RFR(S): 8204157: Compiler.sunflow hangs after JDK-8192992

[Nils Eliasson]

Hi,

I didn't fix the load scheduling properly in JDK-8192992. I introduced a 
loop check on phi-nodes when doing scheduling. The missing part was that 
non-loop-phis fell through and didn't affect the load placement at all. 
(Phis are not pushed on the search stack so any store happening after 
was not observed.)

With this fix loop-phis still force the load into the block preceding 
the phi (so that any store in the loop can't interfere with the load 
from the backedge). The new part is that regular phis fall through to 
next block and affects the load scheduling as a normal store would have.

Bug: https://bugs.openjdk.java.net/browse/JDK-8204157

Webrev: http://cr.openjdk.java.net/~neliasso/8204157/webrev.01/

Please review,

Nils