Skip to content

docs(audit): T7-4 — quarterly upstream-backlog re-audit (2026-04-29)#205

Merged
lusoris merged 1 commit intomasterfrom
docs/t7-4-quarterly-upstream-audit-2026-04-29
Apr 29, 2026
Merged

docs(audit): T7-4 — quarterly upstream-backlog re-audit (2026-04-29)#205
lusoris merged 1 commit intomasterfrom
docs/t7-4-quarterly-upstream-audit-2026-04-29

Conversation

@lusoris
Copy link
Copy Markdown
Owner

@lusoris lusoris commented Apr 29, 2026 •
edited
Loading

Summary

T7-4 quarterly re-audit of the Netflix/vmaf upstream-backlog. Walks the
12 upstream commits landed since the fork's last chore(upstream): port boundary (PR #181, 798409e3 / 314db130).

Test plan

  • git fetch upstream master succeeded.
  • git log upstream/master --oneline 798409e3^..upstream/master
    returned 12 commits matching the audit table.
  • Per-commit fork cross-check via
    git log master --oneline --grep=... and grep -rn for marker
    symbols (vif_sigma_nsq, motion_v2,
    VMAF_FEATURE_EXTRACTOR_PREV_REF, etc.).
  • Pre-commit hooks pass on touched files (CHANGELOG.md,
    docs/upstream-backlog-audit-2026-04-29.md).

Deep-dive deliverables (ADR-0108)

  • no research digest needed: this PR is itself the quarterly audit doc (docs/upstream-backlog-audit-2026-04-29.md); not a research-NNNN-shaped fresh investigation.
  • no alternatives: scheduled audit, not a non-trivial decision per CLAUDE §12 r8.
  • no rebase-sensitive AGENTS invariants in this PR.
  • Reproducer / smoke-test command — see Test plan.
  • CHANGELOG.md entry — Unreleased § Changed (audit refresh).
  • no rebase note needed: doc-only, no upstream-merge surface.

🤖 Generated with Claude Code

@claude
docs(audit): T7-4 — quarterly upstream-backlog re-audit (2026-04-29)
2f16fc6 
Walk the 12 upstream Netflix/vmaf commits landed since the fork's last
chore(upstream): port boundary (PR #181, 798409e / 314db13). 8 are
already on fork; 4 are flagged for fork action with 4 recommended new
T-rows. No code changes; no docs/state.md changes (no upstream commit
ruled in/out a fork bug). Doc doubles as the ADR-0108 research digest.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced Apr 29, 2026
Merged
Merged
Merged
@lusoris lusoris merged commit 10a71ac into master Apr 29, 2026
52 of 53 checks passed
@lusoris lusoris deleted the docs/t7-4-quarterly-upstream-audit-2026-04-29 branch April 29, 2026 13:16
@github-actions github-actions Bot mentioned this pull request Apr 29, 2026
Open
@lusoris lusoris mentioned this pull request Apr 29, 2026
Closed
4 tasks
lusoris pushed a commit that referenced this pull request Apr 30, 2026
@claude
feat(upstream): port c70debb — adm+vif test deltas
29a3b12 
Cherry-pick the additive halves of Netflix upstream commit c70debb
("libvmaf/test: port new adm/vif/speed tests", Kyle Swanson,
2026-04-28). Adds:

- libvmaf/src/feature/adm_csf_tools.h (verbatim, declares the inline
  adm_native_csf DLM-CSF helper).
- libvmaf/test/test_adm_csf.c (verbatim modulo `static` on
  almost_equal + clang-format reflow; 2 mu_assert cases).
- libvmaf/test/test_barten_csf.c (verbatim modulo `static` on
  almost_equal + clang-format reflow; 23 mu_assert cases over
  barten_rod_cone_sens / barten_mtf / barten_csf /
  linear_interpolate / barten_watson_blend_csf).
- libvmaf/test/meson.build registers the two executables and
  test() entries.

The other two halves of c70debb are deliberately deferred:

- test_vif_tools.c depends on the upstream `vif` runtime-helper
  chain (NUM_KERNELSCALES, vif_validate_kernelscale,
  vif_get_filter, speed_get_antialias_filter, a [_][5][65] filter
  table) that the fork skips per Research-0024 Strategy E to
  protect the ADR-0138 / 0139 / 0142 / 0143 SIMD bit-exactness
  contract.
- test_speed_chroma.c #includes feature/speed.c, which the fork
  does not yet ship (audit row T-NEW-1).

