#67: bulk copies for the remaining flat list element types - #80
Merged
Conversation
list<u8> went bulk in #63; every other flat element type still moved one interpreted store()/load() per element (~13-45 ns/el of despecialize + asserts + DataView), and a stream<u32> paid that on every rendezvous. cabi/bulk_lists.ts: TypedArray-backed lift/store for bool, s8, u16/s16, u32/s32, u64/s64, f32/f64 — decision (a) from the issue: host shapes are UNCHANGED (number[]/bigint[]/boolean[]; no contract edit), only the copy mechanics change. Semantics parity with the per-element path is exact: * integers keep storeInt's assert texts ("int store", "64-bit store requires bigint"), assert-before-write ordering, and wrap-on-overflow (TypedArray element coercion == the matching DataView setter); * floats keep the deterministic profile's NaN canonicalization in BOTH directions (float.ts decodeI32AsFloat/encodeFloatAsI32 semantics: every lifted NaN is the canonical one, every stored number NaN writes the canonical bit pattern; non-NaN values are bit-exact, f32 narrowing is the same IEEE RTNE); * bool stores normalize by truthiness and lift nonzero as true, as before. The platform's little-endianness is a NAMED assumption (PLATFORM_LITTLE_ENDIAN, checked once): wasm memory is LE by spec, JS TypedArrays are platform-endian; on a BE platform (and for misaligned test-only views) the helpers decline and the DataView per-element loops remain as the endianness-correct fallback. u8 keeps its dedicated Uint8Array-shaped path; char stays per-element (its lift IS per-element USV validation); compound types fall through unchanged. Issue-checklist deviation, reviewed: "PayloadChunk generalization" is N/A under decision (a) — non-u8 stream chunks stay plain arrays and hit these fast paths inside GuestBuffer read/write anyway. 16 KiB lists, this box: store 15-22 -> 1.5-2.9 ns/el (10-14x); lift u32 17 -> 1.4 ns/el, f64 23 -> 3.3, u64 23 -> 8.7 (bigint allocation bound). The conformance values directory (165 cases over every list kind) passes unchanged through the new paths. Also closes #63 review F3: stream-pass grows `take` (bounded consumption, returns the sum of what it took), and a new test drives a 256 KiB typed offer against it — the writeAll re-offer rounds against a real guest's GuestBuffer, with data integrity pinned exactly and the take count only bounded (wit-bindgen buffers reads). Tests: bulk_list_test.ts pins per-type round-trips, wrap parity, assert-text parity (incl. Infinity/NaN), bigint wrap mod 2^64, f32 narrowing, canonical-NaN bit patterns both directions (f32 + f64 payload NaNs), bool normalization, empty/fixed-length lists, the misaligned-view fallback, and that the fast path actually engages on LE platforms. docs/architecture.md §7 records the policy. Gates: build, test-rust, test-runtime (380), test-wasi-shims (50), test-ct-runner, test-bundle, test-translate, examples, conformance (1254/0, no stale xfails), sched-seeds (1, 4242), test-ports, test-webrtc, shells (sm-pinned; jsc-pinned arch-skipped), browsers (chromium, firefox), websocket-conformance (55/0), smoke-tls, smoke-c0 legs 2+4. Reviewed against definitions.py (canonicalize_nan32/64, DETERMINISTIC_PROFILE, store/load discipline) + float.ts/memory.ts; advisories addressed in-diff (bytesOf defense-in-depth on the bool paths, extra assert cases, f64 payload-NaN twin, fast-path-engages smoke). Closes #67.
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.
Closes #67. Companion PR: the #68 bench lanes (separate, bench-only).
Decision (a) from the issue: host shapes are unchanged (
number[]/bigint[]/boolean[]— no contract edit); only the copy mechanics change.cabi/bulk_lists.tsadds TypedArray-backed lift/store for bool, s8, u16/s16, u32/s32, u64/s64, f32/f64, with exact per-element semantics parity:"int store","64-bit store requires bigint"), same assert-before-write order, same wrap-on-overflow (TypedArray coercion ≡ DataView setters);decodeI32AsFloat/encodeFloatAsI32semantics; canonical bit patterns pinned by test against absolute bytes, including payload-NaN lift→re-store round trips for f32 and f64);Endianness is a named assumption (
PLATFORM_LITTLE_ENDIAN, one check): on a BE platform the helpers decline and the DataView per-element loops remain as the correct fallback (also exercised by a misaligned-view test). u8 keeps itsUint8Array-shaped path from #63;charstays per-element (USV validation is the cost); compound types fall through.Checklist deviation, reviewed as sound: "PayloadChunk generalization" is N/A under decision (a) — non-u8 stream chunks stay plain arrays and hit these paths inside
GuestBufferread/write anyway.Numbers (16 KiB lists, this box): store 15–22 → 1.5–2.9 ns/el (10–14×); lift u32 17 → 1.4, f64 23 → 3.3, u64 23 → 8.7 ns/el (bigint-allocation bound).
Also closes #63 review F3:
stream-passgrowstake(bounded consumption returning the sum of what it took), driven by a 256 KiB typed offer — thewriteAllre-offer rounds against a real guest, data integrity pinned exactly, take count bounded (wit-bindgen buffers reads).Gates: test-runtime (380), conformance (1254/0 — the values directory exercises every list kind through the new paths), sched-seeds, wasi-shims (50), ct-runner, bundle, translate, examples, ports, webrtc, smoke-tls, websocket-conformance (55/0), smoke-c0 legs 2+4, shells, browsers, build/test-rust. Reviewed against definitions.py (
canonicalize_nan32/64,DETERMINISTIC_PROFILE) + float.ts/memory.ts; all advisories addressed in-diff.