ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 20:51:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix the GFX 950 Universal GEMM (#2597)

Browse Source 
* solve the gfx950 error

* clang format

* fix a typo error

---------

Co-authored-by: ThomasNing <thomasning@amd.com>
This commit is contained in:
Thomas Ning
2025-08-01 09:32:24 -07:00
committed by GitHub
parent e6104daecc
commit 7c44a763fa
3 changed files with 30 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
include/ck_tile/ops/gemm/warp/warp_gemm.hpp
View File

@@ -265,17 +265,25 @@ using WarpGemmMfma_f32_16x16x64_bf8_bf8 = WarpGemmImpl<WarpGemmAtrributeMfmaIter
WarpGemmAttributeMfmaImpl_f32_16x16x32_bf8_bf8<WGAttrCtlEnum::Default_>,
2>>;
using WarpGemmMfma_f32_16x16x128_fp8_fp8 = WarpGemmImpl<WarpGemmAtrributeMfma<
WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_fp8<WGAttrCtlEnum::Default_>>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp8_fp8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_fp8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
using WarpGemmMfma_f32_16x16x128_fp8_bf8 = WarpGemmImpl<WarpGemmAtrributeMfma<
WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_bf8<WGAttrCtlEnum::Default_>>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp8_bf8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_bf8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
using WarpGemmMfma_f32_16x16x128_bf8_fp8 = WarpGemmImpl<WarpGemmAtrributeMfma<
WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_fp8<WGAttrCtlEnum::Default_>>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_bf8_fp8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_fp8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
using WarpGemmMfma_f32_16x16x128_bf8_bf8 = WarpGemmImpl<WarpGemmAtrributeMfma<
WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_bf8<WGAttrCtlEnum::Default_>>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_bf8_bf8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_bf8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_32x32x64_fp8_fp8 = WarpGemmImpl<

16
include/ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp
View File

@@ -92,10 +92,10 @@ template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float,
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 64, false> { using Type = WarpGemmMfma_f32_16x16x64_bf8_bf8; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 32, 32, 16, true> { using Type = WarpGemmMfma_f32_32x32x16_bf8_bf8_CTransposed; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_fp8_fp8; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_fp8_bf8; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_bf8_fp8; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_bf8_bf8; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_fp8_fp8<>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_fp8_bf8<>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_bf8_fp8<>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_bf8_bf8<>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 32, 32, 64, false> { using Type = WarpGemmMfma_f32_32x32x64_fp8_fp8<>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 32, 32, 64, false> { using Type = WarpGemmMfma_f32_32x32x64_fp8_bf8<>; };
@@ -110,6 +110,14 @@ template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float,
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 32, 32, 64, false, false, false, WGAttrNumAccessEnum::Quad> {
using Type = WarpGemmMfma_f32_32x32x64_bf8_bf8<WGAttrNumAccessEnum::Quad>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 16, 16, 128, false, false, false, WGAttrNumAccessEnum::Quad> {
using Type = WarpGemmMfma_f32_16x16x128_fp8_fp8<WGAttrNumAccessEnum::Quad>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 16, 16, 128, false, false, false, WGAttrNumAccessEnum::Quad> {
using Type = WarpGemmMfma_f32_16x16x128_fp8_bf8<WGAttrNumAccessEnum::Quad>; };
template<> struct WarpGemmMfmaDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float, 16, 16, 128, false, false, false, WGAttrNumAccessEnum::Quad> {
using Type = WarpGemmMfma_f32_16x16x128_bf8_fp8<WGAttrNumAccessEnum::Quad>; };
template<> struct WarpGemmMfmaDispatcher<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>; };
// int8
// ADataType, BDataType, AccDataType, MPerWave, NPerWave, KPerWave, TransposeC, SwizzleA, UseStructuredSparsity
template<> struct WarpGemmMfmaDispatcher<ck_tile::int8_t, ck_tile::int8_t, ck_tile::int32_t, 32, 32, 16, false> { using Type = WarpGemmMfma_i32_32x32x16_i8_i8; };