C2 compiler bug closed as incomplete

[Julien Nicoulaud]

Hi,

Here are some more tests as asked on the JIRA issue:
 -XX:-PartialPeelLoop : no crash
 -XX:-LoopUnswitching : still crashes
 -XX:-SplitIfBlocks : still crashes
 -XX:LoopUnrollLimit=0 : no crash
 -XX:-RangeCheckElimination : still crashes

=> so it looks like PartialPeelLoop is the cause of the crash.

I also tried these tests:
 -XX:-PartialPeelAtUnsignedTests : no crash
 -XX:+TracePartialPeeling : no crash
 -XX:-BackgroundCompilation : no crash
 -XX:+TracePartialPeeling -XX:CICompilerCount=8 : no crash

It looks like a concurrency issue in PartialPeelLoop, but I can't find a
way to reproduce it when logging is on.

Regards,
Julien


2014-04-11 11:39 GMT+02:00 Julien Nicoulaud <julien.nicoulaud at gmail.com>:

> Hi all, thanks for your answers.
>
> I did some further testing and here is the results:
>  - JDK 8u20-b05 : still crashes, but not always - logs at
> https://dl.dropboxusercontent.com/u/62997/JDK-8038985/8u20-b05.zip - core
> dump at
> https://dl.dropboxusercontent.com/u/62997/JDK-8038985/8u20-b05/core.16369.tar.bz2
>   - JDK 8u0 with -XX:-EliminateAutoBox : still crashes, but not always -
> logs at
> https://dl.dropboxusercontent.com/u/62997/JDK-8038985/8u0-no-EliminateAutoBox.zip
>
> Also, since the crash always seems to occur when inlining Trove's
> TCustomHashMap (http://trove.starlight-systems.com/) into my code, I
> tried to:
>   - Upgrade Trove from 3.0.3 to 3.1a1 (from
> https://bitbucket.org/robeden/trove/downloads) => still crashes
>  - Replace Trove by other hash map implementations => no crash
>
> Regards,
> Julien
>
>
> 2014-04-07 22:04 GMT+02:00 Christian Thalinger <
> christian.thalinger at oracle.com>:
>
>
>> On Apr 7, 2014, at 2:39 AM, Volker Simonis <volker.simonis at gmail.com>
>> wrote:
>>
>> > Really seems like a duplicate of
>> > https://bugs.openjdk.java.net/browse/JDK-8032566 "8032566: Crash in
>> > JIT when running Scala compiler (and compiling Scala std lib)"
>> > although that one should have been fixed in jdk8 b127. On the other
>> > hand it was not really fixed - just worked around by switched off
>> > autoboxing by default (you can double check with
>> > -XX:-EliminateAutoBox).
>> >
>> > Maybe the problem also appears even with autoboxing switched off?
>> >
>> > By the way, JIT replay data doesn't help without access to the involved
>> classes.
>>
>> Right.  We would need the classes extracted from the core file.  Also
>> I’ve moved the replay and hs_err data into files and attached them to the
>> bug report because the bug was unreadable before.
>>
>> >
>> > Regards,
>> > Volker
>> >
>> >
>> >
>> > On Sun, Apr 6, 2014 at 8:55 AM, Ivan Krylov <ivan at azulsystems.com>
>> wrote:
>> >> bcc: discuss at openjdk.java.net
>> >> cc: hotspot-compiler-dev at openjdk.java.net
>> >>
>> >>
>> >> Hi Julien,
>> >>
>> >> Hotspot-compiler-dev is the right alias from discussing C2 crashes. In
>> the meantime may you can try the current the latest EA build like 8u20 b05
>> [1]?
>> >>
>> >> Thanks,
>> >>
>> >> Ivan
>> >>
>> >> [1] - https://jdk8.java.net/download.html
>> >>
>> >> On 05 Apr 2014, at 13:57 , Julien Nicoulaud <
>> julien.nicoulaud at gmail.com> wrote:
>> >>
>> >>> Hi all,
>> >>>
>> >>> I don't know if it's the right place to discuss this, but here is my
>> issue :
>> >>>
>> >>> I reported a VM crash in C2 compilation, but it was closed as
>> incomplete
>> >>> with the following comment "We need a clear instructions how to
>> reproduce
>> >>> the problem".
>> >>>
>> >>> I clearly explained in the report I could not provide a simple,
>> >>> self-contained test case to reproduce the bug, and I provided hs_err
>> and
>> >>> JIT replay log file instead.
>> >>>
>> >>> So what am I supposed to do in this case ? Debug it myself ? Or are
>> there
>> >>> another options so I can provide useful debug output ?
>> >>>
>> >>> Here is the bug report:
>> >>> https://bugs.openjdk.java.net/browse/JDK-8038985
>> >>> http://bugs.java.com/bugdatabase/view_bug.do?bug_id=8038985
>> >>>
>> >>> Regards,
>> >>> Julien Nicoulaud
>> >>>
>> >>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20140604/4e3d41ad/attachment.html>