Abstract
For many computer programs today the performance bottleneck isn’t in how quickly the processor can execute the program on the data it’s given — it’s in how fast data can be pulled from memory into the processor. Reducing the number of round trips from the memory to the processor is usually the best thing that can be done to speed up the execution time. The standard way to do this is to hire several very expensive experts to rewrite the program so that it is very careful about how memory is used. My thesis will focus on ways to do this automatically, in ways more approachable for folks who are not necessarily experts in CPU memory hierarchies, that don’t require rewriting the program.