glingyan commented on a change in pull request #14492: use mkl sparse matrix to improve performance URL: https://github.com/apache/incubator-mxnet/pull/14492#discussion_r273369361
########## File path: src/operator/tensor/dot-inl.h ########## @@ -775,13 +777,35 @@ inline void DotCsrDnsDnsImpl(const OpContext& ctx, } using nnvm::dim_t; - +#if (MSHADOW_USE_MKL == 1) + TShape lhs_shape = lhs.shape(); + TShape rhs_shape = rhs.shape_; +#endif const TBlob data_l = lhs.data(); const TBlob indptr_l = lhs.aux_data(csr::kIndPtr); const TBlob col_idx_l = lhs.aux_data(csr::kIdx); const TBlob& data_r = rhs; const TBlob data_out = *ret; +#if (MSHADOW_USE_MKL == 1) + if (data_l.type_flag_ == mshadow::kFloat32 Review comment: MKL is the same with mxnet https://software.intel.com/en-us/mkl-developer-reference-c-sparse-blas-csr-matrix-storage-format it do not allow duplicate entries & expect the col indices of any input/output csr matrices to be sorted， Element i of the integer array columns is the number of the column in A that contains the i-th value in the values array. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
