simd: fuse wrapping i32 add into masked sum

[AdaWorldAPI]

Adds one semantics-preserving reduction primitive for expression folding: masked_sum_wrapping_add_i32(a, b, mask). For each selected row it performs i32::wrapping_add first, then widens the wrapped row value to i64 and accumulates. This is deliberately not masked_sum_i32(a)+masked_sum_i32(b), which changes semantics on row overflow.

Why this belongs in ndarray: lance-graph's mask-risc law routes all row/mask arithmetic through ndarray::simd; the new kernel erases a derived i32 lane without materialising it and without creating a consumer-side hand-rolled loop. Like masked_sum_i32, the implementation walks selected bits and is intentionally scalar because exact widening semantics matter more than an I32x16::reduce_sum that accumulates in i32.

Coverage: facade re-export plus the existing cross-realization masking parity group, with the reference explicitly computing vals[i].wrapping_add(rhs[i]) before widening. Public docs include the overflow boundary example i32::MAX + 1 -> i32::MIN.

Consumer follow-up: lance-graph mask-risc terminal/lowering for fused SUM(a+b) after this lands.

Summary by CodeRabbit

• New Features

  • Added a masked integer reduction that combines paired 32-bit values with wrapping arithmetic and returns a 64-bit total.
  • The operation is now available through the public SIMD interface.
  • Mask selections beyond the available input rows are ignored.

• Bug Fixes

  • Added validation to ensure masked reduction results match the selected-element reference calculation, including wrapping overflow behavior.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Essentials

Run ID: ed339002-cb9b-4753-b1ac-a3abffe5d704

📥 Commits

Reviewing files that changed from the base of the PR and between ecb2f2c and 932729e.

📒 Files selected for processing (3)

• crates/simd-masking-parity/src/lib.rs
• src/simd.rs
• src/simd_masking_ops.rs

Included review availability: 1 review is currently available. Your included PR review attempts over the past 7 days set your current allowance at 2 reviews per hour.

---

📝 Walkthrough

Walkthrough

The change adds masked_sum_wrapping_add_i32, exports it through simd, and validates its masked pairwise wrapping behavior in the SIMD parity checks.

Changes

Masked wrapping reduction

Layer / File(s)	Summary
Implement and export the reduction src/simd_masking_ops.rs, src/simd.rs	Adds masked_sum_wrapping_add_i32. It validates input lengths and mask size, performs wrapping i32 pairwise addition, widens results to i64, and exposes the function through the public SIMD facade.
Validate masked pairwise sums crates/simd-masking-parity/src/lib.rs	Extends check_masked_reductions with a second operand vector and compares each masked reduction with a wrapping reference accumulation.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Feature

Suggested reviewers: claude

Merge Risk: ⚪ Minimal · up to 93272

The new masked wrapping reduction follows the repository’s slice-level API contract and preserves the documented wrapping and masking behavior. No actionable merge-blocking risk remains.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 3 functions across 3 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: adding a fused wrapping i32 addition to the masked sum operation.

• Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Commit to this branch
• Create a new PR

---

_{A rabbit checks each masked sum,
With wrapping hops where values run.
Two vectors meet, then widen wide,
While parity guards stand by their side.
The SIMD trail now passes bright.}

_{Comment @coderabbitai help to get the list of available commands.}

[AdaWorldAPI]

@coderabbitai review

Please attack the semantic boundary rather than style:

1. masked_sum_wrapping_add_i32 must compute i32::wrapping_add per selected row before widening to i64; any implementation equivalent to independent widened sums is wrong on overflow.
2. The new function must preserve masked_sum_i32's mask-tail and sparse-set-bit behavior, with no read past either input lane.
3. The parity reference must be independent enough to catch an accidental widen-before-add implementation, not merely repeat the function under test.

This is intentionally scalar inside the masking facade: the existing masked_sum_i32 rationale against I32x16::reduce_sum applies here too.

[coderabbitai]

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo '=== Changed files ==='
git diff --name-status HEAD^ HEAD

echo '=== Pull request diff ==='
git diff --unified=80 HEAD^ HEAD

echo '=== Candidate masking symbols ==='
rg -n -C 8 --glob '*.rs' 'masked_sum_(wrapping_add_i32|i32)\s*\(' . || true

Length of output: 42363

---

@AdaWorldAPI Reviewed the requested semantic boundary.

• masked_sum_wrapping_add_i32 performs a[i].wrapping_add(b[i]) before the i64 cast. It does not implement independent widened sums.
• It retains masked_sum_i32 behavior for required mask coverage, sparse set-bit iteration, and partial final-word masking. The clamp prevents a dirty tail bit from indexing past either input slice.
• The parity reference is separate from the kernel. It folds selected rows with vals[i].wrapping_add(rhs[i]) as i64.
• The parity corpus includes overflow-capable i32::MIN and i32::MAX rows. It also covers lengths 0, 1, 63, 64, 65, and 130 with empty, full, random, and sparse masks.

I found no defect in these semantic boundaries.

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_dd39f13c-1849-4e7f-beb0-5f18e35964e8)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_6766b071-6bed-42fb-821b-084c349aba74)