gemm_internal_cublaslt Class — pytorch Architecture

Source Code

aten/src/ATen/cuda/CUDABlas.cpp lines 1289–1295

template <typename Dtype, typename C_Dtype = Dtype>
inline void gemm_internal_cublaslt(CUDABLAS_GEMM_ARGTYPES_AND_C_DTYPE(Dtype, C_Dtype)) {
  // forward to bgemm implementation but set strides and batches to 0
  if (!bgemm_internal_cublaslt(transa, transb, m, n, k, alpha, a, lda, 0, b, ldb, 0, beta, c, ldc, 0, 0)) {
    gemm_internal_cublas(CUDABLAS_GEMM_ARGS(Dtype));
  }
}