Fix the GFX 950 Universal GEMM (#2597)

* 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

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<