Heads up: XmlValidation fails with Shenandoah JDK 9, possible compiler bug

[Roland Westrelin]

> I think we got ourselves a regression. This is the first time I see it, probably
> because of tiered compilation turned off, which puts more stress on C2.

Should be fixed now.

> Bisect points at mammoth change:
>
> changeset:   13043:b1d87ff3b89a
> user:        roland
> date:        Mon Feb 27 13:57:51 2017 +0100
> summary:     remove useless call to recompute_dom_depth + move loop opts to
> shenandoah barrier classes + rework post expansion loop opts

FYI, it's unrelated to that change eventhough I also found that I can
reproduce it only with it.

Roland.