Quantum Software on HPC Systems

Overview

This page describes how a user can successfully install specific quantum software tools to run on select OLCF HPC systems. This allows a user to utilize our systems to run a simulator, or even as a “frontend” to a vendor’s quantum machine (“backend”).

For most of these instructions, we use conda environments. For more detailed information on how to use Python modules and conda environments on OLCF systems, see our Python on OLCF Systems page, and our Conda Basics guide.

Example Python scripts for each QCUP vendor are located at our new Hello QCUP Scripts page.

Warning

Currently, the install steps listed below only work for our x86_64 based systems (Andes, Frontier, etc.). The steps can be explored on Summit, but – due to Summit’s Power architecture – is not recommended or guaranteed to work.

Note

Working installations are not limited to versions shown below. Versions for libraries listed below are what was tested most recently.

Qiskit

Qiskit is open-source software for working with quantum computers at the level of circuits, pulses, and algorithms. Installing Qiskit provides access to the IBM Quantum backends by default, but installing additional provider plugins enables access to other vendors’ backends as well.

IBM Quantum

Installing qiskit and qiskit-ibm-runtime provides access to IBM Quantum backends. Since cloud-based simulators have been deprecated, installing qiskit-aer enables access to local simulators.

See these links for more information:

• https://docs.quantum.ibm.com/start/install

• https://docs.quantum.ibm.com/start/hello-world

• https://docs.quantum.ibm.com/lab

• https://docs.quantum.ibm.com/verify/local-testing-mode

Andes

module load gcc/9.3.0
module load python
conda create -n ENV_NAME python=3.10 -c conda-forge
source activate ENV_NAME
conda install qiskit=1.0.2 qiskit-ibm-runtime=0.22.0 qiskit-aer=0.13.3 psutil -c conda-forge

Frontier

module load PrgEnv-gnu/8.3.3
module load miniforge3/23.11.0
conda create -n ENV_NAME python=3.10 -c conda-forge
source activate ENV_NAME
conda install qiskit=1.0.2 qiskit-ibm-runtime=0.22.0 qiskit-aer=0.13.3 psutil -c conda-forge

Qiskit IonQ Provider

The qiskit-ionq extension can be installed to access IonQ QPUs through Qiskit.

See these links for more information:

• https://docs.ionq.com/introduction

• https://ionq.com/resources

• https://ionq.com/resources/anthology/developers/hello-many-worlds-in-7-quantum-languages

• https://docs.ionq.com/guides/managing-api-keys

• https://docs.ionq.com/guides/sdks/qiskit

Andes

module load gcc/9.3.0
module load python
conda create -n ENV_NAME python=3.10 numpy scipy -c conda-forge
source activate ENV_NAME
pip install qiskit==1.0.2 qiskit-ionq==0.5.0 --no-cache-dir

Frontier

module load PrgEnv-gnu/8.3.3
module load miniforge3/23.11.0
conda create -n ENV_NAME python=3.10 numpy scipy -c conda-forge
source activate ENV_NAME
pip install qiskit==1.0.2 qiskit-ionq==0.5.0 --no-cache-dir

PyQuil/Forest SDK (Rigetti)

Quil is the Rigetti-developed quantum instruction/assembly language. PyQuil is a Python library for writing and running quantum programs using Quil.

Installing pyQuil requires installing the Forest SDK. To quote Rigetti: “pyQuil, along with quilc, the QVM, and other libraries, make up what is called the Forest SDK”. Because we don’t have Docker functionality and due to normal users not having sudo privileges, this means that you will have to install the SDK via the “bare-bones” method. The general info below came from:

• https://pyquil-docs.rigetti.com/en/stable/

• https://docs.rigetti.com/qcs

• https://docs.rigetti.com/qcs/getting-started/set-up-your-environment/installing-locally

• https://pyquil-docs.rigetti.com/en/stable/getting_started.html#run-your-first-program

The bare-bones installation only contains the executable binaries and manual pages, and doesn’t contain any of the requisite dynamic libraries. As such, installation doesn’t require administrative or sudo privileges. This method of installation requires one, through whatever means, to install shared libraries for BLAS, LAPACK, libffi, and libzmq3. Some download methods are listed here:

• Lapack (with BLAS) download: http://www.netlib.org/lapack/

• libffi download:

  • Older versions: https://sourceware.org/ftp/libffi/

  • Newer: https://github.com/libffi/libffi/releases/

• ZMQ download: https://github.com/zeromq/libzmq/releases

• Forest SDK download: https://qcs.rigetti.com/sdk-downloads

PyQuil - Installing Dependencies

Below is an example script that will download, build, and install all dependencies for pyQuil. Before running the script, make sure to have these modules loaded:

Andes

module load gcc/9.3.0 cmake

Frontier

module load PrgEnv-gnu/8.3.3 cmake

Warning

By default this will install the libraries into your $HOME directory. Also note that you will be prompted where to install the Forest SDK while the script is executing (just press enter to stick with the default $HOME location when prompted).

