It's clever, but is it ART?
Rudyard Kipling
In an ideal microprocessor, one that can issue one machine instruction per clock cycle, the best performance that can be achieved is that, for every cycle, one useful instruction is completed. The processor's choice of instructions then leads to an efficient implementation of the task at hand. The two goals of optimization are: (1) to reduce the total number of cycles required to achieve a particular task by reducing the number of cycles wherein no useful work is done and (2) to make the most effective use of the cycles consumed.
In real-world microprocessors, many factors conspire to cause deviation from ideal performance. The art of optimization lies in figuring out how to work around a particular factor or how to mitigate its effect on performance.
The overall strategy for optimization is straightforward and can be summarized by the following recipe:
Identify the hotspots.
Identify the cause of the performance bottleneck and assess the suitability of using SIMD techniques.
Select a suitable SIMD optimization and method of applying it.
Repeat steps 1 3 until the performance goal is reached.
This chapter introduces you to the concepts and the terminology associated with performance optimization, laying the foundation for the...
TABLE OF CONTENTS 
