Code Optimization

CODE OPTIMIZATION. The target code generated by compilers is still not as good as code produced by an expert assembly language programmer.

• Compilers need to perform code improving transformations in order to reduce the running time of the target code.
• Such transformations are called optimization though very rarely will the code be optimal in the formal sense of the word.
• The transformations performed by an optimizing compiler should satisfy the following properties:
  • the transformation should preserve the meaning (semantics) of programs,
  • the transformation should, on average, speed up programs by a measurable amount,
  • though the size of programs is not generally an issue in itself, it should remain reasonable.
  • the amount of compiler effort put into optimizing a section of code should be relative to
    • the number of times the code will be run
    • to the entire program

In general code optimization problems

• are NP-complete and
• thus we cannot expect to find a global optimum in a reasonable amount of time.
• Hence the strategy is mainly based on heuristics, that is
• identify hot areas and put a lot of resources into optimizing these.

Some examples of hot areas

• loops and nested loops
• optimize or eliminate procedure calls
• recognize common subexpressions
• propagate constants
• and lots lots more