Port your Jacobi and matrix inversion codes to the Origin using OpenMP. Run on a grid of 1000 - 5000 points. Use 4 processors max. For fixed grid size, give speedup on 1, 2, 4 processors.
DUE October 12
To submit a job, create a file, say jobfile. Some things you may want to include in it (change if using cc):
/tt
#!/bin/csh -f #PBS -l ncpus=4 #PBS -V #PBS -q medium_p #PBS -M pitman@math.buffalo.edu cp /CCRSGI/home/pitman/fortran/sph/sph* $PBSTMPDIR cd $PBSTMPDIR setenv OMP_NUM_THREADS 4 f90 -O2 -LNO -IPA -n32 -mips4 -r10000 -mp -o sph sphomp.f date time ./sph > outfile date cp outfile /CCRSGI/home/pitman/fortran/sph/ cp positions /CCRSGI/home/pitman/fortran/sph/ cp position_init /CCRSGI/home/pitman/fortran/sph/ # # remove scratch directory cd /FCScratch \rm -r $PBSTMPDIR
To submit,
qsub jobfile
The command qstat, with flags -Q or -q gives status of your job. See Tom Furlani's OpenMP/PBS training talk on the CCR training page Introduction to OpenMP and PBS