#!/bin/bash

set -e

WORKDIR=${HOME}/quil_build_dir
mkdir -p ${WORKDIR} && cd ${WORKDIR}

VERSION_LAPACK="3.10.0"
VERSION_LIBFFI="3.4.6"
VERSION_ZEROMQ="4.3.5"
VERSION_FOREST="2.23.0"

URL_LAPACK="https://github.com/Reference-LAPACK/lapack/archive/refs/tags/v${VERSION_LAPACK}.tar.gz"
URL_LIBFFI="https://github.com/libffi/libffi/releases/download/v3.4.6/libffi-${VERSION_LIBFFI}.tar.gz"
URL_ZEROMQ="https://github.com/zeromq/libzmq/releases/download/v4.3.5/zeromq-${VERSION_ZEROMQ}.tar.gz"
URL_FOREST="https://downloads.rigetti.com/qcs-sdk/forest-sdk-${VERSION_FOREST}-linux-barebones.tar.bz2"

NAME_LAPACK="lapack-${VERSION_LAPACK}"
NAME_LIBFFI="libffi-${VERSION_LIBFFI}"
NAME_ZEROMQ="zeromq-${VERSION_ZEROMQ}"
NAME_FOREST="forest-sdk-${VERSION_FOREST}-linux-barebones"

MACHINE=$(uname -m)

INSTALL_DIR_LAPACK=$HOME/rigetti/${NAME_LAPACK}_${MACHINE}
INSTALL_DIR_LIBFFI=$HOME/rigetti/${NAME_LIBFFI}_${MACHINE}
INSTALL_DIR_ZEROMQ=$HOME/rigetti/${NAME_ZEROMQ}_${MACHINE}
INSTALL_DIR_FOREST=$HOME/rigetti/forest-sdk_${VERSION_FOREST}-linux-barebones

# download and unzip the dependencies if they don't exist
pyquil_lib_get() {
        local gz_name="${1}.tar.gz"
        if [[ -f "${gz_name}" ]]; then
                # continue
                echo "${1} already downloaded"
        else
                curl -L -o "${gz_name}" "${2}"
        fi

        if [[ -d "${1}" ]]; then
                echo "${1} already unzipped"
        else
                tar -xf "$gz_name"
        fi

}

pyquil_lib_get $NAME_LAPACK $URL_LAPACK
pyquil_lib_get $NAME_LIBFFI $URL_LIBFFI
pyquil_lib_get $NAME_ZEROMQ $URL_ZEROMQ

if [[ -f "${NAME_FOREST}.tar.bz2" ]]; then
        echo "forest already downloaded"
else
        curl -L -o "${NAME_FOREST}.tar.bz2" "${URL_FOREST}"
fi

if [[ -d "${NAME_FOREST}" ]]; then
        echo "forest already unzipped"
else
        tar -xf "${NAME_FOREST}.tar.bz2"
fi

# install zmq
cd $NAME_ZEROMQ
./configure --prefix="${INSTALL_DIR_ZEROMQ}/" --with-libsodium=no
make
make install
cd $WORKDIR

# install libffi
cd $NAME_LIBFFI
./configure --prefix=$INSTALL_DIR_LIBFFI
make
make install
mkdir -p $INSTALL_DIR_LIBFFI/include
cp include/ffi*.h $INSTALL_DIR_LIBFFI/include/
# only necessary when using libffi==3.4
if [[ ! -f "$INSTALL_DIR_LIBFFI/lib64/libffi.so.6" ]]; then
        ln -s $INSTALL_DIR_LIBFFI/lib64/libffi.so.8 $INSTALL_DIR_LIBFFI/lib64/libffi.so.6
fi
cd $WORKDIR

# install lapack
cd $NAME_LAPACK
mkdir -p build
cd build
cmake \
        -DBUILD_SHARED_LIBS=ON \
        -DCMAKE_INSTALL_LIBDIR=$INSTALL_DIR_LAPACK \
        -DCMAKE_Fortran_COMPILER=gfortran \
        -DCMAKE_C_COMPILER=gcc \
        ..

