To get started on how to setup and run cairo-native check the getting started section.

To read more in-depth documentation, visit this page.

• Implemented Library Functions
• Getting started
• Command Line Interface
• Benchmarking
• API usage example
• From MLIR to native binary
• Cairo Test CLI Tool
• Scarb Test CLI Tool
• Scarb Dump CLI Tool
• Ahead of time compilation remarks

⚠️ Disclaimer

🚧 cairo-native is still being built therefore API breaking changes might happen often so use it at your own risk. 🚧

For versions under 1.0 cargo doesn't comply with semver, so we advise to pin the version the version you use. This can be done by adding cairo-native = "0.1.0" to your Cargo.toml

Implemented Library Functions

Cairo Native works by leveraging the intermediate representation of Cairo called Sierra. Sierra uses a list of builtin functions that implement the language functionality, those are called library functions, short: libfuncs. Basically every statement in a sierra program is a call to a libfunc, thus they are the core of Cairo Native progress towards feature parity.

This is a list of the current progress implementing each libfunc.

Footnotes on the libfuncs list:

1. It is implemented but we're not handling potential issues like lifetimes yet.
2. It is implemented but we're still debating whether it should be a Rust-like Box<T> or if it's fine treating it like another variable.
3. It is implemented but side-effects are not yet handled (ex. array cloning/dropping).
4. Not supported by the Cairo to Sierra compiler.
5. Implemented with a dummy. It doesn't do anything yet.

Getting Started

Dependencies

• Linux or macOS (aarch64 included) only for now
• LLVM 18 with MLIR: On debian you can use apt.llvm.org, on macOS you can use brew
• Rust 1.78.0 or later, since we make use of the u128 abi change.
• Git

Setup

This step applies to all operating systems.

Run the following make target to install the dependencies (both Linux and macOS):

make deps

Linux

Since Linux distributions change widely, you need to install LLVM 18 via your package manager, compile it or check if the current release has a Linux binary.

If you are on Debian/Ubuntu, check out the repository https://apt.llvm.org/ Then you can install with:

sudo apt-get install llvm-18 llvm-18-dev llvm-18-runtime clang-18 clang-tools-18 lld-18 libpolly-18-dev libmlir-18-dev mlir-18-tools

If you decide to build from source, here are some indications:

# Go to https://github.com/llvm/llvm-project/releases
# Download the latest LLVM 18 release:
# The blob to download is called llvm-project-18.x.x.src.tar.xz

# For example
wget https://github.com/llvm/llvm-project/releases/download/llvmorg-18.1.7/llvm-project-18.1.7.src.tar.xz
tar xf llvm-project-18.1.7.src.tar.xz

cd llvm-project-18.1.7.src.tar
mkdir build
cd build

# The following cmake command configures the build to be installed to /opt/llvm-18
cmake -G Ninja ../llvm \
   -DLLVM_ENABLE_PROJECTS="mlir;clang;clang-tools-extra;lld;polly" \
   -DLLVM_BUILD_EXAMPLES=OFF \
   -DLLVM_TARGETS_TO_BUILD="Native" \
   -DCMAKE_INSTALL_PREFIX=/opt/llvm-18 \
   -DCMAKE_BUILD_TYPE=RelWithDebInfo \
   -DLLVM_PARALLEL_LINK_JOBS=4 \
   -DLLVM_ENABLE_BINDINGS=OFF \
   -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DLLVM_ENABLE_LLD=ON \
   -DLLVM_ENABLE_ASSERTIONS=OFF

ninja install

Setup a environment variable called MLIR_SYS_180_PREFIX, LLVM_SYS_181_PREFIX and TABLEGEN_180_PREFIX pointing to the llvm directory:

# For Debian/Ubuntu using the repository, the path will be /usr/lib/llvm-18
export MLIR_SYS_180_PREFIX=/usr/lib/llvm-18
export LLVM_SYS_181_PREFIX=/usr/lib/llvm-18
export TABLEGEN_180_PREFIX=/usr/lib/llvm-18

Alternatively, if installed from Debian/Ubuntu repository, then you can use env.sh to automatically setup the environment variables.

source env.sh

MacOS

The makefile deps target (which you should have ran before) installs LLVM 18 with brew for you, afterwards you need to execute the env.sh script to setup the needed environment variables.

source env.sh

Env Variables

The script env.sh automatically sets the necessary variables for MacOS and Ubuntu/Debian. If it doesn't fit your specific environment you can copy it to .env/.envrc and adapt it acordingly.

source env.sh

The script is compatible with direnv

Make commands:

Running make by itself will list available targets.

