yum-mirror/slang

Making it easier to work with shaders

git clone https://git.yummers.dev/yum-mirror/slang

Harsh Aggarwal (NVIDIA)Updated support to enable batch3 (#8219)e0c20a076

master
branchesmasterall refs →
1.4 KiB33 linesraw
1// byte-address-half-atomics.slang
2// test the atomics on half types.
3
4// Disabled because validation layer doesn't like vector atomics, although nv driver does allow it.
5//DISABLED_TEST(compute, vulkan):COMPARE_COMPUTE(filecheck-buffer=CHECK): -vk -compute -profile cs_6_2 -render-features half -shaderobj -emit-spirv-directly -output-using-type
6//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK): -cuda -compute -profile cs_6_2 -render-features half -shaderobj -emit-spirv-directly -output-using-type
7//TEST:SIMPLE(filecheck=SPIRV):-target spirv -entry computeMain -stage compute -emit-spirv-directly -skip-spirv-validation
8//TEST:SIMPLE(filecheck=HLSL):-target hlsl -entry computeMain -profile cs_6_3
9//TEST_INPUT:set tmpBuffer = ubuffer(data=[0 0 0 0], stride=4)
10RWByteAddressBuffer tmpBuffer;
11
12//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name=outputBuffer
13RWStructuredBuffer<float> outputBuffer;
14
15[numthreads(1, 1, 1)]
16void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
17{
18    half originalValue;
19    
20    // SPIRV: OpAtomicFAddEXT
21    // HLSL: NvInterlockedAddFp16x2
22    tmpBuffer.InterlockedAddF16(0, 1.0h, originalValue);
23    tmpBuffer.InterlockedAddF16(2, 2.0h, originalValue);
24    
25    half2 v = tmpBuffer.Load<half2>(0);
26    
27    // CHECK: 1.0
28    outputBuffer[0] = v.x;
29    // CHECK: 2.0
30    outputBuffer[1] = v.y;
31    // CHECK: 0.0
32    outputBuffer[3] = originalValue;
33}