Tracked alongside the 2026-04-29 quarterly upstream-backlog
re-audit (PR #205); resolves audit row T-NEW-2.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
lusoris added a commit that referenced this pull request Apr 30, 2026
@lusoris @claude
feat(upstream): port c70debb — adm+vif test deltas (#211)
e66b510 
* feat(upstream): port c70debb — adm+vif test deltas

Cherry-pick the additive halves of Netflix upstream commit c70debb
("libvmaf/test: port new adm/vif/speed tests", Kyle Swanson,
2026-04-28). Adds:

- libvmaf/src/feature/adm_csf_tools.h (verbatim, declares the inline
  adm_native_csf DLM-CSF helper).
- libvmaf/test/test_adm_csf.c (verbatim modulo `static` on
  almost_equal + clang-format reflow; 2 mu_assert cases).
- libvmaf/test/test_barten_csf.c (verbatim modulo `static` on
  almost_equal + clang-format reflow; 23 mu_assert cases over
  barten_rod_cone_sens / barten_mtf / barten_csf /
  linear_interpolate / barten_watson_blend_csf).
- libvmaf/test/meson.build registers the two executables and
  test() entries.

The other two halves of c70debb are deliberately deferred:

- test_vif_tools.c depends on the upstream `vif` runtime-helper
  chain (NUM_KERNELSCALES, vif_validate_kernelscale,
  vif_get_filter, speed_get_antialias_filter, a [_][5][65] filter
  table) that the fork skips per Research-0024 Strategy E to
  protect the ADR-0138 / 0139 / 0142 / 0143 SIMD bit-exactness
  contract.
- test_speed_chroma.c #includes feature/speed.c, which the fork
  does not yet ship (audit row T-NEW-1).

Tracked alongside the 2026-04-29 quarterly upstream-backlog
re-audit (PR #205); resolves audit row T-NEW-2.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(adm): make M_PI portable for MinGW + MSVC (T-NEW-2 / PR #211 follow-up)

MinGW64's <math.h> does not expose M_PI by default, causing
adm_csf_tools.h:36 to fail with "'M_PI' undeclared" on the Windows CI
job. Define _USE_MATH_DEFINES before <math.h> (so MSVC and conforming
MinGW configurations expose the constant) and add a literal fallback
after the include so any non-conforming libm still compiles.

Linux/glibc was already fine; verified test_adm_csf still passes.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Lusoris <lusoris@pm.me>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
lusoris pushed a commit that referenced this pull request Apr 30, 2026
docs(upstream): T-NEW-3 — CHANGELOG + rebase-notes for 32-bit ADM/cpu…
ec661aa 
… port

Adds the lusoris-fork CHANGELOG entry and rebase-notes ledger row
0075 covering the verbatim port of upstream Netflix/vmaf
8a28970 + 1b6c388 (32-bit ADM/cpu fallbacks).

Pairs with the two preceding commits in this branch
(feat(x86/cpu): port 1b6c388, feat(adm): port 8a28970).
Closes audit row T-NEW-3 from PR #205
(docs/upstream-backlog-audit-2026-04-29.md).
lusoris added a commit that referenced this pull request Apr 30, 2026
@lusoris @1480c1 @claude
feat(upstream): port 8a28970 + 1b6c388 — 32-bit ADM/cpu fallbacks (#212)
968cb3f 
* feat(x86/cpu): port upstream 1b6c388 — remove avx+ 32-bit limit

Cherry-pick of Netflix/vmaf 1b6c388 by
Christopher Degawa. Drops the `#if ARCH_X86_64` guard around the
AVX/AVX2/AVX-512 detection block in `vmaf_get_cpu_flags_x86`, so 32-bit
i686 builds can now report and dispatch >128-bit SIMD when the CPU and
kernel support it. Pairs with 8a28970 which adds the 32-bit
`extract_epi64` fallback the AVX2/AVX-512 ADM paths need.

Original-commit: 1b6c388
Original-author: Christopher Degawa <ccom@randomderp.com>

* feat(adm): port upstream 8a28970 — fallback for extract_epi64 on 32-bit

Cherry-pick of Netflix/vmaf 8a28970 by
Christopher Degawa. Adds a fallback for `_mm256_extract_epi64` on 32-bit
x86 builds (where the intrinsic is unavailable), implemented via two
`_mm_extract_epi32` calls per 64-bit lane. Pairs with 1b6c388 to enable
the AVX2/AVX-512 ADM SIMD paths on i686.

