RFR: 8165661: SPECjvm2008-crypto.signverify regression in 9-b105

[Vladimir Kozlov]

On 10/20/16 11:24 AM, Rahul Raghavan wrote:
> Hi,
>
> Please review following patch for JDK- 8165661.
> <webrev.00> :  http://cr.openjdk.java.net/~rraghavan/8165661/webrev.00/
>
> 1. <bug details> : https://bugs.openjdk.java.net/browse/JDK-8165661
> Performance regression reported with Internal run benchmark harness (SPECjvm2008-crypto.signverify) for x86_64 with JDK9-b105 build onwards.

Bug is 'confidential' and not visible outside. but it simple says that 
2% regression was found.

>
> 2. Found following change done as sub-part of old JDK-8145322 Enhancement is the root cause of the reported Performance regression -
>         [the (CastII (AddI x const)) -> (AddI (CastII x) const) transformation change block in CastIINode::Ideal() - castnode.cpp]
>
>       (https://bugs.openjdk.java.net/browse/JDK-8145322 - Code generated from unsafe loops can be slightly improved)
>       (http://hg.openjdk.java.net/jdk9/jdk9/hotspot/rev/c8b709902e0e)
>       (http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2015-December/020381.html)
>
> 3. Understood there was no performance numbers with old 8145322 fix.
> But now with proposed <webrev.00> change (reverting old sub part of 8145322),
>   got no jprt functional failures, no other performance regressions and confirmed reported signverify regression is solved.

Good.

>
> 4. So to go forward as Roland W suggested, request for review and approval of this proposed <webrev.00> to close this 8165661.
> Afterwards I will create a  new separate enhancement CR to investigate
>   why this reverted old 8145322 part caused issues and for a possible better version of the same.

Agree.

Thanks,
Vladimir

>
>
> Thanks,
> Rahul
>