https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122456
--- Comment #19 from Andrew Macleod <amacleod at redhat dot com> --- Created attachment 62788 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62788&action=edit patch to ignore new code This simply removes the "return" from the new code added by the patch which triggered this issue. This means the new chunk of code remains in the function, but any values it produces are overwritten by the original code. The original results are always returned. With this patch applied, VRP/Ranger produces 100% identical results to what it did before. The bootstrap continues to fail even though VRP is no longer effecting the IL from before the triggering patch. That seems to be pretty clear that something latent was uncovered. Perhaps the slightly increased size of wi_fold is causing LTO to do something different?
