I have compared binary_search before this PR and after, and my own logics.
The PR says that it can process &[usize; 1000] in 15ns and three times faster than old logic. However when i tried same logic, it tooks about 40ns and improvement from the old logic was negligible.
I also cloned the master branch and tried x.py bench, and it tooks 60ns!
Why this happen?
Something is wrong with me? Updates on compiler breaked some criteria?