fix qlinear nan bug#4914
Draft
aarushjain29 wants to merge 7 commits into
Draft
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adjusts MIGraphX’s quantizelinear operator to handle NaN inputs deterministically during quantization, aligning behavior with common ONNX runtime expectations (notably ONNX Runtime CPU EP).
Changes:
- Add explicit NaN handling for integral quantization outputs: NaN inputs now saturate to the quantized type’s minimum value.
- Prevent NaN from flowing through the clamp logic (
std::min/std::max) where it would otherwise silently select the max bound.
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files
🚀 New features to boost your workflow:
|
Regressions detected 🔴 |
|
| output[i] = min_value; | ||
| return; | ||
| } | ||
| auto rounding_mode = fegetround(); |
Collaborator
There was a problem hiding this comment.
But this and the next line out of the visit_all since this does an assignment for each and not needed.
| { | ||
| output[i] = min_value; | ||
| return; | ||
| } |
Collaborator
There was a problem hiding this comment.
If this is consistent with GPU kernels, I think you just need to add this If block and thats it.
| } | ||
| auto rounding_mode = fegetround(); | ||
| fesetround(FE_TONEAREST); | ||
| auto rounded = std::nearbyint(input[i] / scales[i]); |
Collaborator
There was a problem hiding this comment.
Remove this, and just do the nearbyint in the static cast for quantized below.
| auto rounding_mode = fegetround(); | ||
| fesetround(FE_TONEAREST); | ||
| auto rounded = std::nearbyint(input[i] / scales[i]); | ||
| fesetround(rounding_mode); |
Collaborator
There was a problem hiding this comment.
Why are we changing rounding mode again? Is this because we're assigning festround() above per item?
TedThemistokleous
requested changes
May 28, 2026
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
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.
Motivation
ONNX Runtime's CPUExecutionProvider and MIGraphX disagreed on QuantizeLinear output when the input is NaN:
input: [nan]
scale: 0.5
onnxruntime out: [-128]
migraphx out: [ 127] ← MIGraphX
Technical Details
Changelog Category
Add a
CHANGELOG.mdentry for any option other thanNot Applicable