[11] RTM tests fail
Gustavo Romero
gromero at linux.vnet.ibm.com
Fri Aug 31 15:53:23 UTC 2018
Hi Goetz,
On 08/31/2018 04:22 AM, Lindenmaier, Goetz wrote:
> Hi Gustavo,
>
> I'm fine with having a third property. Your description makes sense.
>
> But the property is true when the compiler does not implement rtm,
> e.g., on solarissparc. Why don't you implement something like
> boolean isRTMCompiler = false
> if (!isGraalEnabled) && (ppc || x86) isRTMCompiler = true
Well... it looks like that when my mind was exercising behind the
Power / ARM / X86 microcosm something has gone haywire. Thanks a lot
for reviewing it. I fixed that and sent a RFR:
http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2018-August/030398.html
Best regards,
Gustavo
> Best regards,
> Goetz.
>
>> -----Original Message-----
>
>> From: Gustavo Romero <gromero at linux.vnet.ibm.com>
>> Sent: Donnerstag, 30. August 2018 00:28
>> To: Christian Thalinger <cthalinger at twitter.com>; Lindenmaier, Goetz
>> <goetz.lindenmaier at sap.com>
>> Cc: vladimir.kozlov at oracle.com; hotspot compiler <hotspot-compiler-
>> dev at openjdk.java.net>
>> Subject: Re: [11] RTM tests fail
>>
>> Hi Götz, Christian,
>>
>> Cristian, I don't know how easy it's to trigger an ad hoc test on your system to
>> test the RTM suite against a Graal JVM, but I'm wondering if you could kindly
>> test
>> a fix for JDK-8209972:
>>
>> http://cr.openjdk.java.net/~gromero/misc/8209972_GRAAL_dont_run_RTM
>> _tests_with_Graal.patch
>>
>>
>> Götz, I've tried to fix that in vm.rtm.cpu property but it got cumbersome
>> around
>> TestUseRTMLockingOptionOnUnsupportedCPU.java. There, vm.rtm.cpu
>> would be false
>> for Graal JVM and so the test runs and expects some failure message from
>> the JVM
>> regarding unsupported CPU or OS, however the problem is rather that the
>> compiler
>> itself doesn't support RTM. I could adapt that test but tweaking it locally
>> seems odd to me. Also I think that the other tests (*Unsupported*.java) are
>> irrelevant when the selected compiler does not support RTM (please see the
>> commit
>> message where I try to explain it with other words). What do you think?
>>
>> Thanks.
>>
>>
>> Best regards,
>> Gustavo
>>
>> On 08/29/2018 10:52 AM, Gustavo Romero wrote:
>>> Hi Christian,
>>>
>>> A quick question:
>>>
>>> What's the way you recommend to run the compiler/rtm tests with Graal
>> JVM?
>>>
>>> I'm using something like that but I would like know if there is a better way
>> to
>>> do that. I don't want to run all the tests, only the RTM ones, so I would like
>>> to avoid 'make hostpot-*':
>>>
>>> JT_JAVA=/usr/lib/jvm/java-8-openjdk-amd64
>> /home/gromero/jtreg/bin/jtreg -javaoption:-
>> XX:+UnlockExperimentalVMOptions -javaoption:-XX:+EnableJVMCI -
>> javaoption:-XX:+UseJVMCICompiler -v1 -conc:6 -jdk:./build/linux-x86_64-
>> normal-server-release/jdk/ ./test/hotspot/jtreg/compiler/rtm
>>>
>>> Vladimir, thanks for filing JDK-8209972.
>>>
>>> Thanks.
>>>
>>> Regards,
>>> Gustavo
>>>
>>> On 08/24/2018 10:53 AM, Gustavo Romero wrote:
>>>> Hi Goetz,
>>>>
>>>> On 08/24/2018 04:44 AM, Lindenmaier, Goetz wrote:
>>>>> Hi Gustavo,
>>>>>
>>>>>> If it's possible would that be the best solution in this case?
