[Bug tree-optimization/80155] [7 regression] Performance regression with code hoisting enabled

[thopre01 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80155

--- Comment #28 from Thomas Preud'homme <thopre01 at gcc dot gnu.org> ---
(In reply to rguenther@suse.de from comment #27)
> On Fri, 7 Apr 2017, thopre01 at gcc dot gnu.org wrote:
> 
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80155
> > 
> > --- Comment #26 from Thomas Preud'homme <thopre01 at gcc dot gnu.org> ---
> > Yes true.
> > 
> > By the way, your first patch did improve things a bit. If it's not too invasive
> > could it be possible to commit it for GCC 7?
> 
> It's not really in the shape for being committable.  I'll see if I can
> come up with sth better early in stage1 that might be back-portable.

Thanks.

> 
> > I think I'll try the approach you suggested in
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77498#c10 and see if I get some
> > better results.
> 
> For this case distance is minimal already so I don't expect you can
> do anything (there's also not really separate dataflow for hoisting).

If I remember well the key in the benchmark was that the basic block where the
code was hoisted from could be reached by a much longer path. So although it
was hoisted to a direct predecessor, sometime that value had to be kept for a
very long time.