cmake --build . -j --target install
cd $WORKDIR

# install forest sdk
cd $NAME_FOREST
./${NAME_FOREST}.run

echo ""
echo "Export these variables before running pyquil, the qvm, or quilc:"
echo "export LD_LIBRARY_PATH=\"${INSTALL_DIR_LAPACK}:\$LD_LIBRARY_PATH\""
echo "export LD_LIBRARY_PATH=\"${INSTALL_DIR_LIBFFI}/lib64:\$LD_LIBRARY_PATH\""
echo "export LD_LIBRARY_PATH=\"${INSTALL_DIR_ZEROMQ}/lib:\$LD_LIBRARY_PATH\""
echo "export PATH=\"${INSTALL_DIR_FOREST}:\$PATH\""
echo ""

echo "You can clean up and remove ${WORKDIR} if desired"

To verify your installation, try calling quilc and qvm (after exporting your $PATH and $LD_LIBRARY_PATH with your newly installed libraries):

$ quilc —-version
1.23.0 [e6c0939]
$ qvm —-version
1.17.1 [cf3f91f]

Finally, you are ready to install pyQuil:

Andes

module load gcc/9.3.0
module load python
conda create -n ENV_NAME python=3.10 numpy scipy -c conda-forge
source activate ENV_NAME
pip install pyquil==4.8.0 typing_extensions --no-cache-dir

Frontier

module load PrgEnv-gnu/8.3.3
module load miniforge3/23.11.0
conda create -n ENV_NAME python=3.10 numpy scipy -c conda-forge
source activate ENV_NAME
pip install pyquil==4.8.0 typing_extensions --no-cache-dir

PyQuil - Setting up Servers

Now that everything is installed properly, the rest of the instructions follow Rigetti’s Documentation .

With the way pyQuil works, you need to launch its compiler server, launch the virtual machine / simulator QVM server, and then launch your pyQuil Python program on the same host. Running a Python script will ping and utilize both the compiler and QVM servers. As a proof of concept, this has been done on a single login node and the steps are outlined below.

Using your already created ENV_NAME virtual environment (outlined above):

(ENV_NAME)$ quilc -P -S > quilc.log 2>&1 & qvm -S > qvm.log 2>&1 & python3 script.py ; kill $(jobs -p)

PennyLane

PennyLane is a cross-platform Python library for programming quantum computers. Its differentiable programming paradigm enables the execution and training of quantum programs on various backends.

General information of how to install and use PennyLane can be found here:

• https://docs.pennylane.ai/en/stable/introduction/pennylane.html

• https://pennylane.ai/qml/demonstrations/

• https://pennylane.ai/install/

• https://docs.pennylane.ai/en/stable/index.html

On our systems, the install method is relatively simple:

Andes

module load gcc/9.3.0
module load python
conda create -n ENV_NAME python=3.10 pennylane -c conda-forge
source activate ENV_NAME

Frontier

module load PrgEnv-gnu/8.3.3
module load miniforge3/23.11.0
conda create -n ENV_NAME python=3.10 pennylane -c conda-forge
source activate ENV_NAME

Pytket

The tket framework is a software platform for the development and execution of gate-level quantum computation, providing state-of-the-art performance in circuit compilation. Pytket is a python module for interfacing with tket, and installing the Quantinuum pytket extension allows pytket circuits to be executed on Quantinuum’s quantum devices.

For more information please see:

• https://tket.quantinuum.com/api-docs/

• https://cqcl.github.io/pytket-quantinuum/api/

• https://tket.quantinuum.com/examples/Getting_started.html

• https://github.com/CQCL/pytket-quantinuum/tree/main/examples

Andes

module load gcc/9.3.0
module load python
conda create -n ENV_NAME_qtm python=3.10 numpy -c conda-forge
source activate ENV_NAME_qtm
pip install pytket==1.26.0 pytket-quantinuum==0.31.0 scipy --no-cache-dir

Frontier

module load PrgEnv-gnu/8.3.3
module load miniforge3/23.11.0
conda create -n ENV_NAME_qtm python=3.10 numpy scipy -c conda-forge
source activate ENV_NAME_qtm
pip install pytket==1.26.0 pytket-quantinuum==0.32.0 --no-cache-dir

Batch Jobs

Although lightweight code can be run on the login nodes, more computationally intensive code should be run on the compute nodes through the use of a batch job. See the relevant HPC System Guide for more examples and best practices when running on the compute nodes for a given system.

