RFR: 8309130: x86_64 AVX512 intrinsics for Arrays.sort methods (int, long, float and double arrays) [v40]

[iaroslavski]

On Wed, 20 Sep 2023 22:46:16 GMT, Srinivas Vamsi Parasa <duke at openjdk.org> wrote:

>> Srinivas Vamsi Parasa has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   change variable names of indexPivot* to pivotIndex*
>
> Hi Vladimir,
> 
> Just trying to understand: is there a reason to use `DualPivotQuicksort_RadixForParallel.java` and `DualPivotQuicksort_RadixForAll.java`?
> 
> Would it not be sufficient to do the following two runs:
> 
> 1. Baseline (Stock JDK) vs. AVX512 sort for` sort() `and `parallelSort()` ?
> 2. AVX512 sort vs. Radix sort for `sort()` and `parallelSort()` ?
> 
>> [1] current implementation in JDK [2] your AVX12 version based on [1], from this PR [3] my new version with Radix sort for parallel case plus your AVX12 changes [4] my new version with Radix sort for all cases plus your AVX12 changes 
> 
> Thanks,
> Vamsi

Hi @vamsi-parasa , @sviswa7 and @jatin-bhateja !

I agree with you that bot PRs are independent and radix sort discussion should not be here, my fault. I asked to run benchmarking because actual AVX512 changes are in C++ code and I don't have ready environment for testing.
I'm moving this discussion to PR https://github.com/openjdk/jdk/pull/13568.

Many thanks,
Vladimir

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

PR Comment: https://git.openjdk.org/jdk/pull/14227#issuecomment-1728996248