[vectorIntrinsics+compress] RFR: 8276083: Incremental patch to further optimize new compress/expand APIs over X86 [v2]

[Jatin Bhateja]

On Mon, 1 Nov 2021 22:19:46 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:

>> How did the perf tests get generated? I don't see any related changes to see how they would be generated.
>
>> @PaulSandoz , your comments are addressed. Please let me know if there are further comments.
> 
> Did you run `gen-tests.sh`? 
> 
> I tried running that on a clone of your branch and it generates more code, some of which does not compile:
> 
> 1. `maskCompress` iterates over a non-existent array `a`.
> 2. Benchmarks for ROR and ROL are generated and the scalar benchmarks refer to non-existent methods `ROR_scalar` and `ROL_scalar`.
> 3. The gather/scatter benchmarks are deleted. (They got removed when i switched the scatter/gather unit tests over to the load/store files. See https://github.com/openjdk/jdk17/pull/48/) 
> 
> So it looks like we got out of sync, likely via merges from mainline. We should at least fix 2 and possibly reconsider how to do 3.

> > @PaulSandoz , your comments are addressed. Please let me know if there are further comments.
> 
> Did you run `gen-tests.sh`?
> 
> I tried running that on a clone of your branch and it generates more code, some of which does not compile:
> 
> 1. `maskCompress` iterates over a non-existent array `a`.
> 2. Benchmarks for ROR and ROL are generated and the scalar benchmarks refer to non-existent methods `ROR_scalar` and `ROL_scalar`.
> 3. The gather/scatter benchmarks are deleted. (They got removed when i switched the scatter/gather unit tests over to the load/store files. See [8266518: Refactor and expand scatter/gather tests jdk17#48](https://github.com/openjdk/jdk17/pull/48))
> 
> So it looks like we got out of sync, likely via merges from mainline. We should at least fix 2 and possibly reconsider how to do 3.

Hi @PaulSandoz ,
I have fixed 1 it was a typo,  as already mentioned I did some manual editing since existing generation is broken.

I will post the fixes for items 2 and 3 in a immediate subsequent patch once this gets integrated, changes in this patch are related to compress/expand and its performance improvements.

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

PR: https://git.openjdk.java.net/panama-vector/pull/157