Anyone have PARDISO working on OS X?

[ringlery]

Ok,
I've tried to get the sample PARDISO code (from www.pardiso-project.org, slightly modified) running on two machines running OSX ( a core duo laptop using IA32 libraries and a dual quad core mac pro using both 32 and 64 bit libraries). Regardless of whether I try the fortran or c sample code it compiles and links successfully but gives a segmentation fault when executing the first call to PARDISO (with phase = 11). I also get a segmentation fault when trying to execute my own code.

Has anyone successfully used the PARDISO routine with MKL 10.0.2.018 and the intel c/c++/fortran compiler version 10.1.012?

[ringlery]

Just an update...
I installed MKL on a linux virtual machine running on one of the above mentioned OSX machines. Using the same sample code, and linking as prescribed in the manual, the program compiles, links and executes as it should. I have tried almost every conceivable combination of linking (even in groups of three, as suggested in a previous post for a different problem).

So there's nothing wrong with the sample code, and it appears that it links properly in OSX. I should mention that I'm running OSX 10.5.2. Any insight would certainly be appreciated.

[Sergey_K_Intel1]

Hello,

It is well known problem for 32-bit MKL 10.0 Update 2. MKL64-bit binaries for OSXshould be Ok. The reason of failures is that MKL was built byan old Intel C++ compiler which generates codes incompatible with the latest gcc libraries

$ gcc --version
i686-apple-darwin9-gcc-4.0.1 (GCC) 4.0.1 (Apple Inc. build 5465)
Copyright (C) 2005 Free Software Foundation, Inc.

As Intel compiler folk reported theseproblems with incomaptibility were fixed in10.0.022and 10.1.007 Intel C++ compiler or later versions. The coming MKL 10.0 Update 3 will contain binaries rebuiltbyone of thesecompilers.In any case youhave to use at least 10.0.022 or 10.1.007 or later versions ofIntel C++/Fortran compilers.

There is another workaroundfor your MKL 10 Update 2 whichmight work with the latestIntel compilers. The workaround is not to use DSS/PARDISO statistics. It can be done by setting parameter MSGLVL (message level information) to 0. Hope it helps.

All the best

Sergey

[ringlery]

Thanks for the answer. I have gotten it working with the 64 bit libraries. Changing the MSGLVL and using the 32 bit libraries didn't seem to work, though.