RFR(S) 8007270: Make IsMethodCompilable test work with tiered
Nils Eliasson
nils.eliasson at oracle.com
Tue Feb 11 13:35:22 PST 2014
cutoff may increase if the method was compiled for a lower level and we
want another attempt for full opt, the second check is to limit the
amount of attempts.
//N
On 2014-02-11 20:06, Vladimir Kozlov wrote:
> (i < PER_METHOD_RECOMPILATION_CUTOFF*2) should be always true since (i
> < cutoff):
>
> + for (long i = 0L; (i < cutoff) &&
> + (i < PER_METHOD_RECOMPILATION_CUTOFF*2) &&
>
> Thanks,
> Vladimir
>
> On 2/11/14 5:15 AM, Nils Eliasson wrote:
>> Hi,
>>
>> Thanks for your contribution Igor.
>>
>> After some additional testing I noticed some inconsistencies and
>> filed https://bugs.openjdk.java.net/browse/JDK-8034188
>>
>> This test now runs faster, and targets c2-compiles only, in server
>> and with tiered.
>>
>> Webrev: http://cr.openjdk.java.net/~neliasso/8007270/webrev.02/
>>
>> Thanks,
>> //Nils
>>
>>
>> On 2014-01-29 11:11, Igor Ignatyev wrote:
>>> Hi Nils,
>>>
>>> you can skip running on client by yourself, see attached diff.
>>> Igor
>>>
>>> On 01/28/2014 07:02 PM, Nils Eliasson wrote:
>>>> Hi all,
>>>>
>>>> I need a review for this change.
>>>>
>>>> http://cr.openjdk.java.net/~neliasso/8007270/webrev.01/
>>>>
>>>> This test was disabled since it didn't work very well with tiered (or
>>>> client). It tests the PerMethodRecompilationCutoff that was introduced
>>>> to disable c2-compilations of a method when it has been deoptimized
>>>> too
>>>> many times. The bug report suggested we should disable c1
>>>> compilations
>>>> as well but I don't think that was the intent of the cutoff feature.
>>>>
>>>> I have changed the following in the test
>>>> * skip test when running client only (not supported by jtreg at the
>>>> moment)
>>>> * check what compilation level was used when compiling so that it can
>>>> keep track of the number of c2 compiles (and deopts) correctly in
>>>> tiered
>>>> mode
>>>> * compile and deopt up to the cutoff limit only once
>>>> * added PerMethodRecompilationCutoff=4 flag to commandline to reduce
>>>> wasted time in test (default 400)
>>>>
>>>> Now the test works and the running time has been reduced to seconds
>>>> instead of minutes.
>>>>
>>>> Kind regards,
>>>> Nils Eliasson
>>>>
>>>>
>>
More information about the hotspot-compiler-dev
mailing list