Do not dealloc the scratch space - #100
Merged
Merged
Conversation
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #100 +/- ##
==========================================
+ Coverage 90.78% 90.79% +0.01%
==========================================
Files 60 60
Lines 3179 3183 +4
Branches 389 389
==========================================
+ Hits 2886 2890 +4
Misses 209 209
Partials 84 84 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This is a memory bug in the compute/assemble kernels. (The evaluate kernel is unaffected.) A recent PR (#99) called realloc at the end of the assemble/evaluate kernels to deallocate any unused space. This is technically wrong in the assemble/compute case because compute still writes 0s to the scratch space, even if that space is ultimately not used (because there were no nonzeros to write to it). These zeros may be written in compute after being deallocated in assemble.
This PR fixes this by always leaving the scratch space allocated. This is slightly memory wasteful for evaluate kernels, but the alternative is to either desync how evaluate and assemble work or move the scratch space to some temporary location and only copy the result to vals if it was written to.