Skip to content

Latest commit

 

History

History
1190 lines (821 loc) · 48.8 KB

CONTRIBUTING.md

File metadata and controls

1190 lines (821 loc) · 48.8 KB
Raw

Contributing

First read the overall project contributing guidelines. These are all included in the Qiskit documentation:

https://qiskit.org/documentation/contributing_to_qiskit.html

Contributing to Aer

In addition to the general guidelines, there are specific details for contributing to Aer. These are documented below.

Pull request checklist

When submitting a pull request and you feel it is ready for review, please ensure that:

  1. The code follows the code style of the project and successfully passes the tests. For convenience, you can execute tox locally, which will run these checks and report any issues.

    If your code fails the local style checks, you can use tox -eblack and tox -eclang to automatically fix and update the code formatting in python and C++, respectively.

  2. The documentation has been updated accordingly. In particular, if a function or class has been modified during the PR, please update the docstring accordingly.

  3. If it makes sense for your change that you have added new tests that cover the changes.

  4. Ensure that if your change has an enduser-facing impact (new feature, deprecation, removal, etc.), you have added a reno release note for that change and that the PR is tagged for the changelog.

Changelog generation

The changelog is automatically generated as part of the release process automation. This works through a combination of the git log and the pull request. When a release is tagged and pushed to GitHub, the release automation bot looks at all commit messages from the git log for the release. It takes the PR numbers from the git log (assuming a squash merge) and checks if that PR had a Changelog: label on it. If there is a label, it will add the git commit message summary line from the git log for the release to the changelog.

If there are multiple Changelog: tags on a PR, the git commit message summary line from the git log will be used for each changelog category tagged.

The current categories for each label are as follows:

PR Label Changelog Category
Changelog: Deprecation Deprecated
Changelog: New Feature Added
Changelog: API Change Changed
Changelog: Removal Removed
Changelog: Bugfix Fixed

Release Notes

When making any end user-facing changes in a contribution, we have to make sure we document that when we release a new version of qiskit-aer. The expectation is that if your code contribution has user-facing changes that you will write the release documentation for these changes. This documentation must explain what was changed, why it was changed, and how users can either use or adapt to the change. The idea behind the release documentation is that when a naive user with limited internal knowledge of the project is upgrading from the previous release to the new one, they should be able to read the release notes, understand if they need to update their program which uses Qiskit, and how they would go about doing that. It ideally should explain why they need to make this change too, to provide the necessary context.

To make sure we don't forget a release note if the details of user-facing changes over a release cycle, we require that all user facing changes include documentation at the same time as the code. To accomplish this, we use the reno tool which enables a git-based workflow for writing and compiling release notes.

Adding a new release note

Making a new release note is quite straightforward. Ensure that you have reno installed with::

pip install -U reno

Once you have reno installed, you can make a new release note by running in your local repository checkout's root::

reno new short-description-string

where short-description-string is a brief string (with no spaces) that describes what's in the release note. This will become the prefix for the release note file. Once that is run, it will create a new yaml file in releasenotes/notes. Then open that yaml file in a text editor and write the release note. The basic structure of a release note is restructured text in yaml lists under category keys. You add individual items under each category and they will be grouped automatically by release when the release notes are compiled. A single file can have as many entries in it as needed, but to avoid potential conflicts you'll want to create a new file for each pull request that has user-facing changes. When you open the newly created file, it will be a full template of the different categories with a description of a category as a single entry in each category. You'll want to delete all the sections you aren't using and update the contents for those you are. For example, the end result should look something like::

features:
  - |
    Introduced a new feature ``foo``, that adds support for doing something to
    ``AerProvider`` objects. It can be used by using the ``foo`` function,
    for example::

      from qiskit_aer import foo
      from qiskit_aer import AerProvider
      foo(AerProvider())

  - |
    The ``qiskit_aer.AerProvider`` module has a new method ``foo()``. This is
    the equivalent of calling the ``qiskit_aer.foo()`` to do something to your
    ``AerProvider``. This is the equivalent of running ``qiskit_aer.foo()`` on
    your provider, but it has the convenience of running it natively on
    an object. For example::

      from qiskit_aer import AerProvider

      provider = AerProvider()
      provider.foo()

deprecations:
  - |
    The ``qiskit_aer.bar`` module has been deprecated and will be removed in a
    future release. Its sole function, ``foobar()`` has been superseded by the
    ``qiskit_aer.foo()`` function which provides similar functionality but with
    more accurate results and better performance. You should update your calls
    ``qiskit_aer.bar.foobar()`` calls to ``qiskit_aer.foo()``.

You can also look at other release notes for other examples.

You can use any restructured text feature in them (code sections, tables, enumerated lists, bulleted list, etc.) to express what is being changed as needed. In general, you want the release notes to include as much detail as needed so that users will understand what has changed, why it changed, and how they'll have to update their code.

After you've finished writing your release notes, you'll want to add the note file to your commit with git add and commit them to your PR branch to make sure they're included with the code in your PR.

