Arm backend: Improve dtype validation #15871
Open
+185
−55
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
Improve dtype validaiton in NodeVisitors. Signed-off-by: Oscar Andersson <[email protected]> Change-Id: Ieb9ced1ae8d2db916e6c8bc0b45773a640d330db
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15871
Note: Links to docs will display an error until the docs builds have been completed. ❌ 4 New Failures, 3 Unrelated FailuresAs of commit 4399aa8 with merge base 9952aef ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
oscarandersson8218
added
partner: arm
Copilot finished reviewing on behalf of
mergennachin
November 18, 2025 15:27
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR improves dtype validation in the ARM backend's node visitors by making dtype support conditional on TOSA specifications and extensions, adding bool dtype support to several operators, and refining dtype lists for better consistency and correctness.
- Enhanced dtype validation to conditionally support INT16 and INT48 based on TOSA 1.0 "int16" extension availability
- Added BOOL dtype support to operators like permute, repeat, expand, slice, and cat
- Removed INT8/INT16 from comparison operators (eq, lt, le, gt, ge) to align with TOSA spec constraints
- Added test cases with bool tensors and xfail markers for U55 hardware which doesn't support bool
Reviewed Changes
Copilot reviewed 24 out of 24 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| backends/arm/test/ops/test_repeat.py | Added bool test case with xfail for U55 |
| backends/arm/test/ops/test_permute.py | Added bool test case with xfail for U55, renamed int16 quantization test functions for clarity (int16→16a8w) |
| backends/arm/test/ops/test_expand.py | Added bool test case with xfail for U55 |
| backends/arm/operators/ops_identity.py | Added conditional dtype validation based on TOSA spec and int16 extension support |
| backends/arm/operators/op_where.py | Deduplicated BOOL from supported dtypes list (moved to base list) |
| backends/arm/operators/op_tosa_transpose.py | Reordered dtype list for consistency (BOOL first, then INT types, then FP types) |
| backends/arm/operators/op_tosa_table.py | Refactored dtype validation to conditionally support INT16 input/INT32 output based on int16 extension |
| backends/arm/operators/op_tosa_resize.py | Improved dtype validation with conditional support for INT16/INT48 based on int16 extension |
| backends/arm/operators/op_tosa_matmul.py | Added conditional support for INT16 input and INT48 output based on int16 extension |
| backends/arm/operators/op_sum.py | Added explicit dtype validation for INT32 and FP32 |
| backends/arm/operators/op_slice.py | Added BOOL dtype support |
| backends/arm/operators/op_repeat.py | Added BOOL dtype support |
| backends/arm/operators/op_permute.py | Added BOOL dtype support |
| backends/arm/operators/op_mul.py | Added INT8 and INT16 dtype support |
| backends/arm/operators/op_max_pool2d.py | Added conditional INT16 support based on int16 extension |
| backends/arm/operators/op_lt.py | Removed INT8 and INT16 from supported input dtypes |
| backends/arm/operators/op_le.py | Removed INT8 and INT16 from supported input dtypes |
| backends/arm/operators/op_index_select.py | Added validation utilities and renamed unused variable from 'index' to '_' |
| backends/arm/operators/op_gt.py | Removed INT8 and INT16 from supported input dtypes |
| backends/arm/operators/op_ge.py | Removed INT8 and INT16 from supported input dtypes |
| backends/arm/operators/op_eq.py | Removed INT8 and INT16 from supported input dtypes |
| backends/arm/operators/op_clamp.py | Added conditional INT16 support based on int16 extension |
| backends/arm/operators/op_cat.py | Added BOOL dtype support and conditional INT16 support, improved validation with proper TosaArg conversion |
| backends/arm/operators/op_avg_pool2d.py | Added conditional INT16 support based on int16 extension |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| class Expand(torch.nn.Module): | ||
| # (input tensor, multiples) | ||
| test_parameters = { | ||
| "randbool_1d": lambda: (torch.randint(0, 1, (1,), dtype=torch.bool), (5,)), |
There was a problem hiding this comment.
The test case name "randbool_1d" uses torch.randint(0, 1, ...) which only generates 0s since randint excludes the upper bound. This should be torch.randint(0, 2, ...) to generate both 0 and 1 values for proper bool testing.
Suggested change
| "randbool_1d": lambda: (torch.randint(0, 1, (1,), dtype=torch.bool), (5,)), | |
| "randbool_1d": lambda: (torch.randint(0, 2, (1,), dtype=torch.bool), (5,)), |
Collaborator
|
This seem to get fails in the Arm test :( |
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.
Improve dtype validation in node-visitors.
cc @freddan80 @per @zingo @digantdesai