[vectorIntrinsics] RFR: 8285281: [x86] Add C2 mid-end and back-end implementation for COMPRESS_BITS and EXPAND_BITS operations [v2]

[Jatin Bhateja]

On Thu, 21 Apr 2022 15:14:09 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:

> Is it worth the additional complexity in C2 over just depending on the fallback? Actually I misunderstood, you are making intrinsic the scalar implementations. I think that should be a PR against the jdk repository.

I initially tried doing this in Java side, by bringing the fall back implementation to forefront and basing that over lane/withLane and scalar compress/expand post intrinsification, but C2 based inline expansion generated better code and almost 2X the performance.  I can rebase this patch on JDK mainline as suggested.

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

PR: https://git.openjdk.java.net/panama-vector/pull/195