We are confronted today with the increasing complexity of our C++ software systems. To manage this complexity and build larger applications and systems, C++ strives to form emergent structures (often found in nature, such as snowflakes' symmetrical structures), where simple patterns combine to form a remarkably complex and powerful system. These structures provide both a means to limit the complexity of each component and the essential economies of scale we rely on when developing software.
From handheld devices to warehouse-sized data centers, motivated by smaller devices and increased concerns over power consumption, we are relying upon C++ to deliver these complex systems with unmatched efficiency. Our optimizing compilers today are more important than ever before and are utterly opaque to most practicing programmers. Compounding matters, the very emergent structures which allow C++ systems to scale for humans often provide unique and unsolved challenges to optimization.
In this talk, I will start with a brief overview of how modern optimizing compilers work with C++ code at a very high level. I will then walk through the specific structures and patterns of C++ code, which are at the core of forming emergent structures out of simple, elegant elements. I will also address how these interactions can be effectively modeled and analyzed by a compiler to produce efficient final programs. All of this will be illustrated by a collection of real world case studies, which are broadly applicable and show up throughout modern C++ code bases. My goal is to give a framework for understanding these interactions both in the C++ code and the optimizing compiler, so that programmers are aware of the implications posed by these patterns. Finally, I will introduce a set of principles and techniques for designing and implementing C++ programs and libraries to specifically clear the way for modern optimizers while retaining the simplicity of each component and the power of the combined whole.
