[Ipopt] problems on OS X 10.7 Lion

[Ray Zimmerman]

Hmmm ... thanks for the tip about checking the BLAS/LAPACK linking with "otool -L". I thought I was getting the same behavior with the netlib BLAS/LAPACK, but now I see that it was still linking to my system BLAS/LAPACK.

otool -L Ipopt/test/hs071_cpp
Ipopt/test/hs071_cpp:
	/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 159.1.0)
	/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libLAPACK.dylib (compatibility version 1.0.0, current version 1.0.0)
	/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib (compatibility version 1.0.0, current version 1.0.0)
	/usr/lib/libstdc++.6.dylib (compatibility version 7.0.0, current version 52.0.0)

Lemme see if I can force it to use the netlib versions ... ok, have to use "--with-blas=BUILD" and "--with-lapack=BUILD" ... ok, I'm back, and the verdict is ...

That was it! Forcing it to use the netlib BLAS allowed me to build a working Ipopt, including a working Matlab MEX file, on Lion.

There must be an issue with the system BLAS on OS X Lion [1]. This is not good.

    Ray

[1] I'm guessing it's related to this https://github.com/mxcl/homebrew/issues/6649#issuecomment-1666673



On Oct 18, 2011, at 12:44 PM, Thomas Vacek wrote:

> Passing in -fdefault-integer-8 is probably not going to work unless you've done some major hacking; ipfint is typedefed in config.h, and if it is a 4-byte type things will not be good for you.
> 
> Have you used "otool -L" to make sure that the blas/lapack linking is working the way you think it is?  I would focus on getting it to work with Netlib blas first, then veclib.
> 
> Tom
> 
> On 10/18/2011 09:18 AM, Ray Zimmerman wrote:
>> 
>> Thanks for the suggestion.
>> 
>> I'm inclined to think that I'm encountering a single problem (when it manifests, the symptom is the same) and it is not related to Matlab, since it occurs at the 'make test' stage of the Ipopt build when using MUMPS, before even attempting to build the Matlab interface. In this case, adding ' -fdefault-integer-8' to the FFLAGS passed to configure allows it to compile and link just fine, but it results in the following error when running make test:
>> 
>> Testing C++ Example...
>>  
>>  ---- 8< ---- Start of test program output ---- 8< ----
>>   ** Instance Error 1 in DMUMPS_F77                    0
>>  ** MPI_ABORT called
>>  ---- 8< ----  End of test program output  ---- 8< ----
>>  
>>     ******** Test FAILED! ********
>> Output of the test program is above.
>> 
>> Likewise for the C and Fortran tests. I'm not sure what this means, but I get the same result with both the system BLAS/LAPACK (-framework vecLib) or with the ones I can download with get.Blas and get.Lapack in the ThirdParty directory.
>> 
>> So, I'm thinking that if I can get 'make test' to pass consistently for a build with MUMPS, it will probably fix the issue with the Matlab MEX file as well.
>> 
>> -- 
>> Ray Zimmerman
>> Senior Research Associate
>> 419A Warren Hall, Cornell University, Ithaca, NY 14853
>> phone: (607) 255-9645

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20111018/f3c30e50/attachment.html>