Is Music Audio Processing Embarrassingly Parallel?

Designing languages and systems for many-core computers is in some respects just like designing for uniprocessors. The challenge is, on the one hand, to develop an efficient organization of data, hardware, and software to accomplish the desired computation and, on the other hand, to invent a notation that is expressive and can be translated automatically to executable instructions. With many-core computers, it is far from obvious how to organize computation to run efficiently, but it seems important to think about this aspect of the problem. If nothing else, we need to have some idea of what a running program will look like so that we can begin to imagine how we might express that program in a high-level language. It is also good to know what a near-optimal program looks like in order to evaluate alternatives. Note that we routinely sacrifice execution efficiency to make programming easier. Careful assembly language programming can often yield a 3x to 10x speedup over compiled languages, but we usually forgo the extra performance to avoid the cost of developing and maintaining carefully tuned, hardwarespecific implementations. It may be that with many-core processors, we will routinely give up a factor of 100 in performance in order to simplify programming.