[aarch64-port-dev ] Outstanding patches to review

[Paul Sandoz]

Hi,

I think you have the gist of it regarding vectorizedMismatch. It requires some cross platform work in the intrinsics and Java code, and likely requires some performance analysis to ensure the thresholds are correct. In addition of course to an arm impl. My hope was we could opportunistically get more bang for the buck in the arm work but I would not want to derail any plans for you to make progress in the near term.

I am happy to help regarding the cross platform work and performance analysis. Perhaps in parallel I should prototype something to help move this forward?

Paul.

> On May 15, 2018, at 09:18, Dmitrij Pochepko <dmitrij.pochepko at bell-sw.com> wrote:
> 
> 
> 
>> On 15.05.2018 19:09, Andrew Haley wrote:
>>> On 05/15/2018 04:07 PM, Dmitrij Pochepko wrote:
>>>> On 15.05.2018 16:12, Andrew Haley wrote:
>>>> 
>>>> I'm looking at
>>>> 
>>>> 8189112 - AARCH64: optimize StringUTF16 compress intrinsic
>>>> 
>>>> 8189113: AARCH64: StringLatin1 inflate intrinsic doesn't use prefetch instruction
>>>> 
>>>> I'm not looking at
>>>> 
>>>> AARCH64: optimize string compare intrinsic
>>>> 
>>>> because it's still in progress.
>>> I don't think it's in progress.
>> Really?  No-one replied to
>> 
>> http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2018-May/028979.html
>> 
> I think conclusion here was that it needs additional measurements while trying different approaches for vectorizedMismatch implementation(and/or changes in only existing x86 implementation). And separate CR was filed to cover this work: https://bugs.openjdk.java.net/browse/JDK-8202783 because it might require trying different approaches and changes in common java part to achieve maximum performance. And Paul stated, that it's still unknown which possible direction to take for vectorizedMismatch.
> 
> Paul, do I understand it right?
> 
> Thanks,
> Dmitrij