[Ipopt] Ipopt Matlab Interface Crashes: DPOTRF, limited memory Hessian

[Anya Castillo]

This message initially began as a query for help but after some debugging, I
found a solution--so I'm just posting the results for the benefit of others,
since I've seen that others have had this similar
problem<http://www.google.com/search?client=ubuntu&channel=fs&q=blas%2C+lapack%2C+matlab+64-bit%0A&ie=utf-8&oe=utf-8#q=ipopt,+matlab+interface,+On+entry+to+DPOTRF+parameter+number++4+had+an+illegal+value&hl=en&client=ubuntu&channel=fs&prmd=ivns&filter=0&bav=on.2,or.r_gc.r_pw.&fp=44b3839bd44fe786>
.

I'm running Matlab 7.10 glnxa64 on Ubuntu 10.10.  I am getting the following
error when running the examplehs051.m from Matlab:  "On entry to DPOTRF
parameter number  4 had an illegal value"

I have tried the following suggestions from the mailing list, with no
success:
(1)  Configure IPOPT using: --with-blas=BUILD and --with-lapack=BUILD. Make
clean, recompile from scratch.
(2)  Configure IPOPT using: g++-4.2 and gcc-4.2 (compatible for Matlab 7.10
glnxa64 on Linux). Make clean, recompile from scratch.
(3)  Configure IPOPT using: gfortran -fdefault-integer-8. Make clean,
recompile from scratch.
(4)  Configure IPOPT using: --with-blas=BUILD and --with-lapack=BUILD. Make
clean, recompile from scratch.

All of the above compile successfully and run the given Matlab examples,
except the example with limited memory hessian (examplehs051.m). Matlab
still crashes.

(Another suggestion from the mailing list that I haven't tried:  Modify the
wrappers in Ipopt/src/LinAlg/IpBlas.cpp and IpLapack.cpp and ma27)

WHAT FINALLY WORKED-->
Configure IPOPT using: ADD_CXXFLAGS='-DFUNNY_LAPACK_FINT'.  Make clean,
recompile from scratch.

Therefore, my configuration command is the following:
./configure --prefix=/apps/Ipopt-3.9.3/Install --enable-static
--with-matlab-home=/apps/R2010a CXX='g++-4.2 -m64' CC='gcc-4.2 -m64'
F77='gfortran -m64' ADD_CXXFLAGS="-fPIC -fexceptions -DFUNNY_LAPACK_FINT"
ADD_CFLAGS="-fPIC -fexceptions" ADD_FFLAGS="-fPIC -fexceptions"

Andreas Waechter suggested the working solution in a thread from 3/14/2011.
Thanks so much, and I didn't see any response to this old thread that the
suggestion worked, but it does work at least in my environment.

Many Thanks!  -Anya
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.coin-or.org/pipermail/ipopt/attachments/20110507/b7157810/attachment.html