Contents

• List of Tables
• List of Figures
• 1 Introduction
  • 1 Challenges of Method-based JIT Compilation
  • 2 Challenges of Efficient Interpretation
  • 3 What We Need
  • 4 Overview of Our Solution
  • 5 Thesis Statement
  • 6 Contributions
  • 7 Outline of Thesis
  
  
• 2 Background
  • 1 High Level Language Virtual Machine
    • 1 Overview of a Virtual Program
    • 2 Interpretation
    • 3 Early Just in Time Compilers
  • 2 Challenges to HLL VM Performance
    • 1 Polymorphism and the Implications of Object-oriented Programming
    • 2 Late binding
  • 3 Early Dynamic Optimization
    • 1 Manual Dynamic Optimization
    • 2 Application specific dynamic compilation
    • 3 Dynamic Compilation of Manually Identified Static Regions
  • 4 Dynamic Object-oriented optimization
    • 1 Finding the destination of a polymorphic callsite
    • 2 Smalltalk and Self
    • 3 Java JIT as Dynamic Optimizer
    • 4 JIT Compiling Partial Methods
  • 5 Traces
  • 6 Hotpath
  • 7 Branch Prediction and General Purpose Computers
    • 1 Dynamic Hardware Branch Prediction
  • 8 Chapter Summary
  
  
• 3 Dispatch Techniques
  • 1 Switch Dispatch
  • 2 Direct Call Threading
  • 3 Direct Threading
  • 4 The Context Problem
  • 5 Subroutine Threading
  • 6 Optimizing Dispatch
    • 1 Superinstructions
    • 2 Selective Inlining
    • 3 Replication
  • 7 Chapter Summary
  
  
• 4 Design and Implementation of Efficient Interpretation
  • 1 Understanding Branches
  • 2 Handling Linear Dispatch
  • 3 Handling Virtual Branches
  • 4 Handling Virtual Call and Return
  • 5 Chapter Summary
  
  
• 5 Evaluation of Context Threading
  • 1 Experimental Set-up
    • 1 Virtual Machines and Benchmarks
    • 2 Performance and Pipeline Hazard Measurements
  • 2 Interpreting the data
    • 1 Effect on Pipeline Branch Hazards
    • 2 Performance
  • 3 Inlining
  • 4 Limitations of Context Threading
    • 1 Heavyweight Virtual Instruction Bodies
    • 2 Context Threading and Profiling
    • 3 Development using SableVM
  • 5 Chapter Summary
  
  
• 6 Design and Implementation of YETI
  • 1 Structure and Overview of Yeti
  • 2 Region Selection
    • 1 Initiating Region Discovery
    • 2 Linear Block Detection
    • 3 Trace Selection
  • 3 Trace Exit Runtime
    • 1 Trace Linking
  • 4 Generating code for traces
    • 1 Interpreted Traces
    • 2 JIT Compiled Traces
    • 3 Trace Optimization
  • 5 Other implementation details
  • 6 Chapter Summary
  
  
• 7 Evaluation of Yeti
  • 1 Experimental Set-up
  • 2 Effect of region shape on dispatch
  • 3 Effect of region shape on performance
  • 4 Early Pentium Results
  • 5 Identification of Stall Cycles
    • 1 Identifying Causes of Stall Cycles
    • 2 Stall Cycle results
    • 3 Trends
  • 6 Chapter Summary
  
  
• 8 Conclusions and Future Work
  • 1 Conclusions and Lessons Learned
  • 2 Future work
    • 1 Virtual instruction bodies as nested functions
    • 2 Extension to Runtime Typed Languages
    • 3 New shapes of region body
    • 4 Vision for new language implementation
  • 3 Summary
  
  
• Bibliography