From quadratic to linear complexity.

[lordnn]

Because tidx is a monotonic sequence.

[kpchoi]

It is difficult to understand why this loop index changes can lead to reduce complexity

[lordnn]

It is difficult to understand why this loop index changes can lead to reduce complexity

while (1) is a cycle from 1 to num_tiles plus for cycle inside it (from 1 to num_tiles) - totally gives quadratic complexity
Inner loop iterations:
0
0 1
0 1 2
0 1 2 3
...
0 1 2 3 ... num_tiles - 1
QUADRATIC
After index was changed inner loop has only one iteration totally from 1 to 'num_tiles' during all outer loop iterations.
New inner loop iterations:
0
1
2
3
...
num_tiles - 1

[lordnn]

I think It can be easily understand if you imaging only one thread execution (no thread in a thread_pool).
But in last commit I have completely removed inner loop.