lesburg at a.chem.upenn.edu
Mon May 6 09:22:54 EST 1996
Hello multiprocessing X-PLOR users,
I would like to know how your multiprocessing experiences correlate
We are currently running X-PLOR 3.1 on a 4 processor SGI
PowerChallenge L running IRIX 6.1. We seem to get only a 5% speed
increase in running a simulated annealing job on 4 processors relative
to running the same job on a single CPU.
While the job is running (the only significant job on the machine),
all four processors show ~100% usage, but again, the job takes just
about as long (in real time) as on a single processor.
One note here: many months ago (on a previously compiled version of
X-PLOR 3.1), I recall that our "multiprocessing" jobs would utilize a
single CPU most of the time, and only use CPU time on a second
processor for a small proportion of the total CPU time. I attributed
this phenomenon to the notion that only the FFT was parallelized (is
this true?) and therefore additional processors got used only during
the FFT calculation. Now, however, all processors are going full
speed all of the time in a seemingly unproductive manner.
Here are our relevant environmental variables:
NUM_THREADS=1 or 4
MP_SET_NUMTHREADS=1 or 4
XPLOR_COMP=-O3 -mips4 -sopt,-r=2,-sy=V -OPT:fold_arith_limit=1678
Also, in the $COMS/link.com file, we have altered the final linking
line to read (for successful linking):
f77 -o xplor.exe *.o -s -lcomplib.sgimath_mp -mips4 -lblas_mp -mp
We would appreciate any help with the compilation process to further
parallelize the executable and your own experiences with X-PLOR
Thank you for your time.
lesburg at chem.upenn.edu
More information about the X-plor