Re: [PATCH 2/2] ivopts: Consider number of invariants when calculating register pressure.

[Richard Biener <richard.guenther@gmail.com>]

On Tue, Oct 25, 2022 at 3:00 PM Dimitrije Milosevic
<Dimitrije.Milosevic@syrmia.com> wrote:
>
> Hi Richard,
>
> > don't you add n_invs twice now given
> >
> >  unsigned n_old = data->regs_used, n_new = n_invs + n_cands;
> >  unsigned regs_needed = n_new + n_old, available_regs = target_avail_regs;
> >
> > ?
>
> If you are referring to the "If we have enough registers." case, correct. After c18101f,
> for that case, the returned cost is equal to 2 * n_invs + n_cands.

It's n_invs + 2 * n_cands?  And the comment states the reasoning.

 Before c18101f, for
> that case, the returned cost is equal to n_invs + n_cands. Another solution would be
> to just return n_invs + n_cands if we have enough registers.

The comment says we want to prefer eliminating IVs over invariants.  Your patch
undoes that by weighting invariants the same so it does no longer have
the effect
of the comment.

> Regards,
> Dimitrije
>
>
> From: Richard Biener <richard.guenther@gmail.com>
> Sent: Tuesday, October 25, 2022 1:07 PM
> To: Dimitrije Milosevic <Dimitrije.Milosevic@Syrmia.com>
> Cc: gcc-patches@gcc.gnu.org <gcc-patches@gcc.gnu.org>; Djordje Todorovic <Djordje.Todorovic@syrmia.com>
> Subject: Re: [PATCH 2/2] ivopts: Consider number of invariants when calculating register pressure.
>
> On Fri, Oct 21, 2022 at 3:57 PM Dimitrije Milosevic
> <dimitrije.milosevic@syrmia.com> wrote:
> >
> > From: Dimitrije Milošević <dimitrije.milosevic@syrmia.com>
> >
> > This patch slightly modifies register pressure model function to consider
> > both the number of invariants and the number of candidates, rather than
> > just the number of candidates. This used to be the case before c18101f.
>
> don't you add n_invs twice now given
>
>   unsigned n_old = data->regs_used, n_new = n_invs + n_cands;
>   unsigned regs_needed = n_new + n_old, available_regs = target_avail_regs;
>
> ?
>
> > gcc/ChangeLog:
> >
> >         * tree-ssa-loop-ivopts.cc (ivopts_estimate_reg_pressure): Adjust.
> >
> > Signed-off-by: Dimitrije Milosevic <dimitrije.milosevic@syrmia.com>
> > ---
> >  gcc/tree-ssa-loop-ivopts.cc | 6 +++---
> >  1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/gcc/tree-ssa-loop-ivopts.cc b/gcc/tree-ssa-loop-ivopts.cc
> > index d53ba05a4f6..9d0b669d671 100644
> > --- a/gcc/tree-ssa-loop-ivopts.cc
> > +++ b/gcc/tree-ssa-loop-ivopts.cc
> > @@ -6409,9 +6409,9 @@ ivopts_estimate_reg_pressure (struct ivopts_data *data, unsigned n_invs,
> >            + target_spill_cost [speed] * (n_cands - available_regs) * 2
> >            + target_spill_cost [speed] * (regs_needed - n_cands);
> >
> > -  /* Finally, add the number of candidates, so that we prefer eliminating
> > -     induction variables if possible.  */
> > -  return cost + n_cands;
> > +  /* Finally, add the number of invariants and the number of candidates,
> > +     so that we prefer eliminating induction variables if possible.  */
> > +  return cost + n_invs + n_cands;
> >  }
> >
> >  /* For each size of the induction variable set determine the penalty.  */
> > --
> > 2.25.1
> >