Nasty bugs in custom update kernels #439
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
|
Ironically, in parallel, I was adding some new functionality and extending the tests to cover it and these were starting to fail because of the second of these issues so the testing was nearly good enough 😟 |
Codecov Report
@@ Coverage Diff @@
## master #439 +/- ##
==========================================
+ Coverage 87.91% 87.94% +0.03%
==========================================
Files 78 78
Lines 16445 16494 +49
==========================================
+ Hits 14457 14506 +49
Misses 1988 1988
Continue to review full report at Codecov.
|
tnowotny
reviewed
Jul 20, 2021
tnowotny
approved these changes
Jul 20, 2021
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Outdated suggestions cannot be applied.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.
Two really nasty bugs have come to light as I've started building models which do more complex stuff with custom updates:
Firstly, we were building the group start indices (the arrays that get binary searched to find the index to the right data if groups get merged) across all custom update groups. This is clearly wrong as separate kernel(s) are generated for each update group.
Secondly, in the CUDA kernel, the group start indices for custom updates and custom weight updates weren't being generated together so if there were enough of either to require binary searching there would be problems.