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
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
|
// Tests automatic synthesis of Differential type requirement for generic types.
//
// This specifically tests a synthesis path that occurs when the lookup of the Differential type happens before the conformance-check.
// If this path doesn't construct the generic differential type correctly, it will throw an error when constructing the array
// in this line: Feature<3>.Differential b = {0.2, 0.3, 0.4};
//
//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type
//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -shaderobj -output-using-type
//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
//TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer
RWStructuredBuffer<float> outputBuffer;
__generic<let C : int>
struct Feature: IDifferentiable
{
float vals[C];
}
struct Linear<let C : int>
{
typedef Feature<C> Input;
typedef Feature<C> Output;
[BackwardDerivative(eval_bwd)]
Output eval(Input in_feature)
{
Output out_feature;
for (int i = 0; i < C; i++)
{
out_feature.vals[i] = in_feature.vals[i] * 2.0;
}
return out_feature;
}
void eval_bwd(inout DifferentialPair<Input> in_feature_pair, Feature<C>.Differential d_output)
{
/* empty.. doesn't really matter */
}
}
[Differentiable]
Feature<3> f(Feature<3> a, Linear<3> layer)
{
return layer.eval(a);
}
[numthreads(1, 1, 1)]
void computeMain(uint3 dispatchThreadID: SV_DispatchThreadID)
{
Feature<3> a = {1.0, 2.0, 3.0};
Feature<3>.Differential b = {0.2, 0.3, 0.4};
Linear<3> layer;
var dpA = diffPair(a, b);
var result = fwd_diff(f)(dpA, layer).d;
outputBuffer[0] = result.vals[0];
outputBuffer[1] = result.vals[1];
outputBuffer[2] = result.vals[2];
}
|
