This is an implementation of a super simple stack based VM in Rust, as a learning exercise for myself. Absolutely do not use this for anything real or production worthy.

The language and APIs will change over time.

A lot of this is based on https://csl.name/post/vm, but while that is written in Python, this is in not :)

I'm using the simplicity of the language to play around with my understanding of Rust, for code organization, performance optimizations, and other idioms. If you check out the benchmarking section, you can see that I've spent a bunch of time on trying to optimize the Fibonnaci program in the examples/ directory, and have made a ton of prorgress. As of d474ca45365c2f6dc3d15c3357f64fa439db5be, fib_10 runs in about 33% of the original runtime.

Right now this is a super simple stack based language that supports:

• Basic values: Num, Bool, and String
• Labels (for jmp, call, and return)
• Operations: swap, rot, over, drop, dup
• Comparison and boolean operations
• Arithmetic operations
• string to int and int to string parsing (no error handling for this)

Assuming you have rustup.

cargo test

Run the fib program for the 5th fibonacci number (debug).

cargo run -- examples/fib 5

Run this in release mode.

cargo run --release -- examples/fib 5

Use cargo benchcmp for bench comparisons.

All of these are run on an Early 2015 13-inch Macbook Pro, 2.9 GHz Intel Core i5, 16GB RAM.

git stash && cargo bench > bench-current && git stash pop && cargo bench > bench-new
cargo benchcmp bench-current bench-new

The idea here is to run a profiler inside a docker container to actually get some flamegraphs out, since running them on MacOS is pretty unweildy, and just straight up might not work at all, but it is usable elsewhere.

We are also using BrendanGregg/FlameGraph, so that should be somewhere in your $PATH.

prof/run.sh fib fib_10
cat perf.script | stackcollapse-perf.pl | flamegraph.pl > bench.svg