Loop Level Parallelization

• Execute different iterations of a loop in parallel on multiple processors
• Loop iterations must not depend on each other

Compiler Directives for C and FORTRAN

• OpenMP platform-independent parallelizing directives/pragmas
• Add pragmas to C code and directives to FORTRAN code
• Platform dependent compiler option for multi-processing

OpenMP on Linux with Portand Group Compilers

• Use -mp option for FORTRAN,C,C++
• Environment variable OMP_NUM_THREADS for number of processes

OpenMP on Irix with SGI Compilers

• Use -mp option for FORTRAN,C,C++
• Environment variable OMP_NUM_THREADS for number of processes

Automatic Parallelization SGI (C/C++/Fortran)

• C compiler - SGI Irix command 'cc -pca '
• FORTRAN compiler - SGI Irix command 'f77 (f90) -pfa'
• Delta SGI O200 has automatic compilers for C and Fortran
• Useful for knowing which loops can be easily parallelized.
• Tries to parallelize every loop - can be inefficient

SGI GUI for Parallelization Analysis of FORTAN ( Workshop Pro MPF )

• compare original source and parallelized code
• provides reasons why certain loops cannot be parallelized
• interactive additions of directives or code changes to facilitate parallelization