[Bug target/39723] [4.8/4.9/5 Regression][cond-optab] worse code with long long shifts on v850

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=39723

--- Comment #10 from Jeffrey A. Law <law at redhat dot com> ---
Yea, as is the case with coloring algorithms order of popping off the coloring
stack will impact the final allocation in either a positive or negative way. 
I'm not at all worried about those kind of effects.

Adding the hard register costing propagating stuff would probably help. Given
that register based calling conventions as well as the oddities in the x86 ISA
that there's a ton of hard register costing information we could exploit.

I'm certainly far less concerned with this specific BZ, but I think it
highlights a minor generic improvement that might be made to IRA.  Your call
if/when/how to attack.