Cortex-M: lower transcendental unary operators through the activation LUT - #22155
Merged
Conversation
… LUT cortex_m::quantized_activation is a 256-entry int8 table with the qparams folded in. Only sigmoid, tanh, silu and gelu were registered; this adds log, log2, log10, log1p, sqrt and rsqrt, functional and in-place. Not exp: its codomain is unbounded, so an output scale covering the calibrated maximum leaves almost no resolution elsewhere -- over [-8, 8] the table holds two distinct levels across x in [-2, 2]. Poles saturate, and the undefined region beyond a pole continues the boundary value, so log below zero reads -128 and rsqrt below zero reads 127. Emitting the output zero point there would sit a mid-range value below the rail and break monotonicity. That case is reachable rather than theoretical: a shared quantization spec can widen an operand's grid across zero even where the tensor never goes. The tables are evaluated through torch rather than math, which raises where these functions are undefined instead of returning the -inf or nan the table needs. Silero VAD's magnitude sqrt now lowers, so its expected counts move with this. Authored with Claude Code.
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22155
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 3dd9bb3 with merge base baafd7e ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
AdrianLundell
approved these changes
Aug 26, 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
2 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.
Summary
cortex_m::quantized_activation is a 256-entry int8 table with the qparams folded in. Only sigmoid, tanh, silu and gelu were registered; this adds log, log2, log10, log1p, sqrt and rsqrt, functional and in-place.
Not exp: its codomain is unbounded, so an output scale covering the calibrated maximum leaves almost no resolution elsewhere -- over [-8, 8] the table holds two distinct levels across x in [-2, 2].
Poles saturate, and the undefined region beyond a pole continues the boundary value, so log below zero reads -128 and rsqrt below zero reads 127. Emitting the output zero point there would sit a mid-range value below the rail and break monotonicity. That case is reachable rather than theoretical: a shared quantization spec can widen an operand's grid across zero even where the tensor never goes.
The tables are evaluated through torch rather than math, which raises where these functions are undefined instead of returning the -inf or nan the table needs.
Silero VAD's magnitude sqrt now lowers, so its expected counts move with this.
Authored with Claude Code.