#quantum #circuit #simulator #wrapper


Wrapper around the Quantum Exact Simulation Toolkit (QuEST)

8 releases

0.3.7 Sep 8, 2023
0.3.6 Aug 23, 2023
0.3.4 Jul 28, 2023

#55 in Science

MIT license

27K SLoC

C++ 9K SLoC // 0.1% comments C 8K SLoC // 0.2% comments Rust 7K SLoC // 0.0% comments JavaScript 3K SLoC // 0.2% comments


Test Docs

A wrapper around QuEST v3.5.0.

Quantum Exact Simulation Toolkit (QuEST) is a no-fluff, bent-on-speed quantum circuit simulator [1]. It is distributed under MIT License.

How to use it

Initialize a new binary crate:

cargo new tryme
cd tryme/

Add quest_bind to your project's dependencies:

cargo add quest_bind

Now write some code and put it in ./src/main.rs:

use quest_bind::*;

fn main() -> Result<(), QuestError> {
    // Initialize QuEST environment and report to screen
    let env = &QuestEnv::new();

    // Create a 2-qubit register and report its parameters
    let mut qureg = Qureg::try_new(2, env).expect("cannot allocate new Qureg");
    // Initialize |00> state and print out the state to screen

    // Prepare a Bell state `|00> + |11>`: apply Hadamard gate
    // on qubit 0, then NOT on qubit 1, controlled by qubit 0.
    println!("---\nPrepare Bell state: |00> + |11>");
    qureg.hadamard(0).and(qureg.controlled_not(0, 1))?;

    // Measure both qubits
    let outcome0 = qureg.measure(0)?;
    let outcome1 = qureg.measure(1)?;
    println!("Qubit \"0\" measured in state: |{outcome0}>");
    println!("Qubit \"1\" measured in state: |{outcome1}>");

    // Because the state was entangled, the outcomes
    // should always be the same
    if outcome0 == outcome1 {
        println!("They match!");
    } else {
        panic!("qubits in Bell state should be perfectly correlated");

    // At this point both `qureg` and `env` are dropped and
    // the allocated memory is freed.

The documentation is available online, as well as locally:

cargo doc --open

Lastly, compile and run the program:

cargo run

You should be able to see something like:

Running locally on one node
Number of ranks is 1
OpenMP enabled
Number of threads available is 8
Precision: size of qreal is 8 bytes
Number of qubits is 2.
Number of amps is 4.
Number of amps per rank is 4.
Prepare Bell state: |00> + |11>
Qubit "0" measured in state: |0>
Qubit "1" measured in state: |0>
They match!

Distributed and GPU-accelerated mode

QuEST support for MPI and GPU-accelerated computation ca be enabled in quest_bind by setting appropriate feature flags. To enable QuEST's MPI mode, set the mpi feature for quest_bind. Simply edit Cargo.toml of your binary crate:

name = "tryme"
version = "0.1.0"
edition = "2021"

quest_bind = { features = ["mpi"] }

Now if you compile and run the above program again, the output should be:

Running distributed (MPI) version
Number of ranks is 1

The feature "gpu" enables the GPU-accelerated mode. These two features are mutually exclusive and in case both flags are set, the feature "mpi" takes precedence.


To run unit tests for this library, first clone the repository together with QuEST source code as submodule:

git clone --recurse-submodules https://github.com/marek-miller/quest_bind.git
cd quest_bind

Then run:

cargo test

Note that quest_bind will not run QuEST's test suite, nor will it check QuEST's correctness. The tests here are intended to check if the C API is invoked correctly, and if Rust's types are passed safely back and forth across the FFI boundary.

If you want to run the test suite in the single-precision floating point mode, make sure the build script recompiles libQuEST.so with the right type definitions:

cargo clean
cargo test --features=f32

By defualt, quest_bind uses Rust's double precision floating-point type: f64. See Numercal types section below.

You can also try the available examples by running, e.g.:

 cargo run --release --example grovers_search

To see the list of all available examples, try:

cargo run --example

Note on performance

In the typical case when it's the numerical computation that dominates the CPU usage, and not API calls, there should be no discernible difference in performance between programs calling QuEST routines directly and analogous applications using quest_bind. Remember, however, to enable optimizations for both quest_bind and QuEST by compiling your code using the "release" profile:

cargo run --release

Handling exceptions

On failure, QuEST throws exceptions via user-configurable global invalidQuESTInputError(). By default, this function prints an error message and aborts, which is problematic in a large distributed setup.

We opt for catching all exceptions early by reimplementing invalidQuESTInputError() to unwind the stack using Rust's panic mechanism.

Additionally, all error messages reported by QuEST are logged as errors. To be able to see them, add a logger as a dependency to your crate, e.g.:

cargo add env_logger

Then enable logging in your application:

fn main()  {
    // (...)

and run:

RUST_LOG=info cargo run

See log crate for more on logging in Rust.

The type QuestError doesn't contain (possibly malformed) data returned by the API call on failure. Only successful calls can reach the library user. This is intentional, following guidelines from the QuEST documentation:

[Upon failure] Users must ensure that the triggered API call does not continue (e.g. the user exits or throws an exception), else QuEST will continue with the valid [sic!] input and likely trigger a seg-fault.

See Quest API for more information.

Numerical types

For now, numerical types used by quest_bind match exactly the C types that QuEST uses on x86_64. This is a safe, but not very portable strategy. We pass Rust types directly to QuEST without casting, assuming the following type definitions:

pub type c_float = f32;
pub type c_double = f64;
pub type c_int = i32;
pub type c_longlong = i64;
pub type c_ulong = u64;

This should work for many different architectures. If your system uses slightly different numerical types, quest_bind simply won't compile and there is not much you can do besides manually altering the source code.

To check what C types are defined by your Rust installation, see the local documentation for the module std::ffi in Rust's Standard Library:

rustup doc


Here's a few things to know, if you'd like to contribute to quest_bind.

  • The Rust codebase is formatted according to the settings in ./rustfmt.toml. We enable some unstable features of rustfmt. To format your patches correctly, you will need the nightly version of the Rust compiler. Before opening a pull request, remove lint from the code by running:

    just lint