On Tue, Mar 21, 2017 at 08:41:43AM +0100, Jakub Jelinek wrote:
> On Tue, Mar 21, 2017 at 08:38:20AM +1000, Richard Henderson wrote:
> > On 03/21/2017 07:15 AM, Jakub Jelinek wrote:
> > > Not really sure what we should do if both i1 and i2 are frame related,
> > > shall
> > > we check for each of the CFA reg notes if they are available and equal?
> > > Or punt if either of the insns is frame related?
> >
> > I would punt if either is frame related.
>
> Ok, I'll test then the following patch and gather some statistic on how
> often we trigger this.
The statistics I've gathered unfortunately shows that at least on
powerpc64le-linux it is very important to not give up if both i1 and i2
are frame related and have rtx_equal_p notes.
I've set on unpatched old_insns_match_p flags when returning non-dir_none
and checked those flags in the various callers of these when about to
successfully perform cross-jumping, head-merging etc.
With /f vs. non-/f, the only 3 hits were on the new pr80102.C testcase
during powerpc64le-linux bootstrap/regtest, but /f vs. /f there were
167601 hits.
Jakub