Linking to issues

If you need to link to an issue or other GitHub artifact as part of the release note, this should be done using an inline link with the text being the issue number. For example you would write a release note with a link to issue 12345 as:

fixes:
  - |
    Fixes a race condition in the function ``foo()``. Refer to
    `#12345 <https://github.com/Qiskit/qiskit-aer/issues/12345>` for more
    details.

Generating the release notes

After release notes have been added, if you want to see the full output of the release notes, you'll get the output as an rst (ReStructuredText) file that can be compiled by sphinx. To generate the rst file, you use the reno report command. If you want to generate the full Aer release notes for all releases (since we started using reno during 0.9), you just run::

reno report

but you can also use the --version argument to view a single release (after it has been tagged::

reno report --version 0.5.0

At release time, reno report is used to generate the release notes for the release and the output will be submitted as a pull request to the documentation repository's release notes file

Building release notes locally

Building The release notes are part of the standard qiskit-aer documentation builds. To check what the rendered HTML output of the release notes will look like for the current state of the repo, you need to install pandoc, then you can run: tox -edocs which will build all the documentation into docs/_build/html and the release notes in particular will be located at docs/_build/html/release_notes.html

Style and lint

Aer uses 3 tools for verifying code formatting and lint checking. The first tool is black which is a Python code formatting tool that will automatically update the code formatting to a consistent style. The second tool is pylint which is a code linter which does a deeper analysis of the Python code to find both style issues and potential bugs and other common issues in Python. The third tool is clang-format which is a C++ code formatting tool that will automatically update codes with a consistent style.

You can check that your local modifications conform to the style rules by running tox -elint which will run black, pylint and clang-format to check the local code formatting and lint. If black returns a code formatting error you can run tox -eblack to automatically update the code formatting to conform to the style. However, if pylint returns any error you will have to fix these issues by manually updating your code.

Development Cycle

The development cycle for Aer is all handled in the open using the project boards in GitHub for project management. We use milestones in GitHub to track work for specific releases. The features or other changes that we want to include in a release will be tagged and discussed in GitHub. As we're preparing a new release, we'll document what has changed since the previous version in the release notes.

Branches

  • main:

The main branch is used for development of the next version of Aer. It will be updated frequently and should not be considered stable. The API can and will change on main as we introduce and refine new features.

  • stable/* branches: Branches under stable/* are used to maintain released versions of qiskit-aer. It contains the version of the code corresponding to the latest release for that minor version on pypi. For example, stable/0.4 contains the code for the 0.4.0 release on pypi. The API on these branches are stable and the only changes merged to it are bugfixes.

Release cycle

When it is time to release a new minor version of qiskit-aer, we will:

  1. Create a new tag with the version number and push it to github
  2. Change the main version to the next release version.

The release automation processes will be triggered by the new tag and will perform the following steps:

  1. Create a stable branch for the new minor version from the release tag on the main branch
  2. Build and upload binary wheels to pypi
  3. Create a GitHub release page with a generated changelog
  4. Generate a PR on the meta-repository to bump the Aer version and meta-package version.

The stable/* branches should only receive changes in the form of bug fixes.

Install from Source

Note: The following are prerequisites for all operating systems

We recommend using Python virtual environments to cleanly separate Qiskit from other applications and improve your experience.

  • The simplest way to use environments is by using Anaconda in a terminal window
    $ conda create -y -n QiskitDevEnv python=3
    $ conda activate QiskitDevEnv
  • Clone the Aer repo via git.
    $ git clone https://github.com/Qiskit/qiskit-aer

  • Next, install the platform-specific dependencies for your operating system Linux | macOS | Windows.

  • The common dependencies can then be installed via pip, using the requirements-dev.txt file, e.g.:

    $ cd qiskit-aer
    $ pip install -r requirements-dev.txt

This will also install Conan, a C/C++ package manager written in Python. This tool will handle most of the dependencies needed by the C++ source code. Internet connection may be needed for the first build or when dependencies are added/updated, in order to download the required packages if they are not in your Conan local repository.

Note: Conan use can be disabled with the flag or environment variable DISABLE_CONAN=ON. The Python package conan is still required as a build dependency, it just will not called or used.

This is useful for building from source offline, or to reuse the installed package dependencies.

If we are only building the standalone version and do not want to install all Python requirements you can just install Conan:

$ pip install conan

You're now ready to build from source! Follow the instructions for your platform: Linux | macOS | Windows

Linux

Aer is officially supported on Red Hat, CentOS, Fedora, and Ubuntu distributions, as long as you can install a GCC version that is C++14 compatible and a few dependencies we need.

Dependencies

To get most of the necessary compilers and libraries, install the development environment tools from your Linux distribution by running

CentOS/Red Hat

$ yum groupinstall "Development Tools"

Fedora

$ dnf install @development-tools

Ubuntu

$ sudo apt install build-essential

Although the BLAS and LAPACK library implementations included in the build-essential package are sufficient to build all of the Aer simulators, we recommend using OpenBLAS, which you can install by running

CentOS/Red Hat

$ yum install openblas-devel

Fedora

$ dnf install openblas-devel

Ubuntu

$ sudo apt install libopenblas-dev

And of course, git is required to build from repositories

CentOS/Red Hat

$ yum install git

Fedora

$ dnf install git

Ubuntu

$ apt-get install git

Build

There are two ways of building Aer simulators, depending on your goal:

  1. Build a Python extension that works with Terra.
  2. Build a standalone executable.

Python extension

As any other Python package, we can install from source code by just running:

qiskit-aer$ pip install .

