Support rankRange for op output tensors in opSupportLimits #857
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
|
/cc @philloooo |
reillyeon
approved these changes
Jun 4, 2025
|
FYI, I am prototyping it in Chromium, will inform whether this PR is good to merge. |
huningxin
force-pushed
the
output_tensor_rank
branch
from
f474e81 to
573f901
Compare
huningxin
commented
Sep 15, 2025
There was a problem hiding this comment.
@fdwr @reillyeon , adding new changes that are based on Chromium prototype: https://chromium-review.googlesource.com/c/chromium/src/+/6903617, please take another look. Thanks!
| @@ -2237,8 +2237,8 @@ partial dictionary MLOpSupportLimits { | |||
| </tr> | |||
| <tr> | |||
| <td>*output*</td> | |||
| <td>{{MLArgMinMaxOptions/outputDataType}}</td> | |||
| <td>{{input}}'s [=MLOperand/rank=] - 1 to {{input}}'s [=MLOperand/rank=]</td> | |||
| <td>{{MLOperandDataType/"int32"}}, {{MLOperandDataType/"int64"}}</td> | |||
There was a problem hiding this comment.
Note: This should list all allowed data types. The original outputDatatype is more useful for output data type calculation in the algorithm steps.
| @@ -3421,7 +3421,7 @@ partial dictionary MLOpSupportLimits { | |||
| <tr> | |||
| <td>*output*</td> | |||
| <td>[=/same type as|same as=] {{a}}</td> | |||
| <td>maximum of {{a}}'s [=MLOperand/rank=] and {{b}}'s [=MLOperand/rank=]</td> | |||
| <td>[=/any rank|N=]</td> | |||
There was a problem hiding this comment.
Note: Same here and the following, the allowed ranks, not the output rank calculation.
1. Add allowed ranks for all lstm and gru output tensors. 2. Validate new shape's rank of expand and reshape operations. 3. Simplify allowed ranks for some operations based on implementation experience. 4. Other minor changes for consistency.
huningxin
force-pushed
the
output_tensor_rank
branch
from
9e651d8 to
356febd
Compare
Move the output tensor rank range info from constraints table to Returns section for argMin/Max and reduction ops. The element-wise binary and logical ops, matmul and where ops already have such info in the Returns section. Fix issues of output rank description of prelu and scatterND.
reillyeon
approved these changes
Sep 16, 2025
chromium-wpt-export-bot
pushed a commit
to web-platform-tests/wpt
that referenced
this pull request
Sep 17, 2025
This CL prototypes WebNN spec change [1] that supports tensor rank range for graph input, constant, output and each operation's output. This CL adds new shape rank validation according to allowed output tensor rank of expand and reshape operations. The sequence output of gru and cell operations has different rank, this CL extends context properties to support them. For logical ops, the output rank ranges are set to each op's input rank ranges, while the output data types are set to unified logical output data type of context properties. [1]: webmachinelearning/webnn#857 Bug: 442209350 Change-Id: Ie5ca5c794cebb4586bcccaaa8d237be1dffea458
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.
This PR also supports rankRange for graph input, constant and output in opSupportLimits.
Fix #835
Preview | Diff