Specify OMR_CUDA_HOME when CUDA is enabled #374
Conversation
The search mechanism for CUDA in OMR is changing. Defining OMR_CUDA_HOME will allow OMR to continue to locate CUDA with the new mechanism. Signed-off-by: Robert Young <rwy0717@gmail.com>
| @@ -493,7 +493,7 @@ else | |||
| endif | |||
|
|
|||
| ifeq (true,$(OPENJ9_ENABLE_CUDA)) | |||
| CMAKE_ARGS += -DJ9VM_OPT_CUDA=ON | |||
| CMAKE_ARGS += -DJ9VM_OPT_CUDA=ON -DOMR_CUDA_HOME="$(CUDA_HOME)" | |||
| CMAKE_CUDA_ENV := CUDA_BIN_PATH="$(CUDA_HOME)" | |||
There was a problem hiding this comment.
We don't won't need CMAKE_CUDA_ENV any more: can you remove its definitions and uses?
There was a problem hiding this comment.
I was thinking, In order to keep the repository working, that could come in a later PR. So we'd accept this PR and it's counterpart in jdk11, then accept the OMR patch, then I'd open two new PRs cleaning up OpenJ9.gmk. That way, there's no window where cmake builds are broken. Does that sound OK?
|
The copyright and line endings checks are broken, they both failed to check out the PR. the copyright date is already up-to-date, and this PR is coming from an osx machine, so I doubt there are line ending issues. Maybe someone can rerun the tests, or we can just ignore the failures? |
|
Jenkins copyright check |
|
Jenkins line endings check |
|
Same failure, weird. The tool is trying to check out the wrong commit. Here's some choice output: The checkout (wrong): Setting the status (correct): EDIT: Sorry, my mistake, |
|
Those are the expected commit SHAs ( |
|
Jenkins copyright check |
1 similar comment
|
Jenkins copyright check |
|
Jenkins line endings check |
|
Jenkins compile xlinuxcm jdk8 |
The search mechanism for CUDA in OMR is changing. Defining OMR_CUDA_HOME will allow OMR to continue to locate CUDA with the new mechanism.
Signed-off-by: Robert Young rwy0717@gmail.com