diff --git a/deep_gemm/include/deep_gemm/mma_utils.cuh b/deep_gemm/include/deep_gemm/mma_utils.cuh index bb6284e..0cc554a 100644 --- a/deep_gemm/include/deep_gemm/mma_utils.cuh +++ b/deep_gemm/include/deep_gemm/mma_utils.cuh @@ -867,16 +867,6 @@ struct SM90_64x192x32_F32E4M3E4M3_SS { static constexpr int kNumAccum = M * N / 128; }; -template -struct SM90_U32x1_STSM_N { - __device__ __forceinline__ static void - copy(dtype_t src_0, void* smem_dst) { - const uint32_t src[1] = {*reinterpret_cast(&src_0)}; - asm volatile("stmatrix.sync.aligned.x1.m8n8.shared.b16 [%0], {%1};\n" - :: "l"(smem_dst), "r"(src[0])); - } -}; - template struct SM90_U32x2_STSM_N { __device__ __forceinline__ static void