Do something to parse assembly code or object code and turn it into some approximation to the Sparse linearized instruction format. This will help with analysis of the Linux kernel.

Ideally, this would also support inline assembly blocks. Those may even prove easier to start on.