https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100169
--- Comment #1 from Segher Boessenkool <segher at gcc dot gnu.org> --- The SMS pass is notorious for slight changes making its cost model decide too do things the other way around. Is that what is happening here? Was SMS' decision a good decision (or at least reasonable)? If so, we just should disable the test for p10.