diff options
| author | Yong He <yonghe@outlook.com> | 2023-01-23 06:59:25 -0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-01-23 06:59:25 -0800 |
| commit | 46a4d98baa1d43b33717b4377aefeeaf46b9c2ff (patch) | |
| tree | c89f3a1c416330f859887d00f896b18bcc7488a5 /tests/compute | |
| parent | 263ca18ea516cfce43fda703c0a411aaf1938e42 (diff) | |
Full address insts elimination for backward autodiff. (#2604)
Co-authored-by: Yong He <yhe@nvidia.com>
Diffstat (limited to 'tests/compute')
| -rw-r--r-- | tests/compute/half-texture.slang.glsl | 15 | ||||
| -rw-r--r-- | tests/compute/half-texture.slang.hlsl | 22 |
2 files changed, 21 insertions, 16 deletions
diff --git a/tests/compute/half-texture.slang.glsl b/tests/compute/half-texture.slang.glsl index 88f585378..0eccccaaf 100644 --- a/tests/compute/half-texture.slang.glsl +++ b/tests/compute/half-texture.slang.glsl @@ -21,20 +21,23 @@ layout(std430, binding = 0) buffer _S1 { int _data[]; } outputBuffer_0; -layout(local_size_x = 4, local_size_y = 4, local_size_z = 1) in;void main() +layout(local_size_x = 4, local_size_y = 4, local_size_z = 1) in; +void main() { ivec2 pos_0 = ivec2(gl_GlobalInvocationID.xy); const float _S2 = 1.00000000000000000000 / 3.00000000000000000000; - ivec2 pos2_0 = ivec2(3 - pos_0.y, 3 - pos_0.x); + int _S3 = pos_0.y; + int _S4 = pos_0.x; + ivec2 pos2_0 = ivec2(3 - _S3, 3 - _S4); float16_t h_0 = (float16_t(imageLoad((halfTexture_0), ivec2((uvec2(pos2_0)))).x)); f16vec2 h2_0 = (f16vec2(imageLoad((halfTexture2_0), ivec2((uvec2(pos2_0)))).xy)); f16vec4 h4_0 = (f16vec4(imageLoad((halfTexture4_0), ivec2((uvec2(pos2_0)))))); - imageStore((halfTexture_0), ivec2((uvec2(pos_0))), f16vec4(h2_0.x + h2_0.y, float16_t(0), float16_t(0), float16_t(0))); - imageStore((halfTexture2_0), ivec2((uvec2(pos_0))), f16vec4(h4_0.xy, float16_t(0), float16_t(0))); - imageStore((halfTexture4_0), ivec2((uvec2(pos_0))), f16vec4(h2_0, h_0, h_0)); + imageStore((halfTexture_0), ivec2((uvec2(pos_0))), f16vec4(h2_0.x + h2_0.y, float16_t(0), float16_t(0), float16_t(0))); + imageStore((halfTexture2_0), ivec2((uvec2(pos_0))), f16vec4(h4_0.xy, float16_t(0), float16_t(0))); + imageStore((halfTexture4_0), ivec2((uvec2(pos_0))), f16vec4(h2_0, h_0, h_0)); - int index_0 = pos_0.x + pos_0.y * 4; + int index_0 = _S4 + _S3 * 4; ((outputBuffer_0)._data[(uint(index_0))]) = index_0; return; diff --git a/tests/compute/half-texture.slang.hlsl b/tests/compute/half-texture.slang.hlsl index c606703a4..2d04ee17f 100644 --- a/tests/compute/half-texture.slang.hlsl +++ b/tests/compute/half-texture.slang.hlsl @@ -8,19 +8,21 @@ RWStructuredBuffer<int > outputBuffer_0 : register(u0); [shader("compute")][numthreads(4, 4, 1)] void computeMain(uint3 dispatchThreadID_0 : SV_DISPATCHTHREADID) { - int2 pos_0 = (int2) dispatchThreadID_0.xy; + int2 pos_0 = int2(dispatchThreadID_0.xy); float _S1 = 1.00000000000000000000 / 3.00000000000000000000; - int2 pos2_0 = int2(int(3) - pos_0.y, int(3) - pos_0.x); + int _S2 = pos_0.y; + int _S3 = pos_0.x; + int2 pos2_0 = int2(int(3) - _S2, int(3) - _S3); - half h_0 = halfTexture_0[(uint2) pos2_0]; - vector<half,2> h2_0 = halfTexture2_0[(uint2) pos2_0]; - vector<half,4> h4_0 = halfTexture4_0[(uint2) pos2_0]; + half h_0 = halfTexture_0[uint2(pos2_0)]; + vector<half, 2> h2_0 = halfTexture2_0[uint2(pos2_0)]; + vector<half, 4> h4_0 = halfTexture4_0[uint2(pos2_0)]; - halfTexture_0[(uint2) pos_0] = h2_0.x + h2_0.y; - halfTexture2_0[(uint2) pos_0] = h4_0.xy; - halfTexture4_0[(uint2) pos_0] = vector<half,4>(h2_0, h_0, h_0); + halfTexture_0[uint2(pos_0)] = h2_0.x + h2_0.y; + halfTexture2_0[uint2(pos_0)] = h4_0.xy; + halfTexture4_0[uint2(pos_0)] = vector<half, 4>(h2_0, h_0, h_0); - int index_0 = pos_0.x + pos_0.y * int(4); - outputBuffer_0[(uint) index_0] = index_0; + int index_0 = _S3 + _S2 * int(4); + outputBuffer_0[uint(index_0)] = index_0; return; } |