 marfrit
|
67494ae7ee
|
Iteration 4 close — Track A locked, three-iteration carryover resolved
The iter1+iter2+iter3 frame-11 EINVAL is empirically eliminated. mpv
direct stress test on ohm via patched libva-v4l2-request-fourier:
RequestBeginPicture: 2130
RequestSyncSurface: 4254
S_EXT_CTRLS EINVAL: 0
Unable to set control(s): 0
Generic EINVAL: 0
ENETDOWN: 0
2130 frames at 24 fps = real-time HW decode (>98% of 2160-frame max
in 90 seconds wall time). Track A's Phase 1 success criterion crushed.
Three correctness fixes (4 fork commits):
- 74d8dd1: DPB fields=V4L2_H264_FRAME_REF + skip stale entries
- 385dee1: fresh request_fd per frame (THE load-bearing fix)
- b81ce69: B-slice L1 reflist .fields copy-paste
Plus diagnostic instrumentation (a12d299, 4892656, f21bdf0) deferred
to iter5 sweep alongside earlier iter1/iter3 instrumentation.
Three new memory entries: kernel obfuscation extends to compound TRY,
request_fd lifecycle (fresh per frame), FFmpeg as empirical authority.
README iteration table updated.
Carries to iter5 substrate: DEBUG sweep, mpv libplacebo segfault,
multi-context libva safety, PGO Firefox rebuild, eventual upstream
prep (Mozilla bug + bootlin libva-v4l2-request).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-05-05 14:29:43 +00:00 |
|