https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120006
--- Comment #11 from Sam James <sjames at gcc dot gnu.org> --- Created attachment 61249 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=61249&action=edit small.c This cvise-mangled one (tidied up a fair bit after) seems to do it, but it's weird, e.g. swapping g/h makes it fail with GCC 10..14 at least w/ -fipa-pta. I think this extra reduction may be a different bug? We always have initialised b here because f = 1, so...?
