add limit for CDE0BlockTransferSrcScalarPerVector

ea2c27ca · ltqin · fc50b641 · ea2c27ca
Commit ea2c27ca authored Jul 12, 2023 by ltqin
Hide whitespace changes
Inline Side-by-side

Showing with 4 additions and 0 deletions

include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp ...tched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp +4 -0

No files found.
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
@@ -734,6 +734,10 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
        const auto wave_id     = GetGemm0WaveIdx();
        const auto wave_m_n_id = GetGemm0WaveMNIdx(wave_id[I2]); // I2: 0~63

+        static_assert(CDE0BlockTransferSrcScalarPerVector <= n4,
+                      "vector load must be not greater than n4");
+        static_assert(n4 % CDE0BlockTransferSrcScalarPerVector == 0);
+
        auto d0s_threadwise_copy = generate_tuple(
            [&](auto i) {
                return ThreadwiseTensorSliceTransfer_v2<