[jira] [Commented] (RNG-80) SplitMix64 to natively generate nextInt

["Alex D Herbert (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/RNG-80?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16784400#comment-16784400 ] 

Alex D Herbert commented on RNG-80:
-----------------------------------

I created a new simple class that implements the methods shown above. These match those found in {{SplittableRandom}}.

Here are some results:
||method||Method||Score||Error||Median||
|cached|nextInt|279.09|5.07|279.66|
|cached|nextLong|267.71|3.23|267.30|
|native|nextInt|269.01|0.84|269.07|
|native|nextLong|268.93|0.92|269.02|

The {{nextLong}} is there to show the new class works as expected for generating {{long}} values. It is the same speed.

The results show that the cache is working for the {{SplitMix}} algorithm. So the overhead of storing and checking the cache outperforms generating new {{int}} values.

 

> SplitMix64 to natively generate nextInt
> ---------------------------------------
>
>                 Key: RNG-80
>                 URL: https://issues.apache.org/jira/browse/RNG-80
>             Project: Commons RNG
>          Issue Type: Improvement
>          Components: core
>    Affects Versions: 1.3
>            Reporter: Alex D Herbert
>            Assignee: Alex D Herbert
>            Priority: Minor
>
> The {{SplitMix64}} currently uses the {{LongProvider}} implementation of {{nextInt}} to split the {{long}} into two {{int}} values.
> However it is possible to generate {{int}} values using a variant of the algorithm which uses a different mixing function:
> {code:java}
> /**
>  * Computes Stafford variant 13 of 64bit mix function.
>  *
>  * @return the long
>  */
> public final long nextLong() {
>     long z = state += 0x9e3779b97f4a7c15L;
>     z = (z ^ (z >>> 30)) * 0xbf58476d1ce4e5b9L;
>     z = (z ^ (z >>> 27)) * 0x94d049bb133111ebL;
>     return z ^ (z >>> 31);
> }
> /**
>  * Returns the 32 high bits of Stafford variant 4 mix64 function as int.
>  *
>  * @return the int
>  */
> public final int nextInt() {
>     long z = state += 0x9e3779b97f4a7c15L;
>     z = (z ^ (z >>> 33)) * 0x62a9d9ed799705f5L;
>     return (int)(((z ^ (z >>> 28)) * 0xcb24d0a5c88c35b3L) >>> 32);
> }
> {code}
> This variant is used in {{java.util.SplittableRandom}}. It changes the second shift operation and omits a xor operation.
> A benchmark should be made to test if the variant is faster than the current method to cache the long and split it into two values. 
> Note that the investigation of the speed of caching was performed in [RNG-57|https://issues.apache.org/jira/browse/RNG-57]. The first entry on this post shows the cache only marginally improves (5%) the speed of the {[SplitMix64}} generator. Updating the native generation using a faster algorithm with less mixing may outperform the cache.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)