Installation

Download

The latest version of the PYATB package can be obtained from the GitHub repository:

git clone https://github.com/pyatb/pyatb.git

Prerequisites

Currently, PYATB is only supported on Linux systems and has not been tested on Windows or Mac systems. To use PYATB, the following prerequisites must be installed:

• Python 3.7 or newer

• C++ compiler

• Intel MKL

• pybind11

• Eigen3

• NumPy

• SciPy

• mpi4py

• Matplotlib

Install

1. Before installing PYATB, you need to install the pybind11 module and mpi4py module:

   pip install pybind11
   

   and

   pip install mpi4py
   

2. You can set the addresses of the C++ compiler (non-MPI version), MKL library, and Eigen library through the siteconfig.py file, and modify the relevant variables to point to the correct paths for your system.

   • To set the C++ compiler, you can modify the variable “compiler” in the siteconfig.py. For example, you can set compiler = 'icpc'.

   • To set the MKL library, you can modify the variable “mkl_library_dir” and “mkl_include_dir” in the siteconfig.py.

   • To set the Eigen library, you can modify the variable “eigen_include_dir” in the siteconfig.py. Eigen is a C++ template library for linear algebra. You can find and download it on the official website Eigen.

3. After completing the above preparatory work, you can install PYATB with the following command:

   pip install ./
   

4. After completing the installation process, you can access the pyatb executable and corresponding module, which can be imported using the import pyatb command.

Related to the Intel MKL library

If you encounter the following problem when running pyatb using the intel oneapi MKL library under Anaconda:

undefined symbol: mkl_sparse_optimize_bsr_trsm_i8

This problem may be caused by a conflict between the mkl library in Anaconda and the mkl library in intel oneapi. You can solve it by using the following method:

export LD_PRELOAD=/opt/intel/oneapi/mkl/2022.0.2/lib/intel64/libmkl_def.so.2:\
/opt/intel/oneapi/mkl/2022.0.2/lib/intel64/libmkl_avx2.so.2:\
/opt/intel/oneapi/mkl/2022.0.2/lib/intel64/libmkl_core.so:\
/opt/intel/oneapi/mkl/2022.0.2/lib/intel64/libmkl_intel_lp64.so:\
/opt/intel/oneapi/mkl/2022.0.2/lib/intel64/libmkl_intel_thread.so:\
/opt/intel/oneapi/compiler/2022.0.2/linux/compiler/lib/intel64_lin/libiomp5.so

Note that you need to replace the MKL path with your own environment. For libiomp5.so, its location may sometimes be in the directory where the intel compiler is located.

Another solution is to use the MKL library in the Anaconda virtual environment. If you cannot find it, you can install the MKL library and its include files using the following commands:

conda install -c conda-forge mkl
conda install -c conda-forge mkl-include