`ArrayRetainableByteBufferPool` inefficiently calculates bucket indices

[lorban]

Jetty version(s)
10, 11, 12

Enhancement Description
The ArrayRetainableByteBufferPool.bucketFor() method is on the fast path of every served request, and it delegates its index calculation to a Function<Integer, Integer> lambda. This implies boxing and unboxing are happening each time a bucket must be chosen, and this inefficiency started appearing in profiling reports.

This code should be modified so that no boxing/unboxing is done anymore. A simple way would be to replace the Function<Integer, Integer> lambda with a custom IntIntFunction one for instance.

[lorban]

Here's an async profiler flame graph that shows ArrayRetainableByteBufferPool.bucketFor() is showing up almost 3% of the time: async-profiler-cpu.zip

[joakime]

@lorban do you want to do this for the upcoming Jetty 10 / Jetty 11 releases?
Or postpone this? (if so, move this to the next project)

[sbordet]

@joakime @lorban I'll take this, there is java.util.function.IntUnaryOperator that has exactly the right signature and should have been used from the beginning.

[lorban]

@joakime It's not going to have much impact, no in terms of reliability (we're just replacing Integers with ints) nor in terms of performance (the AdaptiveExecutionStrategy bottleneck is hiding this one) so it doesn't matter too much.

But we did agree that it is too late for #8762 to be included in the upcoming release.