https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84272
--- Comment #5 from ktkachov at gcc dot gnu.org --- (In reply to Jakub Jelinek from comment #4) > Created attachment 43382 [details] > gcc8-pr84272-2.patch > > Or defer deletion of all the fma_nodes until the end, whether they are root > or not. I'd fear that even removing just non-root nodes might mean the > following processing of the children might still look at the parent nodes. > > Untested as well. I'd appreciate feedback from the pass author. Thomas (the author of the pass) tells me that he prefers the second approach but he has a few extra comments on the patch that he'll provide later today. In the meantime I can help with testing the patch on aarch64. Thanks for investigating this Jakub. --- Comment #6 from ktkachov at gcc dot gnu.org --- (In reply to Jakub Jelinek from comment #4) > Created attachment 43382 [details] > gcc8-pr84272-2.patch > > Or defer deletion of all the fma_nodes until the end, whether they are root > or not. I'd fear that even removing just non-root nodes might mean the > following processing of the children might still look at the parent nodes. > > Untested as well. I'd appreciate feedback from the pass author. Thomas (the author of the pass) tells me that he prefers the second approach and he has a few extra comments on the patch that he'll provide later today. In the meantime I can help with testing the patch on aarch64. Thanks for investigating this Jakub.