[A little bit faster Arrays.parallelSort for long/int arrays]

[David Holmes]

Hi Rodion,

Please take this to core-libs-dev at openjdk.java.net.

But note that contributions must be hosted on OpenJDK services due to IP 
considerations and so cannot be picked up from external services like 
github. Please attach a text-based patch (so the mailing list doesn't 
strip it) or inline the patch directly if not too large.

Thanks,
David

On 14/06/2020 8:58 pm, Rodion Efremov wrote:
> Hello,
> 
> I have this <https://github.com/coderodde/ParallelMSDRadixSort> implementation
> of MSD radix sorts for *int/long *arrays. Both run faster than respective
> java.util.Arrays.parallelSort methods in light of my benchmarking:
> 
> seed = 1592131844387
> Warming up on long arrays...
> 
> seed = 1592131852073
> Benchmarking on long arrays...
> coderodde's Arrays.parallelSort: 845
> Java's Arrays.parallelSort     : 1119
> Java's Arrays.sort             : 4141
> true
> 
> seed = 1592131859902
> Warming up on int arrays...
> 
> seed = 1592131866867
> Benchmarking on int arrays...
> coderodde's Arrays.parallelSort: 601
> Java's Arrays.parallelSort     : 1023
> Java's Arrays.sort             : 4384
> true
> 
> (Above all figures in milliseconds while sorting (random) arrays of length
> 50 million array components. Also, warming arrays are 50 million components
> as well.)
> 
> Do we need the above implementations of Arrays.parallelSort in JDK?
> 
> Best regards,
> Rodion E.
>