emirge - environment for MirgeCom

Emirge is a repository with some tools that captures a set of dependencies for mirgecom, as well as mirgecom itself.

The mirgecom dependencies that emirge installs are:

1. Miniforge/Conda and conda packages (e.g., pocl)
2. Pip packages (e.g., pyopencl)

Installation

In most cases, running ./install.sh should be sufficient to install all packages and their dependencies.

./install.sh takes several arguments:

• --install-prefix=DIR: Install mirgecom and git pip packages to DIR instead of the default (./).
• --conda-prefix=DIR: Install conda in DIR instead of the default directory (./miniforge3).
• --env-name=NAME: Create conda environment named NAME instead of the default (ceesd).
• --modules: Install a modules.zip file that contains a copy of all python packages that are installed through git (see below for details).
• --branch=NAME: Install the NAMEd branch of mirgecom instead of the default branch (main).
• --fork=NAME : Install mirgecom from a fork (default=illinois-ceesd).
• --conda-pkgs=FILE: Install additional conda packages from the list of package names specified in FILE.
• --conda-env=FILE: Obtain conda package versions from conda environment file FILE.
• --pip-pkgs=FILE: Install additional pip packages from the pip requirements file specified in FILE.
• --git-ssh: Use SSH-based URL to clone mirgecom.
• --debug: Show debugging output of this script (set -x).
• --skip-clone: Skip cloning mirgecom, assume it will be manually copied to the selected installation prefix.
• --help: Print this help text.

Testing the installation

Testing can be done by:

$ mirgecom/examples/run_examples.sh mirgecom/examples/

Running on systems with lots of nodes (>256)

On large systems, the file system can become a bottleneck for loading Python packages. On these systems, it is recommended to create a zip file with the modules to speed up the startup process. This can be done by specifying the --modules parameter to install.sh, or by running makezip.sh after installation.

See https://github.com/illinois-ceesd/planning/issues/26 for more details.

Manual installation

Please use the instructions above instead.

Running wavelet0

Prerequesites

Install POCL

Installation with conda-miniforge

$ wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
# For Power8/9:
# $ wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-ppc64le.sh
# For MacOS:
# $ wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-MacOSX-x86_64.sh

# Install Miniforge/conda:
$ bash ./Miniforge3-Linux-x86_64.sh

# Optional: create conda environment
$ export MY_CONDA=/path/to/installed/conda # Default installation path: $HOME/miniforge3
$ $MY_CONDA/bin/conda create -n ceesd
$ . $MY_CONDA/bin/activate ceesd

# Install required conda packages:
$ conda install pip pocl numpy pyopencl islpy flake8 mypy pudb

# Install optional conda packages:
$ conda install clinfo

# In a new session, you may reactivate this environment using:
. $MY_CONDA/bin/activate ceesd

Installation with Spack

$ git clone [email protected]:spack/spack
$ source spack/share/spack/setup-env.sh
# Maybe edit your Spack config
# $ spack config edit packages
$ spack install pocl

Install Python packages

$ pip install pyvisfile
$ for m in pytools pymbolic dagrt leap loopy meshmode grudge mirgecom; do cd $m && pip install -e . && cd ..; done

Run wavelet0

$ cd mirgecom/examples; python wave-eager.py