Bill Stoddard wrote: > > Please lets not go overboard with these types of optimizations. This could get out >of > control quickly. Having code that is easy to read, understand and maintain is -much- >more > important than saving a a few extra cycles.
In a previous job where I had a lot of low level hardware profiling information available, we found that on average, instructions which incurred cache misses took something like 26 times longer to execute than instructions which didn't miss. That difference is likely to grow as CPUs get faster relative to main memory. There is a fair amount of research and development in feedback-directed compilation techniques to automate such optimizations. For example, http://www.research.ibm.com/journal/sj/372/schmidt.html . But since this bothers people, I will move the infrequently executed instructions back into the hot path. Greg
