[gmx-users] crash in double precision GROMACS

[Claudio Arlandini]

Hi, has someone any experience in running GROMACS in double precision on a 
HP box?

In single precision everything runs smoothly. I wanted to compare 
performances in the double precision case. Compilation produces no error, 
but when I run grompp_d it crashes after reading and checking consistency 
of input files for a bus error
---------------------------------------------------------
.....
Or. Res. Fit     has 1 element(s): rest
Checking consistency between energy and charge groups...
splitting topology...
Walking down the molecule graph to make shake-blocks
There are 60416 charge group borders and 24576 shake borders
There are 24576 total borders
Division over nodes in atoms:
   7600  7600  7600  7600  7600  7600  7625  7626  7626  7626  7626  7626 
7626  7626  7626  7623
writing run input file...
Bus error (core dumped)
----------------------------------------------------------------
.....
I attached tusc to the process, obtaining:

write(2, "u n   i n p u t ", 8) 
....................................................................... = 8
write(2, "  ", 1) 
..................................................................................... 
= 1
write(2, "f i l e . . . \n", 8) 
....................................................................... = 8
open("1md.tpr", O_RDONLY, 0666) 
....................................................................... 
ERR#2 ENOENT
open("1md.tpr", O_RDWR|O_CREAT|O_TRUNC|O_LARGEFILE, 0666) 
............................................. = 3
ioctl(3, TCGETA, 0x7b000140) 
.......................................................................... 
ERR#25 ENOTTY
   Received signal 10, SIGBUS, in user mode, [SIG_DFL], partial siginfo
     Siginfo: si_code: I_NONEXIST, faulting address: 0x40045d6c, si_errno: 0
     PC: 0x5bebf, instruction: 0x50c40002
exit(10) [implicit] 
................................................................................... 
WIFSIGNALED(SIGBUS)|WCOREDUMP
------------------------------------------------------------------


I tried everything I could imagine:
- different machines:  a 64CPU HP Superdome running HPUX11.11
         a 20 CPU HP V2500 running 11.00
- different compilers: HP ansiC and gcc
- different options:  with and without MPI
                 with and without fortran
                 some more
- 32 and 64 bit compilation
- my own runs and gmxbench runs

It definitely seems a bug to me.
Any hint?

Best regards

Claudio Arlandini


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Dr.  Claudio Arlandini,  PhD                \ C onsorzio
    System Administration                     \ I nteruniversitario
                                               \ L ombardo per la
    e-mail arlandini at cilea.it                  \ E laborazione
    ph. +39 02 26995.246 (fax +39 02 2135520)  \ A utomatica

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\/~~~~~~~~~~~~~~~~~~~~~~~~~~~~~