Fix unmatched K size of WarpGemmMfmaBf16Bf16F32M16N16K32TransposedCDistribution on gfx950 (#2393)

This commit is contained in:
Yi DING
2025-06-25 07:35:54 +08:00
committed by GitHub
parent 778ac24376
commit c5d9181e1b

View File

@@ -172,7 +172,7 @@ using WarpGemmMfmaBf16Bf16F32M32N32K16TransposedCDistribution =
#if defined(__gfx950__)
using WarpGemmMfmaBf16Bf16F32M16N16K32TransposedCDistribution =
WarpGemmImpl<WarpGemmAtrributeMfmaTransposedCDistribution<
WarpGemmAttributeMfmaImplBf16Bf16F32M16N16K16<WGAttrCtlEnum::Default_>>>;
WarpGemmAttributeMfmaImplBf16Bf16F32M16N16K32<WGAttrCtlEnum::Default_>>>;
#else
using WarpGemmMfmaBf16Bf16F32M16N16K32TransposedCDistribution =
WarpGemmImpl<WarpGemmAtrributeMfmaIterateKAndTransposedCDistribution<