avm1: Replace recursion with iterative qsort

[123jjck]

Fixes #18109

Speeds up sorting

var arr = [];
var i = 0;
while(i < 3000)
{
   arr.push({key:i});
   i++;
}

flash.external.ExternalInterface.call("console.log",arr[0]);

var startDate = new Date();
flash.external.ExternalInterface.call("console.log","Sorting started at: " + startDate);

arr.sortOn("key",Array.DESCENDING | Array.NUMERIC);

var endDate = new Date();
flash.external.ExternalInterface.call("console.log","Sorting ended at: " + endDate);

flash.external.ExternalInterface.call("console.log",arr[0]);

var duration = endDate.getTime() - startDate.getTime();
flash.external.ExternalInterface.call("console.log","Sorting duration (ms): " + duration);

In this example:
Recursion qsort duration: 347ms
Iterative qsort duration: 271ms

[adrian17]

Thank you! Sorry for a delay, I added some review comments. I think this looks fine, though I'm kinda surprised at the 20% speedup.

[adrian17]

> isn't normally possible, right? Can we assert that?

[123jjck]

> isn't normally possible, right?

Yes, you're right! Should I use assert! or debug_assert! for this?

[adrian17]

assert does sound nice, unless it kills the performance back.

[123jjck]

I think we can't replace the if condition with assert because low >= high can occur when the pivot partition results in a subarray with no elements. (e.g., [5, 5, 5, 5, 5])

[adrian17]

I know low == high can occur, but I meant that specifically low > high shouldn't ever occur.

[123jjck]

I know low == high can occur, but I meant that specifically low > high shouldn't ever occur.

Oh, I misunderstood then. I tried to assert! low >= high, not low > high. Then that explains why the tests failed.

[adrian17]

I think it looks good, once you update the PR I can probably quickly test and approve it tomorrow c:

[adrian17]

I don't think that assert is strictly critical, so I think this can go in as-is (let's just squash on merge).

[adrian17]

Thank you!