[jdk11u-dev] RFR: 8271567: AArch64: AES Galois CounterMode (GCM) interleaved implementation using vector instructions [v2]

[Volker Simonis]

On Fri, 24 Sep 2021 12:59:23 GMT, Andrew Haley <aph at openjdk.org> wrote:

>> This is a backport of the big AES/GCM patch from JDK head. It's a
>> major change and it's had very little time (almost a day) to mature in
>> head, so perhaps it shouldn't be backported to 11 for some time; I
>> wouldn't be at all surprised if some reviewers' reaction was "What
>> have you been smoking?" However, there is a good reason for a
>> backport: OpenJDK on x86 has a major advantage. AES/GCM is an
>> important cipher, the current AArch64 implementation is much slower
>> than x86, and some workloads are severely impacted.
>> 
>> I'm open to all arguments about why this should or shouldn't be pushed,
>> and I'm quite happy to wait for another release cycle or two if people
>> think that's the best course of action.
>
> Andrew Haley has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Whitespace

I have no problem with you smoking whatsoever you like :)

But more serious. This isn't a full review but from a quick check it looks like this feature can easily turned off with `-XX:-UseAESCTRIntrinsics`, right? So in that case I have no objections of downporting this instantly to 11u. We might even be ultra-conservative and make `-XX:-UseAESCTRIntrinsics` the default in the downport, but I'm not sure it that's necessary.

Thanks for implementing this intrinsic and dowporting to 11u. It closes a real gap between x64 and aarch64,
Volker

-------------

PR: https://git.openjdk.java.net/jdk11u-dev/pull/410