[Bug tree-optimization/63844] [4.8/4.9/5 Regression] open mp parallelization prevents vectorization

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63844

--- Comment #11 from rguenther at suse dot de <rguenther at suse dot de> ---
On Tue, 18 Nov 2014, jakub at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63844
> 
> --- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
> (In reply to Richard Biener from comment #9)
> > Valid testcase (well, assign sth useful to g...)
> > 
> > #define N 10000000
> > int a[N], b[N], c[N];
> > 
> > main()
> > {
> > 
> >   int i, g;
> > 
> > #pragma omp parallel for num_threads(4)
> >   for (i = 0; i < N; i++) {
> >       a[i] = b[i] + c[i] + g;
> >   }
> > }
> > 
> > in the OMP fn we fail to hoist the load of .omp_child_i->g out of the
> > loop and thus generate inferior code.
> 
> And does making receiver_decl restrict and/or reference type help with that?

Yes.

> BTW, we probably should add some IPA omp pass that would try to constant
> propagate across from GOMP_parallel*/GOMP_task* callers to the *.omp_fn.*
> functions (or teach IPA-SRA to do that?).

I think aggregate IPA-CP does that, IPA-SRA cannot as the function has
its address taken.

Richard.