[SYCL][libdevice] return qNaN for (float/double)sNaN to bfloat16 conversion - #23084
Merged
Merged
Conversation
…ersion intel#22114 introduced a fix to return NaN for some specific float or double sNaN bit patterns when converting to bfloat16. Together with that change it also changed the behavior and added a test specifically for the case of converting sNaN to bfloat16. The new behaviour in that PR is to return sNaN for all sNaN inputs. This is not the correct behaviour, IEEE 754 specifies conversions should signal and then return a quiet NaN. We do not implement raising exceptions in the device library, so we cannot signal, but we should still return a quiet NaN for all sNaN inputs. The new behaviour matches the host conversion (from glibc/compiler-rt), see this example on compiler explorer: https://godbolt.org/z/E7ezaPxEd (assisted by Claude Opus 5 <noreply@anthropic.com>).
jinge90
approved these changes
Sep 3, 2026
KseniyaTikhomirova
approved these changes
Sep 3, 2026
Contributor
Author
|
I believe the L0 Windows Arc GPU failure is unrelated, I see it on other workflow runs too. Reported to: #23100 |
Contributor
Author
|
@intel/llvm-gatekeepers I believe this can be merged :) |
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
4 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
#22114 introduced a fix to return NaN for some specific float or double sNaN bit patterns when converting to bfloat16.
Together with that change it also changed the behavior and added a test specifically for the case of converting sNaN to bfloat16. The new behaviour in that PR is to return sNaN for all sNaN inputs.
This is not the correct behaviour, IEEE 754 specifies conversions should signal and then return a quiet NaN. We do not implement raising exceptions in the device library, so we cannot signal, but we should still return a quiet NaN for all sNaN inputs.
The new behaviour matches the host conversion (from glibc/compiler-rt), see this example on compiler explorer: https://godbolt.org/z/E7ezaPxEd (assisted by Claude Opus 5 noreply@anthropic.com).