Cell Processor (aka Broadband Engine)

The general idea

The Cell has a general-purpose PowerPC processor and 8 specialized vector cores, each similar in style to a multimedia instruction set. The vector units each have 256KB of local memory, to and from which data and instructions must be explicitly transferred. DMA operations are available so data transfers can happen in the background.

The challenge

The Cell design incorporates the following, each of which would present a programming challenge in itself:

multiprocessor programming
specialist vector instruction set
programmer-controlled memory hierarchy

You are required to select an example published by IBM (or another source) and understand this example, both in its strengths and limitations, and explore how this example may be generalised. For example, if an example only works for a data set with very specific characteristics, how would you generalize the code for arbitrary data?