Repository navigation
Implement direct signed carry streams for BIT high routes - #3343
Merged
Mister-Meeseeks merged 1 commit intoOct 18, 2026
Merged
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Change
PR2915 expands a negative high-field carry into repeated sign/shift/sign calls. This adds an opt-in encoded stream rotation with complete prefix/suffix preservation, distinct output ownership and erased private buckets. The +1 path stays with the existing implementation after a tiny-packet comparison exposed a setup regression there.
The −4 kernel fixture (486 bits) uses 54,064 physical tape operations instead of 1,105,367, a 95.1% reduction. A complete signed-scale route uses 410,237 instead of 881,379, a 53.5% reduction. Three alternating fresh-process Python 3.12 benchmark pairs give median instrumented execution speedups of 19.68× and 2.13× respectively. Setup, rewinds, cleanup and native intervals are measured; these are reference fixtures, not multiplication timings.
Scope and evidence
The written source argument gives
O(BQS + I)complete local work, eight additional scratch roles and zero numerical words for the direct kernel. The opt-in adapter preserves the predecessor execute lifecycle except for carry dispatch. Frozen sources, the default adapter, recursive native calls, selected record and κ are unchanged.Verification covers 96 complete before/after kernel pairs, seven complete route pairs, 2,056 independent modular split identities, full input-column provenance, borrowed/dirty state, source corruption, extra input, wrong direction on all-zero payloads, one-role-short allocation and endpoint-preserving extra traffic. The selected registry census covers all 1,123 chart types and 2,246 orientations at q=3 and identity low class: 34,450,048 old full-field primitive calls become 11,858,368 kernel calls. This specialization is not a runtime ratio or a complete GL64 payload execution.
Base:
6f100d4bb5083f8c35d96fd95a539811c62f245c. Executable/archive source pins and complete typed receipts are inresearch/high-carry-streams/. Reproduce with:Expected: all four
PASS_HIGH_CARRY_*suites and successful registered replay.VALIDATION.mdrecords the committed replay, source-export checks and benchmark environment. Whole-caller backing/stock, native refinement, formal verification and supplier admission remain separate obligations. Concurrent #3340's fitted-B1 bridge is complementary and unused.Credit
Jose Manuel / jmbarrancoml supplied #2915's executor, kernels and registry and #2899's backend; David Arias supplied #2753. The physical observer follows #3330. Retain Nick Petalas, commdata2338, Sagar Thakrar, Arham Tahir, ywz65, Ross Richardson and Krzysztof Wójt's cost and caller work and the #2157/#589/#555 lineage. Tekalign (
tekuuu), with substantial OpenAI Codex assistance. New material is Apache-2.0; no independent human review is claimed.