Use proper alignment when copying data to a variable. #20975
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
Using align=1 is very expensive on hardware without unaligned ld/st (eg. GPUs).
maleadt
added
bugfix
|
Any comments? If not, I'd like to merge this some time soon, as it drastically improves performance on some GPU workloads. |
vtjnash
approved these changes
Mar 14, 2017
yuyichao
approved these changes
Mar 14, 2017
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Outdated suggestions cannot be applied.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.
align=1 is very expensive on hardware without unaligned ld/st (eg. GPUs).
ref #20593 (comment)
I'm not really familiar with these codepaths, so please review. A quick check revealed that
lhs.typ == rhs.typ, so just picking the RHS's alignment seemed OK. And as per the LLVM docs,llvm.memsettreats an alignment of 0 or 1 the same (unaligned), in caselayout->alignmentwould ever be 0.