Skip to content

GitLab

Commit 762cbddf authored by danyao12's avatar danyao12
Browse files

fix shuffle GemmORepeat

parent 17774771
Hide whitespace changes
Inline Side-by-side
Showing with 2 additions and 2 deletions
include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_pt3.hpp
View file @ 762cbddf
...@@ -2164,7 +2164,7 @@ struct GridwiseBatchedMultiheadAttentionBackward_Xdl_CShuffle_V1 ...@@ -2164,7 +2164,7 @@ struct GridwiseBatchedMultiheadAttentionBackward_Xdl_CShuffle_V1
// shuffle dK&dV and write // shuffle dK&dV and write
{ {
static_assert(Gemm2Params_N_O_M::GemmNRepeat % CShuffleMXdlPerWavePerShuffle == 0 && static_assert(Gemm2Params_N_O_M::GemmNRepeat % CShuffleMXdlPerWavePerShuffle == 0 &&
Gemm2Params_N_O_M::GemmNRepeat % CShuffleNXdlPerWavePerShuffle == 0, Gemm2Params_N_O_M::GemmORepeat % CShuffleNXdlPerWavePerShuffle == 0,
"wrong!"); "wrong!");
constexpr index_t MWave = Gemm2Params_N_O_M::GemmNWave; constexpr index_t MWave = Gemm2Params_N_O_M::GemmNWave;
...@@ -2363,7 +2363,7 @@ struct GridwiseBatchedMultiheadAttentionBackward_Xdl_CShuffle_V1 ...@@ -2363,7 +2363,7 @@ struct GridwiseBatchedMultiheadAttentionBackward_Xdl_CShuffle_V1
// space filling curve for threadwise C in VGPR // space filling curve for threadwise C in VGPR
constexpr auto sfc_c_vgpr = SpaceFillingCurve<Sequence<Gemm2Params_N_O_M::GemmNRepeat, constexpr auto sfc_c_vgpr = SpaceFillingCurve<Sequence<Gemm2Params_N_O_M::GemmNRepeat,
Gemm2Params_N_O_M::GemmNRepeat, Gemm2Params_N_O_M::GemmORepeat,
1, 1,
1, 1,
1, 1,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment