Inline In Rust

There’s a lot of tribal knowledge surrounding #[inline] attribute in Rust. I often find myself teaching how it works, so I finally decided to write this down.

Caveat Emptor: this is what I know, not necessary what is true. Additionally, exact semantics of #[inline] is not set in stone and may change in future Rust versions.

Why Inlining Matters?

Inlining is an optimizing transformation which replaces a call to a function with its body.

To give a trivial example, during compilation the compiler can transform this code:

fn f(w: u32) -> u32 {
    inline_me(w, 2)

fn inline_me(x: u32, y: u32) -> u32 {
    x * y

Into this code:

fn f(w: u32) -> u32 {
    w * 2
There are many obvious advantages to inlining; two are:

a. There is no function call overhead whatsoever.
b. Caller and callee are optimized together. Advantage can be taken
   of particular argument values and relationships: constant arguments
   can be folded into the code, invariant instructions in the callee
   can be moved to infrequently executed areas of the caller, etc.

In other words, for an ahead of time compiled language inlining is the mother of all other optimizations. It gives the compiler the necessary context to apply further transformations.

Inlining and Separate Compilation

Inlining is at odds with another important idea in compilers — that of separate compilation. When compiling big programs, it is desirable to separate them into modules which can be compiled independently to:

  • Process everything in parallel.

  • Scope incremental recompilations to individual changed modules.

To achieve separate compilation, compilers expose signatures of functions, but keep function bodies invisible to other modules, preventing inlining. This fundamental tension is what makes #[inline] in Rust trickier than just a hint for the compiler to inline the function.

Inlining in Rust

In Rust, a unit of (separate) compilation is a crate. If a function f is defined in a crate A, then all calls to f from within A can be inlined, as the compiler has full access to f. If, however, f is called from some downstream crate B, such calls can’t be inlined. B has access only to the signature of f, not its body.

That’s where the main usage of #[inline] comes from — it enables cross-crate inlining. Without #[inline], even the most trivial of functions can’t be inlined across the crate boundary. The benefit is not without a cost — the compiler implements this by compiling a separate copy of the #[inline] function with every crate it is used in, significantly increasing compile times.

Besides #[inline], there are two more exceptions to this. Generic functions are implicitly inlinable. Indeed, the compiler can only compile a generic function when it knows the specific type arguments it is instantiated with. As that is known only in the calling crate, bodies of generic functions have to be always available.

The other exception is link-time optimization. LTO opts out of separate compilation — it makes bodies of all functions available, at the cost of making compilation much slower.

Inlining in Practice

Now that the underlying semantics is explained, it’s possible to infer some rule-of-thumbs for using #[inline].

First, it’s not a good idea to apply #[inline] indiscriminately, as that makes compile time worse. If you don’t care about compile times, a much better solution is to set lto = true in Cargo profile (docs).

Second, it usually isn’t necessary to apply #[inline] to private functions — within a crate, the compiler generally makes good inline decisions. There’s a joke that LLVM’s heuristic for when the function should be inlined is “yes”.

Third, when building an application, apply #[inline] reactively when profiling shows that a particular small function is a bottleneck. Consider using lto for releases. It might make sense to proactively #[inline] trivial public functions.

Fourth, when building libraries, proactively add #[inline] to small non-generic functions. Pay special attention to impls: Deref, AsRef and the like often benefit from inlining. A library can’t anticipate all usages upfront, it makes sense to not prematurely pessimize future users. Note that #[inline] is not transitive: if a trivial public function calls a trivial private function, you need to #[inline] both. See this benchmark for details.

Fifth, mind generic functions. It’s not too wrong to say that generic functions are implicitly inline. As a result, they often are a cause for code bloat. Generic functions, especially in libraries, should be written to minimize unwanted inlining. To give an example from wat:

// Public, generic function.
// Will cause code bloat if not handled carefully!
pub fn parse_str(wat: impl AsRef<str>) -> Result<Vec<u8>> {
  // Immediately delegate to a non-generic function.

// Separate-compilation friendly private implementation.
fn _parse_str(wat: &str) -> Result<Vec<u8>> {


  1. Language reference.

  2. Rust performance book.

  3. @alexcrichton explains inline. Note that, in reality, the compile time costs are worse than what I described — inline functions are compiled per codegen-unit, not per crate.

  4. More @alexcrichton.

  5. Even more @alexcrichton.

Discussion on /r/rust.

There is now a follow up post: It’s Not Always iCache.

This post is a part of One Hundred Thousand Lines of Rust series.