Contents

• List of Tables
• List of Figures
• Introduction
  • Research Objectives
  • Organization of This Dissertation
  
  
• Background
  • Historical Review of Data Parallel Languages
    • Multi-processing and Distributed Data
  • High Performance Fortran
    • The Processor Arrangement and Templates
    • Data Alignment
  • Message-Passing for HPC
    • Overview and Goals
    • Early Message-Passing Frameworks
    • MPI: A Message-Passing Interface Standard
  • High Level Libraries for Distributed Arrays
    • PARTI
    • The Global Array Toolkit
    • NPAC PCRC Runtime Kernel - Adlib
  • Discussion
  
  
• The HPspmd Programming Model
  • Motivations
  • HPspmd Language Extensions
  • Integration of High-Level Libraries
  • The HPJava Language
    • Multiarrays
    • Processes
    • Distributed Arrays
    • Parallel Programming and Locations
    • More Distribution Formats
    • Array Sections
    • Low Level SPMD Programming
  • Adlib for HPJava
  • Discussion
  
  
• Compilation Strategies for HPJava
  • Multiarray Types and HPspmd Classes
  • Abstract Syntax Tree
  • Type-Checker
    • Type Analysis
    • Reachability and Definite (Un)assignment Analysis
  • Pre-translation
  • Basic Translation Scheme
    • Translation of a multiarray declaration
    • Translation of multiarray creation
    • Translation of the on construct
    • Translation of the at construct
    • Translation of the overall construct
    • Translation of element access in multiarrays
    • Translation of array sections
  • Test Suites
  • Discussion
  
  
• Optimization Strategies for HPJava
  • Partial Redundancy Elimination
  • Strength Reduction
  • Dead Code Elimination
  • Loop Unrolling
  • Applying PRE, SR, DCE, and LU
    • Case Study: Direct Matrix Multiplication
    • Case Study: Laplace Equation Using Red-Black Relaxation
  • Discussion
  
  
• Benchmarking HPJava, Part I: Node Performance
  • Importance of Node Performance
  • Direct Matrix Multiplication
  • Partial Differential Equation
    • Background on Partial Differential Equations
    • Laplace Equation Using Red-Black Relaxation
    • 3-Dimensional Diffusion Equation
  • Q3 - Local Dependence Index
    • Background on Q3
    • Data Description
    • Experimental Study - Q3
  • Discussion
  
  
• Benchmarking HPJava, Part II: Performance on Parallel Machines
  • Direct Matrix Multiplication
  • Laplace Equation Using Red-Black Relaxation
  • 3-Dimensional Diffusion Equation
  • Q3 - Local Dependence Index
  • Discussion
  
  
• Related Systems
  • Co-Array Fortran
  • ZPL
  • JavaParty
  • Timber
  • Titanium
  • Discussion
  
  
• Conclusions
  • Conclusion
  • Contributions
  • Future Works
    • HPJava
    • High-Performance Grid-Enabled Environments
    • Java Numeric Working Group from Java Grande
    • Web Service Compilation (i.e. Grid Compilation)
  • Current Status of HPJava
  • Publications
  
  
• Bibliography