Enola Compiler

Compilation for Dynamically Field-Programmable Qubit Arrays with Efficient and Provably Near-Optimal Scheduling. Open source under the BSD 3-Clause license.

Logistics

• We recommend to run the compiler in a Python3 virtual environment.
• The file enola/router/codegen.py is based on a file from OLSQ-DPQA, so you need to install networkx and matplotlib.
• To compile circuits in the QASM format, you need qiskit installed. We used the API in qiskit-1.1.0 specifically.
• The code for Misra&Gries coloring algorithm in enola/scheduler/ is from this Misra-Gries-coloring.

Repo structure

• run.py is an example of using the compiler on a circuit specified by a qubit interaction graph. Refer to python run.py -h for options.
• run_qasm.py is an example of using the compiler for circuits in the QASM format. Refer to python run_qasm.py -h for options.
• enola/ contains the source files implementing Enola.
• graphs.json contains all the random 3-regular graphs we used as qubit interaction graphs.
• animation.py contains the class Animator that generates animations from the full code produced by Enola based on the implementation from Animation.py in OLSQ-DPQA. Refer to python animation.py -h for options.
• simulator.py contains the class Simulator that calculate the circuit fidelity based on Enola instructions. Refer to python simulator.py -h for options.
• qasm_exp/ is the directory containing one generic circuit example from QASMBench.
• results/ is the default directory for the results.
  • results/code/ contains the code files generated from compilation results.
  • results/animations/ contains animation generated from full code files.
  • results/fidelity/ contains fidelity estimation based on code files.

How to use

• Run python run.py <S> <I> where <S> is the number of vertices in the random 3-regular graph, <I> is the id of the graph. To try other graphs, please edit run.py as needed.
  • Default setting: simulated annealing with dynamic placement, maximal independent set with sorting heuristic for routing strategy, no restriction for vectex number in solving MIS, and no animation code generation.
  • The most scalable setting: trivial layout (--trivial_layout), return to initial mapping after each Rydberg stage (--r2i), using maximal independent set for routing (--routing_strategy=maximalis), and limit the number of vertices to 1000 in solving MIS (--window).
    For example, python run.py 30 0 --trivial_layout --r2i --routing_strategy=maximalis --window
• (Optional) To generate animation, run python animation.py <F> where <F> is the full code file, e.g., results/code/rand3reg_30_0_code_full.json.