Cairo M is a brand new Cpu AIR leveraging M31 as its native prime field, unleashing the Maximum power of Starkware's S-TWO to enable Mobile proving.
Cairo Max looks like Cairo Zero, just tastes better.
Cairo M is designed to provide efficient computation on consumer hardware, especially mobile phone. It includes several components that work together to compile, run, and prove Cairo programs.
The main design choices are
- keep registers count at their minimum, using only
pc(program counter) andfp(frame pointer) - each frame is of constant deterministic size
- read-write memory
- variable-size instruction encoding (x86 style)
- support of native types (u8, u32, etc.) thanks to an heavy use of Stwo's component system
- Compiler: The
cairo-m-compilercrate handles the compilation of Cairo code into Cairo Assembly for execution. - Runner: The
cairo-m-runnercrate is responsible for executing the compiled Cairo programs and generate the trace. - Prover: The
cairo-m-provercrate generates proofs of correct execution for the programs run using the runner.
To get started with Cairo M, clone the repository and build the project using
Cargo. Make sure you have the Rust nightly toolchain installed as specified in
rust-toolchain.toml.
Stwo is added as a git submodule to allow for an easier debugging of the AIRs. Use
git submodule update --init --recursiveto pull the correct pinned version.
Note for MacOS users: You need to have lld installed, which can be done
easily with brew install lld. Also, make sure that your clang is the
Homebrew one (brew install llvm), not the Xcode one.
The project uses trunk.io for managing all the linters, so make sure to install both the CLI and the VScode extension.
RUSTFLAGS="-C target-cpu=native" cargo bench --bench vm_benchmark -- --verboseThe command will run all benchmark functions from the VM and display the throughput results in your terminal.
Samply can be used
cargo install --locked samplyCompile the program you want to run and launch samply:
cargo build -r
target/release/cairo-m-compiler --input crates/prover/tests/test_data/fibonacci.cm --output crates/prover/tests/test_data/fibonacci.json
samply record target/release/cairo-m-prover --entrypoint fib --arguments 100000 crates/prover/tests/test_data/fibonacci.jsonFor memory, it leverages DHAT.
CARGO_PROFILE_RELEASE_DEBUG=true cargo test --release --package cairo-m-prover --test prover --features dhat-heap -- test_memory_profile_fibonacci_prover --nocaptureYou can visualize with dh_view