[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [bgl-discuss] crashes in VN mode
At 05:01 PM 2/8/2005, Boyana Norris wrote:
I've been running some (16 and 32-processor) GYRO simulations, built with
/home/norris/software/bin/mpif90 (rip-off of Satish's version, just
changed from f77 to f90), and the following optimization options:
-qbgl -qarch=440d -qtune=440
-qcache=level=1:type=i:size=32:line=32:assoc=64:cost=8
-qcache=level=1:type=d:size=32:line=32:assoc=64:cost=8
-qcache=level=2:type=c:size=4096:line=128:assoc=8:cost=40 -qhot=simd -O3
This runs fine in coprocessor mode but crashes (with core dump) in vn mode
after the initialization of the simulation otherwise.
Another possible source of problems is memory exhaustion - in VN mode, you
have only half as much memory per MPI process. I seem to remember that
running out of memory did not generate "nice" results (i.e., a clean death).
Does anyone know if there are alternate entry points for the C library
routines? It would be nice to replace malloc with
void *malloc( size_t s )
{
void *p = _malloc( s );
if (!p) { reliably scream and die; }
return p;
}
Ditto for brk (I'm not sure which Fortran may use).
Bill
I'm starting to experiment with different compiler options, but wanted to
ask in the meanwhile if compiling with the intention of running in virtual
node mode is supposed to be done differently than when running in
coprocessor mode?
Also, what debugger should I be using and how (in conjunction with
mpirun)?
Thanks!
Boyana
- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.
William Gropp
http://www.mcs.anl.gov/~gropp
- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.