I built a programming language from raw assembly — and it beats C by 6.6 on LZ77

So I had a slightly unreasonable idea two years ago:

What if I built a programming language without depending on C, LLVM, or any existing toolchain?

The result is Jda — a high-performance systems language bootstrapped from raw x86-64 assembly. The compiler now compiles itself. No GC. No runtime. Single static binaries.

And on real benchmarks, it’s doing things I genuinely didn’t expect.

The numbers that made me double-check

I ran a 6-language benchmark suite (C, Rust, Go, Jda, Python, Ruby) on real algorithms — not microbenchmarks.

• Sudoku (500 puzzles): C 62ms, Rust 62ms, Go 66ms, Jda 41ms

• LZ77 (1 MB compress): C 1,830ms, Rust 2,185ms, Go 2,721ms, Jda 277ms

That’s 1.5× faster than C on Sudoku and 6.6× faster than C on LZ77. Running via Rosetta 2, no less — C/Rust/Go were native ARM64, Jda was x86-64 emulated.

How? Source-level optimizations the C compiler can’t easily do: MOD→AND strength reduction, hash-chain hoisting, aggressive DCE.

Compile times are kinda absurd too

33× faster compilation than Rust, 16× faster than Go. Skipping LLVM saves a lot of time.

A taste of the syntax

fn search_file(path: &i8, pattern: &i8) -> i64 {

let fd = file_open(path, 0)

let buf = file_read_all(fd)

let matches = 0

for i in range(str_len(buf)) {

if substr_match(buf, i, pattern) {

matches += 1

}

}

ret matches

}

A real ripgrep-style search tool, ~400 lines total, compiles to a 1 MB static binary with zero dependencies.

What’s in the box

• Self-hosted compiler — byte-identical fixed point reached April 2026

• 117 stdlib packages — HTTP, JSON, crypto, tensors, neural networks

• Built-in concurrency — goroutine-style green threads, no GC

• 388 conformance tests passing

• Native installers for Windows, macOS, Linux

Try it

Repo: https://www.github.com/jdalang/jda-lang

Happy to answer questions about the bootstrap process, the codegen, or why on earth I did this in the comments.