`limit_diff` used to apply to each thread, so you would get NUM_THREADS*DIFF_LIMIT diffs, potentially much higher than expected. This new implementation can still overshoot the requested limit, but has a much tighter bound.
`limit_diff` used to apply to each thread, so you would get NUM_THREADS*DIFF_LIMIT diffs, potentially much higher than expected. This new implementation can still overshoot the requested limit, but has a much tighter bound.