https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124029

--- Comment #9 from Jeffrey A. Law <law at gcc dot gnu.org> ---
So just so it gets recorded.  Andrew's patch twiddles the cost model slightly
and as a result we get slightly different unrolling decisions in a couple RVV
tests.

Based on what I've seen in the debugger, Andrew's change is correct from a
costing standpoint.  Furthermore the additional unrolling appears to be
reasonable from a codegen standpoint.

When the time to upstream comes we'll adjust those tests appropriately.

Reply via email to