This project provides an OpenQASM source-to-source compiler which supports optimization for different NISQ (Noisy Intermediate-Scale Quantum) devices. It currently contains the CODAR qubit mapping algorithm accepted by DAC 2020. The prototype was developed in C++ by Quanxi Li and Haowei Deng under the guidance of Prof. Yu Zhang in the S4Plus Quantum team at USTC.

The CODAR (COntext-sensitive and Duration-Aware Remapping algorithm) remapper is built on the abstraction of different NISQ devices, called Multi-architecture Adaptive Quantum Abstract Machine (maQAM), and is aware of gate duration difference and quantum program context. You can read the preprint for details of the maQAM and the CODAR algorithm.

If you find the CODAR useful in your research, please consider citing:

@inproceedings{ deng2020codar,
    title	= {{CODAR}: A Contextual Duration-Aware Qubit Mapping for Various {NISQ} Devices},
    author	= {Deng, Haowei and Zhang, Yu and Li, Quanxi},
    booktitle={Proceedings of the 57th Annual Design Automation Conference (DAC 2020)},
	month	= {July}，
    year	={2020},
    organization={IEEE}，
}

@article{deng2020codar,
    title     = {{CODAR}: A Contextual Duration-Aware Qubit Mapping for Various {NISQ} Devices},
    author    = {Deng, Haowei and Zhang, Yu and Li, Quanxi},
    journal   = {arXiv preprint arXiv:2002.10915},
    year      = {2020}，
}

• This project should be built and run on *nix system.
• Run make to build the CODAR source-to-source compiler.
• Run make all to build the compiler, the qasm analyzer tool and the simulator for debugging.
• All executable files will be generated in directory bin/.

First please build the CODAR compiler. By default, the path of it will be bin/codar.

Note: CODAR does not support flattening a program and make sure the module syntax is not used in your OpenQASM program; otherwise please flatten the code before compiling.

Run

    bin/codar

and you will get how to use CODAR as follows

usage: ./codar <filename> [args]
args:
        -o <out_file>   default: output/*.qasm  set output file name
        -m <M>          default: 2              set side length for grid device
        -n <N>          default: 8              set side length for grid device
        -r <run_time>   default: 1              run <run_time> times and output the best one
        -init <file>                            use a specific initial mapping from file
        --init                                  use a specific initial mapping from stdin
        --tokyo                                 use ibm-tokyo 20-qubit device model
        --ibm_q53                               use ibm-Rochester 53-qubit device model
        --google_q54                            use google-Sycamore 54-qubit device model

Compile an OpenQASM program based on a specified grid model

    bin/codar before.qasm -m 6 -n 6 -r 10 -o after.qasm

You will compile before.qasm which is unfit to the device qubit coupling limitation on a "6 x 6" grid device model for 10 times. You will get output to after.qasm which fits the coupling limitation, and the corresponding initial mapping after.qasm.map.

Compile program based on other supported model

bin/codar before.qasm --tokyo -r 10 -o after.qasm

you will compile your program on IBM-Tokyo 20-qubit device model instead of a grid model.

Compile program with a custom initial mapping

bin/codar before.qasm -m 6 -n 6 -o after.qasm --init

Then you enter your initial mapping

please input initial mapping:
0 => 2; 1 => 8; 2 => 7; 3 => 0; 4 => 5;

which means you map the first qubit to q[2], the second to q[8], etc. In 6x6 gird device, q[2] is at (0, 2) and q[8] is at (1, 2). In general, q[i] is at (i / 6, i % 6).

If you hate to enter such inital mapping in the comand line every time, you can save it to a file and run

bin/codar before.qasm -m 6 -n 6 -o after.qasm -init /path/to/initial/mapping/file

and get a same effect.

Attention that if you also set -r for more than once, We will iteratively improve the initial mapping and sometimes give you a better one.

Have fun with CODAR!

CODAR is liscensed under the Apache 2.0.