This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] for PR 31676
Hello,
> Eric Botcazou wrote:
> >> PR rtl-optimization/31676
> >> * df-scan.c (record_nonlocal_goto_receiver_defs): New function.
> >> (df_refs_record): Call it.
> >
> > OK with the additional test:
> >
> > if (current_function_has_nonlocal_label)
> > ...
> >
> > for the invocation of the new function.
> >
> > Thanks for fixing this.
>
> So, Zdenek, if I understand correctly, we now need this patch checked
> in, and then this and the previous patch backported to 4.2?
the patch already is in mainline.
> If you would be able to work on that, I would appreciate it. I would
> also appreciate your thoughts (and other people as well) about the
> safety of that for 4.2.0. If we don't do it 4.2.0, I think we'll almost
> certainly want to do it for 4.2.1, so effort doing the backport will not
> be wasted.
both patches cleanly apply to 4.2, so backporting is not that much
difficult. I am now bootstrapping & testing the patch.
Regarding the safety, neither of the patches is too intrusive, and they
do not implement any new functionality; the patch for PR 31360 just
tunes the heuristics for register pressure evaluation (the danger being
that the new setting is more aggressive, making loop invariant motion
work more often, possibly triggering latent bugs like PR31676). In
general, I think they are fairly safe to include in 4.2, although of
course I would be happier if they were in mainline for longer period of
time.
Zdenek