Allow 0-stride dimensions for cublas input/output #400
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
Currently, if a tensor contains 0-stride dimensions in one of the last two dimensions, then that tensor will not be used directly as input/output to cublas and a copy will be made transparently. This generates copies during matvec operations because matvec clones a tensor and delegates to matmul, which then copies the tensor because the clone creates a 0-stride dimension. This change allows 0-stride dimensions if the dimension size is 1 (and thus no index will stride in that dimension) and thus eliminates allocations/frees and copies when calling matvec.
luitjens
approved these changes
Mar 30, 2023
|
/blossom-ci |
|
thanks for adding all the tests! |
|
/blossom-ci |
2 similar comments
|
/blossom-ci |
|
/blossom-ci |
tylera-nvidia
approved these changes
Mar 31, 2023
|
/blossom-ci |
4 similar comments
|
/blossom-ci |
|
/blossom-ci |
|
/blossom-ci |
|
/blossom-ci |
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Outdated suggestions cannot be applied.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.
Currently, if a tensor contains 0-stride dimensions in one of the last two dimensions, then that tensor will not be used directly as input/output to cublas matmul and a copy will be made transparently. This generates copies during matvec operations because matvec clones a tensor and delegates to matmul, which then copies the tensor because the clone creates a 0-stride dimension. This change allows 0-stride dimensions if the dimension size is 1 (and thus no index will stride in that dimension) and thus eliminates allocations/frees and copies when calling matvec.