mfma_32x32x64_fp8/bf8 (#2148)

* support for mfma_32x32x64_fp8

* clang-formatted

* Fixing sparsity in codegen
This commit is contained in:
Khushbu Agarwal
2025-05-01 13:36:24 -07:00
committed by GitHub
parent 619fba3134
commit d58f2b8bd0
4 changed files with 147 additions and 22 deletions

View File

@@ -228,6 +228,18 @@ using WarpGemmMfma_f32_16x16x128_bf8_fp8 = WarpGemmImpl<WarpGemmAtrributeMfma<
using WarpGemmMfma_f32_16x16x128_bf8_bf8 = WarpGemmImpl<WarpGemmAtrributeMfma<
WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_bf8<WGAttrCtlEnum::Default_>>>;
using WarpGemmMfma_f32_32x32x64_fp8_fp8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_32x32x64_fp8_fp8<WGAttrCtlEnum::Default_>>>;
using WarpGemmMfma_f32_32x32x64_fp8_bf8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_32x32x64_fp8_bf8<WGAttrCtlEnum::Default_>>>;
using WarpGemmMfma_f32_32x32x64_bf8_fp8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_32x32x64_bf8_fp8<WGAttrCtlEnum::Default_>>>;
using WarpGemmMfma_f32_32x32x64_bf8_bf8 = WarpGemmImpl<
WarpGemmAtrributeMfma<WarpGemmAttributeMfmaImpl_f32_32x32x64_bf8_bf8<WGAttrCtlEnum::Default_>>>;
using WarpGemmMfma_f32_32x32x16_fp8_fp8_CTransposed =
WarpGemmImpl<WarpGemmAtrributeMfmaTransposedCDistribution<
WarpGemmAttributeMfmaImpl_f32_32x32x16_fp8_fp8<WGAttrCtlEnum::Default_>>>;