The Convex SPP at the Supercomputing Center of the JKU ====================================================== See http://origin.edvz.uni-linz.ac.at/hardware/SPP.html Login ----- telnet spp.edvz.uni-linz.ac.at General Tools ------------- (See man pages) System Information scm sysinfo Load situation on subcomplex "parallel" cnx_ps -s parallel Switch shell to subcomplex "parallel" mpa -sc parallel Compilation and Execution ------------------------- Automatic parallelization /usr/convex/bin/fc -O3 matmult.f -o matmult /usr/convex/bin/cc -O3 matmult.c -o matmult Execute program on cluster "parallel" with exactly 8 threads mpa -sc parallel -min 8 -max 8 matmult Time execution from command line timex mpa -sc parallel -min 8 -max 8 matmult (within a program, only measure wall clock times, e.g. by time()") Cxpa performance visualization ------------------------------ Compile for parallel execution /usr/convex/bin/fc -cxpa -O3 matmult.c -o matmult Fix execution parameters mpa -m -sc parallel -min 8 -max 8 matmult Start cxpa cxpa matmult - use "Subcomplex Selection" to select subcomplex "parallel" - use "Profile Selection" to "Select Region to Profile" (select all) - "Start" program - Investigate e.g. "3D Profile" (various "Region Types") MPI --- Switch shell to cluster "parallel" mpa -sc parallel Compilation /usr/convex/mpich/bin/mpicc -O2 prog.c -o prog Execution with 8 processes mpirun -np 8 prog