[Bug debug/83758] ICE building gccgo on powerpc64le --with-cpu=power8

[acsawdey at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83758

--- Comment #12 from acsawdey at gcc dot gnu.org ---
Another data point: go1 built with AT10.0 on willow2 does not fail on genoa,
whereas go1 built with AT 10.0 on genoa does fail. 

Going to see if I can nail down which .o file is the carrier for failing
tomorrow. 

Also with the change previously noted in ira-costs.c why do I see a difference
already in the fixup_cfg4 dump which should be way before ira?