Merge branch 'attn-bwd-combined-test' of...

Merge branch 'attn-bwd-combined-test' of https://github.com/ROCmSoftwarePlatform/composable_kernel into attn-bwd-combined-test

Merge branch 'attn-bwd-combined-test' of...
Merge branch 'attn-bwd-combined-test' of https://github.com/ROCmSoftwarePlatform/composable_kernel into attn-bwd-combined-test
37433ee3 · danyao12 · 326e9ee8 · 2e1f2c37 · 37433ee3
Commit 37433ee3 authored Mar 02, 2023 by danyao12
Show whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_v2.hpp ..._batched_multihead_attention_backward_xdl_cshuffle_v2.hpp +1 -1

No files found.
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_v2.hpp
@@ -908,7 +908,7 @@ struct GridwiseBatchedMultiheadAttentionBackward_Xdl_CShuffle_V2
    template <index_t BlockSize_, index_t BlockSliceLength_M_, index_t BlockSliceLength_O_>
    struct YDotYGrad_M_O_
    {
-        static constexpr index_t SrcScalarPerVector = 16 / sizeof(FloatGemmAcc);
+        static constexpr index_t SrcScalarPerVector = 16 / sizeof(DataType);
        static constexpr auto ThreadClusterLength_O =
            Number<BlockSliceLength_O_ / SrcScalarPerVector>{};
        static constexpr auto ThreadClusterLength_M = Number<BlockSize_ / ThreadClusterLength_O>{};