Panama build from "foreign+vector" branch have very unstable compiler

[Lev Serebryakov]

 I try to run JMH benchmarks on my Panama build from "foreign+vector"
branch and results is VERY unstable. Looks like compiler stuck in some
(most) runs.

 Same system, same benchmarks JAR, same command line, everything is the
same. Sometimes (often, 9 times out of 10) it looks like this:

# Warmup: 5 iterations, 10 s each
# Measurement: 10 iterations, 2 s each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Throughput, ops/time
# Benchmark: vector.VectorBenchmarks.VOVec_cs_div_cv
# Parameters: (callSize = 16, startOffset = 0)

# Run progress: 0,00% complete, ETA 00:14:00
# Fork: 1 of 2
# Warmup Iteration   1: 7448,058 ops/s
# Warmup Iteration   2: 7671,674 ops/s
# Warmup Iteration   3: 7632,154 ops/s
# Warmup Iteration   4: 7591,418 ops/s
# Warmup Iteration   5: 7645,616 ops/s
Iteration   1: 7629,146 ops/s
Iteration   2: 7644,162 ops/s
Iteration   3: 7640,662 ops/s
Iteration   4:^C

 If I add "-XX:+PrintCompilation" it changes results dramatically (After
clean up of output):

# Warmup Iteration   1:     467
# Warmup Iteration   2:   11584
# Warmup Iteration   3:   21533
# Warmup Iteration   4:   31547
# Warmup Iteration   5:   41582
Iteration   1:   50571
Iteration   2:   54600
Iteration   4:   58630
Iteration   5:   59635
Iteration   6:   61649
Iteration   7:   65679
Iteration   8:   67694
Iteration   9:   67694
Iteration  10:   69710

  I'm not sure, is it JMH bug or JDK bug and I don't understand how to
check. I can not test this on other JDK as I need Vector API for benchmarks.

-- 
// Black Lion AKA Lev Serebryakov