[CPMD-list] Coredump on SP4

Balasubramanian Sundaram bala at jncasr.ac.in
Sat Aug 13 13:12:42 CEST 2005


Dear Axel,
       Thanks for all your suggestions. The executable that was compiled without 
SMP is able to run the Si64-10ryd (inp.wf) as well as other input files which are 
specific to our research work. Thanks a million!
        If you require my help to further investigate the issue(s) with respect to 
the SMP version on this platform, please let me know. I would require a lot of 
handholding though...

Regards,
Bala
===========Final Makefile that worked=========================

#--------------- Default Configuration for IBM-PWR5-AIX-ESSL-MPI ---------------
SRC  = .
DEST = .
BIN  = .
#QMMM_FLAGS = -D__QMECHCOUPL
#QMMM_LIBS  = -L. -lmm
FFLAGS =  -q64 -qmaxmem=32768 -qtune=pwr4 -qarch=pwr4
LFLAGS = -q64 -L/usr/local/lib -llapack64 -lessl \
             -qarch=pwr4  $(QMMM_LIBS)
CFLAGS =  -q64 -qarch=pwr4
CPP =  /usr/ccs/lib/cpp -P
CPPFLAGS =  -D__IBM -DLAPACK -DFFT_ESSL  -DMALLOC8 -DPOINTER8 -DPARALLEL=PARALLEL 
-DMP_LIBRARY=__MPI
NOOPT_FLAG =  -O0
CC = xlc_r -c -O3 -qstrict
FC = mpxlf_r -c -O3 -qstrict
LD = mpxlf_r  -O3 -qstrict
AR = /usr/bin/ar ruv
#----------------------------------------------------------------------------




Axel Kohlmeyer wrote:
> On Sat, 13 Aug 2005, Balasubramanian Sundaram wrote:
> 
> dear bala,
> 
> BS> Dear Ari and Axel,
> BS>       Thanks for your suggestions. I used Axel's suggestion of configuring a 
> BS> Makefile with the "IBM-PWR5-AIX-ESSL-MPI-SMP" keyword, changed "pwr5" to "pwr4" 
> BS> everywhere, compiled the whole thing and ran it and it produced an output which is 
> BS> attached with this mail as file.out_02_392. It did not coredump.
> 
> hmmm, i've seen that one before. i think it was related to OpenMP.
> please try the IBM-PWR5-AIX-ESSL-MPI (withOUT SMP) config and 
> use pwr4 instead of pwr5.
> 
> i noticed recently that this bulk silicon example can behave 
> a little strange sometimes. you many want to try a different 
> input, too.
> 
> regards,
> 	axel.
> 
> [...]
> 



More information about the CPMD-list mailing list