[Clp] Do I need Lapack/BLAS

[John Forrest]

Haroldo,

Not a perfect answer ....

If Lapack is built in then you can see if it is activated on a
particular problem by setting the log level high enough.  So - 

clp nug12.mps -log 3 -duals  (nug12 from miplib)
gives output such as -

0  Obj 0 Primal inf 24 (24)
        length of U 2383, length of L 202
138  Obj 8.1224 Primal inf 593.6 (385)
        length of U 2653, length of L 357
276  Obj 24.1951 Primal inf 682.15 (485)
        length of U 2931, length of L 545
414  Obj 40.1535 Primal inf 637.5 (496)
        length of U 3352, length of L 706
and later
88658  Obj 522.894 Dual inf 1.07841 (275)
    length of U 17298, length of L 15214 plus 80089 from 283 dense rows
88666  Obj 522.894 Dual inf 2.27984 (293)
    length of U 17538, length of L 15454 plus 80656 from 284 dense rows
88683  Obj 522.894 Dual inf 5.77515 (324)
    length of U 18382, length of L 16063 plus 79524 from 282 dense rows
88700  Obj 522.894

which enables you to see that Lapack is being used on dense section.

You can also see that it is not used on dfl001.mps from netlib, but is
used on pilot87.

John Forrest 



On Tue, 2010-11-30 at 23:00 -0200, Haroldo Santos wrote:
> Dear All,
> 
> Some time ago I asked what differences could have a CLP build
> including Lapack from another without.
> I was informed that Lapack would allow a better performance in the
> case of having some dense matrix.
> 
> Since Lapack increases the build size/dependencies, I would like to
> know if there is an easy way to determine if it is being activated for
> a given linear program (e.g., by looking at the terminal output).
> 
> Anyone has benchmarks indicating how much does it helps ?
> In common benchmarks, for instance:
> http://plato.asu.edu/ftp/milpf.html
> It is activated ? Would it help in these problems ?
> 
> Thanks,
> 
> Haroldo
> 
> -- 
> =============================================================
> Haroldo Gambini Santos
> Computing Department - Universidade Federal de Ouro Preto - UFOP
> email: haroldo [at ] iceb.ufop.br
> home/research page: http://www.iceb.ufop.br/decom/prof/haroldo/
>  
> "Computer science is no more about computers than astronomy 
> is about telescopes." Edsger Dijkstra
>  
> _______________________________________________
> Clp mailing list
> Clp at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/clp