Preface

The art of high-performance programming is making a comeback. I started programming in the days when the programmer had to know where every bit of data went (sometimes quite literally – with switches on the front panel). Now, computers have more than enough power for everyday tasks. Sure, there have always been domains where there is never enough computing power. But most programmers could get away with writing inefficient code. This is not a bad thing, by the way: free from performance constraints, the programmer could focus on making the code better in other ways.

The very first thing this book explains, then, is why more and more programmers are forced to pay attention to performance and efficiency again. This will set the tone for the entire book because it defines the methodology we will be using in subsequent chapters: knowledge about performance must ultimately come from measurements, and every performance-related claim must be supported by data.

There are five components, five elements that together determine the performance of a program. First, we delve into the details and explore the low-level foundation of all things performance: our computing hardware (no switches – promise, those days are gone). From the individual components – processors and memory – we work our way up to multiprocessor computing systems. Along the way, we learn about the memory model, the cost of data sharing, and even lock-free programming.

The second component of high-performance programming is an efficient use of the programming language. It is at this point that the book becomes much more C++-specific (other languages have their own favorite inefficiencies). Following closely is the third element, the skill to help the compiler improve the performance of your programs.

The fourth component is the design. Arguably, it should be the first one: if the design is not done with performance as one of its explicit goals, it is almost impossible to add good performance later as an afterthought. We study designing for performance last, however, since this is a high-level concept and it brings together all the knowledge we will have acquired earlier.

The final, fifth element of high-performance programming is you, the reader. Your knowledge and skill will ultimately determine the result. To help you learn, the book includes many examples that can be used for hands-on exploration and self-study. The learning does not have to stop after you turn over the last page.