yum-mirror/slang

Making it easier to work with shaders

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

Yong HeVarious WGSL fixes. (#5490)7c2ff5475

master
branchesmasterall refs →
3.0 KiB106 linesraw
1//TEST(compute):COMPARE_COMPUTE_EX:-dx12 -compute -shaderobj -output-using-type
2//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -shaderobj -output-using-type
3//TEST(compute):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
4//TEST(compute):COMPARE_COMPUTE_EX:-wgpu -compute -shaderobj -output-using-type
5//TEST:SIMPLE(filecheck=CHECK): -target hlsl -profile cs_5_0 -entry computeMain -line-directive-mode none
6//DISABLE_TEST:SIMPLE(filecheck=CHK):-target glsl -stage compute -entry computeMain -report-checkpoint-intermediates
7
8//TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer
9RWStructuredBuffer<float> outputBuffer;
10
11typedef DifferentialPair<float> dpfloat;
12typedef float.Differential dfloat;
13
14// Test that compute does not have a context.
15// CHECK-NOT: struct {{[a-zA-Z0-9_]*}}_compute_{{[a-zA-Z0-9_]*}}
16
17[BackwardDifferentiable]
18[PreferRecompute]
19float compute(float x, float y)
20{
21    return x * y;
22}
23
24[BackwardDifferentiable]
25[ForceInline]
26float infinitesimal(float x)
27{
28    return x - detach(x);
29}
30
31// Test that computeLoop compiles to just return 0.
32// CHECK: float3 computeLoop{{[_0-9]*}}(float y{{[_0-9]*}})
33// CHECK-NOT: for{{.*}}
34// CHECK: return (float3)0
35
36// Test that computeLoop's intermediates have no float sitting
37// around (must not cache the outvar from 'compute()')
38// CHECK: struct s_bwd_prop_computeLoop_Intermediates
39// CHECK-NEXT: {
40// CHECK-NOT: {{[A-Za-z0-9_]+}} {{[A-Za-z0-9_]+}}[{{.*}}]
41// CHECK: }
42
43[PreferRecompute]
44[BackwardDifferentiable]
45[ForceInline]
46float3 infinitesimal(float3 x)
47{
48    return x - detach(x);
49}
50
51//CHK: note: checkpointing context of 20 bytes associated with function: 'computeLoop'
52[BackwardDifferentiable]
53[PreferRecompute]
54float3 computeLoop(float y)
55{
56    //CHK: note: 4 bytes (float) used to checkpoint the following item:
57    float w = 0;
58
59    //CHK: note: 12 bytes (Vector<float, 3> ) used to checkpoint the following item:
60    float3 w3 = float3(0, 0, 0);
61
62    //CHK: note: 4 bytes (int32_t) used for a loop counter here:
63    for (int i = 0; i < 8; i++)
64    {
65        float k = compute(i, y);
66        float g = select(k > 0.0, k, 0.0);
67        w += g;
68        w3 += float3(g) * y;
69    }
70
71    var p = (w3 / float3(w));
72    return infinitesimal(p);
73}
74
75// Since computeLoop is recomputed, test_simple_loop should have nothing to store
76// therefore we check that there is no intermediate context type generated for test_simple_loop.
77
78// CHECK-NOT: struct {{[a-zA-Z0-9_]*}}test_simple_loop{{[a-zA-Z0-9_]*}}
79[BackwardDifferentiable]
80float test_simple_loop(float y)
81{  
82    float3 x = computeLoop(y);
83    return y + x.x;
84}
85
86[numthreads(1, 1, 1)]
87void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
88{
89    {
90        dpfloat dpa = dpfloat(1.0, 0.0);
91
92        __bwd_diff(test_simple_loop)(dpa, 1.0f);
93        outputBuffer[0] = dpa.d; // Expect: 1.0
94    }
95
96    {
97        dpfloat dpa = dpfloat(0.4, 0.0);
98
99        __bwd_diff(test_simple_loop)(dpa, 0.5f);
100        outputBuffer[1] = dpa.d; // Expect: 2.0
101    }
102
103    outputBuffer[2] = computeLoop(1.0).x;
104}
105
106//CHK-NOT: note