Skip to content
This repository has been archived by the owner on Feb 5, 2024. It is now read-only.

Latest commit

 

History

History
214 lines (133 loc) · 8.63 KB

README.rst

File metadata and controls

214 lines (133 loc) · 8.63 KB

PennyLane-Lightning-GPU Plugin

❗ Warning

The PennyLaneAI organization has archived this repository, which is a snapshot of PennyLane-Lightning-GPU v0.32. This backend was integrated into PennyLane-Lightning where its development continues. We direct users and developers to PennyLane-Lightning to report issues, make pull requests, etc.

About

Read the Docs PyPI PyPI - Python Version

The PennyLane-Lightning-GPU plugin extends the Pennylane-Lightning state-vector simulator written in C++, and offloads to the NVIDIA cuQuantum SDK for GPU accelerated circuit simulation.

PennyLane is a cross-platform Python library for quantum machine learning, automatic differentiation, and optimization of hybrid quantum-classical computations.

Features

  • Combine the NVIDIA cuQuantum SDK high-performance GPU simulator library with PennyLane's automatic differentiation and optimization.
  • Direct support for GPU-enabled quantum gradients with the adjoint differentiation method.

Installation

PennyLane-Lightning-GPU requires Python version 3.9 and above. It can be installed using pip:

pip install pennylane-lightning[gpu]

Use of PennyLane-Lightning-GPU also requires explicit installation of the NVIDIA cuQuantum SDK. The SDK library directory may be provided on the LD_LIBRARY_PATH environment variable, or the SDK Python package may be installed within the Python environment site-packages directory using pip or conda. Please see the cuQuantum SDK install guide for more information.

To build a wheel from the package sources using the direct SDK path:

cmake -BBuild -DENABLE_CLANG_TIDY=on -DCUQUANTUM_SDK=<path to sdk>
cmake --build ./Build --verbose
python -m pip install wheel
python setup.py build_ext --cuquantum=<path to sdk>
python setup.py bdist_wheel

To build using the PyPI/Conda installed cuQuantum package:

python -m pip install wheel cuquantum
python setup.py build_ext
python setup.py bdist_wheel

The built wheel can now be installed as:

python -m pip install ./dist/PennyLane_Lightning_GPU-*.whl

To simplify the build, we recommend using the following containerized build process, which creates manylinux2014 compatible wheels.

Build locally with Docker

To build using Docker, run the following from the project root directory:

docker build . -f ./docker/Dockerfile -t "lightning-gpu-wheels"

This will build a Python wheel for Python 3.9 up to 3.11 inclusive, and be manylinux2014 (glibc 2.17) compatible. To acquire the built wheels, use:

docker run -v `pwd`:/io -it lightning-gpu-wheels cp -r ./wheelhouse /io

which mounts the current working directory, and copies the wheelhouse directory from the image to the local directory. For licensing information, please view docker/README.md.

Build PennyLane-Lightning-GPU with multi-node/multi-gpu support

Use of PennyLane-Lightning-GPU with multi-node/multi-gpu support also requires explicit installation of the NVIDIA cuQuantum SDK (current supported cuQuantum version: cuquantum-cu11), mpi4py and CUDA-aware MPI (Message Passing Interface). CUDA-aware MPI allows data exchange between GPU memory spaces of different nodes without the need for CPU-mediated transfers. Both MPICH and OpenMPI libraries are supported, provided they are compiled with CUDA support. Path to the libmpi.so should be added to the LD_LIBRARY_PATH environment variable. It's recommended to install NVIDIA cuQuantum SDK and mpi4py Python package within the Python environment site-packages directory using pip or conda. Please see the cuQuantum SDK , mpi4py, MPICH, or OpenMPI install guide for more information.

To build a wheel with multi-node/multi-gpu support from the package sources using the direct SDK path:

cmake -BBuild -DENABLE_CLANG_TIDY=on -DPLLGPU_ENABLE_MPI=on -DCUQUANTUM_SDK=<path to sdk>
cmake --build ./Build --verbose
python -m pip install wheel
python setup.py build_ext --define="PLLGPU_ENABLE_MPI=ON" --cuquantum=<path to sdk>
python setup.py bdist_wheel

The built wheel can now be installed as:

python -m pip install ./dist/PennyLane_Lightning_GPU-*.whl

Testing

Test PennyLane-Lightning-GPU

To test that the plugin is working correctly you can test the Python code within the cloned repository:

make test-python

while the C++ code can be tested with

make test-cpp

Please refer to the GPU plugin documentation as well as to the CPU documentation and PennyLane documentation for further references.

Test PennyLane-Lightning-GPU with multi-node/multi-gpu support

To test that the plugin is working correctly you can test the Python code within the cloned repository:

mpirun -np 2 python -m pytest mpitests --tb=short

while the C++ code can be tested with

rm -rf ./BuildTests
cmake . -BBuildTests -DBUILD_TESTS=1 -DPLLGPU_BUILD_TESTS=1 -DPLLGPU_ENABLE_MPI=On -DCUQUANTUM_SDK=<path to sdk>
cmake --build ./BuildTests --verbose
mpirun -np 2 ./BuildTests/pennylane_lightning_gpu/src/tests/mpi_runner

Contributing

We welcome contributions - simply fork the repository of this plugin, and then make a pull request containing your contribution. All contributors to this plugin will be listed as authors on the releases.

We also encourage bug reports, suggestions for new features and enhancements, and even links to cool projects or applications built on PennyLane.

Support

If you are having issues, please let us know by posting the issue on our Github issue tracker, or by asking a question in the forum.

License

The PennyLane-Lightning-GPU plugin is free and open source, released under the Apache License, Version 2.0. The PennyLane-Lightning-GPU plugin makes use of the NVIDIA cuQuantum SDK headers to enable the device bindings to PennyLane, which are held to their own respective license.