yum-mirror/glslang

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

Saddamvalidate createTensorLayoutNV/createTensorViewNV arguments09cfe5ce

main
branchesmainall refs →
2.5 KiB79 linesraw
1#version 450 core
2#extension GL_KHR_memory_scope_semantics : enable
3#extension GL_KHR_cooperative_matrix : enable
4#extension GL_EXT_shader_explicit_arithmetic_types : enable
5#extension GL_NV_cooperative_matrix2 : enable
6#extension GL_EXT_buffer_reference : enable
7
8layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
9
10buffer BufType {
11   float16_t arr[256][256];
12   float16_t x[];
13} Buf;
14
15layout(buffer_reference, std430, buffer_reference_align = 2) buffer fp16Buf {
16    float16_t f;
17};
18
19
20float16_t decode(const in fp16Buf b, const in uint32_t blockCoords[2], const in uint32_t coordInBlock[2])
21{
22    return b.f;
23}
24
25struct S {
26   f16vec2 x;
27};
28
29
30layout(std430, binding  = 0) buffer SBuf {
31    S s[];
32} sbuf;
33
34layout(constant_id = 0) const uint32_t Clamp = gl_CooperativeMatrixClampModeConstantNV;
35layout(constant_id = 1) const int32_t Clamp_signed = gl_CooperativeMatrixClampModeConstantNV;
36
37void main()
38{
39    coopmat<float16_t, gl_ScopeWorkgroup, 64, 32, gl_MatrixUseA> A;
40
41    tensorLayoutNV<2> t = createTensorLayoutNV(2);
42    tensorLayoutNV<3, 1> t2 = createTensorLayoutNV(3, 1);
43
44    t = setTensorLayoutBlockSizeNV(t, 4, 8);
45    t = setTensorLayoutDimensionNV(t, 256, 512);
46    t = sliceTensorLayoutNV(t, 128, 32, 256, 32);
47
48    tensorViewNV<5> v = createTensorViewNV(5);
49    v = setTensorViewDimensionsNV(v, 10, 11, 12, 13, 14);
50    v = setTensorViewStrideNV(v, 10, 11, 12, 13, 15);
51    v = setTensorViewClipNV(v, 0, 16, 0, 16);
52
53    tensorViewNV<5, true> v2 = createTensorViewNV(5, true);
54    tensorViewNV<2, true, 1, 0> v3 = createTensorViewNV(2, true, 1, 0);
55
56    coopMatLoadTensorNV(A, Buf.x, 0, t);
57    coopMatStoreTensorNV(A, Buf.x, 0, t);
58
59    coopMatLoadTensorNV(A, Buf.x, 0, t, v);
60    coopMatStoreTensorNV(A, Buf.x, 0, t, v);
61
62    coopMatLoadTensorNV(A, Buf.x, 0, t, decode);
63    coopMatLoadTensorNV(A, Buf.x, 0, t, v, decode);
64
65    coopMatLoadTensorNV(A, sbuf.s, 1, t, v, decode);
66
67    tensorLayoutNV<2, Clamp> tc = createTensorLayoutNV(2, Clamp);
68
69    coopmat<float16_t, gl_ScopeWorkgroup, 64, 32, gl_MatrixUseA> Arr[2];
70    coopMatLoadTensorNV(Arr[1], Buf.x, 0, t);
71
72    coopMatLoadTensorNV(A, Buf.arr[8], 0, t);
73    coopMatStoreTensorNV(A, Buf.arr[16], 0, t);
74
75    tensorLayoutNV<2, Clamp_signed> tc_s = createTensorLayoutNV(2, Clamp_signed);
76    tensorViewNV<Clamp, true> v4 = createTensorViewNV(Clamp, true);
77    tensorLayoutNV<Clamp> tc2 = createTensorLayoutNV(Clamp);
78    tensorViewNV<Clamp> v5 = createTensorViewNV(Clamp);
79}