The current hasNegatives intrinsic answers false if a byte[] only has positive bytes and true otherwise. Being aggressively vectorized this is a good foundation for a very quick test for whether a String or byte[] is ASCII-only.

However such a test doesn't help and might even be costly for strings that have a lot of leading ASCII, since we'll have to scan the same data twice (and the second time around in a slow, non-vectorized loop).

If we were to instead calculate the number of leading positive bytes we can do a quick arraycopy of the bytes known to be ASCII-only. This speeds up a number of operations, and in the case of encoding latin1 strings might even reduce allotation pressure. I've prototyped this on x64, with promising results:

https://jmh.morethan.io/?gists=428b487e92e3e47ccb7f169501600a88,3c585de7435506d3a3bdb32160fe8904

Draft PR: https://github.com/openjdk/jdk/pull/7231