[chriswarbo]

Supercompilation is "top down": beginning with the given program code and trying to optimise it until it can't find any more improvements. It basically runs an interpreter to evaluate the code, but unknown values (e.g. runtime data) must be treated symbolically. It's similar to aggressively inlining (essentially "calling" functions), and aggressively optimising static calculations.

Supercompilation is good at removing unused scaffolding and indirection, e.g. for code that's written defensively/flexibly, supporting a bunch of fallbacks, override hooks, etc. A common problem with supercompilation is increasing code size, since it replaces many calls to a single general-purpose function/method, with many inlined versions (specialised to various extents).

Superoptimisation is "bottom up": generating small snippets of code from scratch, stopping when it finds something that behaves the same as the original code.