This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] combine: Allow combining two insns to two insns

On Wed, Jul 25, 2018 at 10:28:30AM +0200, Richard Biener wrote:
> On Tue, Jul 24, 2018 at 7:18 PM Segher Boessenkool
> <> wrote:
> >
> > This patch allows combine to combine two insns into two.  This helps
> > in many cases, by reducing instruction path length, and also allowing
> > further combinations to happen.  PR85160 is a typical example of code
> > that it can improve.
> >
> > This patch does not allow such combinations if either of the original
> > instructions was a simple move instruction.  In those cases combining
> > the two instructions increases register pressure without improving the
> > code.  With this move test register pressure does no longer increase
> > noticably as far as I can tell.
> >
> > (At first I also didn't allow either of the resulting insns to be a
> > move instruction.  But that is actually a very good thing to have, as
> > should have been obvious).
> >
> > Tested for many months; tested on about 30 targets.
> >
> > I'll commit this later this week if there are no objections.
> Sounds good - but, _any_ testcase?  Please! ;)

I only have target-specific ones.  Most *simple* ones will already be
optimised by current code (via 3->2 combination).  But I've now got one
that trunk does not optimise, and it can be confirmed with looking at
the resulting machine code even (no need to look at the combine dump,
which is a very good thing).  And it is a proper thing to test even: it
tests that some source is compiled to properly optimised machine code.

Any other kind of testcase is worse than useless, of course.

Testing it results in working code isn't very feasible or useful either.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]