Allen gives a well-thought-out and well-organized introduction to dependence analysis for parallel compilers. The survey of basic dependence types is clear and concise, except that anti-dependence is not introduced at this stage. The length and pace of the video make it a good first look at high-performance dependence analysis.

I found the graphics and examples to be helpful in complementing the narration. In particular, the example of a simple Diophantine equation illustrates the concept of detecting dependences well, although a vague reference to an anti-dependence at the end is confusing, since this basic dependence was not explained earlier. The talk also includes an excellent example of the development of a piece of code into its program dependence graph and control dependence graph, but the example is not completed by showing exactly where the parallelism can effectively be extracted.

The delivery starts out a bit choppy, but improves as the lecture progresses, despite occasional distracting hand gestures. These flaws are minor ones in a fundamentally sound piece of teaching in a difficult area of high-performance compiler analysis. Overall, this videotape is essential material on the subject, and the concluding discussion of the system as a whole (tools, user database, user input, and so on) refers to many aspects of computer science that all computer scientists would do well to heed.