yum-mirror/slang

Making it easier to work with shaders

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

Harsh Aggarwal (NVIDIA)Enable CUDA support for additional HLSL intrinsic tests (#8293)5ec41675d

master
branchesmasterall refs →
4.9 KiB133 linesraw
1//TEST_CATEGORY(wave, compute)
2//TEST:COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-vk -compute -shaderobj -emit-spirv-directly
3//TEST:COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-cuda -compute -shaderobj -profile sm_6_0
4//TEST:COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-vk -compute -shaderobj -emit-spirv-via-glsl -profile sm_6_0 -Xslang... -capability GL_KHR_shader_subgroup_rotate -X.
5//TEST:COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-vk -compute -shaderobj -emit-spirv-directly -xslang -DUSE_GLSL_SYNTAX -allow-glsl
6//TEST:COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-vk -compute -shaderobj -emit-spirv-via-glsl -profile sm_6_0 -allow-glsl -Xslang... -DUSE_GLSL_SYNTAX -capability GL_KHR_shader_subgroup_rotate -X.
7
8#if defined(USE_GLSL_SYNTAX)
9#define __clusteredRotate subgroupClusteredRotate
10#else
11#define __clusteredRotate WaveClusteredRotate
12#endif
13
14//TEST_INPUT:ubuffer(data=[0], stride=4):out,name=outputBuffer
15RWStructuredBuffer<uint> outputBuffer;
16
17#define SUBGROUP_SIZE 32
18#define DELTA 3
19#define CLUSTER_SIZE 8
20
21static uint threadIndex;
22static uint clusterIndex;
23static uint rotatedValue;
24
25__generic<T : __BuiltinArithmeticType>
26bool test1ClusteredRotate()
27{
28    return __clusteredRotate(T(threadIndex), DELTA, CLUSTER_SIZE) == T(rotatedValue);
29}
30
31__generic<T : __BuiltinArithmeticType, let N : int>
32bool testVRClusteredRotate()
33{
34    typealias gvec = vector<T, N>;
35
36#if defined(USE_GLSL_SYNTAX)
37    return (__clusteredRotate(gvec(T(threadIndex)), DELTA, CLUSTER_SIZE) == gvec(T(rotatedValue)));
38#else
39    return (__clusteredRotate(gvec(T(threadIndex)), DELTA, CLUSTER_SIZE) == gvec(T(rotatedValue)))[0];
40#endif
41}
42
43bool test1ClusteredRotateBool()
44{
45    bool currentValue = (threadIndex % 2 == 0) ? true : false;
46    bool rotatedValueBool = (threadIndex % 2 == 0) ? false : true;
47    return __clusteredRotate(currentValue, DELTA, CLUSTER_SIZE) == rotatedValueBool;
48}
49
50__generic<let N : int>
51bool testVRClusteredRotateBool()
52{
53    typealias gvec = vector<bool, N>;
54    bool currentValue = (threadIndex % 2 == 0) ? true : false;
55    bool rotatedValueBool = (threadIndex % 2 == 0) ? false : true;
56
57#if defined(USE_GLSL_SYNTAX)
58    return (__clusteredRotate(gvec(currentValue), DELTA, CLUSTER_SIZE) == gvec(rotatedValueBool));
59#else
60    return (__clusteredRotate(gvec(currentValue), DELTA, CLUSTER_SIZE) == gvec(rotatedValueBool))[0];
61#endif
62}
63
64bool testClusteredRotate()
65{
66    return true
67        & test1ClusteredRotate<float>()
68        & testVRClusteredRotate<float, 2>()
69        & testVRClusteredRotate<float, 3>()
70        & testVRClusteredRotate<float, 4>()
71        & test1ClusteredRotate<half>()
72        & testVRClusteredRotate<half, 2>()
73        & testVRClusteredRotate<half, 3>()
74        & testVRClusteredRotate<half, 4>()
75        & test1ClusteredRotate<uint>()
76        & testVRClusteredRotate<uint, 2>()
77        & testVRClusteredRotate<uint, 3>()
78        & testVRClusteredRotate<uint, 4>()
79        & test1ClusteredRotate<uint16_t>()
80        & testVRClusteredRotate<uint16_t, 2>()
81        & testVRClusteredRotate<uint16_t, 3>()
82        & testVRClusteredRotate<uint16_t, 4>()
83        & test1ClusteredRotate<int>()
84        & testVRClusteredRotate<int, 2>()
85        & testVRClusteredRotate<int, 3>()
86        & testVRClusteredRotate<int, 4>()
87        & test1ClusteredRotate<int16_t>()
88        & testVRClusteredRotate<int16_t, 2>()
89        & testVRClusteredRotate<int16_t, 3>()
90        & testVRClusteredRotate<int16_t, 4>()
91        & test1ClusteredRotate<uint8_t>()
92        & testVRClusteredRotate<uint8_t, 2>()
93        & testVRClusteredRotate<uint8_t, 3>()
94        & testVRClusteredRotate<uint8_t, 4>()
95        & test1ClusteredRotate<uint64_t>()
96        & testVRClusteredRotate<uint64_t, 2>()
97        & testVRClusteredRotate<uint64_t, 3>()
98        & testVRClusteredRotate<uint64_t, 4>()
99        & test1ClusteredRotate<int8_t>()
100        & testVRClusteredRotate<int8_t, 2>()
101        & testVRClusteredRotate<int8_t, 3>()
102        & testVRClusteredRotate<int8_t, 4>()
103        & test1ClusteredRotate<int64_t>()
104        & testVRClusteredRotate<int64_t, 2>()
105        & testVRClusteredRotate<int64_t, 3>()
106        & testVRClusteredRotate<int64_t, 4>()
107        & test1ClusteredRotateBool()
108        & testVRClusteredRotateBool<2>()
109        & testVRClusteredRotateBool<3>()
110        & testVRClusteredRotateBool<4>()
111        ;
112}
113
114[shader("compute")]
115[numthreads(SUBGROUP_SIZE, 1, 1)]
116void computeMain(uint3 dispatchID : SV_DispatchThreadID)
117{
118    threadIndex = dispatchID.x;
119    clusterIndex = dispatchID.x % CLUSTER_SIZE;
120
121    // Determine expected value of clustered rotate in current invocation.
122    // The values passed in are global invocation ids, and we rotate them withina cluster of size `CLUSTER_SIZE`.
123    uint clusterStart = (threadIndex / CLUSTER_SIZE) * CLUSTER_SIZE;
124    rotatedValue = clusterStart + ((threadIndex - clusterStart + DELTA) % CLUSTER_SIZE);
125
126    bool result = true
127            & testClusteredRotate()
128            ;
129
130    // CHECK: 1
131    outputBuffer[0] = uint(result);
132}
133