topic Thanks for the link mecej4, in Intel® oneAPI Math Kernel Library

LINPACK interface to MKL/LAPACK

j0e — Tue, 29 Dec 2015 19:56:06 GMT

Hi All

I often use older (and sometime new) fortran code that uses LINPACK, so I often replace those calls to LAPACK equivalent routines so that I may use MKL. I have found that this can improve performance in some cases. This also allows easy use of Intel Phi coprocessors for large problems where offloading is warranted.

I was wondering if there exist a set of wrapper routines that in general will allow LINPACK to be replaced by MKL?

thanks,

joe

Linpack was developed in the

mecej4 — Tue, 29 Dec 2015 23:22:14 GMT

Linpack was developed in the 1980s, and was superseded by Lapack, well before Intel became a major compiler vendor. Many Linpack routines do not have an exact Lapack replacement, but here is a guide for doing the conversion: http://www.netlib.org/lapack/lug/node147.html .

The file http://www.netlib.org/linpack/readme contains the statement

 If you are interested in acquiring the entire LINPACK, it may
 make more sense to talk with NAG. NAG distribute the software
 on a mag tape for a nominal charge.

That should give you a good idea of the vintage of Linpack.

Thanks for the link mecej4,

j0e — Tue, 29 Dec 2015 23:34:37 GMT

Thanks for the link mecej4, it is useful.

I'm well aware of the vintage of LINPACK, as I was using it back in the 80's, lol.

Of course, there is still a lot of computational software, such as DAE solvers, benchmarks and the like, that still make extensive use of LINPACK. Fortunately, the LINPACK routines I would like to replace have LAPACK counterparts. I suspect someone else out there has written wrappers for all the LINPACK routines that can be mapped to LAPACK, so I was just scrounging. I have written several myself, but they are sloppy.

cheers

Joe, your comments touch upon

mecej4 — Tue, 29 Dec 2015 23:54:06 GMT

Joe, your comments touch upon an issue that has been of some concern. Netlib is not as well maintained as it used to be. Some of the e-mail links of moderators at netlib.sandia.gov are defunct, and it appears doubtful if will remain possible to submit bug reports and have corrections made in a reasonable period. It is a pity, because there is no other single site with the same breadth of coverage of NA topics as Ma Netlib.