Question regarding 8248521: TestVerifyIterativeGVN.java is failing with timeout

[Christian Hagedorn]

Hi Goetz

You're welcome and thanks for confirming that you have not seen a 
timeout in 15 either in your testing. We only saw the stack overflow a 
few times on SPARC in 11.

Best regards,
Christian

On 10.07.20 10:46, Lindenmaier, Goetz wrote:
> Hi Christian,
> 
> Thanks for your explanation.
> I can confirm that it never timed out in 15 in our test infra.
> It might start again in case code changes again, but
> that is the risk with any test.
> (We never saw the stack overflow, though.)
> 
> Best regards,
>    Goetz.
> 
> 
>> -----Original Message-----
>> From: Christian Hagedorn <christian.hagedorn at oracle.com>
>> Sent: Friday, July 10, 2020 10:28 AM
>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; 'Fairoz Matte'
>> <fairoz.matte at oracle.com>
>> Cc: hotspot-compiler-dev at openjdk.java.net
>> Subject: Re: Question regarding 8248521: TestVerifyIterativeGVN.java is
>> failing with timeout
>>
>> Hi Goetz
>>
>> As Fairoz has mentioned, JDK-8246203 changed the algorithm slightly such
>> that it needs now more time for the verification (we check more nodes
>> than before).
>>
>> JDK-8246203 originally only happened in JDK-11 with
>> -XX:+VerifyIterativeGVN where a stack overflow crash occurred with a
>> more or less HelloWorld test (the old algorithm as a recursive one). It
>> turned out that with JDK-11 it compiled a specific big method which
>> generated quite a lot of nodes in a chain which let it crash. However,
>> with JDK-15 (and 16), this method was not compiled anymore as part of a
>> HelloWorld test. It probably got changed since JDK-11 or is not called
>> anymore when starting up. Therefore, we concluded that it must be an 11
>> only issue and just increased the timeout for the test as we have not
>> seen timing it out in JDK-15 or 16.
>>
>> Best regards,
>> Christian
>>
>> On 10.07.20 10:06, Lindenmaier, Goetz wrote:
>>> Hi Fairoz,
>>>
>>> Thanks for the info.
>>>
>>> It's still unclear to me why the algorithm takes longer in 11
>>> than in 15 ... but no matter.
>>>
>>> Best regards,
>>>     Goetz.
>>>
>>> From: Fairoz Matte <fairoz.matte at oracle.com>
>>> Sent: Friday, July 10, 2020 10:01 AM
>>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
>>> Cc: hotspot-compiler-dev at openjdk.java.net
>>> Subject: RE: Question regarding 8248521: TestVerifyIterativeGVN.java is
>> failing with timeout
>>>
>>> Hi Goetz,
>>>
>>> This issue is only applicable to 11u.
>>> After the fix of JDK-8246203,  which changed the algorithm for the
>> verification used with VerifyIterativeGVN (takes more time)
>>> We have adjusted timeout to 1200 from 600.
>>>
>>> Thanks,
>>> Fairoz
>>>
>>> From: Lindenmaier, Goetz
>> <goetz.lindenmaier at sap.com<mailto:goetz.lindenmaier at sap.com>>
>>> Sent: Friday, July 10, 2020 12:11 PM
>>> To: Fairoz Matte
>> <fairoz.matte at oracle.com<mailto:fairoz.matte at oracle.com>>
>>> Cc: hotspot-compiler-dev at openjdk.java.net<mailto:hotspot-compiler-
>> dev at openjdk.java.net>
>>> Subject: Question regarding 8248521: TestVerifyIterativeGVN.java is failing
>> with timeout
>>>
>>> Hi Fairoz,
>>>
>>> we also see this test timing out on mac. But only so with
>>> jdk11u.
>>> Do you mind sharing how you fixed this? Did you just increase
>>> the timeout, or did you figure out why this fails on mac in 11u?
>>>
>>> Thanks,
>>>     Goetz
>>>