>>>>> Yes, I think so. It's the most local place to change this.
>>>>>
>>>>> But maybe VMProps works fine, too. Then this is the better place.
>>>>> I just don't know whether all tests properly check that.
>>>>>
>>>>> In general, I think such information should go into the
>>>>> test suite, not into the testee... kind of strange to adapt a testee
>>>>> so that tests pass. There should be a single point in the test suite
>>>>> where it is listed which VM configuration supports which features.
>>>>> For a while such things were added to Platform.java, but now
>>>>> there is VMProps ... and there are base classes in the test sub-suites
>>>>> that try to do this ...
>>>>> Some tests then again check for the platform themselves.
>>>>>
>>>>> But basically I thought the CPU feature string is there to
>>>>> report about the actual CPU we are running on, not to
>>>>> tell what the VM exploits of these features.
>>>>
>>>> Sure, that makes total sense. So much that the CPU feature string is
>>>> determined very early in the JVM creation.
>>>>
>>>> Thanks.
>>>>
>>>> Best regards,
>>>> Gustavo
>>>>> Best regards,
>>>>> Goetz.
>>>>>
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Gustavo Romero <gromero at linux.vnet.ibm.com>
>>>>>> Sent: Dienstag, 21. August 2018 15:04
>>>>>> To: Christian Thalinger <cthalinger at twitter.com>; Lindenmaier, Goetz
>>>>>> <goetz.lindenmaier at sap.com>
>>>>>> Cc: hotspot compiler <hotspot-compiler-dev at openjdk.java.net>
>>>>>> Subject: Re: [11] RTM tests fail
>>>>>>
>>>>>> Hi Christian, Goetz,
>>>>>>
>>>>>> On 08/21/2018 08:02 AM, Christian Thalinger wrote:
>>>>>>>
>>>>>>>
>>>>>>>> On Aug 20, 2018, at 5:44 PM, Christian Thalinger
>> <cthalinger at twitter.com>
>>>>>> wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>> On Aug 20, 2018, at 5:02 PM, Lindenmaier, Goetz
>>>>>> <goetz.lindenmaier at sap.com> wrote:
>>>>>>>>>
>>>>>>>>> Hi
>>>>>>>>>
>>>>>>>>> I think it would make more sense to add this in VMProps.java when
>>>>>>>>> vm.rtm.cpu is evaluated.
>>>>>>>>> Or to adapt the CPU feature string not to report rtm. While I think
>>>>>>>>> this is crude, I have seen (and then also done) this before.
>>>>>>>>>
>>>>>>>>> Adding !vm.graal.enabled to each test will require changing this
>>>>>>>>> in all tests when, e.g., RTM support is implemented in graal x86, but
>>>>>>>>> not in graal ppc …
>>>>>>>>
>>>>>>>> Good point regarding different architectures.
>>>>>>>
>>>>>>> Anyone want to pick this up? :-)
>>>>>>
>>>>>> Yes, I can, although not promptly. Is it ok to wait a week or so?
>>>>>>
>>>>>> Also it's been quite some time that I don't check the Graal code...
>>>>>>
>>>>>> Goetz, would it be reasonable to handle this at the JVM side? For
>> instance,
>>>>>> would it be possible (and correct) to keep separate RTM feature status
>> for
>>>>>> JVM + Hotspot and for JVM + Graal for each architecture so for a given
>> arch it
>>>>>> would be possible to advertise "rtm" for JVM + Hotspot but not for
>> JVM +
>>>>>> Graal?
>>>>>>
>>>>>> If it's possible would that be the best solution in this case?
>>>>>>
>>>>>> Thanks.
>>>>>>
>>>>>>
>>>>>> Bbest regards,
>>>>>> Gustavo
>>>>>>
>>>>>>>>
>>>>>>>>>
>>>>>>>>> Best regards,
>>>>>>>>> Goetz.
