[FYI] Segfault with hybrid functionals
nadler
rod... at gmx.ch
Thu Aug 11 11:27:39 UTC 2011
Dear all,
This is just for general information which might help people when they get
this problem. When I wanted to use hybrid functionals I always got a
segfault just before the geometry optimization began. I could solve the
problem by changing -O2 to -O0 (for gfortran-4.3 and ifort-11.1.069) and by
removing the flag -xHost (for ifort) when compiling cp2k. This was
somewhat unsatisfactory and by using the TRACE option I found out that the
last subroutine called was calc_screening_functions in
hfx_screening_methods.F. So I added
hfx_screening_methods.o: hfx_screening_methods.F
$(FC) -c $(FCFLAGS2) $<
to the arch file where FCFLAGS2 are those with -O0 set and -xHost or
-march=native removed, the FCFLAGS contains all the optimization flags I
wanted for the rest of the program, i.e. -O2 and -xHost for ifort,
-march=native and -O2 for gfortran. I attached the arch files, just in
case.
With the intel compiler I use the MKL libraries, the MKL wrapper for the
FFTW3 libraries and I had to use the libint_cpp_wrapper contained in the
hfx_tools directory. With the gfortran compiler I used the ACML 4.4.0
libraries, FFTW3 was compiled apart and again I needed
the libint_cpp_wrapper. Libint is version 1.1.4 and FFTW 3.2.2. I hope that
is will be of use for some people.
Cheers,
Roger
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20110811/f6c59d6e/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: r422e2.popt
Type: application/octet-stream
Size: 1956 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20110811/f6c59d6e/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: x2200.popt
Type: application/octet-stream
Size: 1507 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20110811/f6c59d6e/attachment-0001.obj>
More information about the CP2K-user
mailing list