Rtalk Performance for micro benchmarks (how to improve)

[Mark Roos]

>From Charles
        then could it simply be that your
        indy guard logic and arbitrary precision logic adding all that
        overhead? It seems like a lot indeed.

I had not thought about this in a while so perhaps my call site handling 
is an issue.  I went
for simple in that I use a varArgs style of type matching for my fallback 
used in the GWT processing. 
Since I really don't understand how the jvm or the compiler deal with 
varArgs I could be pretty inefficient. 
Others I believe have specialized based on airity.

So my one bootstrap looks like this:  ( well I have them for constants and 
super but only one for each )
        bootStrapSelf(MethodHandles.Lookup callerLookup, String name, 
MethodType type, Object[] arg)

To drive the fallback I combine the args so I only need one fallback 
method ( airity reduction )
    mh=lookup.findStatic(RtCallSite.class, "fallbackSelf", mt);
     mh2=MethodHandles.insertArguments(mh, 0, site); // insert the call 
site
     mh3=mh2.asCollector(RtObject[].class, site.getAirity()); // combine 
the args into an array
 
My fallback looks like:
        fallbackSelf(RtCallSite site, Object[] args)

And my test path has to add the value to test against and drop the args 
except for the receiver.
My class is based on a field in the object
    MethodHandle test = site._test;
    test=MethodHandles.insertArguments(test, 0, rcvr.classField()); // 
need to insert the old rcvr class at the beginning
    test=MethodHandles.dropArguments(test, 0, site.dropArray()); //  drop 
all but the last arg ( rcvr )

Test is then simple 
  public static boolean test(RtObject oldClass, RtObject receiver) {
    boolean rtn=receiver.classField() == oldClass;
    return rtn;
  }

The GWT is formed as
        gwt=MethodHandles.guardWithTest(test, newTarget, oldTarget);

newTarget is my method which in not varArgs style but a normal method with 
airity. 

So I think my only hit is in the insert and drop in the test path.  But I 
do have to do this for each GWT
in series so it seems repetitive.  I could specialize the test by airity 
but I am not sure of the impact.

thanks
mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/mlvm-dev/attachments/20120917/5ae4ee9c/attachment.html