This will build and install Aer with the default options which is probably suitable for most of the users. There's another Pythonic approach to build and install software: build the wheels distributable file.

qiskit-aer$ pip install build
qiskit-aer$ python -I -m build --wheel

This is also the way we will choose to change default Aer behavior by passing parameters to the build system.

Advanced options

As Aer is meant to be executed in many configurations and platforms, there is a complex underlying build system that offers a lot of options you can tune by setting some parameters.

We are using scikit-build as a substitute for setuptools. This is basically the glue between setuptools and CMake, so there are various options to pass variables to CMake, and the underlying build system (depending on your platform). The way to pass variables is:

qiskit-aer$ python ./setup.py bdist_wheel [skbuild_opts] \
[-- [cmake_opts] [-- build_tool_opts]]

where the elements within square brackets [] are optional, and skbuild_opts, cmake_opts, build_tool_opts are to be replaced by flags of your choice. A list of CMake options is available here. For example,

qiskit-aer$ python ./setup.py bdist_wheel --build-type=Debug -- -DCMAKE_CXX_COMPILER=g++-9 -- -j8

This is passing the --build-type option with Debug parameter to scikit-build, so we are telling it to perform a debug build. The -DCMAKE_CXX_COMPILER=g++-9 option is being passed to CMake so it forces the use of g++-9 compiler, and the -j8 flag is telling the underlying build system, which in this case is Makefile, to build in parallel using 8 processes.

After this command is executed successfully, we will have a wheel package into the dist/ directory, so next step is installing it:

qiskit-aer/dist$ pip install -U dist/qiskit_aer*.whl

As we are using scikit-build and we need some Python dependencies to be present before compiling the C++ code, we install those dependencies outside the regular setuptools mechanism. If you want to avoid automatic installation of these packages set the environment variable DISABLE_DEPENDENCY_INSTALL (ON or 1).

Standalone Executable

If you want to build a standalone executable, you have to use CMake directly. The preferred way CMake is meant to be used, is by setting up an "out of source" build. So in order to build your standalone executable, you have to follow these steps:

qiskit-aer$ mkdir out
qiskit-aer$ cd out
qiskit-aer/out$ cmake ..
qiskit-aer/out$ cmake --build . --config Release -- -j4

Once built, you will have your standalone executable into the Release/ or Debug/ directory (depending on the type of building chosen with the --config option):

qiskit-aer/out$ cd Release
qiskit-aer/out/Release/$ ls
qasm_simulator

Advanced options

Because the standalone version of Aer doesn't need Python at all, the build system is based on CMake, just like most of other C++ projects. So to pass all the different options we have on Aer to CMake, we use its native mechanism:

qiskit-aer/out$ cmake -DCMAKE_CXX_COMPILER=g++-9 -DAER_BLAS_LIB_PATH=/path/to/my/blas ..

macOS

Dependencies

We recommend installing OpenBLAS, which is our default choice:

$ brew install openblas

The CMake build system will search for other BLAS implementation alternatives if OpenBLAS is not installed in the system.

You further need to have Xcode Command Line Tools installed on macOS:

$ xcode-select --install

Build

There are two ways of building Aer simulators, depending on your goal:

  1. Build a Python extension that works with Terra;
  2. Build a standalone executable.

Python extension

As any other Python package, we can install from source code by just running:

qiskit-aer$ pip install .

This will build and install Aer with the default options which is probably suitable for most of the users. There's another Pythonic approach to building and installing software: build the wheels distributable file.

qiskit-aer$ python ./setup.py bdist_wheel

This is also the way we will choose to change default Aer behavior by passing parameters to the build system.

Advanced options

As Aer is meant to be executed in many configurations and platforms, there is a complex underlying build system that offers a lot of options you can tune by setting some parameters.

We are using scikit-build as a substitute for setuptools. This is basically the glue between setuptools and CMake, so there are various options to pass variables to CMake, and the underlying build system (depending on your platform). The way to pass variables is:

qiskit-aer$ python ./setup.py bdist_wheel [skbuild_opts] [-- [cmake_opts] [-- build_tool_opts]]

where the elements within square brackets [] are optional, and skbuild_opts, cmake_opts, build_tool_opts are to be replaced by flags of your choice. A list of CMake options is available here. For example,

qiskit-aer$ python ./setup.py bdist_wheel --build-type=Debug -- -DCMAKE_CXX_COMPILER=g++-9 -- -j8

This is passing the --build-type option with Debug parameter to scikit-build, so we are telling it to perform a debug build. The -DCMAKE_CXX_COMPILER=g++-9 option is being passed to CMake so it forces the use of g++-9 compiler, and the -j8 flag is telling the underlying build system, which in this case is Makefile, to build in parallel using 8 processes.

