https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118505
--- Comment #12 from Andrew Pinski <pinskia at gcc dot gnu.org> --- Created attachment 60310 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60310&action=edit Original benchmark Just uploading the standalone benchmark without any changes to the constant formation. Note There is a small regression even at -O3. I have an idea on how to fix that. poor_ifcvt_pred needs to return false if the statement traps. Let me do that first.
