With the current code structure, specifying -post-RA-scheduler=true has different behavior than
enabling it via the scheduling model. The issue is that the mere presence of this option on the
command line suppresses the post scheduler code that would otherwise select an antidependency breaker.
I discovered this when chasing http://llvm.org/PR27681, which should have been reproducible
with -post-RA-scheduler=true. With this change, that will be the case, and should help avoid
such surprises down the road.
I added the "private:" specification simply to make it a little more clear that PostRAScheduler::enablePostRAScheduler
is only referenced from within its class.