• Install the necessary dependencies (on Linux, you need to get LLVM 18 manually):

make deps

• Build a release version:

make build

Or with your native CPU Architecture for even more performance (usually):

make build-native

• Install the cairo-native-dump and cairo-native-run commands:

make install

• Build a optimized development version:

make build-dev

• View and open the docs:

make doc-open

• Run the tests:

make test

• Generate coverage:

make coverage

• Run clippy and format checks:

make check

• Compile the runtime library used for ahead of time compilation:

make runtime

Command Line Interface

cairo-native-dump:

Usage: cairo-native-dump [OPTIONS] <INPUT>

Arguments:
  <INPUT>

Options:
  -o, --output <OUTPUT>  [default: -]
  -h, --help             Print help

cairo-native-run:

This tool allows to run programs using the JIT engine, like the cairo-run tool, the parameters can only be felt values.

echo '1' | cairo-native-run 'program.cairo' 'program::program::main' --inputs - --outputs -

Usage: cairo-native-run [OPTIONS] <INPUT> <ENTRY_POINT>

Arguments:
  <INPUT>
  <ENTRY_POINT>

Options:
  -i, --inputs <INPUTS>
  -o, --outputs <OUTPUTS>
  -p, --print-outputs
  -h, --help               Print help

API usage example

This is a usage example using the API for an easy Cairo program that requires the least setup to get running. It allows you to compile and execute a program using the JIT.

Example code to run a program:

use starknet_types_core::felt::Felt;
use cairo_native::context::NativeContext;
use cairo_native::executor::NativeExecutor;
use cairo_native::values::JitValue;
use std::path::Path;

fn main() {
    let program_path = Path::new("programs/examples/hello.cairo");
    // Compile the cairo program to sierra.
    let sierra_program = cairo_native::utils::cairo_to_sierra(program_path);

    // Instantiate a Cairo Native MLIR context. This data structure is responsible for the MLIR
    // initialization and compilation of sierra programs into a MLIR module.
    let native_context = NativeContext::new();

    // Compile the sierra program into a MLIR module.
    let native_program = native_context.compile(&sierra_program).unwrap();

    // The parameters of the entry point.
    let params = &[JitValue::Felt252(Felt::from_bytes_be_slice(b"user"))];

    // Find the entry point id by its name.
    let entry_point = "hello::hello::greet";
    let entry_point_id = cairo_native::utils::find_function_id(&sierra_program, entry_point);

    // Instantiate the executor.
    let native_executor = NativeExecutor::new(native_program);

    // Execute the program.
    let result = native_executor
        .execute(entry_point_id, params, None)
        .unwrap();

    println!("Cairo program was compiled and executed successfully.");
    println!("{:?}", result);
}

Example code to run a Starknet contract:

use starknet_types_core::felt::Felt;
use cairo_lang_compiler::CompilerConfig;
use cairo_lang_starknet::contract_class::compile_path;
use cairo_native::context::NativeContext;
use cairo_native::executor::NativeExecutor;
use cairo_native::utils::find_entry_point_by_idx;
use cairo_native::values::JitValue;
use cairo_native::{
    metadata::syscall_handler::SyscallHandlerMeta,
    starknet::{BlockInfo, ExecutionInfo, StarkNetSyscallHandler, SyscallResult, TxInfo, U256},
};
use std::path::Path;

/// To run a starknet contract, we need to use a syscall handler, here we show how to implement one (at the end).
#[derive(Debug)]
struct SyscallHandler;

fn main() {
    let path = Path::new("programs/examples/hello_starknet.cairo");

    let contract = compile_path(
        path,
        None,
        CompilerConfig {
            replace_ids: true,
            ..Default::default()
        },
    )
    .unwrap();

    let entry_point = contract.entry_points_by_type.constructor.get(0).unwrap();
    let sierra_program = contract.extract_sierra_program().unwrap();

    let native_context = NativeContext::new();

    let mut native_program = native_context.compile(&sierra_program).unwrap();
    native_program
        .insert_metadata(SyscallHandlerMeta::new(&mut SyscallHandler))
        .unwrap();

