[9] RFR(S): 8179678: ArrayCopy with same src and dst can cause incorrect execution or compiler crash

Tobias Hartmann tobias.hartmann at oracle.com
Tue May 30 12:17:10 UTC 2017


Hi Roland,

okay, looks good. I'll re-run test.

Thanks,
Tobias

On 24.05.2017 09:55, Roland Westrelin wrote:
> 
>> with your fix, compiler/arraycopy/TestEliminatedArrayCopyDeopt fails
>> on all platforms with:
> 
> Thanks. The problem here is that with -XX:-ReduceInitialCardMarks, C2
> adds a g1 post barrier between the arraycopy node and the following
> membar that ArrayCopyNode::may_modify() doesn't expect. Here is a new
> webrev:
> 
> http://cr.openjdk.java.net/~roland/8179678/webrev.04/
> 
> with a fixed ArrayCopyNode::may_modify(). I also added verification code
> that calls ArrayCopyNode::may_modify() after an array copy node is
> expanded to verify that ArrayCopyNode::may_modify() is consistent with
> the just expanded subgraph.
> 
> Roland.
> 


More information about the hotspot-compiler-dev mailing list