topic Re: Looking for the best solution in Intel® Moderncode for Parallel Architectures

Looking for the best solution

ekeom — Wed, 08 Jul 2009 13:30:47 GMT

Dear All,

I have one inversion matrix routine for complex non-symmetric and full matrix, based on Gauss-Jordan algorithm and written in Fortran 95. For large matrix (> 2500 x 2500) it takes a lot of time, thus I would like to use parallel or multithreading computation. There so many solution on the market: OpenMP, MPI, multithreading, MultiCore, GPU (it seemed to be fast)... Can someone tell me what is the best portable solution.

Best regards,

Didace

Re: Looking for the best solution

TimP — Wed, 08 Jul 2009 15:13:20 GMT

No, as you don't care to tell why you are doing this, no one will be able to tell you what is best. Are you not aware that libraries such as MKL and ACML support pre-built OpenMP methods?

Re: Looking for the best solution

ekeom — Fri, 10 Jul 2009 05:13:30 GMT

Quoting - tim18

No, as you don't care to tell why you are doing this, no one will be able to tell you what is best. Are you not aware that libraries such as MKL and ACML support pre-built OpenMP methods?

Dear Tim18,

Thank you very much for your answer. I have try MKL, but I was not able to make it run and I don't know if it yes portable. I mean by protable, if I can compile and linke an executable and send it to someone else for use !

Best regards,

Didace

Re: Looking for the best solution

Tudor — Tue, 14 Jul 2009 11:32:24 GMT

If all you are looking for is portability, then your problem lies not in the threading API / library, bur rather in the language. For good portability, try to implement it in Java.