This repository contains implementations of Conway's Game of Life in different programming languages for benchmarking purposes.
- bash
- rust (tested with 1.85.0)
- lua (tested with 5.2.4)
- luajit (tested with 2.1.1720049189)
- hyperfine (tested with 1.19.0)
- wasmtime (tested with 31.0.0)
NB: no particular effort was taken to write code compatible for those particular versions of tools/languages - they're just what I had installed or had easy access to via Fedora's package manager.
lua- whatever version of lua you have installedluajit- whatever version of luajit you have installedrust- compile Rust using your default toolchain to native coderust-wasip2- compile Rust as above but to wasm and run it via wasmtimerust-wasip2-precompiledas above but also have wasmtime precompile the wasm to native code (no JIT compilation)
The implementations haven't had crazy optimizations applied to them; the goal is to have the implementations be the "obvious idiomatic" code.
Use the run.sh script to execute a simulation:
./run.sh <language> [args...]Each implementation accepts exactly 4 command-line arguments as args:
print_mode- One of: "all", "final", "none"- "all": Print initial state and every generation
- "final": Print only the final state
- "none": No printing
size- Positive integer specifying the size of the square griditerations- Non-negative integer specifying number of generations to simulatepattern_file- Path to a file containing the initial pattern.- The pattern format is that of the plaintext format from the game of life wiki; see the example in patterns/.
Example:
./run.sh rust 80 1024 patterns/10cellinfinitegrowth.cellsUse the build.sh script to build all implementations:
./build.shUse the verify.sh script to verify implementations have identical output:
./verify.shUse the bench.sh script to run benchmarks:
./bench.sh 80 1024 patterns/10cellinfinitegrowth.cellsWhen benchmarking, the implementations skip actually writing each generation's output state to stdout and only the final generation is output; this is to avoid accidentally bottlenecking on IO or string concatenation.
Benchmarking results will be output to stdout and saved to bench_results.json and bench_results.md.
Here's an example run for 1024 generations on a 80x80 grid:
$ ./bench.sh 80 1024 patterns/10cellinfinitegrowth.cells
<earlier output elided>
Summary
./run.sh rust final 80 1024 patterns/10cellinfinitegrowth.cells ran
1.49 ± 0.04 times faster than ./run.sh rust-wasip2-precompile final 80 1024 patterns/10cellinfinitegrowth.cells
1.57 ± 0.06 times faster than ./run.sh rust-wasip2 final 80 1024 patterns/10cellinfinitegrowth.cells
7.33 ± 3.00 times faster than ./run.sh luajit final 80 1024 patterns/10cellinfinitegrowth.cells
89.00 ± 2.43 times faster than ./run.sh lua final 80 1024 patterns/10cellinfinitegrowth.cells
All implementations follow the specification at gol-spec.md and should have identical behavior.
Generally you can point an LLM at that spec and they'll do a decent first pass of implementing a new language.