On Wed, Oct 14, 2015 at 08:54:11PM +0200, Marc Glisse wrote:
> On Wed, 14 Oct 2015, Marek Polacek wrote:
>
> >Evidently, the X - (X / Y) * Y -> X % Y pattern can't change the
> >signedness of X from signed to unsigned, otherwise we'd generate
> >wrong code. (But unsigned -> signed should be fine.)
> >
> >Does anyone see a better fix than this?
>
> I was wondering about producing:
>
> (convert (trunc_mod @0 @1))
That's exactly what we had before r225195.
> Aren't there also problems if the conversion changes the precision? I can
> imagine your patch ending in x % 0, with @1 non-zero but a narrowing cast.
Hmm, I haven't found such a case yet. If you find something, pass
it along ;).
Marek
