[Ipopt] IPOPT Binary Status and linear solver choices.

[Stefan Vigerske]

Hi,

> I am interested in upgrading to the latest IPOPT and would love to use
> the binaries (Ipopt-3.5.5-win32-icl10.0 for example) if they are up and
> running.
> 
> But before I start doing a build and the inevitable head-scratching
> linking battles.
> 0) If I missed out on existing docs for this stuff, I apologize.

There are no docs, sorry.

> 1) What does the binary link into the .lib file these days? Does it have
> the builtin BLAS, LAPACK, and HSL? HSL I can deal with, but BLAS/LAPACK
> will conflict with my intel MKL.

It has Blas, Lapack, and Mumps linked in. HSL is not linked in.

> 2) If it is linked in, any chance of getting a binary build with the
> BLAS/LAPACK not linked in?

No chance ;-). It's a headache to build on Windows, if I find time,
maybe I'll do it.

> 3) Any chance of getting a debug build of the binaries? I notice they
> are built for Ipopt-3.5.3-win32-icl10.0, which I will use instead if the
> above linking issues are not a problem. If I remember, those may have
> been build around the last time I asked this question.

Higher chance here. Since these debug builds are quite large, I did not
made one for every new release. So far, just use 3.5.3 for debugging.

> 4) I remember previously that the Intel MKL version of Paradiso wouldn't
> work with IPOPT. Is this true with the latest MKL? If people think it is
> reasonably likely to work, I can try to do a build/link with it on windows.

I don't know.

> 5) Last, if it is using HSL directly, any chance of having a binary
> distribution (release/debug as well) that is built to link to MUMPS
> (which I would link on my own). Naturally, I would also want to link in
> my own MPI library (or the fake one needed by serial MUMPS). I may try
> this with http://sourceforge.net/projects/winmumps/

See 1).

Stefan

-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan