claude-noether
53df917afb
Module loads + format enumerates (kernel-side wiring works) but every VP9 decode attempt stalls the HW: STA_INT=0x23 (IRQ_RAW + TIMEOUT) when HW timeout enabled; NO IRQ at all when timeout disabled (genuine stall). 6 iterations of register-tuning all failed identically. Hypothesis space narrowed to 3 structural-level possibilities: 1. Probability buffer format mismatch (legacy struct rkvdec_vp9_probs vs BSP hal_vp9d_prob_default format) 2. Missing kernel-side init (cache config, SRAM/QoS, AXI setup that BSP mpp_rkvdec2 does but mainline HEVC happens not to need) 3. vdpu381 register layout doesn't fully expose VP9 control surface (BSP MPP may rely on mpp_dev_set_reg_offset translations we don't replicate) Recommended next path: Sonnet/Janet architect review BEFORE more iteration. The 6 single-bit tuning rounds established that no bit-flip fix is sufficient — structural rethink required. Sibling-campaign close state (HEVC bit-perfect) recoverable on ampere via backup at ~/vp9-iter1-backup/rockchip-vdec.ko.sibling-campaign-close + depmod cycle. Phase 0-2.1 work preserved at boltzmann:~/src/linux-rockchip: vp9-enablement-iter1 (6 commits, 1517 LoC). Format enumeration on /dev/video1 confirms kernel-side wiring is correct. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
ampere-vp9-enablement
Stand-alone port + upstream-targeting work to enable VP9 hardware decode on Rockchip RK3588's rkvdec (vdpu381 register layout).
Status (2026-05-17 ~01:00)
Upstream RK3588 mainline rkvdec (Casanova v7.0 series, landed in Linux 7.0) supports H.264 + HEVC only. VP9 is on Collabora's stated roadmap but no WIP series has been posted to linux-media as of this campaign open. The legacy rkvdec-vp9.c (RK3399 / vdpu341 hardware) is feature-complete at 1042 lines but its register-config logic does not translate directly to vdpu381.
This campaign:
- Ports VP9 enablement to vdpu381 register layout (new file
rkvdec-vdpu381-vp9.c) - Registers VP9 V4L2 controls in
vdpu38x_vp9_ctrl_descs[] - Adds VP9 fmt to
vdpu381_coded_fmts[]with the new ops - Verifies bit-perfect HW vs SW decode (per feedback_compare_hw_against_sw_reference)
- Proposes upstream via linux-media
Sibling campaign: ampere-kernel-decoders closed at HEVC bit-perfect (kernel-agent#14 + #15 are the prerequisite kernel fixes).
Scope (out of)
- VP9 on RK3399 (works via legacy
rkvdec-vp9.calready in mainline) - VP9 on hantro (hantro decoder on RK3588 doesn't expose VP9; this campaign targets rkvdec)
- AV1 on RK3588 (separate work; AV1 is on hantro fdc70000 already + per Collabora)
- VP8 (already works via hantro)
- HEVC (closed in ampere-kernel-decoders)
Process
8-phase loop (per ~/.claude/CLAUDE.md). All commits via claude-noether identity. Patches will be RFC-quality and routed via kernel-agent once ready.