faster `HashCommon.nextPowerOf2`

[richardstartin]

The integer version of this method showed up in a profile, and I was curious enough to look at its implementation. This provides a slightly faster implementation of HashCommon.nextPowerOf2 which takes advantage of numberOfLeadingZeros intrinsics.

I wrote a very simple benchmark to justify the change. While it shows that the existing code is hardly worth optimising, the new code is indeed faster:

Benchmark                     Mode  Cnt  Score   Error  Units
NextPowerOf2Benchmark.after   avgt    5  0.553 ± 0.007  ns/op
NextPowerOf2Benchmark.before  avgt    5  0.916 ± 0.002  ns/op

[richardstartin]

@vigna Since there doesn't appear to be CI running on this change, I feel it's worth mentioning I've run the test suite and everything passed.

[vigna]

Very interesting how it handles the case 0. OK, but, just out of curiosity, did you try to benchmark with GraalVM?

[richardstartin]

Very interesting how it handles the case 0. OK, but, just out of curiosity, did you try to benchmark with GraalVM?

I did not, but I can try that and let you know.

[richardstartin]

I got roughly the same scores (JDK 17.0.5, OpenJDK 64-Bit Server VM, 17.0.5+8-jvmci-22.3-b08, on an M1):

Benchmark                     Mode  Cnt  Score   Error  Units
NextPowerOf2Benchmark.after   avgt    5  0.647 ± 0.109  ns/op
NextPowerOf2Benchmark.before  avgt    5  0.965 ± 0.002  ns/op

[vigna]

OK thanks!

[richardstartin]

Oh no! I thought this was an elegant solution (looking for a problem, of course) :(

[vigna]

Why no? I just merged...

[vigna]

Ah OK for some weird reason I did not merge properly.