Perf: use subtraction instead of multiple compares in compareUint8Arrays

[hildjj]

This looks to have an almost 20% performance win on my machine, mostly by removing one comparison in the tight loop.

• Used Math.sign to clamp the outputs so that the tests still pass.
• Added missing benchmark for compareUint8Arrays

Comparison of benchmark outputs on my machine:
Before: compareUint8Arrays x 479 ops/sec ±0.10% (95 runs sampled)
After: compareUint8Arrays x 574 ops/sec ±3.75% (96 runs sampled)

[sindresorhus]

Use an explicit check

[sindresorhus]

Interesting finding. I would never have guessed that this would be faster. I would have assumed that even two comparisons operators would be faster than a method call.

[hildjj]

The method call only happens when the arrays are different, and the benchmark is rigged so that never happens in the inner loop. A more fair benchmark would probably compare lots of smaller random arrays.

[hildjj]

If we are worried about the overhead of the Math.sign call, you could relax the outputs to be [negative, 0, positive], move the Math.sign call to the tests, change the docs, and this function would still work great as a sort comparison function.

[sindresorhus]

If we are worried about the overhead of the Math.sign call, you could relax the outputs to be [negative, 0, positive], move the Math.sign call to the tests, change the docs, and this function would still work great as a sort comparison function.

Nah, it's fine.

[sindresorhus]

Thanks :)