[CK_TILE] Add fp4 warp gemm 16x16x128 (#2738)

* first commit

* fix format error

* fix vec size error

* fix clang format

* fix type error

* add interface in warp_gemm_impl

* fix interface

* fix bug

* fix bug

---------

Co-authored-by: asleepzzz <hanwen.chang@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
This commit is contained in:
Gino Lu
2025-10-24 01:55:51 +08:00
committed by GitHub
parent 6df69abeef
commit bedade2572
5 changed files with 183 additions and 0 deletions

View File

@@ -300,6 +300,10 @@ using WarpGemmMfma_f32_16x16x64_bf8_bf8 = WarpGemmImpl<WarpGemmAttributeMfmaIter
2>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp4 = WarpGemmImpl<
WarpGemmAttributeMfma<WarpGemmAttributeMfmaScaleImpl_f32_16x16x128_fp4<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp8_fp8 = WarpGemmImpl<
WarpGemmAttributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_fp8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;