    // Call the echo function from the contract using the generated wrapper.
    let entry_point_fn =
        find_entry_point_by_idx(&sierra_program, entry_point.function_idx).unwrap();

    let fn_id = &entry_point_fn.id;

    let native_executor = NativeExecutor::new(native_program);

    let result = native_executor
        .execute_contract(
            fn_id,
            // The calldata
            &[JitValue::Felt252(Felt::ONE)],
            u64::MAX.into(),
        )
        .expect("failed to execute the given contract");

    println!();
    println!("Cairo program was compiled and executed successfully.");
    println!("{result:#?}");
}

// Implement an example syscall handler.
impl StarkNetSyscallHandler for SyscallHandler {
    fn get_block_hash(
        &mut self,
        block_number: u64,
        _gas: &mut u128,
    ) -> SyscallResult<Felt> {
        println!("Called `get_block_hash({block_number})` from MLIR.");
        Ok(Felt::from_bytes_be_slice(b"get_block_hash ok"))
    }

    fn get_execution_info(
        &mut self,
        _gas: &mut u128,
    ) -> SyscallResult<cairo_native::starknet::ExecutionInfo> {
        println!("Called `get_execution_info()` from MLIR.");
        Ok(ExecutionInfo {
            block_info: BlockInfo {
                block_number: 1234,
                block_timestamp: 2345,
                sequencer_address: 3456.into(),
            },
            tx_info: TxInfo {
                version: 4567.into(),
                account_contract_address: 5678.into(),
                max_fee: 6789,
                signature: vec![1248.into(), 2486.into()],
                transaction_hash: 9876.into(),
                chain_id: 8765.into(),
                nonce: 7654.into(),
            },
            caller_address: 6543.into(),
            contract_address: 5432.into(),
            entry_point_selector: 4321.into(),
        })
    }

    fn deploy(
        &mut self,
        class_hash: Felt,
        contract_address_salt: Felt,
        calldata: &[Felt],
        deploy_from_zero: bool,
        _gas: &mut u128,
    ) -> SyscallResult<(Felt, Vec<Felt>)> {
        println!("Called `deploy({class_hash}, {contract_address_salt}, {calldata:?}, {deploy_from_zero})` from MLIR.");
        Ok((
            class_hash + contract_address_salt,
            calldata.iter().map(|x| x + &Felt::ONE).collect(),
        ))
    }

    fn replace_class(
        &mut self,
        class_hash: Felt,
        _gas: &mut u128,
    ) -> SyscallResult<()> {
        println!("Called `replace_class({class_hash})` from MLIR.");
        Ok(())
    }

    fn library_call(
        &mut self,
        class_hash: Felt,
        function_selector: Felt,
        calldata: &[Felt],
        _gas: &mut u128,
    ) -> SyscallResult<Vec<Felt>> {
        println!(
            "Called `library_call({class_hash}, {function_selector}, {calldata:?})` from MLIR."
        );
        Ok(calldata.iter().map(|x| x * Felt::from(3)).collect())
    }

    fn call_contract(
        &mut self,
        address: Felt,
        entry_point_selector: Felt,
        calldata: &[Felt],
        _gas: &mut u128,
    ) -> SyscallResult<Vec<Felt>> {
        println!(
            "Called `call_contract({address}, {entry_point_selector}, {calldata:?})` from MLIR."
        );
        Ok(calldata.iter().map(|x| x * Felt::from(3)).collect())
    }

    fn storage_read(
        &mut self,
        address_domain: u32,
        address: Felt,
        _gas: &mut u128,
    ) -> SyscallResult<Felt> {
        println!("Called `storage_read({address_domain}, {address})` from MLIR.");
        Ok(address * Felt::from(3))
    }

    fn storage_write(
        &mut self,
        address_domain: u32,
        address: Felt,
        value: Felt,
        _gas: &mut u128,
    ) -> SyscallResult<()> {
        println!("Called `storage_write({address_domain}, {address}, {value})` from MLIR.");
        Ok(())
    }

