Optimized C++

Chapter 7. Optimize Hot Statements

The idea is there, locked inside, and all you have to do is remove the excess stone.

Michelangelo di Lodovico Buonarroti Simoni (1475–1564), in response to the question, “How do you create your masterpieces?”

Optimizing at the statement level can be modeled as a process of removing instructions from the stream of execution, in a manner similar to how Michelangelo described the process of sculpting his masterpieces. The problem with Michelangelo’s advice is that it doesn’t address which part of the stone is excess, and which part is the masterpiece.

The problem with optimizing at the statement level is that, aside from function calls, no C++ statement consumes more than a handful of machine instructions. Focusing on such small-scale optimizations doesn’t generally produce enough improvement to make them worth the effort unless the developer can find factors that magnify the cost of the statement, making it hot enough to be worth optimizing. These factors include:

Loops

The cost of statements within a loop is multiplied by the number of times they are repeated. Loops must be identified by the developer. The profiler may point to a function containing a hot loop, but it won’t say which loop in the function is hot. It may point to a function that is hot because it is called from inside one or more loops, but it won’t say which call sites are hot. Since the profiler does not point directly to the loop, the developer must inspect ...