Humboldt-Universität zu Berlin - Mathematisch-Naturwissen­schaft­liche Fakultät - Institut für Physik

Portland compilers

Version:

14.6

Architectures:

i386 (version 11.8-0)
x86_64

Invocation:

pgcc options prog.c C compiler
pgCC options prog.C C++ compiler
pgf77 options prog.f Fortran 77 compiler
pgf90 options prog.f90
pgf90 options prog.f
Fortran compiler
same for fixed source format

(if /usr/global/pgi/bin is in PATH).

pgf90 and pgf95 are both synonyms for pgfortran.

Documentation:

  • man pages
  • /usr/global/pgi/doc/index.htm

Optimisation options:

-O0 disabled
-O1 default
-O -O2  
-O3  
-fast recommended
  • The meaning of -fast[sse] depends on the system type. For details say e.g.
          pgf90 -fast -help
          pgf90 -fastsse -help
    
  • Specify -fast resp. -fastsse for both compile and link steps.
  • Specify the CPU type with -tp xxx, see documentation. It defaults to the system type on which the compiler is running.

Auto-vectorisation

Use of SSE/SSE2 (e.g. with -fast) instructions implies auto-vectorisation.

Successful or failed vectorisation of loops is reported with

      -Minfo=loop
      -Mneginfo=loop

Auto-parallelisation

Parallelisation for multiprocessor systems (SMP) is supported as follows:

      -Mconcur=innermost

Successful or failed parallelisation of loops is reported with

      -Minfo=loop
      -Mneginfo=concur

At runtime, the variable NCPUS is used to specify the number of processors/cores to use:

      NCPUS=<number>; export NCPUS
      a.out

The operating system shows one process close to (ncpus x 100)%, together with the total CPU time.

License:

floating
server: liz.cms.hu-berlin.de


zuletzt geändert: B Bunk, 30.06.2015