https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121298
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> --- I think treating associatable operations in some systematic way would be good. If the whole chain is handled the order usually does not matter, but if only parts of the chain can be combined and recognized this results in missed optimizations otherwise. Better handling of a larger toplevel operation, recognizing a sub-part of it is a bswap and re-materializing the remaining pieces would be another way of viewing this I guess. But I guess we were just lucky before the reassoc change and bswap should be free to re-associate to make a bswap match.
