Instruction count

We begin by looking at the number of instructions each requires to remove and restore a bit (Table 3). The range of instruction counts is one empirical indication of the applications' varying complexity. The excellent performance of LZO is due in part to its implementation as a single function, thus there is no function call overhead. In addition, LZO avoids superfluous copying due to buffering (in contrast with compress and zlib). As we will see, the number of memory accesses plays a large role in determining the speed and energy of an application. Each program contains roughly the same percentage of loads and stores, but the great difference in dynamic number of instructions means that programs such as bzip2 and PPMd (each executing over 1 billion instructions) execute more total instructions and therefore have the most memory traffic.