Support un-fused batchnorm1d/2d on XNNPACK via decomposition #16533
+663
−63
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16533
Note: Links to docs will display an error until the docs builds have been completed. ❌ 3 New Failures, 1 Unrelated FailureAs of commit 1aa7400 with merge base 3dd80c1 ( NEW FAILURES - The following jobs have failed:
UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
Contributor
|
@GregoryComer has exported this pull request. If you are a Meta employee, you can view the originating Diff in D90422630. |
GregoryComer
commented
Jan 10, 2026
GregoryComer
commented
Jan 10, 2026
GregoryComer
commented
Jan 10, 2026
GregoryComer
added
module: xnnpack
…#16533) Summary: Add a new pass - DecomposeBatchNorm - which converts standalone (non-fused) batch norm operators to 1x1 depthwise convolution. This prevents delegation graph breaks when batch norm operators can't be fused. Differential Revision: D90422630
GregoryComer
force-pushed
the
export-D90422630
branch
from
ddd4c1e to
1aa7400
Compare
| """ | ||
|
|
||
| # See https://docs.pytorch.org/docs/stable/generated/torch.nn.BatchNorm1d.html | ||
| denom = torch.sqrt(running_var + torch.Tensor([eps])) |
Contributor
There was a problem hiding this comment.
torch.Tensor([eps]) won't preserve the dtype of running_var. You can do something like:
eps_t = running_var.new_tensor(eps)
denom = torch.sqrt(running_var + eps_t)
|
|
||
| # Compute the equivalent per-channel weights and biases. | ||
| # Note that the batch norm node args are | ||
| # (input, gamma, beta, running_mean, running_var, [training], momentum, eps). |
Contributor
There was a problem hiding this comment.
check that training is False
| # BatchNorm weights have shape [in_c]. | ||
| # So we just need to unsqueeze the [in_c] to to [in_c, 1, 1, [1]]. | ||
| input_meta = bn_node.args[0].meta["val"] | ||
| channel_count = input_meta.shape[1] |
Contributor
There was a problem hiding this comment.
if it's channels-last, we should also return early?
| from executorch.backends.xnnpack.test.tester import Tester | ||
|
|
||
|
|
||
| class TestBatchNorm(unittest.TestCase): |
Contributor
There was a problem hiding this comment.
- Looks like you're testing the no_training variant. Test the other variant too
- Add negative tests too
| .to_edge() | ||
| .check_count({self.bn_name: 1}) | ||
| .run_passes(self.PassStage) | ||
| .check_count({self.conv_name: 1}) |
Contributor
There was a problem hiding this comment.
can you check that conv is in fact depthwise conv with 1×1 kernel?
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: Add a new pass - DecomposeBatchNorm - which converts standalone (non-fused) batch norm operators to 1x1 depthwise convolution. This prevents delegation graph breaks when batch norm operators can't be fused.
Differential Revision: D90422630
cc @digantdesai @cbilgin