Allocate a big output tensor and split in group_index_select_dim0_backward #1764
Conversation
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
✅ Deploy Preview for pytorch-fbgemm-docs canceled.
|
|
This pull request was exported from Phabricator. Differential Revision: D45823864 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Jun 20, 2023
…kward (pytorch#1764) Summary: Pull Request resolved: pytorch#1764 Before this diff, `group_index_select_dim0` backward calls `at::zeros` `group_size` number of times which launches `group_size` elementwise kernels. Since `group_size` can be a large value (up to 55), this can be costly. This diff fixes the problem by allocating one big tensor and splitting it into smaller tensors. This will launch only one elementwise kernel per group. However, this can cause higher overhead on the host side. Reviewed By: jspark1105 Differential Revision: D45823864 fbshipit-source-id: 83832fede9085ff5829f06e7621dfee8621a3e06
|
This pull request was exported from Phabricator. Differential Revision: D45823864 |
sryap
added a commit
to sryap/FBGEMM
that referenced
this pull request
Jun 20, 2023
…kward (pytorch#1764) Summary: Pull Request resolved: pytorch#1764 Before this diff, `group_index_select_dim0` backward calls `at::zeros` `group_size` number of times which launches `group_size` elementwise kernels. Since `group_size` can be a large value (up to 55), this can be costly. This diff fixes the problem by allocating one big tensor and splitting it into smaller tensors. This will launch only one elementwise kernel per group. However, this can cause higher overhead on the host side. Reviewed By: jspark1105 Differential Revision: D45823864 fbshipit-source-id: 92939fbd3801c599c475f45609c55dcc23cedbfc
|
This pull request was exported from Phabricator. Differential Revision: D45823864 |
…kward (pytorch#1764) Summary: Pull Request resolved: pytorch#1764 Before this diff, `group_index_select_dim0` backward calls `at::zeros` `group_size` number of times which launches `group_size` elementwise kernels. Since `group_size` can be a large value (up to 55), this can be costly. This diff fixes the problem by allocating one big tensor and splitting it into smaller tensors. This will launch only one elementwise kernel per group. However, this can cause higher overhead on the host side. Reviewed By: jspark1105 Differential Revision: D45823864 fbshipit-source-id: c44dd60e4707ec25ba9d4d667cdc1a0ef17566ea
|
This pull request was exported from Phabricator. Differential Revision: D45823864 |
|
This pull request has been merged in 0185ac7. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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:
Before this diff,
group_index_select_dim0backward callsat::zerosgroup_sizenumber of times which launchesgroup_sizeelementwisekernels. Since
group_sizecan be a large value (up to 55), this canbe costly.
This diff fixes the problem by allocating one big tensor and splitting
it into smaller tensors. This will launch only one elementwise kernel
per group. However, this can cause higher overhead on the host side.
Differential Revision: D45823864