RFR: 8309130: x86_64 AVX512 intrinsics for Arrays.sort methods (int, long, float and double arrays) [v42]

[iaroslavski]

On Fri, 6 Oct 2023 18:45:59 GMT, Srinivas Vamsi Parasa <duke at openjdk.org> wrote:

>> My tier1-7 testing passed. Good.
>
>> My tier1-7 testing passed. Good.
> 
> Thank you, Vladimir!

Hi @vamsi-parasa,

If DualPivotQuicksort.java is updated, can you improve `partitionDualPivot` and `partitionSinglePivot` methods a little bit also?

We pass `int high` there and then assign `int end = high - 1;`, but later `high` is never used.
It will be better to pass `end` instead of `high`, eliminate `int end = high - 1;` and don't introduce new variable.
This is minor changes but sorting shows better performance.

Please find my suggested code:

1. Replace high -> end, update javadoc: exclusive -> inclusive

` * @param end the index of the last element, inclusive, to be sorted`
` */`
`int[] partition(A a, int low, int end, ...`

` * @param end the index of the last element, inclusive, to be sorted`
` */`
`@IntrinsicCandidate`
`@ForceInline`
`private static <A> int[] partition(Class<?> elemType, A array, long offset, int low, int end, ...`
`    return po.partition(array, low, end, pivotIndex1, pivotIndex2);`
`}`

`private static int[] partitionDualPivot(int[] a, int low, int end,`
`   int lower = low;`
`   int upper = end;`

`private static int[] partitionSinglePivot(int[] a, int low, int end,`
`   int lower = low;`
`   int upper = end;`

2. Javadoc of `int[] partition(A a, int low, )` says 'to be sorted' for paramters a, low and high.
It should be `to be partitioned`. Please correct.

3. The same for method `private static <A> int[] partition(Class<?> elemType,`, update javadoc.

4. Changes from Item 1. require correspondent changes in C++ sources, could please take care of it?

5. And changes from my previous comment: all 4 methods should have @ForceInline and all methods should be without
@ForceInline. Am I right?

I guess new PR should be opened for all mentioned changes. Could you please go ahead?
I appreciate if you help to make sorting more accurate and better.

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

PR Comment: https://git.openjdk.org/jdk/pull/14227#issuecomment-1751739561