uvie

A really fast Vietnamese input method engine in Rust.

Why it is "ultra fast"

Benchmarks (cargo bench) usually put uvie in the ns → low µs range per sequence.

Why it’s fast:

• Minimal allocation (fixed small buffers; optional heapless mode)
• Single-pass transforms (no extra scans / passes)
• Small fixed-capacity buffers (cache-friendly, predictable)
• CPU-friendly control flow: reduce unpredictable branches to lower branch-misprediction stalls
• Branch-reduction techniques: table lookups, bitmasks, and “branchless” selection patterns instead of large match/if ladders
• O(1) query-time operations for the core lookup/transform steps
• Optimized for tight loops.

Features

• Supports Telex and VNI input methods.
• Easy to use: simple API, no dependencies, easy to embed, extensible.
• Default (std): normal Rust String buffers.
• heapless: uses fixed-capacity heapless::String buffers (no heap allocation from the engine itself).
• Can be built in a heapless-friendly configuration for embedded devices, low-resources environments.

Note: in heapless mode, if internal buffers overflow, output may be truncated.

How to use

Add it to your project:

[dependencies]
uvie = { path = "../uvie" }

Telex:

use uvie::{InputMethod, UltraFastViEngine};

let mut e = UltraFastViEngine::new();
e.set_input_method(InputMethod::Telex);

for ch in "phoos".chars() {
    e.feed(ch);
}
assert_eq!(e.feed(' '), "phố ");

VNI:

use uvie::{InputMethod, UltraFastViEngine};

let mut e = UltraFastViEngine::new();
e.set_input_method(InputMethod::Vni);

for ch in "viet65".chars() {
    e.feed(ch);
}
assert_eq!(e.feed(' '), "việt ");

Embedded/heapless check:

cargo check --no-default-features --features heapless

CLI demo

The repository contains a small interactive CLI (enabled only with std).

cargo run -- --mode telex
cargo run -- --mode vni

Controls:

• Press Enter to flush
• Press Ctrl+C to exit

Benchmarks (uvie vs vi)

Benchmarks use criterion.

cargo bench

cargo bench runs benchmarks in the bench profile (release target).

The benchmark file is in benches/perf.rs and currently benchmarks:

• uvie_telex
• uvie_vni

It also includes direct comparisons against the vi crate:

• compare_telex/*
• compare_vni/*

Fairness notes

• uvie is benchmarked by reusing a single UltraFastViEngine instance per benchmark and calling clear() between iterations.
• vi is benchmarked via vi::methods::transform_buffer, reusing a single String output buffer per benchmark iteration.

Results

The exact numbers depend on CPU/OS, but the ratio is stable.

Sample run (Apple Silicon, cargo bench):

See more details at online report

Embedded / heapless build

To build the library without default std and with heapless buffers:

cargo check --no-default-features --features heapless