Add variant enable-cuda-dlopen to configure libfabric with the corresponding parameter
#16
Conversation
…esponding parameter
|
CLA Assistant Lite bot: Thank you for your contribution! Before this PR can be accepted, we require that you read and agree to the Mochi Contributor License Agreement. You can digitally sign the CLA by posting a comment on this Pull Request in the format shown below. This agreement will apply to this PR as well as all future Mochi contributions. I have read the CLA Document and I hereby sign the CLA You can retrigger this bot by commenting recheck in this Pull Request |
There was a problem hiding this comment.
Thank you @thomas-bouvier ! I put a couple of minor comments/suggestions.
| @@ -34,6 +34,8 @@ class Libfabric(BuiltinLibfabric): | |||
| variant('cuda', default=False, description='Configure with cuda support') | |||
| depends_on('cuda', when='+cuda') | |||
|
|
|||
| variant('enable-cuda-dlopen', default=False, description='Use dlopen to open cuda library during runtime and load symbol from the library') | |||
There was a problem hiding this comment.
The spack variant should probably be called cuda-dlopen (or maybe cuda_dlopen? I don't know if dashes are Ok in spack variant names or not). At any rate I think they usually omit the "enable" part when mapping variants to build arguments.
| @@ -74,6 +76,9 @@ def configure_args(self): | |||
| if '+cuda' in spec: | |||
| config_args.append('--with-cuda={0}'.format(self.spec['cuda'].prefix)) | |||
|
|
|||
| if '+enable-cuda-dlopen' in spec: | |||
There was a problem hiding this comment.
are the cuda and enable-cuda-dlopen variants independent, or do you need to have the former to then also use the latter? Just checking if we need some logic to enforce that or not.
I'd like to build libfabric with the
--enable-cuda-dlopenflag. This would allow to remove the requirement of CUDA availability at build time, which I do not have.