COMPARE_COMPUTE_EX:-vk -compute -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 +// WGSL doesn't support wave functions in a dynamic control flow; it works with uniform control flow. +//DISABLE_TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj + //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer RWStructuredBuffer<int> outputBuffer; @@ -14,16 +16,16 @@ RWStructuredBuffer<int> outputBuffer; void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) { int idx = int(dispatchThreadID.x); - + int value = 0; - + if (idx == 2) { // diverge return; } - + value = WaveActiveMin(idx + 1); - + outputBuffer[idx] = value; } diff --git a/tests/hlsl-intrinsic/wave-is-first-lane.slang b/tests/hlsl-intrinsic/wave-is-first-lane.slang index 812e573be..5150c506e 100644 --- a/tests/hlsl-intrinsic/wave-is-first-lane.slang +++ b/tests/hlsl-intrinsic/wave-is-first-lane.slang @@ -4,24 +4,26 @@ //TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -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 -capability cuda_sm_7_0 -shaderobj -//TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj //TEST:COMPARE_COMPUTE_EX:-metal -compute -shaderobj -//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer +// WGSL doesn't support wave functions in a dynamic control flow; it works with uniform control flow. +//TEST:COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj + +//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer RWStructuredBuffer<int> outputBuffer; [numthreads(8, 1, 1)] void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) { int idx = int(dispatchThreadID.x); - + if (idx < 3) { // Diverge!! outputBuffer[idx] = -1; return; } - + int value = 0; outputBuffer[idx] = WaveIsFirstLane(); } diff --git a/tests/hlsl-intrinsic/wave-is-first-lane.slang.expected.txt b/tests/hlsl-intrinsic/wave-is-first-lane.slang.expected.txt index 43debbc9d..fc24c383f 100644 --- a/tests/hlsl-intrinsic/wave-is-first-lane.slang.expected.txt +++ b/tests/hlsl-intrinsic/wave-is-first-lane.slang.expected.txt @@ -6,11 +6,3 @@ FFFFFFFF 0 0 0 -0 -0 -0 -0 -0 -0 -0 -0 |