claude-noether/ampere-av1-enablement
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Phase 3+5 close: amend with Phase 5 review outcome (Amendment 4 kept; Amendment 1 reverted empirically)

Browse Source
This commit is contained in:
Markus Fritsche
2026-05-17 12:19:45 +00:00
parent c6d55bce29
commit f0461f7662
1 changed files with 12 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
phase3_close.md
+12 -3
View File
@@ -1,6 +1,14 @@
# ampere-av1-enablement Phase 3 close — AV1 PASS on all-intra; film_grain+show_existing edge case localized
# ampere-av1-enablement Phase 3+5 close — AV1 PASS on all-intra; film_grain+show_existing edge case localized
Closed 2026-05-17 evening. Phase 2.1 + Phase 3 substantial iteration. Substrate state on ampere: backend tip `c839b94` on `av1-iter1` branch (matches `av1-iter1-imported` on noether). Module installed at `/usr/lib/dri/v4l2_request_drv_video.so`. Kernel `7.0.0-rc3-devices+` unchanged.
Closed 2026-05-17 evening. Phase 2.1 + Phase 3 + Phase 5 review iteration. Substrate state on ampere: backend tip `902d6c1` on `av1-iter1` branch (matches `av1-iter1-imported` on noether, pushed to gitea). Module installed at `/usr/lib/dri/v4l2_request_drv_video.so`. Kernel `7.0.0-rc3-devices+` unchanged.
## Phase 5 review outcome (post-Phase-3)
sonnet-architect review found 3 amendments:
- **Amendment 1 (loop_restoration remap table) — REVERTED**: reviewer proposed a permutation matching Kwiboo + AV1 spec wire encoding. Empirically applied → regressed allintra 10/10 → 0/10 + test_av1 bit-exact → DIFF. Per [[feedback_review_empirical_over_theoretical]] identity mapping wins. Either VAAPI's `yframe_restoration_type` is already in V4L2-enum order, or vpu981's V4L2 enum interpretation differs from the uAPI header docs.
- **Amendment 4 (stale `linked_decode_surface_id`) — KEPT**: clear in BeginPicture after iter2 Fix 3 release. Prevents spurious link-borrows when ffmpeg-vaapi recycles former display surfaces as decode targets. No-op for non-AV1 codecs.
- **Amendment 5 (SEPARATE_UV_DELTA_Q seq flag missing) — noted, not actionable**: VAAPI doesn't expose `color_config.separate_uv_delta_q`. Needs bitstream-side info.
## Test matrix
@@ -42,6 +50,7 @@ Remaining 7/10 divergence on `av1_larger.ivf` localized to:
On `av1-iter1` branch (both ampere `/home/mfritsche/src/libva-v4l2-request-fourier/` + noether `/home/mfritsche/src/libva-multiplanar/libva-v4l2-request-fourier/`):
```
902d6c1 ampere-av1 Phase 5 review: stale linked_decode_surface_id clear; remap fix REVERTED
c839b94 ampere-av1 Phase 3 finding: iter2 Fix 3 release is NOT the divergence cause
d7ef0f6 ampere-av1 Phase 3: SEQUENCE byte-equal kdirect; 3/10 frames PASS bit-exact
5803cbc ampere-av1 Phase 3 progress: film_grain link + UPDATE_GRAIN; frame 0 bit-exact
@@ -54,7 +63,7 @@ ab79ed5 ampere-av1 Phase 3 in-progress notes: UPDATE_GRAIN segfault; 352x288 sti
bed75c0 ampere-av1 Phase 2 step 1: third-device fd scaffolding for vpu981
```
Branch is NOT yet pushed to gitea (per pre-amnesia me's "ssh sideband disconnect on vp9 branch" note — manual push attempt deferred).
Branch pushed to gitea at `marfrit/libva-v4l2-request-fourier` (no ssh sideband disconnect this time — pre-amnesia me's vp9 branch concern not reproduced for av1-iter1).
## Verifier scripts retained