Micro-benchmarking
By the end of the previous section, we figured out where our program spends most of its execution time. We were also surprised when our "obvious" and "foolproof" optimization backfired and made the program run slower, not faster. It is clear now that we have to investigate the performance-critical function in more detail.
We already have the tools for that: the overall program is exercising this code, and we have ways to measure its performance. But we're not really interested in the rest of the program anymore, at least not until we solve the performance issues we already identified.
Working with a large program to optimize just a few lines of code has the following two major drawbacks:
First of all, even though the few lines are identified as performance-critical, it doesn't mean the rest of the program takes no time at all (in our demo example, it does, but recall that this example is supposed to represent the entire large...
