ECEn 424 Review Topics: Chapter 5

---

• What optimization blockers are identified in the text, what form can they take in C code, and how does each limit the ability of the compiler to optimize?
• What measure of performance is used in the text as a guide in improving code? Why is that measure appropriate and how is it computed?
• What kind of optimizations are applied in the running example of "combine" functions in Chapter 5, and what kind of performance improvements result from each?
• Modern CPUs:
  • What is instruction-level parallelism?
  • What is different about a superscalar processor?
  • What is done out of order in an out-of-order processor, and what hardware is required to support this complexity?
  • What does speculative execution refer to, how is it achieved, and what are its advantages and disadvantages?
  • What is retirement and how is it related to speculative execution?
  • What is register renaming and why is it desirable?
  • Given a short loop (in C or assembly) and details about available resources and functional-unit latencies, can you produce the corresponding data-flow graph and identify its critical path? (In other words, can you predict the CPE we would see for that code on a modern Intel CPU?)
• What is loop unrolling and what are the tradeoffs and the limitations of its use?
• How does the use of multiple accumulators lead to increased performance, and what are the limitations of this approach?
• What is a reassociation transformation and how can it result in improved performance?
• In terms of CPE, what is the very best performance that was achieved on the combine functions in Chapter 5, and what techniques were used to obtain that speedup?
• What is register spilling and how would you recognize it in assembly code?
• What can a programmer do in the source code to increase the likelihood that the compiler will use conditional moves rather than conditional branches?
• How can the load latency be measured, and how long is it for processors described in the book?
• What are store buffers and why do they exist? What special concerns arise for stores in the context of speculative execution?
• Why might a dependence between a load and a recent store have a negative impact on performance?
• What basic strategies should we follow in optimizing program performance?
• How does a profiler such as GPROF obtain information about program information, and how can we use such a tool to guide optimization? How accurate is the information GPROF provides?
• Self test: Can you complete all practice problems in Chapter 5?

Updated for 3rd Edition of CS:APP