Compile problem of VASP 4.6 on IBM Power6

Message

deltz · #1 Post by **deltz** » Tue May 06, 2008 10:23 am

I modified the makefile.sp2 and which is successful on IBM Power5, during the compilation, there's no error, and everything seems to be ok, but i try to run some test job, it failed to work, and the error report stated as"segmentation fault",this generally happens when the arrays are ovewdlowed, well, the strange thing is that the old binary file "vasp" which is compiled on Power5 works fine on the new platform. Can anyone give some advices, thanks in advance.

“ERROR: 0031-250 task 1: Segmentation fault
ERROR: 0031-250 task 2: Segmentation fault
ERROR: 0031-250 task 4: Segmentation fault
ERROR: 0031-250 task 6: Segmentation fault
ERROR: 0031-250 task 7: Segmentation fault
ERROR: 0031-250 task 5: Segmentation fault
ERROR: 0031-250 task 3: Segmentation fault
ERROR: 0031-250 task 0: Terminated”

#2 Post by **admin** » Wed May 07, 2008 7:37 am

why at all did you modify the makefile (as the executable for Power5 works) and what modifications were made in the makefile?

deltz · #3 Post by **deltz** » Thu May 08, 2008 6:11 am

well, i just change the OFLAG parameter by "OFLAGS = -O4 -qarch=auto", the original setting is -O3 -qarch=auto, and the modified makefile works for power5.

#4 Post by **admin** » Thu May 08, 2008 11:46 am

just replacing -O4 by -O3 really should work. Are you able to run vasp
-- serial (with the serial executable)
-- parallel with 1 node only
on the Power 6 machine? This is to check whether this error is due to the parallel executable (the different precompiler options generate slightly different code in some subroutines; only the serial executable runs) or due to parallelization problems (hardware, MPI installation; both executables run)

deltz · #5 Post by **deltz** » Mon May 12, 2008 7:52 am

well, serial and parallel with 1 node both work fine.

#6 Post by **admin** » Wed May 14, 2008 9:12 am

this indicates that the error is not actually vasp-related but due to a communication error or a bug/incompatibility of the MPI installation on your machines