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

MX FP GEMM - Test MX FP8 MFMA Instructions (#1902)

Browse Source 
* Refactored `load_A_row_major` to follow scale mapping

* Refactored `load_A_col_major` to follow scale mapping

* Refactored `load_B_col_major` to follow scale mapping

* Verified non-scaled test

* Verified scaled tests

* Used ReferenceMXGemm for verification

* Updated license headers
This commit is contained in:
Andriy Roshchenko
2025-02-21 13:35:54 -07:00
committed by GitHub
parent 3ace125c30
commit ffa13455a2
4 changed files with 897 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
View File

@@ -1,5 +1,5 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
@@ -780,7 +780,6 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32bf8f8>
}
};
// TODO: fix mfma...f8f6f4 instructions
template <>
struct mfma_type<MfmaInstr::mfma_f32_32x32x64f8f6f4>
{
@@ -847,9 +846,14 @@ struct mfma_type<MfmaInstr::mfma_scale_f32_32x32x64f8f6f4>
// clang-format on
template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
__device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
__device__ void run(const FloatA& a,
const int32_t scale_a,
const FloatB& b,
const int32_t scale_b,
FloatC& reg_c) const
{
intrin_mfma_scale_f32_32x32x64f8f6f4<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
intrin_mfma_scale_f32_32x32x64f8f6f4<MPerXdlops, NPerXdlops>::Run(
a, scale_a, b, scale_b, reg_c);
}
};
@@ -871,9 +875,14 @@ struct mfma_type<MfmaInstr::mfma_scale_f32_16x16x128f8f6f4>
// clang-format on
template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
__device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
__device__ void run(const FloatA& a,
const int32_t scale_a,
const FloatB& b,
const int32_t scale_b,
FloatC& reg_c) const
{
intrin_mfma_scale_f32_16x16x128f8f6f4<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
intrin_mfma_scale_f32_16x16x128f8f6f4<MPerXdlops, NPerXdlops>::Run(
a, scale_a, b, scale_b, reg_c);
}
};