This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Programming for Performance (ECE459): Final April 18, 2010 This open-book final has 6 questions, worth 20 points each. Answer all questions. Write the answers in your answer book. You may consult any printed material (books, notes, etc). Question 1: Short Answer One point each. Answer these questions in your exam booklet. (a) Two library functions for doing communication in MPI are and . (b) Doing more work in parallel increases a system’s . (c) Modern CPUs spend most of their time waiting due to . (d) When using OpenCL, you may choose to divide the computation space into . (e) Gustafson’s Law says that parallelization isn’t hopeless when you can increase the . (f) One condition that would impede automatic parallelization is . (g) One primary design goal of DTrace was . (h) A different primary design goal of DTrace was . (i) An sfence instruction prevents reordering of . (j) To effectively parallelize an OpenMP loop where different iterations run for different amounts of time, you want to use . (k) oprofile is an example of a profiler. (l) The three steps in using profile-guided optimization are: ....
View Full Document
This note was uploaded on 09/22/2011 for the course ECE 459 taught by Professor Patricklam during the Spring '11 term at Waterloo.
- Spring '11