Principles of Optimization

Optimization can be a seductive activity leading to endless tweaks and changes of code. The most important part of optimization is knowing when not to do it. Two common cases are:

• Code or system is not run frequently.
• Code is a small fraction of overall time.

In other words, code should only be optimized if it will make a significant effect on the final system and the final system will be used frequently enough to justify the programmer's time and the chance of breaking something.

It helps to have a set of principles to follow in order to guide the process of optimization. The set I use is:

• Make it work before you make it fast.
• Profile everything you do.
• Complexity is bad.
• Preprocessing is good.
• Compute only what you need.