On 1/21/2022 9:52 AM, Andreas Krebbel via Gcc-patches wrote:
When propagating a multi-word register into an access with a smaller
mode the can_change_mode backend hook is already consulted for the
original register. This however is also required for the intermediate
copy in copy_regno which might use a different register class.
Bootstrapped on x86_64 and s390x. No testsuite regressions.
Ok for mainline?
gcc/ChangeLog:
PR rtl-optimization/101260
* regcprop.cc (maybe_mode_change): Invoke mode_change_ok also for
copy_regno.
I'd been sitting on Stefan's patch because it just didn't feel right,
but I also knew I'd need to dig further into it to either resolve my
concerns or be able to express them clearly to Stefan.
I like this patch much better :-) OK for the trunk. Bonus points if
you pick up Stefan's testcase from way back in October.
Jeff