Conflict resolution preserved the fork's clang-format-100col layout and
the `_Alignas(64)` LTO-correctness slot in adm_avx512.c (see
docs/development/known-upstream-bugs.md), substituting `extract_epi64`
for `_mm256_extract_epi64` at the call sites.

Original-commit: 8a28970
Original-author: Christopher Degawa <ccom@randomderp.com>

* docs(upstream): T-NEW-3 — CHANGELOG + rebase-notes for 32-bit ADM/cpu port

Adds the lusoris-fork CHANGELOG entry and rebase-notes ledger row
0075 covering the verbatim port of upstream Netflix/vmaf
8a28970 + 1b6c388 (32-bit ADM/cpu fallbacks).

Pairs with the two preceding commits in this branch
(feat(x86/cpu): port 1b6c388, feat(adm): port 8a28970).
Closes audit row T-NEW-3 from PR #205
(docs/upstream-backlog-audit-2026-04-29.md).

* docs(metrics): note 32-bit ADM portable fallbacks (T-NEW-3 / PR #212 follow-up)

Document the i686-portable scalar fallback for _mm_extract_epi64 and the
removal of the 32-bit AVX gating guard, ported from upstream Netflix
8a28970 and 1b6c388. Link the ADR-0151 i686 CI lane that the fallbacks
unblock for actually scoring frames through integer ADM.

Closes the doc-substance gap flagged by the project-wide CI rule on PR
#212 (SIMD-touched diff missing docs/metrics/ update).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Christopher Degawa <ccom@randomderp.com>
Co-authored-by: Lusoris <lusoris@pm.me>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
lusoris pushed a commit that referenced this pull request May 1, 2026
@claude
docs(state): refresh post-session-2026-04-29 (verify rows, unblock T6…
acf3ad2 
…-1a Netflix Public dataset row)

Update docs/state.md `_Updated:` stamp to 2026-04-29 and rewrite the
"Tiny-AI C1 baseline `fr_regressor_v1.onnx`" deferral row's reopen-trigger
to TRIGGERED — the Netflix Public training corpus that gated C1 is now
locally available at `.workingdir2/netflix/` (9 ref + 70 dis YUVs, ~37 GB,
gitignored; provided by lawrence 2026-04-27), unblocking BACKLOG T6-1a.

Verified the rest of state.md against the 2026-04-29-session merged PR
set (#193–#205, #209). Every merged PR was feature / chore / docs / perf
with no bug-status delta to record per CLAUDE §12 rule 13:
- #193 chore(dnn) T7-12 env override removal — chore.
- #194 docs(research) T7-9 NPU digest — research.
- #195 feat(mcp) T5-2 embedded scaffold — feature.
- #196 feat(vulkan) T7-36 cambi integration — feature.
- #197 feat(motion) Netflix b949ceb port — upstream port.
- #198 chore(backlog) T7-32 micro-investigations — verify-only.
- #199 feat(ai) T6-9 model registry — feature.
- #200 feat(hip) T7-10 HIP scaffold — feature.
- #201 feat(simd) T7-38 SVE2 ports — feature.
- #202 feat(ci) T6-8 parity matrix — feature.
- #203 feat(ai) T6-7 FastDVDnet — feature.
- #205 docs(audit) T7-4 quarterly audit — explicitly notes "no
  state.md changes (no upstream commit ruled in/out a fork bug)".
- #209 perf(sycl) T7-17 fp64-less device — perf.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@lusoris lusoris mentioned this pull request May 1, 2026
Merged
4 tasks
lusoris pushed a commit that referenced this pull request May 1, 2026
@claude
docs(state): refresh post-session-2026-04-29 (verify rows, unblock T6…
f88fb11 
…-1a Netflix Public dataset row)

Update docs/state.md `_Updated:` stamp to 2026-04-29 and rewrite the
"Tiny-AI C1 baseline `fr_regressor_v1.onnx`" deferral row's reopen-trigger
to TRIGGERED — the Netflix Public training corpus that gated C1 is now
locally available at `.workingdir2/netflix/` (9 ref + 70 dis YUVs, ~37 GB,
gitignored; provided by lawrence 2026-04-27), unblocking BACKLOG T6-1a.

