docs: final single-thread numba-vs-rust A/B — gate passed (Phase 5 W4)#259
Merged
Conversation
…sized plan) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…FFI crossing (Phase 5 W3) Add reconstruct_annotated_haplotypes_spliced_fused — the annotated counterpart of reconstruct_haplotypes_spliced_fused. Folds RC in-kernel (bytes RC'd, annotation rows reversed) so the Python _FlatAnnotatedHaps.reverse_masked post-pass is dropped on the rust backend. Byte-identical to the composed numba oracle (new parity backstop). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…combos now single-FFI (Phase 5 W3) Also applies ruff formatting to _haps.py (post-Task-1 residual). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… passed Rust parity-or-better single-thread on every __getitem__ mode (same-session, two tools, two passes): haps/tracks-seqs ~1.65x, annotated/variants ~1.4x, variant-windows ~4.6x, pure tracks-only ~1.05x (fixed-cost-bound, parity). Combined with byte-identical parity (W1-W3 + full suite), no regression risk in removing numba. Gate passed -> proceed to W5 consolidation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Phase 5 W4 — final single-thread
__getitem__A/B (benchmark-only, no code)The migration's final single-thread parity gate before the W5 consolidation (numba deletion + rayon). Gate: rust at parity-or-better single-thread → proceed to consolidation.
Methodology
The shared Carter node makes cross-session wall-clock unreliable, so rust and numba were measured single-thread, in the same back-to-back session, two passes (within-session stable), with the ratio direction pinned (
speedup = numba÷rust, higher ⇒ rust faster). The durable signal is byte-identical parity (already gated across W1–W3 + the full parity suite) plus same-session improve-or-hold. Two independent tools agreed:test_e2e.pypedantic-min andprofile.pysteady-state throughput.Result — GATE PASSED
Rust is parity-or-better on every mode:
Combined with byte-identical parity, there is no single-thread regression risk in removing numba. → Proceed to W5 (golden-snapshot the numba-oracle parity suites, delete numba, add rayon batch parallelism gated byte-identical to the serial golden result).
Full tables + methodology:
docs/roadmaps/phase-5-w4-final-ab.md.Measured on
phase-5-w4≡rust-migration+ W3 fusion (#258); W2 (#257) is test-only and perf-neutral. numba remains present until W5. Phase 5 🚧 (W1–W4 done; W5–W9 remain).🤖 Generated with Claude Code