Collectives™ on Stack Overflow

Find centralized, trusted content and collaborate around the technologies you use most.

Teams

Q&A for work

Connect and share knowledge within a single location that is structured and easy to search.

I'm measuring some single-threaded method invocation (written in Scala) and wanted to analyze the benchmark. Here is how it looks like (implementation details omitted)

@State(Scope.Benchmark)
class TheBenchmarks {
    var data: Array[Byte] = _
    @Param(Array("1024", "2048", "4096", "8192"))
    var chunkSize: Int = _
    @Setup
    def setup(): Unit = {
        data = //get the data
    @Benchmark
    @OutputTimeUnit(TimeUnit.MICROSECONDS)
    @BenchmarkMode(Array(Mode.AverageTime))
    def takeFirstAvroRecord(bh: Blackhole): Unit = {
      val fr = //do computation with data and chunk size
      bh.consume(fr)
Okay, I got some result and wanted to understand it, but the output of the  -prof perfasm is a bit unclear to me. First of all:
....[Hottest Regions]...............................................................................
 44.20%   40.50%        runtime stub  StubRoutines::jbyte_disjoint_arraycopy (205 bytes) 
  6.78%    1.62%         C2, level 4  cats.data.IndexedStateT$$Lambda$21::apply, version 1242 (967 bytes) 
  4.39%    0.79%         C2, level 4  my.pack.age.Mclass::cut0, version 1323 (299 bytes) 
....[Hottest Methods (after inlining)]..............................................................
 44.20%   40.50%        runtime stub  StubRoutines::jbyte_disjoint_arraycopy 
  8.40%    3.93%         C2, level 4  cats.data.IndexedStateT$$Lambda$21::apply, version 1242 
  5.76%    2.67%         C2, level 4  my.pack.age.Mclass::cut0, version 1323 
I found some about jbyte_disjoint_arraycopy. It is declared as follows here as follows.
StubRoutines::_jbyte_disjoint_arraycopy  = generate_disjoint_byte_copy(false, &entry,
                                                                           "jbyte_disjoint_arraycopy");
Judging by the source of the generate_disjoint_byte_copy method it looks like an assembly code generation thing... I could guess that it is some intrinsic array copy for x86...
Question: Can you please give some explanation about the StubRoutines and what may cause it to be the hottest region?
                I see two different questions here: 1) What is stub routine is and how does it work. 2) What is the problem with Hotspot disassembler. Not sure which one to answer - please ask one question per post.
– apangin
                Sep 13, 2018 at 20:39
You guessed right. <type>_disjoint_arraycopy stubs are functions generated in run time specifically for speeding up System.arraycopy calls.
When JVM starts, it produces optimized machine code for certain routines using currently available CPU features. E.g. if CPU supports AVX2, generated arraycopy stubs will make use of AVX2 instructions.
System.arraycopy is HotSpot intrinsic method. When compiled by C2, an invocation of System.arraycopy performs neccessary checks and then calls one of the generated arraycopy stub routines.
If StubRoutines::jbyte_disjoint_arraycopy is the hottest region, it basically means that your benchmark spends most of the time inside System.arraycopy dealing with byte[] arrays. You may try async-profiler to see where this arraycopy is called from.
                Can you get some idea about how this feature check is performed? I went through the generate_disjoint_byte_copy and did not notice any AVX2 related instructions/registers used. Maybe its in some another place...?
– St.Antario
                Sep 14, 2018 at 0:27
                @St.Antario Updated the answer with the links to VM_Version::get_processor_features() and AVX2 usages.
– apangin
                Sep 14, 2018 at 7:46
        
Thanks for contributing an answer to Stack Overflow!
Please be sure to answer the question. Provide details and share your research!
But avoid …
Asking for help, clarification, or responding to other answers.
Making statements based on opinion; back them up with references or personal experience.
To learn more, see our tips on writing great answers.