Recommended BLAS implementation(s)

[Epy]

The importance in choosing a BLAS implementation has come to my attention recently. Basically most CAE software relies on BLAS and LAPACK to solve problems, and so that choice directly governs how fast you can solve a problem. I also understand that this choice is really non-trivial, as the difference between a generic pre-compiled reference BLAS implementation and an optimized, compiled-for-your-system implementation can speed things up by many times.

As such, I'm trying to find the best BLAS implementation to use. Netlib provides the reference implementations of BLAS and LAPACK, so I assume they should be considered the baseline. From what I've read, it looks like OpenBLAS is the best open-source choice. Intel's MKL may possibly be better, but it's not free for commercial use and it doesn't look like it has an F77 interface (just C and F95).

What is the recommended BLAS implementation for use with Elmer? Am I overstating the importance of BLAS? Any comments would be appreciated. Thanks.

[cemg]

Hi,

if you use Linux you can take a look at ATLAS:

http://math-atlas.sourceforge.net/

Debian and Ubuntu (and all major distros, I suppose) have ATLAS packages. A custom compilation will perform even better (and will let you choose between serial and threaded libs).

Regards,

Cesar