For the compute nodes to be able to access external URLs (e.g., trying to connect to IBM backends), you’ll have to use proxy settings in your batch script:

export all_proxy=socks://proxy.ccs.ornl.gov:3128/
export ftp_proxy=ftp://proxy.ccs.ornl.gov:3128/
export http_proxy=http://proxy.ccs.ornl.gov:3128/
export https_proxy=http://proxy.ccs.ornl.gov:3128/
export no_proxy='localhost,127.0.0.0/8,*.ccs.ornl.gov'

Warning

These settings currently do not work for pyQuil; thus, when running pyQuil on the compute nodes, you are unable to connect to Rigetti’s machines and can only run local simulators. To be able to connect to Rigetti’s machines, you’ll have to run on the login nodes instead.

When using Python environments with SLURM, it is always recommended to submit a batch script using the export=NONE flag to avoid $PATH issues and use unset SLURM_EXPORT_ENV in your job script (before calling srun); however, this means that previously set environment variables are NOT passed into the batch job, so you will have to set them again (and load modules again) if they are required by your workflow. Alternatively, you can try submitting your batch script from a fresh login shell.

$ sbatch --export=NONE submit.sl

Below are example batch scripts for running on Andes and Frontier:

Andes

#!/bin/bash
#SBATCH -A ABC123
#SBATCH -J job_name
#SBATCH -N 1
#SBATCH -t 0:05:00
#SBATCH -p batch

unset SLURM_EXPORT_ENV

cd $SLURM_SUBMIT_DIR
date

# Set proxy settings so compute nodes can reach internet (required when not using a simulator)
# Currently, does not work properly with pyQuil
export all_proxy=socks://proxy.ccs.ornl.gov:3128/
export ftp_proxy=ftp://proxy.ccs.ornl.gov:3128/
export http_proxy=http://proxy.ccs.ornl.gov:3128/
export https_proxy=http://proxy.ccs.ornl.gov:3128/
export no_proxy='localhost,127.0.0.0/8,*.ccs.ornl.gov'

# Load python module and virtual environment
module load python
source activate ENV_NAME

# For software like Qiskit,PennyLane,Pytket
#python3 script.py

# For pyQuil
#export LD_LIBRARY_PATH="/ccs/home/YOUR_USERNAME/lapackblas:$LD_LIBRARY_PATH"
#export LD_LIBRARY_PATH="/ccs/home/YOUR_USERNAME/ffi/lib64:$LD_LIBRARY_PATH"
#export LD_LIBRARY_PATH="/ccs/home/YOUR_USERNAME/zmq/lib:$LD_LIBRARY_PATH"
#export PATH="/ccs/home/YOUR_USERNAME/rigetti/forest-sdk_2.23.0-linux-barebones:$PATH"
#quilc -P -S > quilc.log 2>&1 & qvm -S > qvm.log 2>&1 & python3 script.py ; kill $(jobs -p)

Frontier

#!/bin/bash
#SBATCH -A ABC123
#SBATCH -J job_name
#SBATCH -N 1
#SBATCH -t 0:05:00
#SBATCH -p batch

unset SLURM_EXPORT_ENV

cd $SLURM_SUBMIT_DIR
date

# Set proxy settings so compute nodes can reach internet (required when not using a simulator)
# Currently, does not work properly with pyQuil
export all_proxy=socks://proxy.ccs.ornl.gov:3128/
export ftp_proxy=ftp://proxy.ccs.ornl.gov:3128/
export http_proxy=http://proxy.ccs.ornl.gov:3128/
export https_proxy=http://proxy.ccs.ornl.gov:3128/
export no_proxy='localhost,127.0.0.0/8,*.ccs.ornl.gov'

# Load python module and virtual environment
module load miniforge3/23.11.0
source activate ENV_NAME

# For software like Qiskit,PennyLane,Pytket
#python3 script.py

# For pyQuil
#export LD_LIBRARY_PATH="/ccs/home/YOUR_USERNAME/lapackblas:$LD_LIBRARY_PATH"
#export LD_LIBRARY_PATH="/ccs/home/YOUR_USERNAME/ffi/lib64:$LD_LIBRARY_PATH"
#export LD_LIBRARY_PATH="/ccs/home/YOUR_USERNAME/zmq/lib:$LD_LIBRARY_PATH"
#export PATH="/ccs/home/YOUR_USERNAME/rigetti/forest-sdk_2.23.0-linux-barebones:$PATH"
#quilc -P -S > quilc.log 2>&1 & qvm -S > qvm.log 2>&1 & python3 script.py ; kill $(jobs -p)