topic MKL_DCOOMM giving incorrect answers in Intel® oneAPI Math Kernel Library

MKL_DCOOMM giving incorrect answers

Vineet_Y_ — Thu, 31 Jan 2013 19:29:00 GMT

composer_xe_2011_sp1.7.256 and MKL_DCOOMM

In a test program attached as a file MKL_DCOOMM and DGEMM give different answers. The answer given by DGEMM is correct. I would be greatly obliged if anybody can guide me in finding the reason behind this inconsistency.

There is a problem with

mecej4 — Thu, 31 Jan 2013 20:34:44 GMT

There is a problem with passing INTEGER*8 arrays to library routines that expect INTEGER*4 arrays. No such problem occurs with scalar integer arguments as long as the variables do not need more than 32 bits to represent them.

Change INTEGER*4 to INTEGER in your program, and things will work correctly. If you really need to use INTEGER*8 arrays (for some reason that you have not so far explained), you will have to use a version of MKL that supports them (e.g., for X64 targets using the ILP64 libraries).

Either change INTEGER*8 to

Zhang_Z_Intel — Thu, 31 Jan 2013 23:08:16 GMT

Either change INTEGER*8 to INTEGER in your code, or turn on the ILP64 support in MKL. Your program will work fine.

The easiest way to do (in this particular case) is to pass '-i8' option to ifort. For example, ifort -i8 -mkl mkl-dcoomm.f90

Many Thanks

Vineet_Y_ — Fri, 08 Feb 2013 17:20:44 GMT

Many Thanks