🐛 Fix fdiv/fsqrt directed rounding in the legacy divsqrt unit#28
Open
emiliengnr wants to merge 1 commit into
Open
🐛 Fix fdiv/fsqrt directed rounding in the legacy divsqrt unit#28emiliengnr wants to merge 1 commit into
emiliengnr wants to merge 1 commit into
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Hello, here is a pull request for a bug I found.
fpnew_divsqrt_multi drives the legacy PULP div_sqrt_top_mvp unit and passes the
instruction's rounding mode straight to its RM_SI port, but the two use different
encodings. Round-to-nearest-even and round-toward-zero agree, but round-down and
round-up are swapped (RM_SI 2 is toward +inf, 3 is toward -inf), and
round-to-nearest-ties-to-max has no case and falls through to truncate. So fdiv
and fsqrt round one ulp the wrong way under the directed modes, for single and
double precision.
The fix translates the mode into the divsqrt encoding before driving RM_SI
(round-down to 3, round-up to 2, ties-to-max to nearest since the legacy unit has
no ties-to-max mode). Round-to-nearest-even and round-toward-zero are unchanged.
Note: this fixes the round-down / round-up swap at the cvfpu wrapper by translating the mode before RM_SI. pulp-platform/fpu_div_sqrt_mvp/pull/20 fixes the same swap at its source (and also the overflow case). Only one should land, since applying both swaps the modes twice and brings the bug back.