ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 13:11:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix the vector load & fix the gfx950 compv4 error (#2831)

Browse Source
This commit is contained in:
Thomas Ning
2025-09-12 11:48:45 -07:00
committed by GitHub
parent 321627aec5
commit 1894a0dbc3
3 changed files with 35 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
include/ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp
View File

@@ -138,6 +138,7 @@ template<> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float, 16,
using Type = WarpGemmMfma_f32_16x16x128_bf8_fp8<WGAttrNumAccessEnum::Quad>; };
template<> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 128, false, false, false, WGAttrNumAccessEnum::Quad> {
using Type = WarpGemmMfma_f32_16x16x128_bf8_bf8<WGAttrNumAccessEnum::Quad>; };
//WMMA cases
template<bool TransposeC> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 16, 16, 16, TransposeC, false> { using Type =WarpGemmWmma_f32_16x16x16_f8_f8<TransposeC>; };
template<bool TransposeC> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 16, TransposeC, false> { using Type =WarpGemmWmma_f32_16x16x16_bf8_bf8<TransposeC>; };