>>>>>>>>>
>>>>>>>>>> -----Original Message-----
>>>>>>>>>> From: Christian Thalinger <cthalinger at twitter.com>
>>>>>>>>>> Sent: Montag, 20. August 2018 14:55
>>>>>>>>>> To: Gustavo Romero <gromero at linux.vnet.ibm.com>
>>>>>>>>>> Cc: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; hotspot
>>>>>> compiler
>>>>>>>>>> <hotspot-compiler-dev at openjdk.java.net>
>>>>>>>>>> Subject: Re: [11] RTM tests fail
>>>>>>>>>>
>>>>>>>>>> Quick additional question: Graal does not implement RTM (as far
>> as I
>>>>>> know)
>>>>>>>>>> but the tests are always run and I don’t see a:
>>>>>>>>>>
>>>>>>>>>> @requires !vm.graal.enabled
>>>>>>>>>>
>>>>>>>>>> in the test files. Should we add that?
>>>>>>>>>>
>>>>>>>>>>> On Jul 11, 2018, at 4:17 PM, Gustavo Romero
>>>>>>>>>> <gromero at linux.vnet.ibm.com> wrote:
>>>>>>>>>>>
>>>>>>>>>>> Hi Christian,
>>>>>>>>>>>
>>>>>>>>>>> On 07/11/2018 11:10 AM, Christian Thalinger wrote:
>>>>>>>>>>>>> On Jun 29, 2018, at 11:23 AM, Gustavo Romero
>>>>>>>>>> <gromero at linux.vnet.ibm.com> wrote:
>>>>>>>>>>>>>
>>>>>>>>>>>>> Hi Christian,
>>>>>>>>>>>>>
>>>>>>>>>>>>> On 06/29/2018 12:16 PM, Christian Thalinger wrote:
>>>>>>>>>>>>>>> On Jun 25, 2018, at 5:32 PM, Gustavo Romero
>>>>>>>>>> <gromero at linux.vnet.ibm.com
>>>>>> <mailto:gromero at linux.vnet.ibm.com>>
>>>>>>>>>> wrote:
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Hi Christian,
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> On 06/25/2018 09:54 AM, Christian Thalinger wrote:
>>>>>>>>>>>>>>>>> On Jun 25, 2018, at 8:49 AM, Gustavo Romero
>>>>>>>>>> <gromero at linux.vnet.ibm.com
>>>>>> <mailto:gromero at linux.vnet.ibm.com>>
>>>>>>>>>> wrote:
>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>> On 06/25/2018 09:46 AM, Lindenmaier, Goetz wrote:
>>>>>>>>>>>>>>>>>>>> Did you check with or without these fixes?
>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>> Without, unfortunately. Are all of the failures fixed
>> now (in
>>>>>> jdk-
>>>>>>>>>> 11+19)?
>>>>>>>>>>>>>>>>>> I don't know, our machines do not have RTM, only our
>>>>>> Power
>>>>>>>>>> ones do.
>>>>>>>>>>>>>>>>>> But I think Gustavo Romero from IBM claimed so.
>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>> Yup, after the three fixes Goetz mentioned all RTM tests
>> must
>>>>>> pass
>>>>>>>>>> on Intel.
>>>>>>>>>>>>>>>> Ok, I’ll get back to you…
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Yes, please let me know if all went fine :)
>>>>>>>>>>>>>> I did not forget but we can’t merge in jdk-11+19 because of
>>>>>>>>>> https://bugs.openjdk.java.net/browse/JDK-8205616
>>>>>>>>>>>>>> We have to wait until jdk-11+20 is tagged.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Sure!
>>>>>>>>>>>> I’m happy to report that all RTM test failures are fixed. Thanks!
>>>>>>>>>>> Thanks for testing & confirming it. :)
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Regards,
>>>>>>>>>>> Gustavo
>>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>
>>>>
>>>
>
More information about the hotspot-compiler-dev
mailing list