Add nextafter op #2510
Open
Add nextafter op #2510
Conversation
This file contains 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
Test is currently segfaulting for scalar-only inputs.
jacobhinkle
commented
Feb 23, 2023
| if (config.require_full_precision_promoted) { | ||
| TORCH_CHECK( | ||
| common_dtype == c10::ScalarType::Float || | ||
| common_dtype == c10::ScalarType::Double, |
There was a problem hiding this comment.
I am not sure that this is the right place to perform this check.
There was a problem hiding this comment.
Hmm. I got this wrong too. PyTorch actually supports bfloat16, just not float16 for this op. It was added with a manual implementation taken from musl: https://github.com/pytorch/pytorch/pull/61829/files#diff-ece04c31934b3504382e10ed3e9a69f03ffabd81ad1a2a890aab19b1642f53c0R120
jacobhinkle
commented
Feb 23, 2023
jacobhinkle
added a commit
to NVIDIA/Fuser
that referenced
this pull request
Mar 30, 2023
The `nextafter(x, y)` operation provides the nearest distinct representable floating point value to `x` between `x` and `y`. In CUDA these are obtained with the builtins `nextafter` and `nextafterf`. Other types such as bfloat16 are not directly supported, though PyTorch has implemented that case based on some code in musl: https://github.com/pytorch/pytorch/pull/61829/files#diff-ece04c31934b3504382e10ed3e9a69f03ffabd81ad1a2a890aab19b1642f53c0R120. In PyTorch, the torch.nextafter function is defined for any pair of arguments which would normally be promoted to either float32 or float64. So arguments which are both ints, bools, complex, or half-precision floats are not supported. This PR implements a binary op macro with a TypePromotionConfig that enforces that rule. This is a translation/update of csarofeen/pytorch#2510. --------- Co-authored-by: Jacob Hinkle <jhinkle@nvidia.com>
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.
The
nextafter(x, y)operation provides the nearest distinct representable floating point value toxbetweenxandy. In CUDA these are obtained with the builtinsnextafterandnextafterf.In PyTorch, the
torch.nextafterfunction is defined for any pair of arguments which would normally be promoted to eitherfloat32orfloat64. So arguments which are both ints, bools, complex, or half-precision floats are not supported. This PR implements a binary op macro with aTypePromotionConfigthat enforces that rule.