https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105314
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|tree-optimization |target
Keywords| |missed-optimization
Target| |riscv
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
I observed sth similar on arm and x86_64 where the backends were adjusted to
treat two equal variants the same. The issue that is observed from the
GIMPLE level change is that the edge containing the forwarder changes and
so we get slightly different initial RTL out of it.
You have to analyze in more detail what happens.
