ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 21:21:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add padding to 1x1Stride1Pad0 conv specialization (grouped conv bwd weight) (#2675)

Browse Source 
Co-authored-by: Bartłomiej Kocot <barkocot@amd.com>
This commit is contained in:
Enrico Degregori
2025-08-14 00:21:09 +02:00
committed by GitHub
parent b963478759
commit a6f4029276
4 changed files with 90 additions and 166 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp
View File

@@ -1299,13 +1299,6 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffleV3
if constexpr(ConvBackwardWeightSpecialization ==
ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
{
// workaround: disable when K, C is even
#if CK_WORKAROUND_DISABLE_FILTER1x1STRIDE1PAD0_WHEN_K_C_IS_EVEN
if(arg.Conv_C_ % 2 == 0 || arg.Conv_K_ % 2 == 0)
{
return false;
}
#endif
// check if it's 1x1, stride=1 pad = 0 conv
for(int i = 0; i < NDimSpatial; i++)
{