ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-11 17:51:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add new instances for merging multiple fwd conv groups into a single GEMM batch. Allow group merging for C > 1 when vector load/store size is 1 for the output tensor.

Browse Source
This commit is contained in:
Ville Pietilä
2026-01-23 06:07:02 -05:00
parent 2e08a7e5ab
commit f88efddea4
2 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp
View File

@@ -1513,7 +1513,7 @@ struct DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle
if constexpr(NumGroupsToMerge > 1)
{
if(!(C == 1))
if(!(C == 1) && CDEBlockTransferScalarPerVector_NPerBlock > 1)
{
return false;
}