ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-17 09:08:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

remove single rate mfma restriction for f8 blockscale gemm

Browse Source
This commit is contained in:
aska-0096
2025-05-29 10:13:50 +00:00
parent 33085c8458
commit 3c24d690a1
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_d_ab_scale.hpp
View File

@@ -153,9 +153,7 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
static constexpr bool is_single_rate_mfma =
(((is_same<ComputeTypeA, half_t>::value || is_same<ComputeTypeA, bhalf_t>::value) &&
lcm_AK1_BK1 <= 4) ||
(is_same<ComputeTypeA, int8_t>::value && lcm_AK1_BK1 <= 8) ||
((is_same<ComputeTypeA, f8_t>::value || is_same<ComputeTypeA, bf8_t>::value) &&
lcm_AK1_BK1 < 32))
(is_same<ComputeTypeA, int8_t>::value && lcm_AK1_BK1 <= 8))
? true
: false;
static constexpr auto is_scale_mfma = false;