JMH v0.9.1

[Aleksey Shipilev]

Hi there,

JMH 0.9.1 is released and available in Maven Central (thanks Evgeny!).
This is the stabilization release towards JMH 1.0, which should happen
in July.

Thanks to all those people who gave JMH 0.9 a try over two past weeks,
we will keep addressing the found issues over the next few weeks. If you
already migrated to 0.9, then update to 0.9.1 should be painless, and
worthwhile. If you didn't test updated JMH 0.9+ before, please do so as
soon as possible!

Highlights of JMH 0.9.1:

 * Fixed a typo bug in JmhBytecodeGenerator, which locked the generator
strategy to "reflection" always.

 * Control.iterationTime is removed, use BenchmarkParams getters instead
(you can inject BenchmarkParams into @Setup/@TearDown/@Benchmark methods)

 * @Group now has meaningful default label, not empty string when
argument-less @Group is requested.

 * SampleTime now does not print percentiles when there is no reason to
do so due to low number of samples. This makes run log cleaner when only
a single invocation per JVM is measured.

 * "perf" profiler now runs in a very detailed mode, showing more L1
cache, iTLB/dTLB stats, etc.

 * "perf" and "perfasm" profilers now have adjustable delays before the
data collection. Automatic delay calculations were also improved.

 * "perfasm" output was greatly improved: accounting for all events
across compiled code and native code, mapping hot regions to Java and
JVM methods, etc. etc. Just look at JMHSample_08_DeadCode running under
"perfasm": http://cr.openjdk.java.net/~shade/jmh/perfasm-sample.log

 * "perfasm" is now more customizable. Some further work would be in
order to make profiler customizations more accessible via cmdline, not
through properties, but you can look at "perfasm" settings now in the
source code:
 http://hg.openjdk.java.net/code-tools/jmh/file/0.9.1/jmh-core/src/main/java/org/openjdk/jmh/profile/LinuxPerfAsmProfiler.java#l67

Enjoy!

Thanks,
-Aleksey.