[aarch64-port-dev ] aarch64 and arm64 jdk benchmarks result sharing

Zhongwei Yao zhongwei.yao at linaro.org
Mon Apr 10 02:19:11 UTC 2017


Hi Jamsheed,
Thanks for your suggestions! I'll check that option.


On 8 April 2017 at 21:30, Jamsheed C m <jamsheed.c.m at oracle.com> wrote:

> Hi Zhongwei,
>
> ReservedCodeCacheSize of Openjdk Amr64 JDK port is 240M (by default) and
> Openjdk Aarch64 JDK port is 128M(by default), it would be good if we set
> same value for both ports for performance comparison(i.e  set both to 128M).
>
> Thank you!
>
> Best Regards,
>
> Jamsheed
>
> On 3/30/2017 7:13 AM, Zhongwei Yao wrote:
>
>> Hi,
>>    We have been running some Java benchmarks on AArch64 servers with
>> different
>> OpenJDK ports: aarch64 and arm64. You can find the results at [1].
>>    And our JDK version:
>>         - aarch64 OpenJDK: build from latest JDK9 source with config:
>>          "--with-jvm-variants=server --with-debug-level=release"
>>
>>         - arm64 OpenJDK: build from latest JDK9 source with config:
>>          "--with-jvm-variants=server --with-debug-level=release --wi
>> th-cpu-port=arm64"
>>
>>    And from the data, aarch64 OpenJDK is better than arm64 OpenJDK
>> generally, but with some exceptions. So we think it is worth to take a
>> look
>> at following cases for aarch64 OpenJDK:
>>          - Dacapo's sunflow, tradebeans and xalan
>>          - SPECjvm2008's scimark.lu.large
>>
>>    And we also find arm64 is better than aarch64 for byte/short data type
>> vectorization in our JMH cases ([2]), which we are investigating.
>>
>>    Feedbacks are welcome. Thanks!
>>
>>    [1]:
>> https://docs.google.com/spreadsheets/d/18iklOrbaL67i46XHsPTr
>> dTObWqUJJK8EO4-k0eqdLKM/edit?usp=sharing
>>    [2]: https://git.linaro.org/leg/openjdk/jmh-linaro-org.git/
>>
>>
>


-- 
Best regards,
Zhongwei


More information about the aarch64-port-dev mailing list