[Pytorch][Vulkan] Add baddbmm #109851
Closed
[Pytorch][Vulkan] Add baddbmm #109851
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
Summary:
Similar implementation like BMM & ADDMM, the bias tensor is using the packed weights, similar to MM, but increases the index via the z-dim to get more matrices in the batch.
Packed bias (input of MM):
```
ivec3 pos(k_, j_, 0);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# no batch
# k_, j_ has only 1/4 of the range as the original matrix size (H*W matrix i=> H/2*W/2*1 3D Image).
```
Packed bias (input of BMM):
```
ivec3 pos(k_, j_, i);
float v = texelFetch(uInput, pos, 0)
# v.xyzw are 4 numbers in one matrix
# i as batch id
```
**To support broadcasting**, the bias packing of `mm` is slightly different than weight packing, which repeats the single element in height-dim twice to fill the 4 planes (see code for details). The width-dim doesn’t repeat twice, but the code still works, because stacking 3 planes together with the last one empty yields the same 3D image.
However, this doesn’t work for `bmm`, since it’s a series of `{4 planes} {4 planes} … {4 planes}`, and each `{4 planes}` represents a matrix, so only 3 planes completely mess up the indexing. Thus, I repeat the single element in width-dim as well to fill all 4 planes to have the correct indexing.
https://pytorch.org/docs/stable/generated/torch.baddbmm.html
Test Plan:
```
[ttingchulin@27298.od /data/sandcastle/boxes/fbsource (bmm)]$ LD_LIBRARY_PATH=third-party/swiftshader/lib/linux-x64/ buck run fbcode/mode/dev-nosan //xplat/caffe2:pt_vulkan_api_test_bin
```
Reviewed By: yipjustin
Differential Revision: D49402181
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/109851
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (2 Unrelated Failures)As of commit ff78668 with merge base e1d7123 ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
UNSTABLE - The following job failed but was likely due to flakiness present on trunk and has been marked as unstable:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/periodic
|
This pull request was exported from Phabricator. Differential Revision: D49402181 |
yipjustin
approved these changes
Sep 22, 2023
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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:
Similar implementation like BMM & ADDMM, the bias tensor is using the packed weights, similar to MM, but increases the index via the z-dim to get more matrices in the batch.
Packed bias (input of MM):
Packed bias (input of BMM):
To support broadcasting, the bias packing of
mmis slightly different than weight packing, which repeats the single element in height-dim twice to fill the 4 planes (see code for details). The width-dim doesn’t repeat twice, but the code still works, because stacking 3 planes together with the last one empty yields the same 3D image.However, this doesn’t work for
bmm, since it’s a series of{4 planes} {4 planes} … {4 planes}, and each{4 planes}represents a matrix, so only 3 planes completely mess up the indexing. Thus, I repeat the single element in width-dim as well to fill all 4 planes to have the correct indexing.https://pytorch.org/docs/stable/generated/torch.baddbmm.html
Test Plan:
Reviewed By: yipjustin
Differential Revision: D49402181