LLVM now uses something they call the "Greedy Register Allocator". As far as I can tell, it's a variation on the linear allocator with some heuristics. Here's a presentation: https://www.youtube.com/watch?v=hf8kD-eAaxg