Dear Prof. Troyer, did you receive the log file of the make VERBOSE=ON related to the Lapack linking problem of ALPS 2 on AIX ? Do you have any solution for that ? Is there some compiler option which I could try ? Thank you for your help.
Best Jessica
It seems to be a BLAS linker convention issue. Can you please set the variable CMAKE_CXX_FLAG in your CMakeCache.txt file to -DBIND_FORTRAN_LOWERCASE and try again?
Matthias
On Oct 11, 2010, at 10:09 AM, jessica.alfonsi@unipd.it wrote:
Dear Prof. Troyer, did you receive the log file of the make VERBOSE=ON related to the Lapack linking problem of ALPS 2 on AIX ? Do you have any solution for that ? Is there some compiler option which I could try ? Thank you for your help.
Best Jessica
Hello, we tried again following your suggestion but it stops again at the same point. What else can I do ? Best
Jessica
It's very hard to debug your configuration by e-mail. What seems to happen is that it links lapack functions with a trailing underscore although we are on an IBM machine (which does not use the _). That's why it cannot find the lapack function. The file
alps/src/boost/numeric/bindings/detail/config/fortran.hpp
should automatically set the FORTRAN_ID macr to not use an underscore on your IBM machine which is then used in
alps/src/boost/numeric/bindings/lapack/detail/lapack_names.h
Something must go wrong there on your setup although without access to the machine I cannot find out why FORTRAN_ID is set to the wrong value. Please try the following: In both files above replace
# define FORTRAN_ID( id ) id##_
by
# define FORTRAN_ID( id ) id
That should make it work. Once that works we need to go back and figure out why the wrong version was picked on your machine. Some macros must be messed up.
Matthias
On Oct 12, 2010, at 10:06 AM, jessica.alfonsi@unipd.it wrote:
Hello, we tried again following your suggestion but it stops again at the same point. What else can I do ? Best
Jessica
comp-phys-alps-users@lists.phys.ethz.ch