After this command is executed successfully, we will have a wheel package into the dist/ directory, so next step is installing it:

qiskit-aer/dist$ pip install -U dist/qiskit_aer*.whl

As we are using scikit-build and we need some Python dependencies to be present before compiling the C++ code, we install those dependencies outside the regular setuptools mechanism. If you want to avoid automatic installation of these packages set the environment variable DISABLE_DEPENDENCY_INSTALL (ON or 1).

Standalone Executable

If you want to build a standalone executable, you have to use CMake directly. The preferred way CMake is meant to be used, is by setting up an "out of source" build. So in order to build your standalone executable, you have to follow these steps:

qiskit-aer$ mkdir out
qiskit-aer$ cd out
qiskit-aer/out$ cmake ..
qiskit-aer/out$ cmake --build . --config Release -- -j4

Once built, you will have your standalone executable into the Release/ or Debug/ directory (depending on the type of building chosen with the --config option):

qiskit-aer/out$ cd Release
qiskit-aer/out/Release/$ ls
qasm_simulator

Advanced options

Because the standalone version of Aer doesn't need Python at all, the build system is based on CMake, just like most of other C++ projects. So to pass all the different options we have on Aer to CMake, we use its native mechanism:

qiskit-aer/out$ cmake -DCMAKE_CXX_COMPILER=g++-9 -DAER_BLAS_LIB_PATH=/path/to/my/blas ..

Windows

Dependencies

On Windows, you must have Anaconda3 installed. We also recommend installing Visual Studio 2017 Community Edition or Visual Studio 2019 Community Edition.

Anaconda 3 can be installed from their web: https://www.anaconda.com/distribution/#download-section

Visual Studio 2017/2019 Community Edition can be installed from: https://visualstudio.microsoft.com/vs/community/

Once you have Anaconda3 and Visual Studio Community Edition installed, you have to open a new cmd terminal and create an Anaconda virtual environment or activate it if you already have created one:

> conda create -y -n QiskitDevEnv python=3
> conda activate QiskitDevEnv
(QiskitDevEnv) >_

We only support Visual Studio compilers on Windows, so if you have others installed in your machine (MinGW, TurboC) you have to make sure that the path to the Visual Studio tools has precedence over others so that the build system can get the correct one. There's a (recommended) way to force the build system to use the one you want by using CMake -G parameter. We will talk about this and other parameters later.

Build

Python extension

As any other Python package, we can install from source code by just running:

(QiskitDevEnv) qiskit-aer > pip install .

This will build and install Aer with the default options which is probably suitable for most of the users. There's another Pythonic approach to building and installing software: build the wheels distributable file.

(QiskitDevEnv) qiskit-aer > python ./setup.py bdist_wheel

This is also the way we will choose to change default Aer behavior by passing parameters to the build system.

Advanced options

As Aer is meant to be executed in many configurations and platforms, there is a complex underlying build system that offers a lot of options you can tune by setting some parameters.

We are using scikit-build as a substitute for setuptools. This is basically the glue between setuptools and CMake, so there are various options to pass variables to CMake, and the underlying build system (depending on your platform). The way to pass variables is:

qiskit-aer > python ./setup.py bdist_wheel [skbuild_opts] [-- [cmake_opts] [-- build_tool_opts]]

where the elements within square brackets [] are optional, and skbuild_opts, cmake_opts, build_tool_opts are to be replaced by flags of your choice. A list of CMake options is available here. For example,

(QiskitDevEnv) qiskit-aer > python ./setup.py bdist_wheel --build-type=Debug -- -G "Visual Studio 15 2017"

This is passing the --build-type option with Debug parameter to scikit-build, so we are telling it to perform a debug build. The -G "Visual Studio 15 2017" option is being passed to CMake so it forces the use of Visual Studio 2017 C++ compiler to drive the build.

After this command is executed successfully, we will have a wheel package into the dist/ directory, so next step is installing it:

(QiskitDevEnv) qiskit-aer\dist$ pip install -U dist\qiskit_aer*.whl

As we are using scikit-build and we need some Python dependencies to be present before compiling the C++ code, we install those dependencies outside the regular setuptools mechanism. If you want to avoid automatic installation of these packages set the environment variable DISABLE_DEPENDENCY_INSTALL (ON or 1).

Standalone Executable

If you want to build a standalone executable, you have to use CMake directly. The preferred way CMake is meant to be used, is by setting up an "out of source" build. So in order to build our standalone executable, you have to follow these steps:

(QiskitDevEnv) qiskit-aer> mkdir out
(QiskitDevEnv) qiskit-aer> cd out
(QiskitDevEnv) qiskit-aer\out> cmake ..
(QiskitDevEnv) qiskit-aer\out> cmake --build . --config Release -- -j4

Once built, you will have your standalone executable into the Release/ or Debug/ directory (depending on the type of building chosen with the --config option):

(QiskitDevEnv) qiskit-aer\out> cd Release
(QiskitDevEnv) qiskit-aer\out\Release> dir
qasm_simulator

Advanced options

Because the standalone version of Aer doesn't need Python at all, the build system is based on CMake, just like most of other C++ projects. So to pass all the different options we have on Aer to CMake, we use its native mechanism:

(QiskitDevEnv) qiskit-aer\out> cmake -G "Visual Studio 15 2017" -DAER_BLAS_LIB_PATH=c:\path\to\my\blas ..

