[patch] Autoparallelization

Tobias Burnus burnus@net-b.de
Thu Sep 13 19:07:00 GMT 2007


Zdenek Dvorak wrote:
> this patch merges the autoparallelization pass from the parloop branch.
Great.

Some mere user comments:

> + @item -ftree-parallelize-loops=n
> + @opindex ftree-parallelize-loops=n
> + Parallelize loops, i.e., split their iteration space to run in n threads.

Could this be renamed into -fparallelize-loops? I find such flag names
rather clumsy (-ftree-vectorize is another example, but this is going to
disappear for the user as it will be automatically included in -O3 [or
-O2].)

Another question: This option seems to use a fixed number of threads;
while this presumably generates faster code, I wonder whether it makes
sense to also offer a variable number of threads as it is the case for
OpenMP programs. This is what some other vendors do and is useful if one
distributes the program or wants/has to use a different number of cores
depending on resources and problem size.

Tobias



More information about the Gcc-patches mailing list