https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111276
--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Richard Biener from comment #3) > I think the API expects to be guarded to only be called on stmts that > require rewriting. That makes sense, so adding gimple_stmt_with_undefined_signed_overflow and changing all of the places which used arith_code_with_undefined_signed_overflow to use that instead. tree-ssa-ifcombine.cc tested only if lhs was an integral or pointer type, not if it possiblity of overflow. tree-ssa-loop-split.cc, tree-scalar-evolution.cc didn't test lhs at all. tree-if-conv.cc, tree-ssa-ifcombine.cc, tree-ssa-loop-im.cc, tree-ssa-reassoc.cc all tested TYPE_OVERFLOW_UNDEFINED on the lhs already. But moving that into one function definitely cleans up the code too :).
