If you want to develop cudamat, you should clone the github repository instead of downloading a release. Furthermore, it is useful to install it in editable mode. Instead of copying the files somewhere Python can find them, this will point Python directly to the directory you install it from. Either of the following commands will do:
# a) Install for your user in editable mode:
python setup.py develop --prefix=~/.local
# b) Install for your user in editable mode, but with pip:
pip install --user --editable .As for the standard installation, you can set the NVCC_FLAGS
environment variable to compile for a specific architecture.
Your changes to .py files will show up immediately the next time you import
cudamat. Changes to .cu and .cuh files require a recompilation triggered
by just running the above installation command again.
To obtain the latest version, just pull in the remote changes:
git checkout master
git fetch origin
git merge origin/masterThen recompile as per the instructions in the previous section.
If you created a great new feature that is useful to the rest of the world, and it even comes with docstrings and updated tests, we will gladly incorporate it into cudamat. To do that, you will need to send us a pull request from your fork.
If you haven't forked cudamat yet, log in to your github account, go to
https://github.com/cudamat/cudamat and hit the "Fork" button.
Now instead of having origin point to cudamat/cudamat, you will want to have
it point to your fork, and have upstream point to the official project:
git remote rename origin upstream
git remote add origin [email protected]:yourusername/cudamat
git fetch originCreate a branch to house your changes:
git checkout -b my-new-featureHack away, then add your changes, commit and push:
git add all.py my.py changes.py
git commit -m 'Added a new feature that does this and that'
git push origin my-new-featureNow send us a pull request asking to merge yourusername:my-new-feature into
cudamat:master and we will come back to you!