Sources and Types of Parallelism

• Structured: Identical tasks on different data sets
• Unstructured: Different data streams and different instructions
• Algorithm level: Appropriate algorithms and data structures
• Programming:
  • Specify parallelism in parallel languages
  • Write sequential code and use compilers
  • Use course grain parallelism: independent modules
  • Use medium grain parallelism: loop level
  • Use fine grain parallelism: basic block or statement
• Expressing parallelism in programs
  • No good languages
  • Programmers are unable to exploit whatever little is available

Model of Compiler

CODE OPTIMIZATION

High Level Analysis for optimization

• Common subexpression elimination
• Copy propagation
• Dead code elimination
• Code motion
• Strength reduction
• Constant folding
• Loop unrolling
• Induction variable simplification