RR(XS): 8038785: hot workaround fix for a crash in C2 compiler at Node::rematerialize

[Vladimir Kozlov]

I agreed to these changes because it is urgent customer's case.

 From correctness point of view changes are fine but it will introduce a performance regression because loops in next 
cases will not be converted to Counted and as result will not be unrolled, range check eliminated and some some checks 
will not be moved from a loop. In other words no loop optimizations will be done for next cases:

if (a) {
   ...
}
for (; i<l;i++) {}

and when head of the loop (back branch target) is bci 0.

When we find the real cause and a fix we can undo this temporary fix. But it may take time.

Thanks,
Vladimir

On 4/2/14 6:26 AM, Sergey Gabdurakhmanov wrote:
> Hello,
>
> Here is webrev of changes I'm about to integrate:
>
> webrev: http://cr.openjdk.java.net/~sgabdura/8038785/webrev.00/
> bug: https://bugs.openjdk.java.net/browse/JDK-8038785
>
> This is a hot workaround fix for JDK7.
> Proper fix for JDK8 and JDK9 will be done under the bug
> https://bugs.openjdk.java.net/browse/JDK-8039050
>
> Please check JDK-8039050 for additional details.
>
> BR,
> Sergey