I've never used it, but if memory serves they are, like many other companies, using a fork clang. So I expect it will be about as fast if not slower then clang.