https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96272
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- It shouldn't be added to match.pd, the check as written in the source is certainly better for other optimization passes. For PR95853 I've added recently a widening_mul (i.e. very late pass, almost before expansion) pattern matching of another form (very lame in the source code) of the overflow check but in that case I've put a constraint that the overflow check and addition must be in the same bb (and not too far from each other). In this case it isn't in the same bb and again would be very undesirable if it is really too far (would increase register pressure).
