Re: [PATCH][combine][RFC] Don't transform sign and zero extends inside mults

[Segher Boessenkool <segher@kernel.crashing.org>]

[ reordered a bit ]

On Fri, Nov 06, 2015 at 02:14:12PM -0700, Jeff Law wrote:
> On 11/06/2015 07:19 AM, Kyrill Tkachov wrote:
> >>>I think we should also add:
> >>>       && REG_P (XEXP (XEXP (x, 0), 0))
> >>>       && REG_P (XEXP (XEXP (x, 1), 0))
> I tend to agree.

> >Indeed, this looks better but it still needs the REG_P checks for the inner
> >operands of the extends to not screw up the arm case.

> >P.S. Do we want to restrict this to targets that have a widening mul
> >optab like I did in the original patch?
> I don't think it's necessary or desirable.

With the REG_P checks added, now simplification is only stopped for
widening muls of registers, so all is fine -- any such construct _is_
a widening multiplication!

This patch stops combine from generating widening muls of anything else
but registers (immediates, memory, ...).  This probably is a reasonable
tradeoff for all targets, even those (if any) that have such insns.

> >I'll let you put it through it's paces on your setup :)

> I'll let Segher give the final yes/no on this, but it generally looks 
> good to me.

It looks okay to me too.  Testing now, combine patches have the tendency
to do unforeseen things on other targets ;-)


Segher