Parametric Algorithms

A parametric algorithm describes the logic once, then lets compile-time parameters mint a specialized version for each size, type, or width you need. ⚙️

Anything that changes the code's structure, like a vector width or element type, becomes a parameter instead of a runtime argument.

Use a type parameter so the same routine handles Int32, Float64, or any numeric type without you rewriting the body.

For every distinct set of parameter values, Mojo generates a fresh specialized function, each optimized exactly for that combination.

Because parameters are fixed at compile time, the generated code pays zero runtime cost to read them, unlike normal arguments.

Pair a type parameter with a trait so the body can rely on the methods every conforming type promises to provide.

Mix several parameters at once: a type for the data and an Int for the size or width, all resolved together before the program runs.

Mojo can often infer a parameter from the arguments you pass, so you fill in fewer brackets and the compiler completes the rest.

You keep a single readable definition while the compiler quietly emits a fast tailored build for each parameter combination behind the scenes.

Parametric algorithms power Mojo's fast numeric kernels: one matmul or reduction definition, many specialized versions tuned per type and width. 🚀

The big win is staying readable: your logic reads like normal code, while parameters quietly handle every type and size variation for you.

Think about why parametric algorithms stay fast.

A parametric algorithm is written once but specialized many times: parameters carry the variation, and each build is tuned with no runtime cost. 🎯