Fix Issue 15553 - topN very inefficient [slower than sort, even for topN(0)] but should be O(n)

[andralex]

• Replace random selection with getPivot, which is also used by sort.
• Use a local copy instead of swapping range elements around when possible
• Special-case n==0, which may be either requested from the beginning or occur naturally toward the end of the search.

[andralex]

While at it I've also improved performance of partition (which topN uses) to use indexing for arrays.

[andralex]

Ugh, I ran make std.algorithm.sorting.test instead of make std/algorithm/sorting.test and it looked like it's working. We should improve our pattern matching...

[andralex]

minPos is now also faster for certain ranges.

[JackStouffer]

We should improve our pattern matching...

You mean humans'? ;)

[9il]

LGTM

---

For ndslice example (see also update #3925)

• 3x3 blocks: the new topN 90% faster than old topN and 10% slower than sort
• 7x7 blocks: the new topN 110% faster than old topN and 10% faster than sort

[andralex]

@9il I just invited you to the team, I guess passing a library through the review process was testing enough. You'll have the right to pull if you accept the invite. Thanks!

[9il]

@andralex Accepted. Thank you!

[andralex]

ping - including @9il :)

[9il]

ping - including @9il :)

How can I turn on AutoMerge next time?

[andralex]

@9il click on the "details" link by the "auto-tester", create an account, and then you can use the "Toggle" link in there.