Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Build CUDA version on conda-forge #156

Open
mtazzari opened this issue Oct 8, 2018 · 3 comments
Open

Build CUDA version on conda-forge #156

mtazzari opened this issue Oct 8, 2018 · 3 comments
Assignees
Labels

Comments

@mtazzari
Copy link
Owner

mtazzari commented Oct 8, 2018

Now that many users have started using GALARIO on GPUS, it would significantly increase user-friendliness to ship the GPU version through conda.
It seems that conda-forge now supports CUDA packages quite straightforwardily. I think that adding cudatoolkit as a dependency would be sufficient:
https://github.com/conda-forge/cudatoolkit-dev-feedstock

A solution adopted by some packages is to offer two versions of the conda package, i.e. galario and galario-gpu.
Since we structured galario to have the CPU/GPU libraries inside rather than the whole library compiled for CPU or GPU, a possible solution is to use a docker image to prepare the recipe.

Here there is some discussion:
conda-forge/caffe-feedstock#27

Here two example recipes that use cuda 8:
https://github.com/loopbio/caffe-feedstock/blob/f401380a3b63271ddfe2d89a5c68d2448e80ec00/recipe/meta.yaml#L23

this recipe uses a docker image (could be the neatest way for us):
https://github.com/loopbio/caffe-feedstock/blob/f401380a3b63271ddfe2d89a5c68d2448e80ec00/conda-forge.yml#L12-L21

@fredRos
Copy link
Collaborator

fredRos commented Oct 22, 2018

Here two example recipes that use cuda 8:

I need to check this out, we should always use the latest cuda version so people with the latest hardware benefit. I hope the conda forge system allows us to use the latest cuda automatically

@fredRos fredRos self-assigned this Oct 22, 2018
@fredRos fredRos added the build label Oct 22, 2018
@fredRos
Copy link
Collaborator

fredRos commented Oct 29, 2018

I hope the conda forge system allows us to use the latest cuda automatically

Of course nothing comes for free but it seems that conda is supported by nvidia guys, and they already moved cudatoolkit-dev to the latest version 10. That's great

@fredRos
Copy link
Collaborator

fredRos commented Oct 29, 2018

I think it would be easiest to always build on ci using cudatoolkit-dev but not require that for installation. Instead, the user would have to have cuda installed as usual. I wonder if there is a problem if a user has cuda 9 installed and we build galario with cuda 10. But I thought nvidia would strive hard for backwards compatibility.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

2 participants