I love reading about LLVM internals. It's exciting to see LLVM still making major algorithmic improvements to something as important as register allocation- it's a really nicely designed library both to use and to work on, and bringing the generated code's performance closer to compilers like GCC makes it even better.