8234863: Increase default value of MaxInlineLevel

[Jiva, Azeem]

Claes,
  As we discussed, this is an issue with the RxJava 1.3.x microbenchmarks and the delta is quite significant.  Have you had a chance to dive into what's happening with the deeper inlining level?  

-- 
Azeem Jiva 

On 2/21/20, 5:32 AM, "Claes Redestad" <claes.redestad at oracle.com> wrote:

    Hi,
    
    running from the current RxJava tip, the specific microbenchmark you
    refer to does no longer seem to exist.
    
    I ran a few others at random on JDK 14 with -XX:MaxInlineLevel=9 and 15
    and so far seen neither statistically significant improvements nor
    regressions:
    
    -XX:MaxInlineLevel=9
    Benchmark                                Mode  Cnt       Score 
    Error  Units
    CallableAsyncPerf.observeOnCompletable  thrpt    5  141062.960 ± 
    21482.159  ops/s
    CallableAsyncPerf.observeOnFlowable     thrpt    5  129466.284 ± 
    17678.970  ops/s
    CallableAsyncPerf.observeOnMaybe        thrpt    5  141189.313 ± 
    17751.146  ops/s
    CallableAsyncPerf.observeOnObservable   thrpt    5  134234.044 ± 
    20395.257  ops/s
    CallableAsyncPerf.observeOnSingle       thrpt    5  144157.312 ± 
    26030.366  ops/s
    
    -XX:MaxInlineLevel=15
    Benchmark                                Mode  Cnt       Score 
    Error  Units
    CallableAsyncPerf.observeOnCompletable  thrpt    5  146454.659 ± 
    17943.591  ops/s
    CallableAsyncPerf.observeOnFlowable     thrpt    5  133356.045 ± 
    16149.803  ops/s
    CallableAsyncPerf.observeOnMaybe        thrpt    5  136854.532 ± 
    19836.464  ops/s
    CallableAsyncPerf.observeOnObservable   thrpt    5  141694.341 ± 
    11023.596  ops/s
    CallableAsyncPerf.observeOnSingle       thrpt    5  145244.260 ± 
    15513.685  ops/s
    
    /Claes
    
    On 2020-02-20 17:42, Claes Redestad wrote:
    > Thanks for the heads up,
    > 
    > I'll seek approval to try out these microbenchmarks myself and see if
    > I can reproduce.
    > 
    > /Claes
    > 
    > On 2020-02-20 16:43, Jiva, Azeem wrote:
    >> Claes,
    >>    I've been running the RxJava microbenchmarks[1] with JDK8 and JDK11 
    >> with -XX:MaxInlineLevel=15 and there are a number of the 
    >> microbenchmarks, specifically the ComputationSchedulerPerf:observeOn 
    >> microbenchmark that regresses in performance.  In the case of this 
    >> specific benchmark, 30% regression.  I've been looking into why, but 
    >> no definitive answers as of yet.
    >>
    >>
    >>
    >>
    >> [1]:  
    >> https://urldefense.com/v3/__https://github.com/ReactiveX/RxJava__;!!GqivPVa7Brio!PUSXEoQwR0McFMQp4l7gDW0Tj5Y37fu0byZpc58M-kc0dk3z3yow8rUDtu25R0q3gkc$ 
    >>
    >>