Building with GPU support

Aer can exploit GPU's horsepower to accelerate some simulations, specially the larger ones. GPU access is supported either via CUDA® (NVIDIA® chipset) or ROCm® (AMD® GPUs).

Building with CUDA® support

To build with CUDA® support, you need to have CUDA® >= 11.2 preinstalled. See install instructions here. Please note that we only support CUDA® GPU acceleration on Linux platforms at the moment.

Once CUDA® is properly installed, you only need to set a flag so the build system knows what to do:

AER_THRUST_BACKEND=CUDA

For example,

qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA --

If you want to specify the CUDA® architecture instead of letting the build system auto detect it, you can use the AER_CUDA_ARCH flag (can also be set as an ENV variable with the same name, although the flag takes precedence). For example:

qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA -DAER_CUDA_ARCH="7.0" --

or

qiskit-aer$ export AER_CUDA_ARCH="7.0"
qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA --

This will reduce the amount of compilation time when, for example, the architecture auto detection fails and the build system compiles all common architectures.

Few notes on CUDA® GPU builds:

  1. Building takes considerable more time than non-GPU build, so be patient :)
  2. CUDA® >= 11.2 imposes the restriction of building with g++ version not newer than 8
  3. We don't need NVIDIA® drivers for building, but we need them for running simulations
  4. Only Linux platforms are supported

Aer now supports cuQuantum optimized Quantum computing APIs from NVIDIA®. cuStateVec APIs can be exploited to accelerate statevector, density_matrix and unitary methods. cuTensorNet APIs can be exploited to tensor_network merthod. This implementation requires CUDA® toolkit version 11.2 or higher and Volta or Ampare architecture GPUs.

Before building Aer with cuQuantum support, install required components via pip install as following.

qiskit-aer$ pip install nvidia-cuda-runtime-cu11 nvidia-cublas-cu11 nvidia-cusolver-cu11 nvidia-cusparse-cu11 cuquantum-cu11

This example is for CUDA 11. Please replace cu11 to cu12 if your system has CUDA 12.

Then to build with cuQuantum support, set the value AER_PYTHON_CUDA_ROOT=<root of Python env> as following example.

qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA -DAER_PYTHON_CUDA_ROOT=qiskit-aer-venv --

If you want to link cuQuantum library statically, cuQuantum SDK and cuTENSOR should be installed in your system from NVIDIA®. Then set CUQUANTUM_ROOT CUTENSOR_ROOT and CUQUANTUM_STATIC to setup.py.

For example,

qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA -DCUQUANTUM_ROOT=path_to_cuQuantum -DCUTENSOR_ROOT=path_to_cuTENSOR -DAER_ENABLE_CUQUANTUM=true -DCUQUANTUM_STATIC=true --

To run with cuStateVec, set device='GPU' to AerSimulator option and set cuStateVec_enable=True to option in execute method.

sim = AerSimulator(method='statevector', device='GPU')
results = execute(circuit,sim,cuStateVec_enable=True).result()

Also you can accelrate density matrix and unitary matrix simulations as well.

sim = AerSimulator(method='density_matrix', device='GPU')
results = execute(circuit,sim,cuStateVec_enable=True).result()

Building with ROCm® support

ROCm® support has been added matching the CUDA® implementation based on the thrust library. This enables Aer to run on AMD® GPUs, including the AMD® Instinct GPU line based on the CDNA architecture. ROCm® only support linux platforms.

To build the standalone version, the following should be sufficient:

cmake <qiskit-aer source folder> -G Ninja \
   -DCMAKE_INSTALL_PREFIX=<qiskit-aer target instalation folder> \
   -DSKBUILD=FALSE \
   -DAER_THRUST_BACKEND=ROCM \
   -DAER_MPI=<set to ON or OFF depending on whether to activate MPI support> \
   -DAER_ROCM_ARCH=<target AMD GPU list, white-space separated, e.g. 'gfx90a gfx908'> \
   -DCMAKE_BUILD_TYPE=Release \
   -DBUILD_TESTS=True
ninja install

Alternatively, and possibly preferred for most use cases, you can create a Python wheel file that you can install as part of your Python environemnt:

cd <qiskit-aer source folder>

QISKIT_AER_PACKAGE_NAME='qiskit-aer-gpu-rocm' \
   python3 setup.py bdist_wheel -- \
      -DAER_THRUST_BACKEND=ROCM \
      -DAER_MPI=<set to ON or OFF depending on whether to activate MPI support> \
      -DAER_ROCM_ARCH=<target AMD GPU list, white-space separated, e.g. 'gfx90a gfx908'>

pip install --force-reinstall dist/qiskit_aer_gpu_rocm-*.whl

In both cases, the host system needs to have a functional ROCm® instalation and the environment variable ROCM_PATH set pointing to the ROCm® instalation folder if that is not the default /opt/rocm. Depending on how your Python environment is set, you might need to install Aer's required development modules:

cd <qiskit-aer source folder>
pip install -r requirements-dev.txt

To leverage the ROCm® implementations no code changes are needed on top of one already does for CUDA®. Running with cuStateVec, for instance, requires set device='GPU' to AerSimulator option and set cuStateVec_enable=True option, similarly to what is done for CUDA®:

sim = AerSimulator(method='statevector', device='GPU')
results = execute(circuit,sim,cuStateVec_enable=True).result()

Building with MPI support

Aer can parallelize its simulation on the cluster systems by using MPI. This can extend available memory space to simulate quantum circuits with larger number of qubits and also can accelerate the simulation by parallel computing. To use MPI support, any MPI library (i.e. OpenMPI) should be installed and configured on the system.

