https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123070
--- Comment #23 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #22) > (In reply to Andrew Pinski from comment #21) > > Hacking generic_type_for in tree-ssa-loop-ivopts.cc to return the pointer > > type fixes the issue. But I still am not sure that is valid. > > > > generic_type_for used to return pointer_type_node, see PR 20625 > > (r0-67159-gd482f417e4445f). But then an integer type due to PR 27865 > > (r0-76106-g2052721560b232). > > Most of r0-76106-g2052721560b232 was removed when pointer plus was added > (r0-81506-g5be014d5b728cf) too. So maybe we can move back to ptr_type_node. I am going to try that. I do know `gcc.dg/tree-ssa/ivopts-[5689].c` testcases fail but those are the IV-opts testcases for this exact testcase so they can just be adjusted. I am almost positive this will improve more code but is it really too late into stage 3 to do this; I will have to check that out. IV-OPTs is one of the hardest gimple level code to work with too.
