CS 525: GPU Programming - Spring 2012

Andy Johnson

Electronic Visualization Laboratory
2032 Engineering Research Facility (ERF)

TEXTBOOKS:

OpenGL Shading Language 3rd Ed aka 'the orange book' Randi J. Rost Addison-Wesley Professional ISBM-13: 978-0321637635

Heterogeneous Computing with OpenCL Gaster, Howes, et al. Morgan Kaufmann ISBN-13: 978-0123877666

This might also be helpful

Programming Massively Parallel Processors: A Hands-on Approach Kirk and Hwu Morgan Kaufmann ISBN-13: 978-0123814722

SCHEDULE OF TOPICS TO BE COVERED:

PREREQUISITES:

You should feel comfortable programming in C or C++, and have access to a computer with a modern graphics card capable of using the OpenGL Shading Extensions (OpenGL 2.0 compatible) and running OpenCL. Having experience programming OpenGL (at the CS 488 level) will be helpful.

Standard departmental disclaimer: If you do not have the prerequisites for this course make sure that you drop this course right away. The department will verify the prerequisites for all students registered in this course during the first few weeks of the term and if you do not have the prerequisites, you will be notified and dropped from the course after the normal drop/add period. By that time, you will not be able to enroll in any new course.

OVERVIEW:

Programmable GPUs (Graphics Processing Units) emerged in 2001 and since then they have become a dominant topic in computer graphics and scientific visualization, allowing programmers to more directly take control of computer graphics hardware. We will look at the use of GPUs both for graphics processing and general parallel computation through a series of projects and presentations.

ATTENDANCE:

Attendance is very important, but not mandatory. Part of your grade will be based on your in-class participation (i.e. asking good questions, making good comments) so please take that into account.

I make a habit of putting all of my notes on the web, however these notes should not be considered 'official' until the day of class. If I make any major changes to a page afterwards, then I will announce it in class but note that at the bottom of each page is a modification date so you should be able to tell when the notes were last modified. I try to ensure that the notes on the web accurately reflect the lecture, but its what is discussed in class 'that counts', not what's written in the notes.

There is also a piazza page for the class - please sign up at https://piazza.com/uic/spring2012/

UIC wireless is kind of iffy in the classroom. If you want better access to wireless then please fill out the survey at http://www.surveymonkey.com/s/26KJSQ8

ASSIGNMENTS:

• Each student will make one paper presentation during the term on current work with GPUs.
  
• Each student will do three projects to gain familiarity with programming a GPU.

More on the presentations

• The paper presentation will focus on in-depth looks at some modern shaders or the use of GPUs for computation. 
  
• The presentations will be done individually, and are expected to be be professional - you don't need to dress up, but the A/V should be good, it must be rehearsed, it must be on-time, and the audience should learn from the presentation.
• The paper presentation will be 20 minutes long with an additional 5 minutes of questions at the end, the project presentations will be 10-20 minutes long with an additional 5 minutes of questions at the end. These times may be adjusted depending on the number of people in the class.You should plan to use all of your time. Finishing several minutes early shows bad planning, as does finishing late. You should try to finish less than one minute before your time runs out.
  
• See the link on the course web site for the set of html files on how to give good presentations, so you know what I mean by a 'good' presentation.
• Whatever you do, don't read the paper to your audience.
• Have a few slides at the beginning of your talk to give the background of the authors of the paper so we know about their past work and their research interests.
  
• Demonstrations (movies, animations, etc) usually are very helpful in promoting understanding
• If you put a formula or acronym onto your slides you better be able to explain it
• Be sure to have a backup plan in case the technology fails. I will never intentionally take away technology, but technology fails, so be prepared.
• Presentation slots will be randomly assigned, you may trade slots with other students as you wish as long as both agree to the trade and you notify Andy about it at least a week in advance.
• Each person will need to propose a paper topic and have it approved by Andy. If two people ask to present the same material then the person who's email arrives first to Andy wins.
• Everyone in the class is expected to read the appropriate material before coming to class so we can have a good discussion and learn from the presentations and the papers. As part of this each presenter is expected to make their paper available to the class at least one week ahead of time either in pdf or html format via the web or through photocopies
• There are 4 goals for the paper presentations. 1) You should learn a good amount about an area that is of interest to you, 2) The audience should learn a fair amount about that area and learn about what is cutting edge work on GPUs, 3) You get to practice giving a paper presentation for when you present your own papers, and 4) You get to watch other people give good / bad / mediocre presentations and learn from what you see and hear.
  

More on the projects

• There will be 3 programming projects in the class. The first will familiarize students with GLSL; the second will be an OpenCL project, and the third will be up to each student - preferably something that makes use of the GPU in the student's own research area. Make sure you can demonstrate your projects functionality to the class and to me in person at EVL.
  

EXAMINATIONS:

There will be a final exam at the end of the term covering the material presented in class. Most likely this will be a programming project that will be presented during the final exam period.

GRADES:

The final grade will be calculated as follows	Percent	Letter grades
Projects (3 at 15% each)	45%	A: 87% - 100%
Paper Presentation	30%	B: 75% -   86%
In class participation	10%	C: 62% -   74%
Final Exam (most likely another project)	15%	D: 50% -   61%
TOTAL	100%	E: 00% -   49%

Note: that you must get a passing (D or better) grade on _all_ projects and final to get a passing grade in the course.

Also note: I only give Incompletes for serious hospitalization issues which come up suddenly near the end of the term.

Also also note: I have no qualms about giving Ds or Fs in a graduate level course, though the vast majority of the grades have tended to be As and Bs. I also have no qualms about failing graduate students for cheating - so be good.

last modified 4/8/12