[9] RFR (L): 8059623: JEP-JDK-8043304: Test task: command line options tests
Christian Thalinger
christian.thalinger at oracle.com
Thu Dec 18 22:19:30 UTC 2014
Looks good.
> On Dec 10, 2014, at 4:39 AM, Tobias Hartmann <tobias.hartmann at oracle.com> wrote:
>
> Hi Filipp,
>
> On 10.12.2014 13:21, Filipp Zhinkin wrote:
>> Tobias,
>>
>> as I wrote you in a private message, until a fix for 8064940 doesn't affect
>> the way options are processed there is no need to update proposed tests.
>>
>> I've filed 8067135 for new tests that will verify actual alignment of code heaps.
>
> Okay, sounds good.
>
> Thanks,
> Tobias
>
>>
>> Regards,
>> Filipp.
>>
>> On 12/08/2014 06:18 PM, Filipp Zhinkin wrote:
>>> On 12/08/2014 06:12 PM, Tobias Hartmann wrote:
>>>> On 08.12.2014 13:25, Filipp Zhinkin wrote:
>>>>> Hi Tobias,
>>>>>
>>>>> thank you for suggestion.
>>>>>
>>>>> Yes, I think we should take code heap size alignment into account.
>>>>>
>>>>> What alignment policy you're going to implement for 8064940?
>>>> My current fix just large-page-aligns the code heap sizes.
>>>>
>>>>> Maybe instead of checking that values are in
>>>>> (value - page_size, value + page_size) interval we should just check
>>>>> that all values were aligned up to page_size?
>>>> Yes, that's a better solution. However, I don't know how to figure out the
>>>> available page sizes from Java code.
>>> There's Unsafe::pageSize() method. Also, I saw a RFR on hs-rt list
>>> about to add such method to WB API, but it need to check how well
>>> it is going to work with large pages.
>>>
>>> Thanks,
>>> Filipp.
>>>>
>>>> Best,
>>>> Tobias
>>>>
>>>>> Thanks,
>>>>> Filipp.
>>>>>
>>>>> On 12/08/2014 12:37 PM, Tobias Hartmann wrote:
>>>>>> Hi Filipp,
>>>>>>
>>>>>> the actual size of a code heap is affected by alignment and therefore may be
>>>>>> different to the size set via the command line. For example, on Sparc we
>>>>>> have to
>>>>>> make sure that the code heaps are large page (4MB) aligned to reduce the
>>>>>> number
>>>>>> of ITLB misses (will be introduced with [1]).
>>>>>>
>>>>>> Maybe we should check if the actual size of the code heap is within
>>>>>> boundaries,
>>>>>> i.e., within the specified size +- (large) page size.
>>>>>>
>>>>>> Thanks,
>>>>>> Tobias
>>>>>>
>>>>>> [1] https://bugs.openjdk.java.net/browse/JDK-8064940
>>>>>>
>>>>>> On 05.12.2014 18:06, Filipp Zhinkin wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> please take a look at CLI tests for segmented code cache (JDK-8059623).
>>>>>>>
>>>>>>> There are three new tests:
>>>>>>> compiler/codecache/cli/
>>>>>>> codeheapsize/TestCodeHeapSizeOptions
>>>>>>> printcodecache/TestPrintCodeCacheOption
>>>>>>> TestSegmentedCodeCacheOption
>>>>>>>
>>>>>>> All tests consist of several test cases aimed to verify different aspects
>>>>>>> of options' processing.
>>>>>>>
>>>>>>> These tests partially overlapped with c/c/CheckSegmentedCodeCache test,
>>>>>>> but add additional value - these tests actually check final values
>>>>>>> of tested options and verifies PrintCodeCache output.
>>>>>>>
>>>>>>> Bug id: https://bugs.openjdk.java.net/browse/JDK-8059623
>>>>>>> Webrev: http://cr.openjdk.java.net/~fzhinkin/8059623/webrev.00/
>>>>>>> Testing: manual & automated
>>>>>>>
>>>>>>> This change depends on:
>>>>>>> https://bugs.openjdk.java.net/browse/JDK-8054892: Improve compiler's CLI
>>>>>>> tests
>>>>>>> error reporting
>>>>>>> https://bugs.openjdk.java.net/browse/JDK-8066440: Various changes in
>>>>>>> testlibrary
>>>>>>> for JDK-8059613
>>>>>>>
>>>>>>> Thanks
>>>>>>> Filipp.
>>>
>>
More information about the hotspot-compiler-dev
mailing list