Skip to content

Conversation

@ssnl
Copy link
Collaborator

@ssnl ssnl commented Jun 22, 2018

Strategy:

  1. Move all upstream files to Module_CUDA_fix/upstream/*.
  2. Create a custom Module_CUDA_fix/FindCUDA.cmake that first includes upstream/CMakeInitializeConfigs.cmake and then upstream/FindCUDA.cmake so it can be used by submodules too.
  3. Remove all instances of include(CMakeInitializeConfigs).

For easier review process, here's how cmake/Module_CUDA_fix looks like after this patch.

@ezyang @soumith @orionr

cc @li-roy @zdevito

@ssnl
Copy link
Collaborator Author

ssnl commented Jun 22, 2018

@pytorchbot retest this please

@ssnl ssnl merged commit 675b579 into pytorch:master Jun 22, 2018
@ssnl ssnl deleted the findcuda_wrapper branch June 22, 2018 19:29
@ngimel
Copy link
Collaborator

ngimel commented Jun 22, 2018

Thanks for fixing the build! May be it is time to just require newer CMake? (3.10 or whichever has all the FindCUDA stuff).

@ssnl
Copy link
Collaborator Author

ssnl commented Jun 22, 2018

@ngimel IIRC @soumith said we want to keep cmake 3.5 (3.9?) support.

@ngimel
Copy link
Collaborator

ngimel commented Jun 22, 2018

Then may be cmake 3.5 builds should be added to CI?

@ssnl
Copy link
Collaborator Author

ssnl commented Jun 24, 2018

@soumith @ezyang should we use older cmake versions in CI to prevent such regressions?

@ezyang
Copy link
Contributor

ezyang commented Jun 24, 2018

Yes.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants