[CK_TILE] Add FP8xF4 Flatmm (#3401)

* Refactor policy

* fix a bank conflict

* Enable mixed mx flatmm

* Update
This commit is contained in:
Yi DING
2025-12-17 10:01:48 +08:00
committed by GitHub
parent 3dfa794fab
commit 57e1e4a848
9 changed files with 231 additions and 223 deletions

View File

@@ -306,10 +306,9 @@ using WarpGemmMfma_f32_16x16x64_bf8_bf8 = WarpGemmImpl<WarpGemmAttributeMfmaIter
WarpGemmAttributeMfmaImpl_f32_16x16x32_bf8_bf8<WGAttrCtlEnum::Default_>,
2>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp4 = WarpGemmImpl<
WarpGemmAttributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_f8f6f4<pk_fp4_t, pk_fp4_t>,
AttrNumAccess>>;
template <typename A, typename B, WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_f8f6f4 = WarpGemmImpl<
WarpGemmAttributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_f8f6f4<A, B>, AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp8_fp8 = WarpGemmImpl< //
WarpGemmAttributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_f8f6f4<fp8_t, fp8_t>,