Chapter 16: Performance

Level 3 — Experience

Inline functions, restrict qualifiers, and measurement -- squeezing every cycle out of C, and how Auto handles optimization for you.

Performance matters. Not for every program, and not at every moment, but when it matters, C programmers need to understand how to write fast code and how to prove that their code is fast. C gives you fine-grained control over performance through keywords like inline and restrict, through compiler attributes, and through careful measurement. Auto takes a different approach: the compiler handles most optimizations, and you focus on writing clear code.

---

16.1 Inline Functions

The C inline keyword suggests to the compiler that a function's body should be substituted directly at the call site, avoiding the overhead of a function call:

// C Deep Dive: inline functions
static inline int square(int n) {
    return n * n;
}

static inline int max(int a, int b) {
    return (a > b) ? a : b;
}

The rules for inline in C are subtle and frequently misunderstood:

• inline alone does not guarantee inlining -- it is a hint.
• An inline function defined in a header needs a corresponding external definition in exactly one translation unit, or the program has undefined behavior (C99/C11).
• static inline is the practical solution: each TU gets its own copy, and the linker sees no duplicate symbols.
• Compilers are free to ignore inline entirely. Modern compilers often inline functions that are not marked inline and refuse to inline functions that are.

C Deep Dive: The interaction between inline and linkage in C is one of the most confusing aspects of the language. C99 introduced inline with external linkage semantics that differ from C++. In practice, most C programmers use static inline in headers and avoid the complexity. C23 simplifies this by making inline behave more like C++.

Auto's approach. Auto does not have an inline keyword. The compiler decides inlining automatically based on heuristics:

• Small functions are always inlined.
• Functions called in hot loops are candidates for inlining.
• The programmer can annotate performance-critical functions, but the compiler makes the final decision.

// Auto: no inline keyword needed
fn square(n int) int {
    n * n
}

fn max(a int, b int) int {
    if a > b { a } else { b }
}

fn main() {
    print("square(7) =", square(7))
    print("max(3, 7) =", max(3, 7))
}