Parallel/Distributed Processing and LAPACK

[SEspyder]

I'm looking into creating a C++ numerical algebra library with extentions into pattern recognition and other fields. Right now, I am developing in Microsoft Visual C++ 6.0 and GNU G++ on Solaris and incorporating LAPACK routines. Due to its applications, much processing power is required for real-world tasks. As such, I would like to allow distributed and/or parallel processing. I have these questions I haven't been able to answer:

1. Does LAPACK naturally distribute tasks between processors on a dual-processor or larger-scale parallel processing systems?

2. Right now, I am only familiar with threads (pthreads). Are there any other recommend libraries or methods for developing distributed/parallel applications?

Thanks in advance.

[Julien Langou]

Hello

1. Does LAPACK naturally distribute tasks between processors on a dual-processor or larger-scale parallel processing systems?

no, LAPACK does not naturally distribute tasks between processors. LAPACK is purely sequential.

• but LAPACK can benefit of some parallelism on shared memory machine if you use a multithreaded BLAS ( see http://icl.cs.utk.edu/lapack-forum/viewtopic.php?t=55 )
• ScaLAPACK does distribute tasks between processors (any number of procs is ok)
  (see http://www.netlib.org/scalapack/scalapack_home.html )

2. Right now, I am only familiar with threads (pthreads). Are there any other recommend libraries or methods for developing distributed/parallel applications?

I do not get it. threads are used for shared memory machines. If you want to develop
a distributed applications you will certainly go with processes. A recomended library in this case is certainly any MPI library.