[vectorIntrinsics+mask] RFR: 8271313: AArch64: SVE backend support for masking operations with predicate feature

[Xiaohong Gong]

On Fri, 30 Jul 2021 04:08:23 GMT, Jie Fu <jiefu at openjdk.org> wrote:

> So does the performance gain reasonable, especially for something like IntMaxVector.AND_NOTMasked?
> Thanks.

Hi @DamonFool thanks for looking at this PR and sorry for my late reply! Yes, it seems the performance for some of the `lanewise` operations does gain too much. But from the dumped assembly codes, it actually saves two instructions inside the loop. Currently we didn't spend too much time analyzing the exact performance benefit. And I think the showed performance data can just be treated as a reference to the result of the initial masking feature support project.

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

PR: https://git.openjdk.java.net/panama-vector/pull/105