Re: Resulting class performance

Jeremy,

  The HotSpot JVM does certain optimizations based on some assumptions
based on common bytecode patterns. For instance, it looks for certain
bytecode macro sequences to optimize constructs like loops, etc.
  You can read more about that on HotSpot wiki at
http://wikis.sun.com/display/HotSpotInternals/PerformanceTechniques

  regards,
  Eugene


On Fri, May 7, 2010 at 3:40 PM, Jeremy Cowgar <[email protected]> wrote:
> I am writing a language called Josl (http://www.josl.org). Currently it is
> interpreted by it's own VM. I am finding that I spend as much time on the VM
> as I do with the language and this week I have been researching the JVM and
> targeting it. This will provide JIT compilation but more importantly all the
> great libraries that come with Java and are available. Today I began writing
> a compiler in Java for the Josl language using ASM to generate the code. I
> have a test compiler working for a bare bones version of Josl. It does a
> poor job of parsing the code, does string compares for key words, etc... but
> it works. I then setup a benchmark (benchmarking too early? :-D).
>
> Anyway, I am concerned that I am not understanding something correctly. Here
> is my benchmark program. I've used it early on to just see what the raw VM
> loop speed is. i.e. do as minimal amount of work possible. Josl is a stack
> based language, BTW and it's entry word is "main", just like Java's.
>
> main 0 300,000,000 times 1+ next
>
> So, that starts at 0 and adds 1 each time through 300 million times. In Josl
> on my machine it takes 2.561 seconds. In my resulting .class file, It takes
> 3.009 seconds. I was expecting a huge performance increase.
>
> I placed all my Josl-java code on github: http://github.com/jcowgar/josl-jvm
> ... now, bear in mind, I just wanted to quickly get to the part of making
> JVM byte code, pay no attention to the total lack of proper word lookups,
> the poor parser, etc... All that will completely change.
>
> Now, I also made this program in Java, it's source is:
>
> public class LoopTest {
>    public static void main(String args[]) {
>        int i = 0;
>        int current = 0;
>        int finish = 300000000;
>
>        while (current < finish) {
>            i = i+ 1;
>            current++;
>        }
>    }
> }
>
> I tried to do it just as I did in Josl, i.e. I have a loop counter I call
> current, I also have a finish counter. Each time through I check to see if
> current < finish, I increment current and I do whatever else is contained in
> the body of the loop, in this case adding 1 to a variable each time.
>
> The above LoopTest runs in 0.259 seconds. That's the kind of performance I
> was expecting with my .class files. I have used javap -c on both the
> LoopTest class and my class, they do not loop a whole lot different. Does
> anyone know what the performance difference could be caused by?
>
> LoopTest dump:
>
> public class LoopTest extends java.lang.Object{
> public LoopTest();
>  Code:
>   0:   aload_0
>   1:   invokespecial   #1; //Method java/lang/Object."<init>":()V
>   4:   return
>
> public static void main(java.lang.String[]);
>  Code:
>   0:   iconst_0
>   1:   istore_1
>   2:   iconst_0
>   3:   istore_2
>   4:   ldc     #2; //int 300000000
>   6:   istore_3
>   7:   iload_2
>   8:   iload_3
>   9:   if_icmpge       22
>   12:  iload_1
>   13:  iconst_1
>   14:  iadd
>   15:  istore_1
>   16:  iinc    2, 1
>   19:  goto    7
>   22:  return
>
> }
>
> JoslProgram dump:
>
> public class JoslProgram extends java.lang.Object{
> public JoslProgram();
>  Code:
>   0:   aload_0
>   1:   invokespecial   #8; //Method java/lang/Object."<init>":()V
>   4:   return
>
> public static void main(java.lang.String[]);
>  Code:
>   0:   ldc     #11; //int 0
>   2:   ldc     #12; //int 300000000
>   4:   istore_1
>   5:   iconst_0
>   6:   istore_0
>   7:   iload_1
>   8:   iload_0
>   9:   if_icmple       21
>   12:  ldc     #13; //int 1
>   14:  iadd
>   15:  iinc    0, 1
>   18:  goto    7
>   21:  return
>
> }
>
>
>
> --
> You receive this message as a subscriber of the [email protected] mailing list.
> To unsubscribe: mailto:[email protected]
> For general help: mailto:[email protected]?subject=help
> OW2 mailing lists service home page: http://www.ow2.org/wws
>
>

-- 
You receive this message as a subscriber of the [email protected] mailing list.
To unsubscribe: mailto:[email protected]
For general help: mailto:[email protected]?subject=help
OW2 mailing lists service home page: http://www.ow2.org/wws