Purely out of curiosity, would it be more efficient for GHC to use the new built-in exception handling instructions in LLVM?