blob: 32d464d57e948ce4dbaa1137cbf634eea0c2095c (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
|
// No atomic support on CPU
//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -shaderobj
// No support for int64_t on DX11
//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj
//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -profile cs_6_0 -render-features atomic-int64 -compile-arg -O2 -shaderobj
//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -render-features atomic-int64 -shaderobj
//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -shaderobj
// The test doesn't directly use this, but having this defined makes the 0 slot available if NVAPI is going to be used
// Only strictly necessary on the D3D12 path
//TEST_INPUT:ubuffer(data=[0 0 0 0 ], stride=4):name=nvapiBuffer
RWStructuredBuffer<int> nvapiBuffer;
//TEST_INPUT:ubuffer(data=[0 1 2 3 4 5 6 7]):out,name=outputBuffer
RWByteAddressBuffer outputBuffer;
// With only 4 threads there is no contention - which makes for a simple test
// but doesn't actually test for the exchange atomicity
[numthreads(4, 1, 1)]
void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
{
int idx = dispatchThreadID.x;
// Try directly reading
uint2 currentValue2 = outputBuffer.Load2(idx << 3);
uint64_t currentValue = uint64_t(currentValue2.y) | currentValue2.x;
uint64_t readValue = outputBuffer.InterlockedExchangeU64(idx << 3, currentValue + 1);
}
|
