ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 14:59:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Add FP8xF4 Flatmm (#3401)

Browse Source 
* Refactor policy

* fix a bank conflict

* Enable mixed mx flatmm

* Update
This commit is contained in:
Yi DING
2025-12-17 10:01:48 +08:00
committed by GitHub
parent 3dfa794fab
commit 57e1e4a848
9 changed files with 231 additions and 223 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
include/ck_tile/ops/flatmm/kernel/mx_flatmm_kernel.hpp
View File

@@ -414,12 +414,7 @@ struct MXFlatmmKernel : FlatmmKernel<TilePartitioner_, MXFlatmmPipeline_, Epilog
(ScaleM::GranularityMN != -1 && ScaleM::GranularityK == 0) || // per token
(ScaleN::GranularityMN != -1 && ScaleN::GranularityK == 0); // per channel
auto a_block_window_with_distr =
ck_tile::make_tile_window(a_block_window.get_bottom_tensor_view(),
a_block_window.get_window_lengths(),
a_block_window.get_window_origin(),
MXFlatmmPipeline::GetADramTileDistribution());
const auto& c_block_tile = MXFlatmmPipeline{}(a_block_window_with_distr,
const auto& c_block_tile = MXFlatmmPipeline{}(a_block_window,
b_flat_block_window,
scale_a_block_window,
scale_b_block_window,