feat: GF-T trainer size optimizations (scale_q + signmul) - #1839
Merged
Conversation
gft_xorpercep3.t27: fully-on-chip 2-layer XOR trainer with two reusable,
numerically-identical size optimizations (in-spec test PASS):
- scale_q(x,k) = x*2^-k via exponent-offset shift, replaces smul(eta,.) for a
power-of-2 eta (no multiplier)
- signmul(g,h) = sign/zero mux, valid because the perceptron error g is exactly
{-1,0,+1}, replaces smul(g,.)
Shrinks the design 19.5M -> 17.86M fasm (6 magmuls -> 2). Honest note: still
above the ~17M correctness ceiling -- the bulk is magsub (normalize loop in
every sadd), not the magmuls, so multiplier optimizations do not clear the
ceiling; a full 2-layer train step is irreducibly ~17.9M. Working on-chip path
stays the split. scale_q/signmul are reusable for any GF-T trainer near budget.
Refs #1764
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Contributor
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
New spec gft_xorpercep3.t27 — fully-on-chip 2-layer XOR trainer with two reusable, numerically-identical optimizations (in-spec PASS): scale_q(x,k)=x*2^-k via exponent shift (replaces smul(eta,.) for power-of-2 eta); signmul(g,h)=sign/zero mux (valid since perceptron error g is exactly {-1,0,+1}, replaces smul(g,.)). Shrinks 19.5M->17.86M fasm (6 magmuls->2). Honest note: still above the ~17M correctness ceiling — the bulk is magsub (normalize in every sadd), not magmuls; a full 2-layer train step is irreducibly ~17.9M. Working on-chip path stays the split. docs/NOW.md updated. Refs #1764