https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61225
--- Comment #7 from zhenqiang.chen at linaro dot org --- Yes. gcc.target/i386/pr49095.c is still FAIL. I have not found an easy way to fix it. The root cause for the FAIL is that: A register copy is forwarded by copyprop_hardreg_forward called in shrink-wrapping. Then the peephole2 can not optimize it since the adress looks like different. I am trying to add more complexity peephole rules. If not work, I will add some pre-check to skip copyprop_hardreg_forward.