Aer supports MPI both with and without GPU support. Currently following simulation methods are supported to be parallelized by MPI.

  • statevector
  • density_matrix
  • unitary

To enable MPI support, the following flag is needed for build system based on CMake.

AER_MPI=True

For example,

qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_MPI=True

By default GPU direct RDMA is enable to exchange data between GPUs installed on the different nodes of a cluster. If the system does not support GPU direct RDMA the following flag disables this.

AER_DISABLE_GDR=True

For example,

qiskit-aer$ python ./setup.py bdist_wheel -- -DAER_MPI=True -DAER_DISABLE_GDR=True --

Running with multiple-GPUs and/or multiple nodes

Aer parallelizes simulations by distributing quantum states into distributed memory space. To decrease data transfer between spaces the distributed states are managed as chunks that is a sub-state for smaller qubits than the input circuits.

For example, 30-qubits circuit is distributed into 2^10 chunks with 20-qubits.

To decrease data exchange between chunks and also to simplify the implementation, we are applying cache blocking technique. This technique allows applying quantum gates to each chunk independently without data exchange, and serial simulation codes can be reused without special implementation. Before the actual simulation, we apply transpilation to remap the input circuits to the equivalent circuits that has all the quantum gates on the lower qubits than the chunk's number of qubits. And the (noiseless) swap gates are inserted to exchange data.

Please refer to this paper (https://arxiv.org/abs/2102.02957) for more detailed algorithm and implementation of parallel simulation.

So to simulate by using multiple GPUs or multiple nodes on the cluster, following configurations should be set to backend options. (If there is not enough memory to simulate the input circuit, Aer automatically set following options, but it is recommended to explicitly set them)

  • blocking_enable

should be set to True for distributed parallelization. (Default = False)

  • blocking_qubits

this flag sets the qubit number for chunk, should be smaller than the smallest memory space on the system (i.e. GPU). Set this parameter to satisfy sizeof(complex)*2^(blocking_qubits+4) < size of the smallest memory space in byte.

Here is an example how we parallelize simulation with multiple GPUs.

sim = AerSimulator(method='statevector', device='GPU')
circ = transpile(QuantumVolume(qubit, 10, seed = 0))
circ.measure_all()
result = execute(circ, sim, shots=100, blocking_enable=True, blocking_qubits=23).result()

To run Aer with Python script with MPI parallelization, MPI executer such as mpirun should be used to submit a job on the cluster. Following example shows how to run Python script using 4 processes by using mpirun.

mpirun -np 4 python example.py

MPI_Init function is called inside Aer, so you do not have to manage MPI processes in Python script. Following metadatas are useful to find on which process is this script running.

  • num_mpi_processes : shows number of processes using for this simulation
  • mpi_rank : shows zero based rank (process ID)

Here is an example how to get my rank.

sim = AerSimulator(method='statevector', device='GPU')
result = execute(circuit, sim, blocking_enable=True, blocking_qubits=23).result()
dict = result.to_dict()
meta = dict['metadata']
myrank = meta['mpi_rank']

Multiple shots are also distributed to multiple nodes when setting device=GPU and batched_shots_gpu=True. The results are distributed to each processes.

Note : In the script, make sure that the same random seed should be used for all processes so that the consistent circuits and parameters are passed to Aer. To do so add following option to the script.

from qiskit_algorithms.utils import algorithm_globals
algorithm_globals.random_seed = consistent_seed_to_all_processes

Building a statically linked wheel

If you encounter an error similar to the following, you may are likely in the need of compiling a statically linked wheel.

    ImportError: libopenblas.so.0: cannot open shared object file: No such file or directory

However, depending on your setup this can proof difficult at times. Thus, here we present instructions which are known to work under Linux.

In general, the workflow is:

  1. Compile a wheel
    qiskit-aer$ python ./setup.py bdist_wheel
  1. Repair it with auditwheel
    qiskit-aer$ auditwheel repair dist/qiskit_aer*.whl

auditwheel vendors the shared libraries into the binary to make it fully self-contained.

The command above will attempt to repair the wheel for a manylinux* platform and will store it under wheelhouse/ from where you can install it.

It may happen that you encounter the following error:

    auditwheel: error: cannot repair "qiskit_aer-0.8.0-cp36-cp36m-linux_x86_64.whl" to "manylinux1_x86_64" ABI because of the presence of too-recent versioned symbols. You'll need to compile the wheel on an older toolchain.

