LAPACK Archives

[Lapack] Problems about BLAS

Dear Jerry,
the BLAS library that you can download from netlib is just a reference that
basically defines the routine interfaces and the correct behavior. Thus, not
only it is single-threaded but also very inefficient. Open source, fast,
multithreaded BLAS libraries exist but are rather complicated to understand.
You can look at ATLAS or GotoBLAS. Also, the PLASMA project is a good
example of multithreaded numerical library but it implements LAPACK
operations, i.e., higher level operations that those in BLAS.
Good luck

Alfredo



2009/3/24 ?? <jerry.z.mystery@Domain.Removed>

Dear Sir,
         I am a student now using the lapack for my study, something about
performance  analyzing of Lapack on SMP.  But it made me confused when I
tested the BLAS library.

        The BLAS contained in the tar ball (from netlib) seemed to be
single-threaded. I tested them on both Intel Xeon and Intel Pentium, but
found no significant difference in timing.

        I know there are some  tuned BLAS libraries like MKL or ACML, but
they are not open source products while I need the source code to do my
study.  I also read some worknotes about the ScaLapack, but it says
Scalapack is for distributed memory system,  and I'm not going to deal with
the cluster or MPP issues currently.

        Now I have no idea  where I can find opensource BLAS, which is
implemented to run parallelly on SMP structure, could you please give me
some information?

        many thanks


--------------------------------
 Jerry.Zhai

 From China

_______________________________________________
Lapack mailing list
Lapack@Domain.Removed
http://lists.cs.utk.edu/listinfo/lapack




-- 
-----------------------------------------
Alfredo Buttari, PhD
CNRS-IRIT
2 rue Camichel, 31071 Toulouse, France
http://buttari.perso.enseeiht.fr
-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://lists.cs.utk.edu/private/lapack/attachments/20090324/0fbaaf69/attachment.htm
 

<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or