[tpn]

Superseded by #12124, which now owns this implementation within its complete feature family. The original commits, branch and validation evidence remain available. This draft is closed unmerged.

---

Historical scope and validation below refer to the original source heads.

Block kernels can read neighboring values with coop::shuffle(block, &values, options). It shifts blocked-order values by one item, preserves the input, and returns fresh storage with the same type, extent and alignment. Down is the default; Up reads the previous item.

let next = coop::shuffle(block, &values, Default::default());
let previous = coop::shuffle(block, &values,
    coop::ShuffleOptions::default().mode(coop::Up));

The exposed boundary stays invalid: get returns None, and write initializes it before Store. All input slots must be initialized. Includes a generic Load → Shuffle → edge repair → Store example and guide.

Current validation: strict host/docs checks, 17 native requests, shape regression, 12 compiler controls (three accepted, nine rejected), two release builds and complete artifact inspection. GPU validation covers 3,840 example values, ten fixture cases, three clean sanitizers and four expected traps with immediate health checks. Current-head host CI passes (71 tests/doctests; nine native tests excluded), along with the three upstream checks. The draft bot status is not a source review. No current performance comparison is claimed.

Depends on #12124 and Oxide through #1589. Review the owning Shuffle diff; the main-base Files changed view includes earlier prerequisites.