Check out the new USENIX Web site. next up previous
Next: Predictor Parameter Variations Up: Tuning Branch Predictors to Previous: Path History Based Predictor

Experimental Setup

The following experimental strategy was used in this study. The traces of indirect branches corresponding to the virtual method call sites were obtained through modifications to the JDK 1.0.2 source code. The benchmarks shown in Table 2 were executed using the modified JDK 1.0.2 on a Sparc-20 processor under Solaris 2.5 operating system. The javac and javadoc benchmarks are large applications with 25,400 and 28,471 lines of code respectively. The richards and deltablue benchmarks are medium size benchmarks with 410 and 984 lines of code [4]. These two benchmark were chosen as they have been used in earlier studies of polymorphic behavior of object oriented languages [5]. The heap benchmark is an 4495 line applet that implements incremental garbage collection.



Vijaykrishnan Narayanan
1999-02-24