https://github.com/pkgw/dedalus-builder
Make system-tuned Conda packages of the Dedalus differential equation solver.
https://github.com/pkgw/dedalus-builder
Last synced: over 1 year ago
JSON representation
Make system-tuned Conda packages of the Dedalus differential equation solver.
- Host: GitHub
- URL: https://github.com/pkgw/dedalus-builder
- Owner: pkgw
- Created: 2017-10-09T00:36:22.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2017-10-14T12:21:07.000Z (over 8 years ago)
- Last Synced: 2025-01-23T17:35:24.211Z (over 1 year ago)
- Language: Shell
- Size: 46.9 KB
- Stars: 1
- Watchers: 4
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Support: support/check-environment.py
Awesome Lists containing this project
README
# dedalus-builder
The scripts in this repository make it easy to install Dedalus on your system
in a way that leverages your system’s native MPI and/or BLAS implementations.
## The approach
We use Anaconda Python and its `conda` tool to make pre-compiled packages of
Dedalus and its dependencies. For dependencies that do not depend sensitively
on your system’s particular characteristics, we use pre-built packages from
the [conda-forge](https://conda-forge.org/) project. But when we can take
advantage of your system MPI or BLAS installations, we build custom packages.
You can then install these packages into new directory to create a
self-contained Python installation that includes Dedalus and its dependencies.
Or, if you already have an Anaconda installation, you can install the packages
into that directory.
The packages built by this tool include binary code that depends on your
computer's pre-existing MPI setup. So they *might* be portable to other
systems, if they have a compatible setup, but they probably won’t be.
## Instructions
First, set up your shell environment so that the programs can identify any
system libraries you wish to use:
- If you wish to use a system MPI installation, make sure that the desired
`mpicc` program is in your `$PATH`. Otherwise, Anaconda’s prepackaged
MPI implementation will be used.
- If you wish to use a system BLAS installation, pass the argument
`--with-blas=LIBDIR` to the `./configure` program described below, where
your BLAS libraries live in the directory `LIBDIR`.
Then, run `./initialize`. This will download and set up a self-contained
version of the Conda build infrastructure.
Then, run `./configure`, possibly adding a `--with-blas=LIBDIR` argument. This
will validate your build configuration and save its parameters. If you need to
alter your build environment, just rerun this script.
Then run `./compile`. This will compile Dedalus and its the MPI- and
BLAS-sensitive dependencies, creating binary package files in the Conda
format. The goal is that these builds should succeed reliably but there are
likely bugs that need to be shaken out.
Once the packages are built, you have two installation choices. By running
`./install-fresh `, you can create a new, self-contained Python
installation that contains Dedalus and IPython. By running `./install-into
`, you can install Dedalus into a preexisting Anaconda Python
installation.