Automatic Parallelization fail

LAMOME Julien CS-SI julien.lamome-cs-si@irsn.fr
Thu Dec 8 14:47:00 GMT 2011


Hi !
I try to use the automatic parellilazion with the last (4.6.2) gcc version. I use this « tuto » http://openwall.info/wiki/internal/gcc-local-build for building gcc and tools.
I compile with :
my-gcc loop.c -o loop -fopenmp -floop-parallelize-all -ftree-parallelize-loops=8 -fdump-tree-parloops-details -lpthread

When I launch my "loop", I have only one processor used.
If I add a "#pragma " directive (for use explicitely OpenMP), I have some processors used.

In test-c-117t.parloops, loop 5 and 7are not noted "loop is not parallel according to graphite", so I think they could be parallelized ?

Thank for your help



More information about the Gcc-help mailing list