Use thread-local to track CUDA device in JNI [skip ci] #6597
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Please update the changelog in order to start CI tests. View the gpuCI docs here. |
abellina
approved these changes
Oct 26, 2020
revans2
approved these changes
Oct 26, 2020
vyasr
added
4 - Needs Review
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Looking at a recent Nsight profile of a Java application using cudf, I noticed a lot of CPU samples in
cudaGetDevice. This is caused by theauto_set_devicecalls made in each cudf JNI function to ensure each thread is using the same device used when RMM was initialized. Initially we thought callingcudaGetDevicewould be "cheap enough", but this is apparently not the case, at least when profiling under Nsight.This changes the cudf JNI code to track the thread's CUDA device in a thread local rather than needing to call
cudaGetDeviceon each cudf call to obtain it. This saved 0.5ms perTable.contiguousSplitcall in a microbenchmark, and I noticed it also significantly reduced the time dilation we've seen in Nsight profiles of cudf Java applications.