https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86020
--- Comment #8 from Bill Schmidt <wschmidt at gcc dot gnu.org> --- Honza, sorry for being so late to respond! I had to poke the performance team once more on this. Reverting this patch did indeed solve the problem for us on trunk, and we are seeing far better performance than we were before. I think backporting is supported by the evidence we've seen at this point. Thanks very much for the fix!