-
Notifications
You must be signed in to change notification settings - Fork 607
Add more Edge->TOSA lowerings and fixes #475
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
Jerry-Ge
commented
Sep 25, 2023
- FP32 Softmax lowering
- Conv2d lowering with groups
- Add promtote_shape and tranpose helpers for aligning with TOSA rank requirements
- Add more e2e test cases
- Other minor fixes
✅ Deploy Preview for resplendent-gnome-14e531 canceled.
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks Jerry. The PR is looking good at a high level. I left some comments inline. Getting so close to MV2 :)
Also in the future no need to squash the commits, you may have >1 commits per PR. It is good to keep one logical change per commit.
it's weird that I had no
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks nice! Really minor comments and they can be follow up pull request.
* FP32 Softmax lowering * Conv2d lowering with groups * Add promtote_shape and tranpose helpers for aligning with TOSA rank requirements * Add more e2e test cases * Other minor fixes Signed-off-by: Jerry Ge <[email protected]>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks Jerry. If you think this is good to go and no changes are pending on your side for this PR, I can merge this. Please let me know.
@digantdesai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@digantdesai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
1 similar comment
@digantdesai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
I found lintrunner didn't cover all files after local commit so use: |
[conv2d_res.name], | ||
attr, | ||
) | ||
if group.number > 1: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For this to be depthwise conv you need group.number == output_channels == input_channels, right?
@@ -538,6 +536,79 @@ def preprocess( # noqa: C901 | |||
[outp.name], | |||
attr_output_transpose, | |||
) | |||
elif exir_ops.edge.aten._softmax.default == node.target: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
curious, are you doing fp emulation for softmax?
@digantdesai merged this pull request in f6a8d9d. |