Closed bend-n closed 1 month ago
Thanks!
This should use adjust_nan
to avoid making any guarantees about the NaNs being returned.
Ideally, we'd also not return the exact result but perturb it a bit, to drive home the point that these operations are non-deterministic. But the existing fast-math intrinsics don't do that either so we can leave that to https://github.com/rust-lang/miri/issues/3555.
This should use
adjust_nan
to avoid making any guarantees about the NaNs being
the comment on f*_fast
says that wrapping_binary_op
already calls generate_nan
?
the comment on f*_fast says that wrapping_binary_op already calls generate_nan?
Ah, you're right, thanks!
Thanks, looks good. :)
Please squash the commits.
@bors r+
:pushpin: Commit 18379e8e70c6ab5d79bb19d0b37c2a1d964928d6 has been approved by RalfJung
It is now in the queue for this repository.
:hourglass: Testing commit 18379e8e70c6ab5d79bb19d0b37c2a1d964928d6 with merge 24fc363cbcb04e686ff399b9a96babb99e50b45e...
:sunny: Test successful - checks-actions Approved by: RalfJung Pushing 24fc363cbcb04e686ff399b9a96babb99e50b45e to master...
supports the
f*_algebraic
intrinsics.