Topics covered in Special Topics Section 2/16/06

Raymond Doyle

• Definition of compiler optimization, why it is useful
• Concept of compiler-generated code transformations
• Very high level overview of the memory hierarchy and caches, to motivate discussion of optimization to hide memory latencies
• The following optimizations were discussed:
• • Constant folding and propagation
  • Common subexpression elimination
  • Algebraic simplification and strength reduction
  • Dead code elimination
  • Inlining
  • Loop-invariant code motion, and other code motion
  • Loop unrolling
  • Tail Recursion elimination
  • Instruction scheduling
  • Register allocation
  • Cache optimizations
  • Software piplining
• Emphasized that better algorithms are a better way to speed up a program, and a compiler can’t change an algorithm (e.g. replace insertion sort with merge sort).