Enable sequence TBE CPU via AVX #2195
Closed
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
✅ Deploy Preview for pytorch-fbgemm-docs canceled.
|
|
This pull request was exported from Phabricator. Differential Revision: D51918878 |
sryap
force-pushed
the
export-D51918878
branch
from
c1cfda8
to
13764e7
Compare
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Dec 7, 2023
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation by forcing pooling factors of 1 (i.e., passing `at::arange(index_size)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Differential Revision: D51918878
|
This pull request was exported from Phabricator. Differential Revision: D51918878 |
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Dec 7, 2023
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation by forcing pooling factors of 1 (i.e., passing `at::arange(index_size)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Differential Revision: D51918878
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Dec 7, 2023
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation by forcing pooling factors of 1 (i.e., passing `at::arange(index_size)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Differential Revision: D51918878
sryap
force-pushed
the
export-D51918878
branch
from
13764e7
to
7538611
Compare
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Dec 7, 2023
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation of pooled TBE by forcing pooling factors of 1 (i.e., passing `at::arange(index_size + 1)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Reviewed By: jspark1105 Differential Revision: D51918878
|
This pull request was exported from Phabricator. Differential Revision: D51918878 |
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation of pooled TBE by forcing pooling factors of 1 (i.e., passing `at::arange(index_size + 1)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Reviewed By: jspark1105 Differential Revision: D51918878
sryap
force-pushed
the
export-D51918878
branch
from
7538611
to
9f03fc8
Compare
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Dec 7, 2023
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation of pooled TBE by forcing pooling factors of 1 (i.e., passing `at::arange(index_size + 1)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Reviewed By: jspark1105 Differential Revision: D51918878
|
This pull request was exported from Phabricator. Differential Revision: D51918878 |
sryap
force-pushed
the
export-D51918878
branch
from
9f03fc8
to
c11f5c7
Compare
|
This pull request was exported from Phabricator. Differential Revision: D51918878 |
sryap
pushed a commit
to sryap/FBGEMM
that referenced
this pull request
Dec 8, 2023
Summary: Instead of using the ref implementation for sequence embedding on CPU, this diff directs TBE to invoke the AVX implementation of pooled TBE by forcing pooling factors of 1 (i.e., passing `at::arange(index_size + 1)` as offfsets). The performance gained from using the AVX implementation offsets the overhead incurs in creating the new offsets. Reviewed By: jspark1105, YazhiGao Differential Revision: D51918878
|
This pull request has been merged in f8bd441. |
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.
Summary:
Instead of using the ref implementation for sequence embedding on CPU,
this diff directs TBE to invoke the AVX implementation by forcing
pooling factors of 1 (i.e., passing
at::arange(index_size)asofffsets). The performance gained from using the AVX implementation
offsets the overhead incurs in creating the new offsets.
Differential Revision: D51918878