This means that your toolchain uses later versions of system libraries than are allowed by the manylinux* platform specification (see also 1, 2 and 3). If you do not need your wheel to support the manylinux* platform you can resolve this issue by limiting the compatibility of your wheel to your specific platform. You can find out which platform this is through

    qiskit-aer$ auditwheel show dist/qiskit_aer*.whl

This will list the platform tag (e.g. linux_x86_64). You can then repair the wheel for this specific platform using:

    qiskit-aer$ auditwheel repair --plat linux_x86_64 dist/qiskit_aer*.whl

You can now go ahead and install the wheel stored in wheelhouse/.

Should you encounter a runtime error like

    Inconsistency detected by ld.so: dl-version.c: 205: _dl_check_map_versions: Assertion `needed != NULL' failed!

this means that your patchelf version (which is used by auditwheel under the hood) is too old (pypa/auditwheel#103) Version 0.9 of patchelf is the earliest to include the patch NixOS/patchelf#85 which resolves this issue. In the unlikely event that the patchelf package provided by your operating system only provides an older version, fear not, because it is really easy to compile patchelf from source.

Hopefully, this information was helpful. In case you need more detailed information on some of the errors which may occur be sure to read through Qiskit#1033.

Useful CMake flags

There are some useful flags that can be set during CMake command invocation and will help you change some default behavior. To make use of them, you just need to pass them right after -D CMake argument. Example:

qiskit-aer/out$ cmake -DUSEFUL_FLAG=Value ..

In the case of building the Qiskit Python extension, you have to pass these flags after writing -- at the end of the python command line, eg:

qiskit-aer$ python ./setup.py bdist_wheel -- -DUSEFUL_FLAG=Value

These are the flags:

  • USER_LIB_PATH

    This flag tells CMake to look for libraries that are needed by some of the native components to be built, but they are not in a common place where CMake could find it automatically.

    Values: An absolute path. Default: No value. Example: python ./setup.py bdist_wheel -- -DUSER_LIB_PATH=C:\path\to\openblas\libopenblas.so

  • AER_BLAS_LIB_PATH

    Tells CMake the directory to look for the BLAS library instead of the usual paths. If no BLAS library is found under that directory, CMake will raise an error and terminate. It can also be set as an ENV variable with the same name, although the flag takes precedence.

    Values: An absolute path. Default: No value. Example: python ./setup.py bdist_wheel -- -DAER_BLAS_LIB_PATH=/path/to/look/for/blas/

  • BUILD_TESTS

    It will tell the build system to build C++ tests along with the simulator.

    Values: True|False Default: False Example: python ./setup.py bdist_wheel -- -DBUILD_TESTS=True

  • CMAKE_CXX_COMPILER

    This is an internal CMake flag. It forces CMake to use the provided toolchain to build everything. If it's not set, CMake system will use one of the toolchains installed in system.

    Values: g++|clang++|g++-8 Default: Depends on the running platform and the toolchains installed Example: python ./setup.py bdist_wheel -- -DCMAKE_CXX_COMPILER=g++

  • AER_THRUST_BACKEND

    We use Thrust library for GPU support through CUDA. If you want to build a version of Aer with GPU acceleration, you need to install CUDA and set this variable to the value: "CUDA". There are other values that will use different CPU methods depending on the kind of backend you want to use:

    • "OMP": For OpenMP support
    • "TBB": For Intel Threading Building Blocks

    Values: CUDA|OMP|TBB Default: No value Example: python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA

  • AER_CUDA_ARCH

    This flag allows you to specify the CUDA architecture instead of letting the build system auto detect it. It can also be set as an ENV variable with the same name, although the flag takes precedence.

    Values: Auto | Common | All | List of valid CUDA architecture(s). Default: Auto Example: python ./setup.py bdist_wheel -- -DAER_THRUST_BACKEND=CUDA -DAER_CUDA_ARCH="5.2; 5.3"

  • DISABLE_CONAN

    This flag allows disabling the Conan package manager. This will force CMake to look for the libraries in use on your system path, relying on FindPackage CMake mechanism and the appropriate configuration of libraries in order to use it. If a specific version is not found, the build system will look for any version available, although this may produce build errors or incorrect behaviour.

    WARNING: This is not the official procedure to build AER. Thus, the user is responsible of providing all needed libraries and corresponding files to make them findable to CMake.

    This is also available as the environment variable DISABLE_CONAN, which overrides the CMake flag of the same name.

    Values: ON | OFF Default: OFF Example: python ./setup.py bdist_wheel -- -DDISABLE_CONAN=ON

  • AER_MPI

    This flag enables/disables parallelization using MPI to simulate circuits with large number of qubits on the cluter systems. This option requires any MPI library and runtime installed on your system. MPI parallelization can be used both with/without GPU support. For GPU support GPU direct RDMA is enabled by default, see option AER_DISABLE_GDR below.

    Values: True|False Default: False Example: python ./setup.py bdist_wheel -- -DAER_MPI=True

  • AER_DISABLE_GDR

    This flag disables/enables GPU direct RDMA to exchange data between GPUs on different nodes. If your system does not support GPU direct RDMA, please set True to this option. You do not need this option if you do not use GPU support. You may also have to configure MPI to use GPU direct RDMA if you enable (AER_DISABLE_GDR=False) this option.

