Get the latest tech news

How Clang compiles a function (2018)

I’ve been planning on writing a post about how LLVM optimizes a function, but it seems necessary to first write about how Clang translates C or C++ into LLVM. This is going to be fairly high-level: - rather than looking at Clang’s internals, I’m going to focus on how Clang’s output relates to its input - we’re not going to look at any non-trivial C++ features We’ll use this small function that I borrowed from these excellent lecture notes on loop optimizations: bool is_sorted(int *a, int n) { for (int i = 0; i < n - 1; i++) if (a[i] > a[i + 1]) return false; return true; } Since Clang doesn’t do any optimization, and since LLVM IR was initially designed as a target for C and C++, the translation is going to be relatively easy.

Textual IR is bulky and not particularly fast to print or parse; the binary “bitcode” format is always preferred when a human isn’t in the loop. This may all seem confusing and arbitrary but these design decisions have been found to allow many parts of a compiler to be expressed in natural and efficient ways. I haven’t seen the measurements but my understanding is that generating a lot of memory operations and then almost immediately optimizing many of them away isn’t a major source of compile-time overhead.

Get the Android app