ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Fix flatmm on gfx11 and gfx12 (#2790)

Browse Source 
1. Correct shuffle_b and MakeBFlatDramTileDistribution according to WMMA warp layout
2. Add FlatmmConfig16_Wmma for gfx11 and gfx12
This commit is contained in:
linqunAMD
2025-09-10 08:28:00 +08:00
committed by GitHub
parent 82890192dd
commit df4ee556d6
14 changed files with 224 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
include/ck_tile/ops/flatmm/kernel/flatmm_kernel.hpp
View File

@@ -127,7 +127,10 @@ struct FlatmmKernel
return dim3(TilePartitioner::GridSize(M, N), 1, KBatch);
}
CK_TILE_HOST static constexpr auto BlockSize() { return dim3(kBlockSize); }
CK_TILE_HOST static constexpr auto BlockSize()
{
return is_wave32() ? dim3(kBlockSize / 2) : dim3(kBlockSize);
}
CK_TILE_HOST static constexpr KernelArgs
MakeKernelArgs(const FlatmmHostArgs<NumDTensor>& hostArgs)