Why is invokedynamic faster than invokestatic

Question

Recently, I have been working with low-level Java, modifying the JVM bytecode and generating Class Files without compiling .java files. I just learned about invokedynamic, and I was able to generate my own InvokeDynamic constants and BootstrapMethod attributes using my own proprietary JVM bytecode tool to allow me to use invokedynamic properly.

After this, I made simple timing script in java that tests both invokestatic and invokedynamic on the same method. However, my script showed that invokedynamic was about 8% faster than invokestatic, something that I thought would be impossible.

Here is the code that I used to time each method:

import java.lang.invoke.*;

public class InvokeDynamicTest implements Runnable {

    private static void doNothing(){
        //do nothing
    }
    
    private static CallSite getCallSite(MethodHandles.Lookup s, String name, MethodType methodType){
        try{
            return new ConstantCallSite(s.findStatic(InvokeDynamicTest.class, name, methodType));
        }catch(Throwable t){
            throw new RuntimeException(t);
        }
    }
    
    public static void main(String[] args) {
        final Runnable r = (Runnable) new InvokeDynamicTest();
        
        long time = System.currentTimeMillis();
        int counter = 0;
        while(System.currentTimeMillis() - time < 5000){
            doNothing();
            counter++;
        }
        System.out.println(counter);
        
        time = System.currentTimeMillis();
        counter = 0;
        while(System.currentTimeMillis() - time < 5000){
            r.run();
            counter++;
        }
        System.out.println(counter);
    }
    
    public void run() {
        //Use bytecode modifier to insert invokedynamic
    }

}

Here is the output:

514048300
545970671

As for the modified bytecode after compilation, here are the new constants as dumped by javap:

   #9 = Class              #10            // InvokeDynamicTest
   #10 = Utf8              InvokeDynamicTest
...
  #70 = Utf8               getCallSite
  #71 = Utf8               (Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/invoke/MethodType;)Ljava/lang/invoke/CallSite;
  #72 = NameAndType        #70:#71        // getCallSite:(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/invoke/MethodType;)Ljava/lang/invoke/CallSite;
  #73 = Methodref          #9.#72         // InvokeDynamicTest.getCallSite:(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/invoke/MethodType;)Ljava/lang/invoke/CallSite;
  #74 = MethodHandle       6:#73          // REF_invokeStatic InvokeDynamicTest.getCallSite:(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/invoke/MethodType;)Ljava/lang/invoke/CallSite;
  #75 = Utf8               doNothing
  #76 = Utf8               ()V
  #77 = NameAndType        #75:#76        // doNothing:()V
  #78 = InvokeDynamic      #0:#77         // #0:doNothing:()V

Here is the new code for the run method:

  public void run();
    descriptor: ()V
    flags: (0x0001) ACC_PUBLIC
    Code:
      stack=100, locals=100, args_size=1
         0: invokedynamic #78,  0             // InvokeDynamic #0:doNothing:()V
         5: return

Here is the BootstrapMethods table:

BootstrapMethods:
  0: #74 REF_invokeStatic InvokeDynamicTest.getCallSite:(Ljava/lang/invoke/MethodHandles$Lookup;Ljava/lang/String;Ljava/lang/invoke/MethodType;)Ljava/lang/invoke/CallSite;
    Method arguments:

Am I missing something? I cannot possibly see how invokedynamic could be faster than invokestatic

This is a very standard example of a poorly designed Java benchmark that will give you highly misleading results. (But also, invokedynamic can convert itself to invokestatic after the first call.) — Louis Wasserman, Feb 17 '22 at 05:42

Why is invokedynamic faster than invokestatic

0 Answers0