IPT Readings on Instrumentation, Profiling, and Tracing Seminar presentation by Alessandra Gorla University of Lugano December 7, 2006

Java Bytecode Analysis and Optimization Soot BCEL JABA Soot BCEL JABA

Overview  Introduction  Tools  Soot - a Java Bytecode Optimization Framework  BCEL - Byte Code Engineering Library  JABA - JAva Bytecode Analyzer  Introduction  Tools  Soot - a Java Bytecode Optimization Framework  BCEL - Byte Code Engineering Library  JABA - JAva Bytecode Analyzer

Introduction  Java application are usually much slower than C and C++ applications.  Possible approaches to solve the problem  Bytecode optimizers  Significant optimizations. Create new classes  Bytecode annotators  Create new classes with annotations  Bytecode manipulation tools  Manipulate bytecode in its original form  Java application packagers  Compress and/or obfuscate code  Java native compilers  Compile Java to native executables  Java application are usually much slower than C and C++ applications.  Possible approaches to solve the problem  Bytecode optimizers  Significant optimizations. Create new classes  Bytecode annotators  Create new classes with annotations  Bytecode manipulation tools  Manipulate bytecode in its original form  Java application packagers  Compress and/or obfuscate code  Java native compilers  Compile Java to native executables

Soot  Bytecode optimizer framework  Intraprocedural optimization  Whole program optimization  Three intermediate representations  Baf: streamlined representation of the bytecode  Jimple: typed 3-address representation  Grimp: Jimple aggregated version  Bytecode optimizer framework  Intraprocedural optimization  Whole program optimization  Three intermediate representations  Baf: streamlined representation of the bytecode  Jimple: typed 3-address representation  Grimp: Jimple aggregated version

Baf  Constant pool abstraction  Give type to dup and swap instructions  Local vars are given explicit names  Constant pool abstraction  Give type to dup and swap instructions  Local vars are given explicit names

Jimple  3- address code representation (not for jsr)  Stack is replaced by additional local vars (prefixed by $)  3- address code representation (not for jsr)  Stack is replaced by additional local vars (prefixed by $)

Grimp  Much easier to read than Baf of Jimple  Has a representation of the new operator  Aggregate expressions  Much easier to read than Baf of Jimple  Has a representation of the new operator  Aggregate expressions

Optimization framework

Optimizations  Intraprocedural optimizations  Constant propagation and folding  Conditional and unconditional branch elimination  Copy propagation  Dead assignment and unreachable code elimination  Expression aggregation  Whole program optimization (call graph)  Method inlining  (Devirtualization of method calls)  Intraprocedural optimizations  Constant propagation and folding  Conditional and unconditional branch elimination  Copy propagation  Dead assignment and unreachable code elimination  Expression aggregation  Whole program optimization (call graph)  Method inlining  (Devirtualization of method calls)

Experiments

BCEL  Bytecode manipulation library  Written in Java  Opensource  Offers capabilities to inspect, edit and create Java binary classes  Package to represent class  Package to dynamically generate and modify classes  Code examples, utilities  Bytecode manipulation library  Written in Java  Opensource  Offers capabilities to inspect, edit and create Java binary classes  Package to represent class  Package to dynamically generate and modify classes  Code examples, utilities

BCEL Class representation

BCEL Class editing

BCEL - Optimization example  Push 0 or 1 to the stack to evaluate boolean expressions  Combination of boolean expressions:  Keep pushing 0s and 1s to the stack  Algorithm to apply:  Replace IfInstruction branch target with ifne branch target  Push 0 or 1 to the stack to evaluate boolean expressions  Combination of boolean expressions:  Keep pushing 0s and 1s to the stack  Algorithm to apply:  Replace IfInstruction branch target with ifne branch target

BCEL - Optimization example 5: aload_0 6: ifnull #13 9: iconst_0 10: goto #14 13: iconst_1 14: nop 15: ifne #36 18: iload_1 19: iconst_2 20: if_icmplt #27 23: iconst_0 24: goto #28 27: iconst_1 28: nop 29: ifne #36 32: iconst_0 33: goto #37 36: iconst_1 37: nop 38: ifeq #52 41: getstatic System.out 44: ldc "Ooops" 46: invokevirtual println 52: return 5: aload_0 6: ifnull #13 9: iconst_0 10: goto #14 13: iconst_1 14: nop 15: ifne #36 18: iload_1 19: iconst_2 20: if_icmplt #27 23: iconst_0 24: goto #28 27: iconst_1 28: nop 29: ifne #36 32: iconst_0 33: goto #37 36: iconst_1 37: nop 38: ifeq #52 41: getstatic System.out 44: ldc "Ooops" 46: invokevirtual println 52: return if((a == null) || (i < 2)) System.out.println("Ooops"); 10: aload_0 11: ifnull #19 14: iload_1 15: iconst_2 16: if_icmpge #27 19: getstatic System.out 22: ldc "Ooops" 24: invokevirtual println 27: return

JABA  Bytecode analyzer library  Graphs representation  Control Flow Graph  Class Control Flow Graph  Interclass Control Flow Graph  Bytecode analyzer library  Graphs representation  Control Flow Graph  Class Control Flow Graph  Interclass Control Flow Graph

JABA - CFG public void metodo(){ int x = 2; int c = 3; while(x > 0) { c++; x--; }

JABA - ICFG