1.1K Star 1.7K Fork 2.9K

GVPMindSpore / mindquantum

加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
克隆/下载
INSTALL.md 8.93 KB
一键复制 编辑 原始数据 按行查看 历史
donghufeng 提交于 2023-07-27 14:32 . remove experimental

Installation

This file is designed to get you started with building and installing MindQuantum.

Quick-start

Build a binary Python wheel

In order to generate binary wheels of MindQuantum, we recommend that you use the build package from Pypa:

cd mindquantum
python3 -m build .

This will, however, produce binary wheels that may depend on some external libraries found somewhere on your system. Distributing them might therefore run into some issues if the other users do not have the same libraries installed or in different non-standard locations.

Delocating the wheels

In order to make sure that all the required external library dependencies are contained within the binary wheel file, you can instruct the binary wheel building process to delocate the produced wheels. In practice, this means to remove any dependency on external (shared) libraries on your system by integrating those directly into the binary wheel.

Using cibuildwheel

This is the preferred way of building binary wheels as it relies on Docker images (on Linux only) or the standard Python distribution on MacOS and Windows.

cd mindquantum
python3 -m cibuildwheel .

you might need to specify your platform if cibuildwheel is not able to automatically detect your platform or if you want to build Linux wheel on MacOS for example:

cd mindquantum
python3 -m cibuildwheel --platform linux .

See python3 -m cibuildwheel --help for more information about which platforms are available.

Linux

On Linux, you may run the script directly on any machine as it is using Docker images in order to build the delocated binary wheel. Same thing on MacOS or Windows if you would like to build Linux binary wheels.

MacOS

On MacOS, cibuildwheel will install the official Python distribution on your system before building the binary wheel. This makes running the script on your development machine not appropriate unless you understand what you are doing.

Windows

On Windows, cibuildwheel will install the official Python distribution using NuGet on your system before building the binary wheel. This makes running the script on your development machine not appropriate unless you understand what you are doing.

On your local machine

If you do not want to rely on the cibuildwheel machinery (e.g. on MacOS) you can also automatically call auditwheel or delocate after building the wheel by specifying the MQ_DELOCATE_WHEEL environment variable, like so:

cd mindquantum
MQ_DELOCATE_WHEEL=1 python3 -m build .

If you plan on distributing the wheel to other people that might not have the same system as yours, we highly recommend that you try to specify the MQ_DELOCATE_WHEEL_PLAT environment variable. By default, the setup scripts assumes 'linux_x86_64' on 64 bits machines but you may specify any platforms supported by auditwheel. In order to distribute your wheels to a larger audience, we would recomment setting MQ_DELOCATE_WHEEL_PLAT=manylinux2010_x86_64, although this might result in an error when delocating the wheel if the version of your compiler is too recent.

Build MindQuantum locally

You can setup MindQuantum for local development by using one of the local build scripts:

  • build_locally.bat (MS-DOS BATCH script)
  • build_locally.ps1 (PowerShell script)
  • build_locally.sh (Bash script)

Except a few minor differences [1]_, the functionalities of all three scripts are identical. All the scripts accept a flag to display the help message (-h, --help, -H, -Help`, /h, /Help`` for Bash, Powershell and MS-DOS BATCH). Please invoke the script of your choice in order to view the latest set of functionalities provided by it.

  1. Setup a Python virtual environment
  2. Update the virtual environment's packages and install some required dependencies
  3. Add a PTH-file to the Python virtual environment to make sure that MindQuantum will be detected
  4. Create a build directory and run CMake within it
  5. Compile MindQuantum in-place

The next time you run the script, unless you specify one of the cleaning options or force a CMake configuration step, the script will only re-compile MindQuantum.

For more information, have a look at the help message of the scripts which you can access using ./build_locally.sh -h or ./build_locally.sh --help. The output is shown below for reference.

CMake configuration

CMake options

Here is an exhaustive list of all CMake options available for customization

Option name Description Default value
BUILD_SHARED_LIBS Build shared libs OFF
BUILD_TESTING Enable building the test suite OFF
CLEAN_3RDPARTY_INSTALL_DIR Clean third-party installation directory OFF
CUDA_ALLOW_UNSUPPORTED_COMPILER Allow the use of an unsupported comipler version for CUDA OFF
CUDA_STATIC Use static versions of the Nvidia CUDA libraries OFF
DISABLE_FORTRAN_COMPILER Forcefully disable the Fortran compiler for some 3rd party libraries ON
ENABLE_CMAKE_DEBUG Enable verbose output to debug CMake issues OFF
ENABLE_CUDA Enable the use of CUDA code OFF
ENABLE_GITEE Use Gitee instead of GitHub for (some) third-party dependencies OFF
ENABLE_MD Use /MD, /MDd flags when compiling (MSVC only) OFF
ENABLE_MT Use /MT, /MTd flags when compiling (MSVC only) OFF
ENABLE_PROFILING Enable compilation with profiling flags OFF
ENABLE_RUNPATH Prefer RUNPATH over RPATH when linking ON
ENABLE_STACK_PROTECTION Enable stack protection during compilation ON
IN_PLACE_BUILD Build the C++ MindQuantum libraries in-place OFF
IS_PYTHON_BUILD Whether CMake is called from setup.py OFF
LINKER_DTAGS Enable --enable-new-dtags (else use --disable-new-dtags) when linking ON
LINKER_NOEXECSTACK Use -z,noexecstack during linking (if supported) ON
LINKER_RELRO Use -z,relro during linking (if supported) ON
LINKER_RPATH Use RUNPATH/RPATH related flags when compiling ON
LINKER_STRIP_ALL Use --strip-all during linking (if supported) ON
USE_OPENMP Use the OpenMP library for parallelisation ON
USE_PARALLEL_STL Use the parallel STL for parallelisation (using TBB or else) OFF
USE_VERBOSE_MAKEFILE Generate verbose Makefiles (if supported) ON

Here are some more precisions on some of the above options:

CLEAN_3RDPARTY_INSTALL_DIR

This will delete any pre-existing installations within the local installation directory (by default /path/to/build/.mqlibs) except the ones that are currently needed based on the hashes of the third-party libraries.

DISABLE_FORTRAN_COMPILER

This currently only has an effect when installing Eigen3.

CMake variables

In addition to the above CMake options, you may pass certain special CMake variables in order to customize your build. These are described below in more details.

MQ_FORCE_LOCAL_PKGS

This variable's value is case-insensitive. It may be either of:

  • a single string (all)
  • a comma-separated list of CMake package names for one or more of MindQuantum's third-party dependencies (e.g. gmp,eigen3)

Any or all packages listed will be compiled locally during the CMake configuration process.

MQ_XXX_FORCE_LOCAL

Setting this to a truthful value for one of MindQuantum's third-party dependencies will result in that/these packages to be compiled locally during the CMake configuration process. Note that the package name XXX must be all caps.

Python
1
https://gitee.com/mindspore/mindquantum.git
git@gitee.com:mindspore/mindquantum.git
mindspore
mindquantum
mindquantum
master

搜索帮助