Branch removal

[Patrick Metzler]

> My original suggestion is not valid for your case because loop
> optimizations modified the graph in such way that you can not use your
> transformation. Lets do opposite: insert you optimization just after
> first igvn.optimize() in Compile::Optimize(). I would hope that you
> still have "diamond" shape graph at that point.

Ideal Graph Visualizer shows this CFG (non-empty else branch) already 
after parsing. Also after the first igvn.iptimize(), there are two basic 
blocks in the else branch.

Best regards,
Patrick