    fn emit_event(
        &mut self,
        keys: &[Felt],
        data: &[Felt],
        _gas: &mut u128,
    ) -> SyscallResult<()> {
        println!("Called `emit_event({keys:?}, {data:?})` from MLIR.");
        Ok(())
    }

    fn send_message_to_l1(
        &mut self,
        to_address: Felt,
        payload: &[Felt],
        _gas: &mut u128,
    ) -> SyscallResult<()> {
        println!("Called `send_message_to_l1({to_address}, {payload:?})` from MLIR.");
        Ok(())
    }

    fn keccak(
        &mut self,
        input: &[u64],
        _gas: &mut u128,
    ) -> SyscallResult<cairo_native::starknet::U256> {
        println!("Called `keccak({input:?})` from MLIR.");
        Ok(U256(Felt::from(1234567890).to_le_bytes()))
    }

    /*
    ... more code here, check out the full example in examples/starknet.rsd
    */
}

For more examples, check out the examples/ directory.

Benchmarking

Requirements

• hyperfine: cargo install hyperfine
• cairo 2.6.4
• Cairo Corelibs
• LLVM 18 with MLIR

You need to setup some environment variables:

$MLIR_SYS_180_PREFIX=/path/to/llvm18  # Required for non-standard LLVM install locations.
$LLVM_SYS_181_PREFIX=/path/to/llvm18  # Required for non-standard LLVM install locations.
$TABLEGEN_180_PREFIX=/path/to/llvm18  # Required for non-standard LLVM install locations.

make bench

The bench target will run the ./scripts/bench-hyperfine.sh script. This script runs hyperfine commands to compare the execution time of programs in the ./programs/benches/ folder. Each program is compiled and executed via the execution engine with the cairo-native-run command and via the cairo-vm with the cairo-run command provided by the cairo codebase. The cairo-run command should be available in the $PATH and ideally compiled with cargo build --release. If you want the benchmarks to run using a specific build, or the cairo-run commands conflicts with something (e.g. the cairo-svg package binaries in macos) then the command to run cairo-run with a full path can be specified with the $CAIRO_RUN environment variable.

From MLIR to native binary

# to mlir with llvm dialect
sierra2mlir program.sierra -o program.mlir

# translate all dialects to the llvm dialect
"$MLIR_SYS_180_PREFIX/bin/mlir-opt" \
        --canonicalize \
        --convert-scf-to-cf \
        --canonicalize \
        --cse \
        --expand-strided-metadata \
        --finalize-memref-to-llvm \
        --convert-func-to-llvm \
        --convert-index-to-llvm \
        --reconcile-unrealized-casts \
        "program.mlir" \
        -o "program-llvm.mlir"

# translate mlir to llvm-ir
"$MLIR_SYS_180_PREFIX"/bin/mlir-translate --mlir-to-llvmir program-llvm.mlir -o program.ll

# compile natively
"$MLIR_SYS_180_PREFIX"/bin/clang program.ll -Wno-override-module \
    -L "$MLIR_SYS_180_PREFIX"/lib -L"./target/release/" \
    -lcairo_native_runtime -lmlir_c_runner_utils \
    -Wl,-rpath "$MLIR_SYS_180_PREFIX"/lib \
    -Wl,-rpath ./target/release/ \
    -o program

./program

cairo-native-test cli tool

This tool mimics the cairo-test tool and is identical to it, the only feature it doesn't have is the profiler.

You can download it on our releases page.

$ cairo-native-test --help
Compiles a Cairo project and runs all the functions marked as `#[test]`.
Exits with 1 if the compilation or run fails, otherwise 0.

Usage: cairo-native-test [OPTIONS] <PATH>

Arguments:
  <PATH>  The Cairo project path to compile and run its tests

Options:
  -s, --single-file            Whether path is a single file
      --allow-warnings         Allows the compilation to succeed with warnings
  -f, --filter <FILTER>        The filter for the tests, running only tests containing the filter string [default: ]
      --include-ignored        Should we run ignored tests as well
      --ignored                Should we run only the ignored tests
      --starknet               Should we add the starknet plugin to run the tests
      --run-mode <RUN_MODE>    Run with JIT or AOT (compiled) [default: jit] [possible values: aot, jit]
  -O, --opt-level <OPT_LEVEL>  Optimization level, Valid: 0, 1, 2, 3. Values higher than 3 are considered as 3 [default: 0]
  -h, --help                   Print help
  -V, --version                Print version

For single files, you can use the -s, --single-file option.

