This is the mail archive of the cygwin@sourceware.cygnus.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: optimization



> "Reza Habib" <reza@psych.utoronto.ca> writes:
[snip]
> > is windows).  I'm in the process of writing some statistical tools for
> > image analysis.  My data are in floating point and I will mainly be
> > conducting matrix algebra type operations.  I'm trying to pick the
> > compiler which will produce the fastest,  most efficient code.  Thanks.
> 
> For floating point intensive code, MSVC is hard to beat at this time.
> However, for majority of the `real-life' numerical code, it doesn't matter
> all that much as long as the compiler codegen performance is reasonably
> close.

I wonder if anybody has used to Intel Math Kernel Libraries with the 
egcs compiler? It consists of BLAS level 1 to 3 and some LAPACK 
functions, but egcs is not supported. I have seen a very big increase 
in computational performance using this library with the MSVC 
compiler, but we should be able to get the same performance jump when 
using egcs. No matter what compiler you use you should consider this 
library for matrix computations (provided you're planning to run the 
code on a Intel processor).

Have anybody tested this?

  Regards

    Lars Gregersen

--------------------------------------------------------------------- 
Lars Gregersen                         Telephone no.: +45 45 25 28 08
Technical University of Denmark        Fax no.      : +45 45 88 22 58
Department of Chemical Engineering     E-mail       : lg@kt.dtu.dk
Bygning 229
DK-2800 Lyngby
Denmark
---------------------------------------------------------------------
Homepage: http://www.gbar.dtu.dk/~matlg/

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]