Installing Super_Hydro¶
TL;DR¶
Simple installation (no performance boosts)
python3 -m pip install super_hydro
Including the FFTW:
conda env create -n super_hydro -f anaconda-project.yaml
conda install -n super_hydro conda-forge::pyfftw # FFTW support
conda activate super_hydro
python3 -m pip install super_hydro[fftw]
If you have an NVIDIA GPU:
conda env create -n super_hydro -f anaconda-project.yaml
conda install -n super_hydro conda-forge::pyfftw conda-forge::cupy
conda activate super_hydro
python3 -m pip install super_hydro[fftw,gpu] # Choose your options
For notebook support, install the super_hydro [Jupyter] kernel:
python3 -m ipykernel install --user --name "super_hydro" --display-name "Python 3 (super_hydro)"
From source, you can use Anaconda Project:
anaconda-project run init # No GPU support (use for client)
# OR
anaconda-project run init_gpu # If you have an NVIDA GPU
Details¶
Pip¶
The super_hydro application can be installed with Pip using one of the following:
python3 -m pip install super_hydro # Install basic version
python3 -m pip install super_hydro[fftw] # Needs FFTW libraries
python3 -m pip install super_hydro[gpu] # Needs CUDA toolkit
but the fftw and gpu extras that improve performance require specific libraries that
cannot be installed with Pip:
[fftw]: Install the pyFFTW library which requires the FFTW libraries to be installed on your system.[gpu]: Install the CuPy library which requires an NVIDIA GPU as well as the corresponding CUDA toolkit to be installed on your system.
Installing these performance version with pip can be a bit problematic because you must manually manage the correct underlying FFTW and CUDA libraries.
Conda¶
Instead, you can use Conda to first create an environment with appropriate versions of these binary libraries, then install the reset using Pip.
If you do not already have Conda installed, download and install one of the following:
Miniconda: Minimal conda installation. If you want the full anaconda stack later, you can
conda install anaconda.Anaconda: Rather complete conda installation with the full scientific computing stack right from the start.
conda env create -n super_hydro -f anaconda-project.yaml
conda install -n super_hydro conda-forge::pyfftw # FFTW support
conda install -n super_hydro conda-forge::cupy # GPU support
conda activate super_hydro
python3 -m pip install super_hydro[fftw,gpu] # Choose your options
Anaconda Project¶
Alternatively, you can do all of this with Anaconda Project if you have access to the source:
anaconda-project run init # No GPU support or client
# OR
anaconda-project run init_gpu # If you have an NVIDA GPU
These are roughly equivalent to, but more convenient than, the following Conda commands:
conda env create -p envs/super_hydro -f anaconda-project.yaml
conda activate envs/super_hydro
poetry install -E docs -E tests -E fftw
#pip install --use-feature=in-tree-build -e .[docs,tests,fftw]
python3 -m ipykernel install --user --name "super_hydro" --display-name "Python 3 (super_hydro)"
jupyter nbextensions_configurator enable --user
or, if you have an NVIDIA GPU and want to use it:
conda env create -p envs/super_hydro_gpu -f anaconda-project.yaml
conda activate envs/super_hydro_gpu
conda install conda-forge::cupy
poetry install -E docs -E tests -E fftw -E gpu
#pip install --use-feature=in-tree-build -e .[docs,tests,fftw,gpu]
python3 -m ipykernel install --user --name "super_hydro_gpu" --display-name "Python 3 (super_hydro_gpu)"
jupyter nbextensions_configurator enable --user
These create Conda environments in envs/conda_env or envs/conda_env_gpu with the
appropriate FFTW or CUDA binary libraries, then use pip to install the
remaining dependencies. Finally, a jupyter kernel is created so that the notebook
examples can find the correct environment. (All of our documentation uses the
super_hydro kernel, but you can change this to use the super_hydro_gpu kernel if you
are running the server locally and want the performance benefits.)
Once these are done, you can run various commands with
anaconda-project run server # Start a server
anaconda-project run client # Start a client
anaconda-project run shell # Start a shell
The last command starts a [Bash] shell with the appropriate environment activated, similar to what you would get above with
conda activate super_hydro
Jupyter Kernel¶
If you run Jupyter notebook from an environment where you installed super_hydro,
then you can just visit the Demo notebooks. If you launch [Jupyter] from
another environment, however, then you will need to install a super_hydro kernel.
This can be done by activating the virtual environment, then running:
python3 -m ipykernel install --user --name super_hydro --display-name super_hydro
You can then select the super_hydro kernel. To remove the kernel, run:
jupyter kernelspec remove super_hydro
You can install multiple kernels, for example, if you have both super_hydro and
super_hydro_gpu environments:
conda activate super_hydro
python3 -m ipykernel install --user --name super_hydro --display-name super_hydro
conda activate super_hydro_gput
python3 -m ipykernel install --user --name super_hydro_gput --display-name super_hydro_gpu
If you use Anaconda Project, this is done by the following commands, which will create both environments and kernels.
anaconda-project run init
anaconda-project run init_gpu