Optimization best practicesOptimization is not a form of black magic, but a regular part of software development. But with good tools and good process, you will definitely produce faster and better code

---

by Andrew Binstock, principal analyst, Pacific Data Works LLC. Intel Corp.

Want optimized code? Learn optimization methodology and the best known methods used by Intel optimization engineers.

"Programmers waste enormous amounts of time thinking about, or worrying about, the speed of noncritical parts of their programs, and these attempts at efficiency actually have a strong negative impact when debugging and maintenance are considered. We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil." –Donald Knuth (adapted from C. A. R. Hoare)

Donald Knuth’s observation is a way of saying that software optimization is often poorly planned and done without the careful methodology and rigor that accompanies other software-development processes. The widespread, undisciplined view of optimization process derives from several perceptions that are largely incorrect. The first of these is that optimization is more of an art than a science. This perspective, which rankles optimization professionals, suggests there is some magic to good optimization. In fact, the thoughtful application of tools and methodology removes much of the mystery and mystique of optimization. It is hard, disciplined work—much like everything else in software development. The other common misperception is that optimization is an after-thought. This view reflects the reality that developers are so pressed to get tested software out the door that there is little time for the luxury of optimization. Only once a disaster occurs—a month-end that couldn’t close because of processing time or a sale not made because of competitors’ better performance—only then is optimization accorded an elevated (and generally short-lived) prominence. To be good at optimization, a different view is necessary.

This white paper describes the methodology and best known methods used by Intel optimization engineers. These processes have been proven over more than a decade of use and are an excellent template for developing best methods for in-house optimization projects.

Top-down, closed-loop approach
The Intel methodology employs a top-down, closed-loop model. The top-down aspect does not refer to the top-down approach familiar to programmers from structured programming. That concept suggests that code should be designed from a mainline that branches off into successively smaller building blocks. That is a logical, structured design. In the optimization field, top and bottom depend not on a logical relationship but on a purely practical dimension: you must first optimize the elements whose performance masks or affects the performance profile of other elements. For example, you would make sure your memory allocations were optimized before you started tuning block sizes loaded into processor caches. But before tuning memory allocations, you would make sure that the system’s memory bus was overloaded—that factor alone could account for poor performance. Later in the paper, we’ll examine the steps in this top-to-bottom progression.