For a project, it needs to have a cairo_project.toml specifying the crate_roots. You can find an example under the cairo-tests/ folder, which is a cairo project that works with this tool.

cairo-native-test -s myfile.cairo

cairo-native-test ./cairo-tests/

This will run all the tests (functions marked with the #[test] attribute).

cairo-native-stress cli tool

This tool runs a stress test on Cairo Native.

$ cairo-native-stress --help
A stress tester for Cairo Native

It Sierra programs compiles with Cairo Native, caches, and executes them with AOT runner. The compiled dynamic libraries are stored in `AOT_CACHE_DIR` relative to the current working directory.

Usage: cairo-native-stress [OPTIONS] <ROUNDS>

Arguments:
  <ROUNDS>
          Amount of rounds to execute

Options:
  -o, --output <OUTPUT>
          Output file for JSON formatted logs

  -h, --help
          Print help (see a summary with '-h')

To quickly run a stress test and save logs as json, run:

make stress-test

This takes a lot of time to finish (it will probably crash first), you can kill the program at any time.

To plot the results, run:

make stress-plot

To clear the cache directory, run:

make stress-clean

scarb-native-test cli tool

This tool mimics the scarb test command. You can download it on our releases page.

$ scarb-native-test --help
Compiles all packages from a Scarb project matching `packages_filter` and
runs all functions marked with `#[test]`. Exits with 1 if the compilation
or run fails, otherwise 0.

Usage: scarb-native-test [OPTIONS]

Options:
  -p, --package <SPEC>         Packages to run this command on, can be a concrete package name (`foobar`) or a prefix glob (`foo*`) [env: SCARB_PACKAGES_FILTER=] [default: *]
  -w, --workspace              Run for all packages in the workspace
  -f, --filter <FILTER>        Run only tests whose name contain FILTER [default: ]
      --include-ignored        Run ignored and not ignored tests
      --ignored                Run only ignored tests
      --run-mode <RUN_MODE>    Run with JIT or AOT (compiled) [default: jit] [possible values: aot, jit]
  -O, --opt-level <OPT_LEVEL>  Optimization level, Valid: 0, 1, 2, 3. Values higher than 3 are considered as 3 [default: 0]
  -h, --help                   Print help
  -V, --version                Print version

scarb-native-dump cli tool

This tool mimics the scarb build command. You can download it on our releases page.

This tool should be run at the directory where a Scarb.toml file is and it will behave like scarb build, leaving the MLIR files under the target/ folder besides the generated JSON sierra files.

Debugging Tips

Useful environment variables

These 2 env vars will dump the generated MLIR code from any compilation on the current working directory as:

• dump.mlir: The MLIR code after passes without locations.
• dump-debug.mlir: The MLIR code after passes with locations.
• dump-prepass.mlir: The MLIR code before without locations.
• dump-prepass-debug.mlir: The MLIR code before passes with locations.

Do note that the MLIR with locations is in pretty form and thus not suitable to pass to mlir-opt.

export NATIVE_DEBUG_DUMP_PREPASS=1
export NATIVE_DEBUG_DUMP=1

Enable logging to see the compilation process:

export RUST_LOG="cairo_native=trace"

Other tips:

• Try to find the minimal program to reproduce an issue, the more isolated the easier to test.
• Use the debug_utils print utilities, more info here:

#[cfg(feature = "with-debug-utils")]
{
    metadata.get_mut::<DebugUtils>()
        .unwrap()
        .print_pointer(context, helper, entry, ptr, location)?;
}

Ahead of time Compilation Remarks

To use the AOT executor, it needs to know where the static runtime library is located, this can be configured using the following env var pointing to the absolute path to the library:

export CAIRO_NATIVE_RUNTIME_LIBRARY=/absolute/path/to/libcairo_native_runtime.a