-
Notifications
You must be signed in to change notification settings - Fork 4
Installation on HPC
On HPCs (such as NERSC Perlmutter considered below), it is recommended that you install Triumvirate in development mode (see below) in a clean environment without sourcing an external environment, as solutions to installation within external environments are individual to that environment.
🔔 Installation instructions under the cosmodesi
environment on NERSC Perlmutter are in this wiki.
In a clean login-shell session, change to the directory where you want keep the Triumvirate code repository. Clone and enter the repository:
cd <repo-dir>
git clone https://github.com/MikeSWang/Triumvirate.git
cd Triumvirate/
Load the necessary HPC software modules which may include the compiler/package manager and dependencies such as OpenMP, GSL and FFTW.
For NERSC Perlmutter, the following is/are needed:
module load cray-fftw
❕ GSL was treated as a module before on NERSC, but it is now a system-wide library that does not need to be explicitly loaded.
If you use the Python interface of Triumvirate, you also need the Python/Conda module:
module load python
It is recommended that you install the Python package in a separate dedicated Conda environment:
conda create -n <env>
conda activate <env>
In general, execute
make clean
make -j install useomp=true
to install an OpenMP-enabled version of Triumvirate.
Replace install
above with cppinstall
or pyinstall
if you only need the C++ or Python interface respectively. See the official documentation for more details.
To check the Python package is installed, see if
pip list | grep 'Triumvirate'
returns a non-empty string.
To check it is installed correctly, use this function in Python:
>>> from triumvirate import validate_installation as vi
>>> vi()
Installation of Triumvirate has been validated.
True
It should return True
and print out the message "Installation of Triumvirate has been validated.".
Wiki
- Home
- Installation troubleshooting
- Use with HPCs
- Installation on HPC
- Running jobs on HPC
- HPC-specific instructions
- Code snippets