build problems on Mac

[Peterson, Kirk]

Hi,

I wonder if anyone else has run into the following build problem on a Mac running something like OS 10.8.4 (Xcode 4.6.2). I'm using the intel fortran compiler (composer_xe_2011_sp1.6.038) and gcc (4.2.1) with openmpi 1.6.3. My setup command was simply:

./setup --fc=mpif90 --cc=mpicc --cxx=mpicxx --int64=on --mpi=on

everything looks fine until very near the end, where I get:

Linking Fortran executable cfread.x
Linking Fortran executable dirac.x
ipo: warning #11016: Warning unknown option --start-group
ipo: warning #11016: Warning unknown option --end-group
ipo: warning #11016: Warning unknown option --start-group
ipo: warning #11016: Warning unknown option --end-group
ipo: warning #11016: Warning unknown option --start-group
ipo: warning #11016: Warning unknown option --end-group
ipo: warning #11016: Warning unknown option --start-group
ipo: warning #11016: Warning unknown option --end-group
Linking Fortran executable dirac_mointegral_export.x
ipo: warning #11016: Warning unknown option --start-group
ipo: warning #11016: Warning unknown option --end-group
ipo: warning #11016: Warning unknown option --start-group
ipo: warning #11016: Warning unknown option --end-group
ld: unknown option: --start-group
make[2]: *** [cfread.x] Error 1
make[1]: *** [CMakeFiles/cfread.x.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
ld: unknown option: --start-group
make[2]: *** [dirac.x] Error 1
make[1]: *** [CMakeFiles/dirac.x.dir/all] Error 2
ld: unknown option: --start-group
make[2]: *** [dirac_mointegral_export.x] Error 1
make[1]: *** [CMakeFiles/dirac_mointegral_export.x.dir/all] Error 2


At one time I had successfully build Dirac10 in the same way, but I'm guessing it's something simple I've missed this time around.

thanks in advance,

-Kirk

[André Campos]

Hi

 I also got problems on mac using ifort. It only worked properly when I used gfortran instead. I am not quite sure what might be the cause of it.

Best

Andre

--
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Peterson, Kirk]

As an update to this, it turns out that the --start-group and --end-group linker commands are not support by the Apple ld. I then discovered that these link commands are only included if the math libraries are specified via the $MATH_ROOT environment variable. So instead I specified the mkl libraries explicitly with --explicit-lib=-L/opt/intel/mkl/lib -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core . The program then linked with no problems. Unfortunately a few of the test jobs are now failing, e.g., krci_energy . In this case the job completes but the energy is wrong by about a mEh or so. So this seems like a compiler optimization bug. I'd appreciate any hints, but I know that's a bit like shooting in the dark. I suppose it could be my mpi implementation since krci_properties_omega_tdm crashed with what looks like some mpi problems. I'll see if using 32-bit integers helps.

best regards,

-Kirk

[radovan bast]

hi Kirk,

sorry for the troubles and thanks for reporting this.

i didn't know about this (being a linux person ...).

i will read a bit about the start/end group and perhaps soon publish a patch

for this as soon as i have found a general solution.

also we will improve the testing procedure for the next release so

that it gives more useful information if things don't work.

best regards,

  radovan

--

# PDC Center for High Performance Computing &

# Department of Theoretical Chemistry and Biology 

# Royal Institute of Technology, Stockholm

# +46-8-790-6628

[Peterson, Kirk]

Hi Radovan,

it seems for a 32-bit build one can point $MATH_ROOT to the mkl location like in linux and then after setup one just has to delete all the "--start-group;" and "-Wl, --end-group" strings from CMakeCache.txt in /build before invoking make and then it links correctly.  For a 64-bit build the libraries that are chosen for mkl in this case do not seem to correspond to the correct I*8 ones. The program links but does not run (I get the Lapack integer*4/8 inconsistency error).  So then I just use the --explicit-lib command below without setting $MATH_ROOT.  In either case though krci_energy test job fails with an incorrect energy.  

best regards,

-Kirk

[Peterson, Kirk]

Hi Radovan,

one last update on this.  I just made a I*8 serial build, all else being the same as before, and now all test jobs run ok.  So it seems the problems with KRCI before were related to mpi. I was using OpenMPI 1.6.3 just as in my (successful) linux builds.  Doubtful I'll spend any more time trying to track that down since it's probably more appropriate for the developers should they find the time at some point.

best regards,

-Kirk

[Stefan Knecht]

dear Kirk,

thanks for your detailed bug(?) hunting. I have so far only used gfortran on my Mac but I recently discovered that my university has a campus license for the intel software suite. 

however, for family reasons, I can't promise to look into the issue the next three-four weeks but will do so when I am back at work. 

with best regards,

stefan

--

Stefan Knecht

Laboratory of Physical Chemistry - The Reiher Group -

ETH Zuerich

HCIG 230

Wolfgang-Pauli-Str. 10

CH-8093 Zuerich

Switzerland

phone: +41 44 633 46 46   fax: +41 44 633 15 94

email: stefan...@phys.chem.ethz.ch

web: http://www.theochem.uni-duesseldorf.de/users/stefan/index.htm

   http://www.reiher.ethz.ch/people/knechste

[Peterson, Kirk]

Dear Stefan,

no problem, this is not a huge priority for me. If you can get to it at some point that would be great.  We have several new iMacs in my group that I was wanting to throw some Dirac calculations onto to take the load off our linux cluster.  Even a serial build (as I have now) should be helpful.

best regards,

-Kirk