[11u] RFR 8245512: CRC32 optimization using AVX512 instructions
Doerr, Martin
martin.doerr at sap.com
Thu Jan 7 10:26:23 UTC 2021
Hello everybody,
I think 2x performance improvement for CRC32 computations is a good argument. The original change is tested for quite some time and didn't cause follow-up issues. Backport is simple and was already reviewed.
So I'd vote for taking it after running it through testing.
Best regards,
Martin
> -----Original Message-----
> From: Viswanathan, Sandhya <sandhya.viswanathan at intel.com>
> Sent: Mittwoch, 6. Januar 2021 19:49
> To: Andrew Haley <aph at redhat.com>; Langer, Christoph
> <christoph.langer at sap.com>; Doerr, Martin <martin.doerr at sap.com>;
> Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
> Cc: Vladimir Kozlov <vladimir.kozlov at oracle.com>; jdk-updates-
> dev at openjdk.java.net; hotspot-compiler-dev at openjdk.java.net
> Subject: RE: [11u] RFR 8245512: CRC32 optimization using AVX512 instructions
>
> Hi Andrew,
>
> I understand the dilemma.
> We see ~2x performance gain which is significant, and the patch is very self-
> contained with no impact on other architectures.
> CRC calculation is prevalent in big Java frameworks, so the request here.
> I hope exception can be made for this request.
>
> This is my first performance backport request for JDK 11. We do understand
> the balance that you need to maintain in accepting such requests.
>
> Best Regards,
> Sandhya
>
> -----Original Message-----
> From: Andrew Haley <aph at redhat.com>
> Sent: Wednesday, January 06, 2021 1:57 AM
> To: Langer, Christoph <christoph.langer at sap.com>; Viswanathan, Sandhya
> <sandhya.viswanathan at intel.com>; Doerr, Martin
> <martin.doerr at sap.com>; Lindenmaier, Goetz
> <goetz.lindenmaier at sap.com>
> Cc: Vladimir Kozlov <vladimir.kozlov at oracle.com>; jdk-updates-
> dev at openjdk.java.net; hotspot-compiler-dev at openjdk.java.net
> Subject: Re: [11u] RFR 8245512: CRC32 optimization using AVX512 instructions
>
> On 12/23/20 1:09 PM, Langer, Christoph wrote:
> >
> > as we should be cautious about backports to an LTS release if something is
> not just a bugfix, I'm not sure whether this backport for a performance
> enhancement is good to go. Generally, if it can be proven that it's really just a
> potential performance improvement without side effects to other scenarios,
> I would be fine with it. But as I'm not an expert in this area, I'd prefer if it
> there was some further endorsement (or disapproval)...
> >
> > @Andrew Haley; @Lindenmaier, Goetz; @Doerr, Martin, what's your
> assessment?
>
> We could backport for a significant performance bottleneck, but I'm leaning
> toward the idea that JDK 11 is done: we're keeping it going by fixing bugs and
> adapting to changing environments, but that should be all.
>
> However, there are exceptions to every rule!
>
> There is a social problem in that maintainers get very demotivated because
> non-LTS releases are "never used" so people's work doesn't appear in
> production for years. I'm not sure exactly how to address that, except to say
> that using the OpenJDK release stream shouldn't be so very scary.
>
> --
> Andrew Haley (he/him)
> Java Platform Lead Engineer
> Red Hat UK Ltd. <https://www.redhat.com>
> https://keybase.io/andrewhaley
> EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671
More information about the jdk-updates-dev
mailing list