[Coin-discuss] using CLP barrier with OSI
Kish Shen
kish.shen at crosscoreop.com
Fri Dec 1 19:56:31 EST 2006
Hi John,
Thanks for your reply!
On Friday 01 December 2006 09:37, John J Forrest wrote:
> Kish,
>
> If it linked then it should be OK - the fact that you got errors meant it
> was trying to link in code. You can double check with stand-alone version
> i.e. try
>
> clp test.mps -barrier and
> clp test.mps -cholesky uni -barrier
>
> and you should see improved ordering
>
In fact I have already successfully tested the new ClpInterior with my own code, and it
does run significantly faster with the Ufl code. Thanks!
What I was worried about is that I did not insert the flags into the right place, or miss other
places where they should be inserted.
I could not find a test.mps file, so I could not run the test you suggested. I thought this might
be generated by `make test', which I didn't do, but when I tried this, I got a compiler error,
because the compilation of some code did not have the -lamd etc. flags, so obviously my hacks
were not generic enough...but I don't really need this, as all I need is a working libClp.a/.so file.
I decided to try out the CLP_USE_CHOLMOD. Would you recommend this? Does this give any
further performance improvements?
On a related topic: should the ADD_CXXFLAGS (and other similar ADD_*) flags be added
officially to the Makefile for Clp etc.? On another thread (about my problems compiling COIN
projects on x86_64), Andreas Waechter suggest that ADD_CXXFLAGS etc. should be the way
user pass their own flags to compile the COIN projects, so I guess they should be included
in the compiling of all code (and will be empty unless redefined by the user).
Kish Shen
> John Forrest
>
>
>
> Kish Shen
> <K.Shen at crosscore
> op.com> To
> Sent by: Discussions about open source
> coin-discuss-boun software for Operations Research
> ces at list.coin-or. <coin-discuss at list.coin-or.org>
> org cc
>
> Subject
> 11/30/06 11:54 PM Re: [Coin-discuss] using CLP
> barrier with OSI
>
> Please respond
> to
> about open source
> software for
> Operations
> Research
> <coin-discuss at lis
> t.coin-or.org>
>
>
>
>
>
>
> John J Forrest wrote:
> you do not use a
> > third-party ordering and factorization code. One from the University
> > of Florida is suitable. If I get time next week I will try and make
> > it easier to use with the svn version of Coin - It was working in the
> > old cvs version. This what you have to do - add to your configure
> ADD_CXXFLAGS="-DUFL_BARRIER -I(directory with amd.h) -L(directory with
> > libamd.a) -lamd
> for compilation.
> suggested here, but
> have run into some
> Clp directory
> the following way:
>
> 1. added ADD_CXXFLAGS to CXXCOMPILE and LTCXXCOMPILE, which is used to
> compile the code
> after
> libClp.la. This is needed to linking.
> correctly, but I got the code
> compiled.Should I have added it somewhere else?
>
> Kish Shen
> but you have to add an extra -lamd in
> > Makefile (before -lm). As I said I will try and make prettier next
> week.
> > might work well with a problem of the shape you describe. If you then
> > want to convert to Osi the function calls would be very similar.
> Forrest
> coin-discuss-bounces at list.coin-or.org
> to
> > <coin-discuss at list.coin-or.org>
> > To
> > Subject
> >
> nesting problems. I have quite a lot of variables (up to 1000)
> > representing positions of nested parts and approximatly the same
> of magnitude of constaints (up to 5000). Those constraints
> the non-overlapping condition between nested parts.
> those problems as fast as possible. I would like
> algorihtm can speed up my computation time ; and
> the barrier algorithm included in CLP with OSI
> >
> Coin-discuss mailing list
> http://list.coin-or.org/mailman/listinfo/coin-discuss
> >
> >
> list
> http://list.coin-or.org/mailman/listinfo/coin-discuss
> >
>
> _______________________________________________
> list
> http://list.coin-or.org/mailman/listinfo/coin-discuss
>
>
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-discuss
>
More information about the Coin-discuss
mailing list