ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

MX GEMM - Add FP8 GEMM Tests for Different Layouts (#2152)

Browse Source 
* Add gemm_mx_fp8_bf8 example with row-major B

* Add more overloads of MX MFMA instructions

* Add MK_KN (RRR) tests

* Add KM_NK (CCR) tests

* Add more problem sizes to Large tests

* Add test_gemm_mx to the list of regression tests
This commit is contained in:
Andriy Roshchenko
2025-05-01 11:55:48 -06:00
committed by GitHub
parent b9d17bdb11
commit 79b0bfeb41
15 changed files with 642 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
View File

@@ -1129,6 +1129,12 @@ struct MfmaSelector
return MfmaInstr::mfma_scale_f32_32x32x64f8f6f4;
}
template <>
constexpr auto GetMfma<bf8_t, 32, 32, f8_t, false, true>()
{
return MfmaInstr::mfma_scale_f32_32x32x64f8f6f4;
}
template <>
constexpr auto GetMfma<f8_t, 16, 16>()
{
@@ -1147,6 +1153,18 @@ struct MfmaSelector
return MfmaInstr::mfma_scale_f32_16x16x128f8f6f4;
}
template <>
constexpr auto GetMfma<f8_t, 16, 16, bf8_t, false, true>()
{
return MfmaInstr::mfma_scale_f32_16x16x128f8f6f4;
}
template <>
constexpr auto GetMfma<bf8_t, 16, 16, f8_t, false, true>()
{
return MfmaInstr::mfma_scale_f32_16x16x128f8f6f4;
}
template <>
constexpr auto GetMfma<bf8_t, 32, 32>()
{