Data Dependence Analysis

• Check whether program can be restructured
• GCD test
• Banerjee’s Test
• I-test
• Omega test
• Architectural specific tests

Program Restructurer

• Loop restructuring
• Statement reordering
• Loop (distribution) fission
• Breaking dependence cycle: Node splitting
• Loop interchanging
• Loop skewing
• Loop jamming
• Handling while loops

Technique to Improve Detection of Parallelism: Interactive Compilation

• Back-end to user
• Compiler directives
  • Force Parallel
  • No Parallel
  • Max-trips count
  • Task identification
  • No side effects
• Incremental Analysis

Scalar Processors

• Machines use standard CPUs
• Main memory is in hierarchy: RAM and cache
• Most programs exhibit locality of reference
  • Same items are referred very close in time (temporal)
  • Nearby locations are referred frequently (spatial)
• Programmers do not have control over memory
• Programmers can take advantage of cache