[GraphBolt][CUDA] Refactor overlap_graph_fetch, simplify gb.DataLoader.
#7681
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
|
To trigger regression tests:
|
mfbalin
added
the
expedited
mfbalin
commented
Aug 11, 2024
frozenbugs
reviewed
Aug 16, 2024
| @@ -134,16 +134,14 @@ def create_dataloader( | |||
| ############################################################################ | |||
| if args.storage_device != "cpu": | |||
| datapipe = datapipe.copy_to(device) | |||
| datapipe = datapipe.sample_neighbor(graph, args.fanout) | |||
| datapipe = datapipe.sample_neighbor( | |||
| graph, args.fanout, overlap_fetch=args.storage_device == "pinned" | |||
There was a problem hiding this comment.
Did we release the overlap_fetch in last release, if so, we need to highlight the API change if we really believe this is a good api change.
There was a problem hiding this comment.
I think it is in DGL 2.3. In this release, this parameter moved from gb.DataLoader to sample_neighbor.
There was a problem hiding this comment.
https://docs.dgl.ai/en/latest/generated/dgl.graphbolt.DataLoader.html#dgl.graphbolt.DataLoader shows the overlap_graph_fetch option.
This pull request was closed.
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Outdated suggestions cannot be applied.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.
Description
gb.DataLoaderto the sampling classes foroverlap_graph_fetch. It is an argument of the sampler now.gpu_graph_cacheto the FusedCSCSamplingGraph because otherwise it won't be shared across train or validation dataloader for example. That way, there will be a single graph cache for multiple dataloader instances.The end goal is to make sampling and compaction async and hide the latency of GPU CPU synchronizations during these stages. A lot of our optimizations also have inherent GPU CPU synchronization, which will be handled in the same way. I needed to clean the code and refactor before #7682.
I might refactor more and reduce the amount of code in our code base in the future.
Checklist
Please feel free to remove inapplicable items for your PR.
Changes