topic Conflicting documentation for P?GEMM descriptor in Intel® oneAPI Math Kernel Library

Conflicting documentation for P?GEMM descriptor

Raghu_Reddy — Fri, 06 Apr 2012 18:48:15 GMT

The general PBLAS documentation indicates that the descriptor should be dimensioned to be size 8:

http://software.intel.com/sites/products/documentation/hpc/composerxe/en-us/mklxe/mkl_manual_win_mac/pbla/pbla_Overview.htm

But in the documentation for PDGEMM, it is described as an array of size 8:
http://software.intel.com/sites/products/documentation/hpc/composerxe/en-us/mklxe/mkl_manual_win_mac/pbla/functn_pgemm.htm#functn_pgemm

Am I misinterpreting the documentation? And this particular case I am interested in a dense matrix.

Thank you!

Conflicting documentation for P?GEMM descriptor

Gennady_F_Intel — Mon, 09 Apr 2012 05:21:19 GMT

thanks for the report - this is looks like a documentation's error. we will check and fix it asap.

thanks you.

--Gennady

Conflicting documentation for P?GEMM descriptor

Ying_H_Intel — Wed, 25 Apr 2012 08:32:32 GMT

In detial,
In http://software.intel.com/sites/products/documentation/hpc/composerxe/en-us/mklxe/mkl_manual_win_mac/pbla/pbla_Overview.htm

and in MKL PBLAS sample,

/* Definition of matrix descriptor */

typedef MKL_INT MDESC[ 9 ];

the desc was defined as 9 dimension

but in MKL reference manual,

under many functions, like pdgemm, it was defined as 8 dimension

desca

(global and local) INTEGER array of dimension 8. The array descriptor of the distributed matrix A.

Correct number is 9. Documentation should be corrected.

Thanks
Ying