use 3 prefetch

fe5fe8a2 · wangshaojie6 · 1f1ae5bb · fe5fe8a2
Commit fe5fe8a2 authored Jun 16, 2022 by wangshaojie6
Hide whitespace changes
Inline Side-by-side

Showing with 4 additions and 2 deletions

include/ck/tensor_operation/gpu/device/device_gemm_xdl_splitk_c_shuffle.hpp ...operation/gpu/device/device_gemm_xdl_splitk_c_shuffle.hpp +4 -2

No files found.
--- a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -246,7 +246,8 @@ struct DeviceGemmXdlSplitKCShuffle
        CShuffleNRepeatPerShuffle,
        CBlockTransferScalarPerVector_NWaveNPerXDL,
        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-        true>;
+        true, 
+        3>;

    // GridwiseGemm
    using GridwiseGemmAtomicAdd = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
@@ -289,7 +290,8 @@ struct DeviceGemmXdlSplitKCShuffle
        CShuffleNRepeatPerShuffle,
        CBlockTransferScalarPerVector_NWaveNPerXDL,
        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-        true>;
+        true,
+        3>;

    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
        decltype(GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}));