This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [gomp] More IL work
- From: Diego Novillo <dnovillo at redhat dot com>
- To: Sebastian Pop <sebastian dot pop at cri dot ensmp dot fr>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Tue, 2 Aug 2005 11:22:04 -0400
- Subject: Re: [gomp] More IL work
- References: <20050713230235.GA27453@topo.toronto.redhat.com> <20050719095140.GA7735@napoca.cri.ensmp.fr> <20050719134525.GA14690@topo.toronto.redhat.com> <20050719160653.GC6714@napoca.cri.ensmp.fr> <20050719182316.GC16454@topo.toronto.redhat.com> <20050801173918.GA985@napoca.cri.ensmp.fr>
On Mon, Aug 01, 2005 at 07:39:18PM +0200, Sebastian Pop wrote:
> One of the things I have thought about is that it could be possible to
> flag some of the loops to be distributed at the moment where the loops
> are created, and keep the loops representation intact until we
> generate parallel code for these loops, so a part of the gomp pragmas
> are translated into flags on loops until this information is used for
> generating the calls to libgomp.
>
When you say "flag some of the loops", do you mean flag the loop
structures? If so, I think that would be workable.
The only time we need the loops lexically marked is when
processing OpenMP pragmas in the FE. But once we are in the
optimizers, transformations should just work with the loop
structures and feed the loop bodies into the thread creation
code.