[vsipl++] 1.3 Configuration Issues on OpenSUSE 10.2

[Jules Bergmann]

Brad,

Sourcery VSIPL++ incorporates ATLAS version 3.7.11, which is a bit dated 
(circa 2005).  How that compares to a stock blas/lapack library will 
depend quite a bit on what "stock" is:

  * Comparing against a "stock" ATLAS, it will depend on your
    architectures.  For architectures not supported in 3.7.11
    (with a predefined config file), a newer version of ATLAS
    will probably do much better.  You can configure VSIPL++
    to use a pre-existing ATLAS.

  * Comparing against a vendor lapack/blas, such as Intel's MKL,
    IBM's ESSL, or AMD's ACML, my suspicion is the vendor libraries
    will do better.  Again, you can configure VSIPL++ to use these
    libraries if you have them.

  * Comparing against the reference implementation of lapack/blas
    from netlib.org, the builtin ATLAS will win hands down!

I'm sorry that you encountered problems trying to build VSIPL++ for 
OpenSUSE.  However, it sounds like you were able to work around them. 
Would you mind contributing your patch for inclusion in future versions 
of VSIPL++?

I'm not sure what version of ATLAS OpenSUSE ships with, but if it is 
more recent than 3.7.11, it may be worth installing and using.  Let me 
know if you have any questions configuring VSIPL++ to use an external ATLAS.

				thanks,
				-- Jules

Brad Howes wrote:
> On a box running OpenSUSE 10.2, I had some problems configuring vsipl++ 
> 1.3 to use the blas and lapack libraries present on the system. In the 
> 'configure' script, the code that checks for g2c is somewhat broken, 
> missing double-quotes around some variable expansions, and I could not 
> get the test cases to run unless I linked in -lgfortran by manually 
> adding it to the proper action line in the 'configure' file. After those 
> minor changes, I was able to properly configure using 
> --with-lapack=generic. I can post a diff if desired.
> 
> As an aside, does anyone have any information on how stock blas/lapack 
> installs perform compared to the vsipl++ builtin version?
> 
> Thanks!
> 
> Brad
> 
> -- 
> Brad Howes
> Group 42
> MIT Lincoln Laboratory • 244 Wood St. • Lexington, MA 02173
> Phone: 781.981.5292 • Fax: 781.981.3495 • Secretary: 781.981.7420
> 
> 
> 
> 


-- 
Jules Bergmann
CodeSourcery
jules at codesourcery.com
(650) 331-3385 x705