Verified the rest of state.md against the 2026-04-29-session merged PR
set (#193–#205, #209). Every merged PR was feature / chore / docs / perf
with no bug-status delta to record per CLAUDE §12 rule 13:
- #193 chore(dnn) T7-12 env override removal — chore.
- #194 docs(research) T7-9 NPU digest — research.
- #195 feat(mcp) T5-2 embedded scaffold — feature.
- #196 feat(vulkan) T7-36 cambi integration — feature.
- #197 feat(motion) Netflix b949ceb port — upstream port.
- #198 chore(backlog) T7-32 micro-investigations — verify-only.
- #199 feat(ai) T6-9 model registry — feature.
- #200 feat(hip) T7-10 HIP scaffold — feature.
- #201 feat(simd) T7-38 SVE2 ports — feature.
- #202 feat(ci) T6-8 parity matrix — feature.
- #203 feat(ai) T6-7 FastDVDnet — feature.
- #205 docs(audit) T7-4 quarterly audit — explicitly notes "no
  state.md changes (no upstream commit ruled in/out a fork bug)".
- #209 perf(sycl) T7-17 fp64-less device — perf.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
lusoris pushed a commit that referenced this pull request May 1, 2026
@claude
docs(state): refresh post-session-2026-04-29 (verify rows, unblock T6…
b19c14d 
…-1a Netflix Public dataset row)

Update docs/state.md `_Updated:` stamp to 2026-04-29 and rewrite the
"Tiny-AI C1 baseline `fr_regressor_v1.onnx`" deferral row's reopen-trigger
to TRIGGERED — the Netflix Public training corpus that gated C1 is now
locally available at `.workingdir2/netflix/` (9 ref + 70 dis YUVs, ~37 GB,
gitignored; provided by lawrence 2026-04-27), unblocking BACKLOG T6-1a.

Verified the rest of state.md against the 2026-04-29-session merged PR
set (#193–#205, #209). Every merged PR was feature / chore / docs / perf
with no bug-status delta to record per CLAUDE §12 rule 13:
- #193 chore(dnn) T7-12 env override removal — chore.
- #194 docs(research) T7-9 NPU digest — research.
- #195 feat(mcp) T5-2 embedded scaffold — feature.
- #196 feat(vulkan) T7-36 cambi integration — feature.
- #197 feat(motion) Netflix b949ceb port — upstream port.
- #198 chore(backlog) T7-32 micro-investigations — verify-only.
- #199 feat(ai) T6-9 model registry — feature.
- #200 feat(hip) T7-10 HIP scaffold — feature.
- #201 feat(simd) T7-38 SVE2 ports — feature.
- #202 feat(ci) T6-8 parity matrix — feature.
- #203 feat(ai) T6-7 FastDVDnet — feature.
- #205 docs(audit) T7-4 quarterly audit — explicitly notes "no
  state.md changes (no upstream commit ruled in/out a fork bug)".
- #209 perf(sycl) T7-17 fp64-less device — perf.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
lusoris added a commit that referenced this pull request May 1, 2026
@lusoris @claude
docs(state): refresh post-session-2026-04-29 (verify rows, unblock T6…
42b6d35 
…-1a Netflix Public dataset row) (#245)

Update docs/state.md `_Updated:` stamp to 2026-04-29 and rewrite the
"Tiny-AI C1 baseline `fr_regressor_v1.onnx`" deferral row's reopen-trigger
to TRIGGERED — the Netflix Public training corpus that gated C1 is now
locally available at `.workingdir2/netflix/` (9 ref + 70 dis YUVs, ~37 GB,
gitignored; provided by lawrence 2026-04-27), unblocking BACKLOG T6-1a.

Verified the rest of state.md against the 2026-04-29-session merged PR
set (#193–#205, #209). Every merged PR was feature / chore / docs / perf
with no bug-status delta to record per CLAUDE §12 rule 13:
- #193 chore(dnn) T7-12 env override removal — chore.
- #194 docs(research) T7-9 NPU digest — research.
- #195 feat(mcp) T5-2 embedded scaffold — feature.
- #196 feat(vulkan) T7-36 cambi integration — feature.
- #197 feat(motion) Netflix b949ceb port — upstream port.
- #198 chore(backlog) T7-32 micro-investigations — verify-only.
- #199 feat(ai) T6-9 model registry — feature.
- #200 feat(hip) T7-10 HIP scaffold — feature.
- #201 feat(simd) T7-38 SVE2 ports — feature.
- #202 feat(ci) T6-8 parity matrix — feature.
- #203 feat(ai) T6-7 FastDVDnet — feature.
- #205 docs(audit) T7-4 quarterly audit — explicitly notes "no
  state.md changes (no upstream commit ruled in/out a fork bug)".
- #209 perf(sycl) T7-17 fp64-less device — perf.

Co-authored-by: Lusoris <lusoris@pm.me>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@lusoris