Optimizing C/C++ with Inline Assembly Programming

摘要

Assembly programming can be likened to "working without a net" where only the strong dare to venture because many of the tools familiar to C/C++ programmers-such as debuggers and memory tracers-suddenly become near worthless. Fortunately, you can dip your toes into assembly by utilizing assembler instructions. In nearly every instance, the reason for resorting to assembly programming is to increase processing performance. But before you even open your processor opcodes manual, the first question to ask is simple: How fast does the code need to execute? Obviously, there is an upper limit on how fast a particular algorithm or function will execute. Once you determine that the larger work (to which this smaller element is a part of) benefits from a performance boost, it is important to quantify the theoretical limit. Although it varies between processor architectures, the key elements are clock rate and quantity of data. Things get more involved in refining this estimation with today's long pipeline, branch-predicting, and multiple execution-unit processors, so the input from an experienced architecture guru would be worthwhile.