ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-05 14:11:29 +00:00
Code Issues Packages Projects Releases Wiki Activity

Improve XDL to WMMA porting for grouped conv fwd (#3456)

Browse Source 
Refactors the way the number of XDL (matrix multiply-accumulate) instructions per wave is calculated and used in the grouped convolution forward implementations, especially to better support WMMA (Wave Matrix Multiply-Accumulate) instructions and 16x16 tiles. 
The changes use MXdlPerWave instead of NXdlPerWave to increase number of waves per M dim.
This commit is contained in:
Bartłomiej Kocot
2025-12-19 23:58:51 +01:00
committed by GitHub
parent 2d9c962e2c
commit cbc8335964
13 changed files with 226 additions and 133 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
experimental/builder/test/utils/ckb_conv_test_configs.hpp
View File

@@ -68,7 +68,7 @@ constexpr TransferABC FwdTransfer_4x64x1{
{.m_block = 1, .m_wave_per_xdl = 32, .n_block = 1, .n_wave_per_xdl = 8},
.epilogue = {.m_xdl_per_wave_per_shuffle = 1,
.n_per_wave_per_shuffle = 1,
.scalar_per_vector = 8},
.scalar_per_vector = 4},
},
};