topic This documentation error in Intel® oneAPI Math Kernel Library

Documentation typo for mkl_?omatcopy

Ikuma__Takeshi — Fri, 23 Mar 2018 16:29:23 GMT

I'm reporting documentation inaccuracy for mkl_?omatcopy

https://software.intel.com/en-us/mkl-developer-reference-c-mkl-omatcopy

Definitions of `rows` and `cols` seem to be incorrect (at least for column-major calls)

`rows` and `cols` are documented to be the size of "matrix B (the destination matrix)" but they should be for the matrix A (the source matrix).

`lda` is correctly specified as the rows of A (for column major)

`ldb` is said to "must be at least equal to cols" for ordering = 'C' and trans = 'T'" which is only correct if `rows` and `cols` are defined for A.

I'm only speaking from my debugging experience for the case (ordering = 'C', trans = 'T'") so please validate this is true for all possible combinations.

Thanks,

Kesh Ikuma

This documentation error

John_Young — Fri, 25 Jan 2019 19:57:01 GMT

This documentation error wasted a lot of my time today. It would really help to have this fixed.

My understanding from reading the OpenBlas implementation at

https://github.com/xianyi/OpenBLAS/blob/develop/kernel/arm/omatcopy_ct.c

indicates that (at least for column-major ordering), ROWS and COLS should reference the A matrix size and not the B matrix size.

Could someone please confirm this?

Thanks,

John

There is a similar error in

charles_s_4 — Mon, 04 Feb 2019 21:10:30 GMT

There is a similar error for mkl_sparse_?_create_csc. The documentation is correct online:
https://software.intel.com/en-us/mkl-developer-reference-c-mkl-sparse-create-csc

sparse_status_t mkl_sparse_s_create_csc (sparse_matrix_t *A, sparse_index_base_tindexing, MKL_INT rows, MKL_INT cols, MKL_INT *cols_start, MKL_INT *cols_end, MKL_INT*row_indx, float *values);

The same function per mkl_spblas.h (2019)

sparse_status_t mkl_sparse_s_create_csc( sparse_matrix_t *A,
sparse_index_base_t indexing, /* indexing: C-style or Fortran-style */
MKL_INT rows,
MKL_INT cols,
MKL_INT *rows_start,
MKL_INT *rows_end,
MKL_INT *col_indx,
float *values );

Not a big deal, but it was confusing for a while. It would be worthwhile to check if there are similar errors whenever the CSC format is used.

Thanks.