Skip to content

Latest commit

 

History

History
50 lines (30 loc) · 2.3 KB

CONTRIBUTING.md

File metadata and controls

50 lines (30 loc) · 2.3 KB
Raw

Contribution Guide

By contributing to this benchmark, you are agreeing to release it under MIT License.

To contribute new benchmarks, please open a pull request and the maintainer to run it on test machine.

Shell Scripts

The shell script bin/benchmark provides an easy interface to setup benchmarks on arbitrary machine. If you would like to add a new benchmark please:

  • add your new benchmark installation instructions to the shell scripts so one can install it with just running bin/benchmark install.
  • make sure to add your benchmark command to benchmark_all_parallel and benchmark_all, if you are not benchmarking your code via pytest-benchmark please also add a new entry to benchmark function.

Structure of the benchmarks

**Each software/framework benchmark is stored in their corresponding folder with file name benchmarks.<extension name>, e.g cirq/benchmarks.py.

To make sure that we are doing the right measurement on execution time, the benchmarks are done with the following frameworks:

We always use the minimum time in the statistics for all benchmarks, due to Which estimator should I use?

Format

The benchmark data is generated in the following way:

  • .benchmarks/<system info folder>/ pytest-based benchmark results are saved into json file in this folder
  • . julia-based benchmark results are saved as CSV files in the root folder

Labels

Currently we use the following labels in benchmark group:

  • X, H, T, CNOT, Toffoli for single qubit gates
  • QCBM for parameterized quantum circuits
  • QCBM_batch for batched parameterized quantum circuits
  • QCBM_cuda for parameterized quantum circuit simulation on CUDA
  • QCBM_cuda_batch for batched parameterize quantum circuits on CUDA

Circuit

The circuit we benchmark is shown as below:

QCBM

Each benchmark script benchmarks such a circuit with 9 entanglers from 4 qubits to 25 qubits.