-
Notifications
You must be signed in to change notification settings - Fork 407
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
Change to make CudaMallocAsync available via opt-in with CMake option Kokkos_ENABLE_CUDAMALLOCASYNC #4233
Conversation
Can one of the admins verify this patch? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would prefer CUDA_MALLOC_ASYNC
over CUDA_MALLOCASYNC
and you should be able to remove one preprocessor variable. We also might mention somewhere that this causes problems with UCX
as of now.
Otherwise, this looks good to me.
Agree with Daniel regarding naming. |
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
The Workflow run is cancelling this PR. It is an earlier duplicate of 3979050 run. |
The Workflow run is cancelling this PR. It is an earlier duplicate of 3979336 run. |
The Workflow run is cancelling this PR. It is an earlier duplicate of 3979336 run. |
The Workflow run is cancelling this PR. It is an earlier duplicate of 3979336 run. |
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
Co-authored-by: Daniel Arndt <arndtd@ornl.gov>
We have been trying to contain the proliferation of CMake options. Why should we expose this one to the user? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
#4228 mentions the issue reported will be resolved on the UCX side.
Users can use latest release version of Kokkos in the meantime. I do not see a strong rational to add a new option.
For me, the situation is something like:
This implies to me that it should not be used by default, but people might still be interested in using it. That sounds like a good candidate for an experimental feature to me. |
Agreeing with Daniel here, sorry Damien ... |
Lets vote on Wednesday (while Damien is not here anyway :-) ) |
OK to test |
cmake/KokkosCore_config.h.in
Outdated
@@ -41,6 +41,7 @@ | |||
#cmakedefine KOKKOS_ENABLE_CUDA_LAMBDA | |||
#cmakedefine KOKKOS_ENABLE_CUDA_CONSTEXPR | |||
#cmakedefine KOKKOS_ENABLE_CUDA_LDG_INTRINSIC | |||
#cmakedefine KOKKOS_ENABLE_CUDA_MALLOC_ASYNC |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I forgot to add it to the agenda of yesterday's meeting but I will withdraw my objection if we mark the option as "experimental". Something like
#cmakedefine KOKKOS_ENABLE_CUDA_MALLOC_ASYNC | |
#cmakedefine KOKKOS_ENABLE_IMPL_CUDA_MALLOC_ASYNC |
would do.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Will do!
Retest this please. |
#4228
Add the Kokkos_ENABLE_CUDAMALLOCASYNC option in CMake to enable cudaMallocAsync, OFF by default