 marfrit
|
868d854121
|
iter6 Phase 0 lock: Candidate G — Bug 6 VP8 partial output
User pick. 6 boolean criteria locked: VP8 libva==kdirect; no regression
on VP9/MPEG-2/H.264-keyframe/HEVC; control-payload anchors hold.
Scope: src/vp8.c, src/picture.c VP8 dispatch + buffer cases,
src/surface.c surface_bind_slot, cap_pool slot lifecycle.
No kernel work. Backend-side fix expected (decode runs through
kernel cleanly; output diverges in slot rotation or partial fill).
Predicted small: 5-50 LOC once root-caused. Phase 2 + Phase 3
likely take more wallclock than Phase 6 implementation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-05-12 19:37:13 +00:00 |
|
|
marfrit
|
34e1480de5
|
iter6 Phase 0: substrate inventory + 5 candidate research questions
iter5b-β surfaced 3 explicit bugs (Bug 4 H.264 inter, Bug 5 HEVC
DQBUF ERROR, Bug 6 VP8 partial output) plus carried backlog items
(iter4-B1 device discrimination, B2-B6, L3, Q6, COLOR_RANGE).
Candidates F-J laid out for user lock:
- F: Bug 5 HEVC kernel-rejection (highest claim-vs-reality stigma)
- G: Bug 6 VP8 partial output (smallest suspect surface)
- H: Bug 4 H.264 inter race (highest consumer impact)
- I: Re-anchor regression hashes on β substrate
- J: iter4-B1 auto-detect harden
Recommendation: G → H → F sequence if multiple iters planned;
otherwise H for impact or J for architectural-cleanup fit.
Phase 1 lock pending user pick.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-05-12 19:23:58 +00:00 |
|