RFR(XS): 8202123: C2 Crash in Node::in(unsigned int) const+0x14

[Vladimir Kozlov]

On 7/9/18 6:15 AM, Tobias Hartmann wrote:
> Hi Roland,
> 
> On 09.07.2018 15:02, Roland Westrelin wrote:
>> http://cr.openjdk.java.net/~roland/8202123/webrev.00/
> 
> Looks good to me. I'll start some testing.
+1

Vladimir

> 
>> In the last loop of the test, the array store is moved out of loop with
>> control set right above the loop head and below the loop limit check
>> predicate. Then only, the loop limit, that includes a load that depends
>> on the store, becomes loop invariant and the loop is converted to a
>> counted loop. But because the control of the store is below the loop
>> limit check predicate, the loop limit check predicate is updated with a
>> test that doesn't dominate the if node.
> 
> Why are we not able to figure out that counts != arr? Shouldn't EA even remove the allocation and
> scalar replace counts?
> 
>> This is not targeted for 11 but it seems simple and low risk enough that
>> it could be included in 11.
> 
> Yes, we should fix this in JDK 11. Please change the fix version accordingly.
> 
> Thanks,
> Tobias
>