https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83022
--- Comment #1 from Marc Glisse <glisse at gcc dot gnu.org> --- I am pretty sure this was discussed when the patch was reviewed. IIRC the original patch was specifically pattern-matching if(p!=0) (with p the result of malloc) as the only acceptable condition between malloc and memset for this transformation, but the move to the strlen pass made that inconvenient and it was decided that always doing the transformation was ok (I hope I am not rewriting history). If there is a way to check the probability of reaching the call to memset from the call to malloc (preferably conditional to the fact that malloc returned something != 0), that could be checked before generating calloc, but that may not be easy... (compare the local count of the 2 BBs? That's 10000 and 3300 here, but I see you have already filed PR 83023 to improve it) Did you actually hit a measurable slowdown in a real application?
