[RFC] [parloop branch] Supporting reductions, take 2
Zdenek Dvorak
rakdver@kam.mff.cuni.cz
Tue Sep 4 16:45:00 GMT 2007
Hello,
> > > I'm not sure this code (replacing full gimplification of omp atomic
> with
> > > omp
> > > atomic expansion) is stable enough, and it also needs to be adapted to
>
> > > mainline
> > > with the rest of my code, so I'm not sure I'll get it ready in time to
> be
> > > submitted
> > > before September 10th.
> > >
> > > Zdenek, would you say that targeting to get only the reduction code to
>
> > > mainline
> > > (and having a simple omp_expamd_atomic relying on the
> force_gimple_opeand
> > > to produce
> > > code) is good enough?
> >
> > I think it is not worth hurrying; regardless of whether the
> > autoparallelization in 4.3 will support reduction or not, it will not be
> > practically useful. So I would prefer to leave the improvements for the
> > next version, and use the time to test and tune them.
>
> So, having the reduction code with a very simple expand_omp_atomic will do
> for gcc4.3?
not really -- I think it does not make sense to hurry and commit
half-finished work.
Zdenek
More information about the Gcc-patches
mailing list