Reduce shared memory usage in gpuComputePageSizes by 50% #13047
Merged
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
nvdbaranec
added
libcudf
ttnghia
approved these changes
Apr 4, 2023
…ec/cudf into parquet_shared_memory_reduction
Right. Forgot to update all the function comments. Will fix. |
|
/merge |
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Outdated suggestions cannot be applied.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.
In a multithreaded, multi-stream environment (Spark) we were experiencing a performance regression on some benchmark queries. The culprit was gpu scheduling issues related to the
gpuComputePageSizeskernel. Dependent kernels (gpuDecodePages) were getting serialized becausegpuComputePageSizeswasn't running alongside other streams well.The fix was reducing shared memory usage in
gpuComputePageSizes. The kernel shares a lot of code and data structures withgpuDecodePagesbut doesn't actually use several of the large buffers that are stored in shared memory. This PR refactors those buffers out so that they are only declared in thegpuDecodePageskernel, reducing the shared usage by 50% (3kb).This clears up the performance issue on Spark. I am currently experiencing build issues with cudf benchmarks so I'm marking this as do-not-merge until I can verify with them.
Checklist