    Note: GPU direct between GPUs on the same node (peer-to-peer copy) is automatically enabled if supported GPUs are available.

    Values: True|False Default: False Example: python ./setup.py bdist_wheel -- -DAER_MPI=True -DAER_DISABLE_GDR=True

Tests

Code contributions are expected to include tests that provide coverage for the changes being made.

We have two types of tests in the codebase: Qiskit integration tests and Standalone integration tests.

For Qiskit integration tests, you first need to build and install the Qiskit Python extension, and then run unittest Python framework.

qiskit-aer$ pip install .
qiskit-aer$ stestr run

Manual for stestr can be found here.

The integration tests for Qiskit Python extension are included in: test/terra.

C++ Tests

Our C++ unit tests use the Catch2 framework, an include-only C++ unit-testing framework. Catch2 framework documentation can be found here. Then, in any case, build Aer with the extra cmake argument BUILD_TESTS set to true:

python ./setup.py bdist_wheel --build-type=Debug -- -DBUILD_TESTS=True -- -j4 2>&1 |tee build.log

The test executable will be placed into the source test directory and can be run by:

qiskit-aer$ ./test/unitc_tests [Catch2-options]

Platform support

Bear in mind that every new feature/change needs to be compatible with all our supported platforms: Win64, MacOS (API Level >= 19) and Linux-x86_64. The Continuous Integration (CI) systems will run builds and pass all the corresponding tests to verify this compatibility.

Debug

You have to build in debug mode if you want to start a debugging session with tools like gdb or lldb. To create a Debug build for all platforms, you just need to pass a parameter while invoking the build to create the wheel file:

qiskit-aer$> python ./setup.py bdist_wheel --build-type=Debug

If you want to debug the standalone executable, the parameter changes to:

qiskit-aer/out$> cmake -DCMAKE_BUILD_TYPE=Debug

There are three different build configurations: Release, Debug, and Release with Debug Symbols, whose parameters are: Release, Debug, RelWithDebInfo respectively.

We recommend building in verbose mode and dump all the output to a file so it's easier to inspect possible build issues:

On Linux and Mac:

qiskit-aer$ VERBOSE=1 python ./setup.py bdist_wheel --build-type=Debug 2>&1|tee build.log

On Windows:

qisikt-aer> set VERBOSE=1
qiskit-aer> python ./setup.py bdist_wheel --build-type=Debug 1> build.log 2>&1

We encourage you to always send the whole build.log file when reporting a build issue, otherwise we will ask for it :)

Stepping through the code

Standalone version doesn't require anything special, just use your debugger like always:

qiskit-aer/out/Debug$ gdb qasm_simulator

Stepping through the code of a Python extension is another story, trickier, but possible. This is because Python interpreters usually load Python extensions dynamically, so we need to start debugging the Python interpreter and set our breakpoints ahead of time, before any of our Python extension symbols are loaded into the process.

Once built and installed, we have to run the debugger with the Python interpreter:

$ lldb python

That will get us into the debugger (lldb in our case) interactive shell:

(lldb) target create "python"
Current executable set to 'python' (x86_64).
(lldb)

Then we have to set our breakpoints:

(lldb) b AER::controller_execute
Breakpoint 1: no locations (pending).
WARNING:  Unable to resolve breakpoint to any actual locations.

Here the message is clear, it can't find the function: AER::controller_execute because our Python extension hasn't been loaded yet by the Python interpreter, so it's "on-hold" hoping to find the function later in the execution. Now we can run the Python interpreter and pass the arguments (the python file to execute):

(lldb) r test_qiskit_program.py
Process 24896 launched: '/opt/anaconda3/envs/aer37/bin/python' (x86_64)
3 locations added to breakpoint 1
Executing on QasmSimulator for nq=16
Process 24896 stopped
* thread #12, stop reason = breakpoint 1.1
     frame #0: 0x000000012f834c10 controller_wrappers.cpython-37m-darwin.so`AER::Result AER::controller_execute<AER::Simulator::QasmController>(qobj_js=0x00007000032716b0) at controller_execute.hpp:48:16
     45
     46  	template <class controller_t>
     47  	Result controller_execute(const json_t &qobj_js) {
 ->  48  	  controller_t controller;
     49
     50  	  // Fix for MacOS and OpenMP library double initialization crash.
     51  	  // Issue: https://github.com/Qiskit/qiskit-aer/issues/1
Target 0: (python) stopped.

After this, you can step through the code and continue with your debug session as always.

Dealing with the git blame ignore list

In the Qiskit/qiskit-aer repository we maintain a list of commits for git blame to ignore. This is mostly commits that are code style changes that don't change the functionality but just change the code formatting (for example, when we migrated to use black for code formatting). This file, .git-blame-ignore-revs just contains a list of commit SHA1s you can tell git to ignore when using the git blame command. This can be done one time with something like

git blame --ignore-revs-file .git-blame-ignore-revs qiskit/version.py

from the root of the repository. If you'd like to enable this by default you can update your local repository's configuration with:

git config blame.ignoreRevsFile .git-blame-ignore-revs

which will update your local repositories configuration to use the ignore list by default.