Uh oh!
There was an error while loading. Please reload this page.
This repository was archived by the owner on Aug 20, 2026. It is now read-only.
fix: give the field regulariser _jrls kernels a real per-channel weight stride - #70
Merged
Merged
Conversation
…ht stride
matvec/diag for absolute, membrane, and bending in the JRLS family
(regularisers/field/{1,2,3}d.h) read the weight value once, outside any
per-channel loop (`w = *wgt`), and applied that single value to every
channel. They had no per-channel stride parameter at all, unlike their
_rls siblings whose `wgt[wsc*c]`-style indexing (fixed for the wc=1
broadcast case by #62/kernels#48) is exactly the per-channel read the
_jrls family was missing.
Give each _jrls kernel a `wsc` parameter and move the weight load inside
the per-channel loop, mirroring the corresponding _rls kernel's structure
exactly (including the neighbour-guard `alt` fallback for membrane/bending
at Zero/DST2 boundaries). relax_*_jrls_ is unaffected here since it's
built on matvec/diag at the driver layer.
part of fastfields-cpu-lib#65
_Workstream: claude-jitfields-to-fastfields_
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016AjQcY78NgbagPSbPJRr6ZUh oh!
There was an error while loading. Please reload this page.
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 freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
matvec/diagforabsolute/membrane/bendingin the JRLS family(
regularisers/field/{1,2,3}d.h) read the weight value once, outsideany per-channel loop (
w = *wgt), and applied that single value toevery channel. They had no per-channel stride parameter at all —
unlike their
_rlssiblings, whosewgt[wsc*c]-style indexing (fixed forthe
wc=1broadcast case by fastfields-kernels#48 / fastfields-cpu-lib#62)is exactly the per-channel read the
_jrlsfamily was missing.Fix
Give each
_jrlskernel awscparameter and move the weight load insidethe per-channel loop, mirroring the corresponding
_rlskernel'sstructure exactly — including the neighbour-guard
altfallback formembrane/bending at Zero/DST2 boundaries (from #52).
relax_*_jrls_isunaffected here since it's built on
matvec/diagat the driver layer,so fixing those two propagates automatically.
Spans 6 kernel functions (
matvec/diag×absolute/membrane/bending)× 3 dimensionality files (
1d.h/2d.h/3d.h) = 18 functions.Verification
Header-only — no build of its own. Downstream driver changes (the actual
call sites that pass
wsc) land in afastfields-cpu-implPR that bumpsthis branch as its
kernelspin; the ground-truth regression test and thefull
make -C fastfields-cpu-lib test CXX=clang++run (12/12 suites,44988 checks, 0 failures) live in the
fastfields-cpu-libPR that closesthe tracking issue.
part of fastfields/fastfields-cpu-lib#65
Workstream: claude-jitfields-to-fastfields
🤖 Generated with Claude Code
Generated by Claude Code