[ET-VK] Update partitioner to account for custom packed arguments #6753
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
## Problem Convolution operators, especially for pointwise convolution, may have sizes like ``` W=1, H=1, C=320, N=1280 ``` When represented as a texture, this tensor would normally require a texture with extents ``` (1, 1, 320 / 4 * 1280 = 102400) ``` which would normally exceed texture limits. The new partitioner system detects this and prevents nodes with similar weights from being lowered to Vulkan. However, the partitioner system does not account for the fact that the operator implementation uses a specialized prepacking algorithm which results in valid texture limits for the packed weights. ## Changes * Add field to `OpFeatures` class to annotate that some arguments in an op should be skipped when checking against texture limits * Update metadata tagging pass to ignore annotating constant tensor nodes so that they don't influence memory layout and storage type proposals. Without this change, the tagging pass will try to use buffer storage for the pointwise convolution since the weight can only be represented as a buffer under normal circumstances. Differential Revision: [D65759236](https://our.internmc.facebook.com/intern/diff/D65759236/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/6753
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit d0d34e0 with merge base 793f17e ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
SS-JIA
added a commit
that referenced
this pull request
Nov 11, 2024
## Problem Convolution operators, especially for pointwise convolution, may have sizes like ``` W=1, H=1, C=320, N=1280 ``` When represented as a texture, this tensor would normally require a texture with extents ``` (1, 1, 320 / 4 * 1280 = 102400) ``` which would normally exceed texture limits. The new partitioner system detects this and prevents nodes with similar weights from being lowered to Vulkan. However, the partitioner system does not account for the fact that the operator implementation uses a specialized prepacking algorithm which results in valid texture limits for the packed weights. ## Changes * Add field to `OpFeatures` class to annotate that some arguments in an op should be skipped when checking against texture limits * Update metadata tagging pass to ignore annotating constant tensor nodes so that they don't influence memory layout and storage type proposals. Without this change, the tagging pass will try to use buffer storage for the pointwise convolution since the weight can only be represented as a buffer under normal circumstances. Differential Revision: [D65759236](https://our.internmc.facebook.com/intern/diff/D65759236/) ghstack-source-id: 252885980 Pull Request resolved: #6753
|
This pull request was exported from Phabricator. Differential Revision: D65759236 |
copyrightly
approved these changes
Nov 11, 2024
facebook-github-bot
merged commit d5a0743
into
gh/SS-JIA/140/base
39 of 41 checks passed
facebook-github-bot
temporarily deployed
to
cherry-pick-bot
GitHub Actions
Inactive
kirklandsign
pushed a commit
that referenced
this pull request
Nov 11, 2024
) ## Problem Convolution operators, especially for pointwise convolution, may have sizes like ``` W=1, H=1, C=320, N=1280 ``` When represented as a texture, this tensor would normally require a texture with extents ``` (1, 1, 320 / 4 * 1280 = 102400) ``` which would normally exceed texture limits. The new partitioner system detects this and prevents nodes with similar weights from being lowered to Vulkan. However, the partitioner system does not account for the fact that the operator implementation uses a specialized prepacking algorithm which results in valid texture limits for the packed weights. ## Changes * Add field to `OpFeatures` class to annotate that some arguments in an op should be skipped when checking against texture limits * Update metadata tagging pass to ignore annotating constant tensor nodes so that they don't influence memory layout and storage type proposals. Without this change, the tagging pass will try to use buffer storage for the pointwise convolution since the weight can only be represented as a buffer under normal circumstances. Differential Revision: [D65759236](https://our.internmc.facebook.com/intern/diff/D65759236/) ghstack-source-id: 252885980 Pull Request resolved: #6753 Co-authored-by: Stephen Jia <[email protected]>
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.
Stack from ghstack (oldest at bottom):
Problem
Convolution operators, especially for pointwise convolution, may have sizes like
When represented as a texture, this tensor would normally require a texture with extents
which would normally exceed texture limits. The new partitioner system detects this and prevents nodes with similar weights from being lowered to Vulkan. However, the partitioner system does not account for the fact that the operator implementation uses a specialized prepacking algorithm which results in valid texture limits for the packed weights.
Changes
OpFeaturesclass to annotate that some arguments in an op should be skipped when checking against texture limitsDifferential Revision: D65759236