[vectorIntrinsics+compress] RFR: 8274839: Intrinsic support for compress & expand VectorAPI

[Jatin Bhateja]

On Thu, 7 Oct 2021 10:34:50 GMT, Joshua Zhu <jzhu at openjdk.org> wrote:

> > Hi @JoshuaZhuwj , Thanks for posting the patches, I have added the complete implementation for intrinsification of these APIs and X86 backend implementation for AVX512 and have added you as co-contributor. Please review and share your feedback on following pull-request. #146
> 
> @jatin-bhateja I prefer to seperate the change into smaller patches for others' easy review instead of a large commit. I suggest you could submit based on my changes so that I could easily see the changes and improvements your patch brings.

Hi @JoshuaZhuwj,

In PR146 we have added:-
1) Java side changes for common C2 intrinsic entry point which handles both the flavors of expand/compress APIs i.e.  Vector.compress/expand(m) and Vector.compress/expand(m, v).
2) Inline expansion and common graph creation routine for compress/expand.
3) Selector side changes for compress/expand.
4) X86 backend assembler extensions.
5) Tested patch at different AVX levels 0,1,2,3,KNL though tier3-4 regressions. 
6) Compiler changes uses consistent nomenclature which is based on new expand/compress APIs.

Common portion with this patch is 2) 4) and part of 3) (compress change) 

As suggested I can create an incremental PR with changes 1, 3(expand matcher changes) and 6) once you commit this patch.
OR 
We can jointly push PR 146 if you can kindly review it.

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

PR: https://git.openjdk.java.net/panama-vector/pull/145