libva-v4l2-request-fourier

marfrit/libva-v4l2-request-fourier

Fork 1

Commit Graph

Author	SHA1	Message	Date
claude-noether	692eaa0053	fresnel-fourier iter4 Phase 7 fix-forward: gate ANNEX-B start-code prepend on H.264/HEVC profiles Root cause for VP9 criterion-4 failure traced via runtime instrumentation: context.c:194 unconditionally set context_object->h264_start_code = true for every CreateContext, regardless of codec profile. picture.c:70 then prepends 0x00 0x00 0x01 (ANNEX-B start code) to ALL slice data including VP9 frames. VP9 has no start codes — its uncompressed_header begins with the raw frame_marker byte (0x10 in the high 2 bits). The 3-byte prefix shifted the rkvdec driver's bitstream-read by 24 bits, producing a silent decode failure (frame_marker mismatch -> driver fails to locate a valid frame -> CAPTURE slot stays at cap_pool init pattern, the dim 0x4c green visible in Phase 7 hwdownload PNGs). iter4 fix: switch on config_object->profile in RequestCreateContext. Set h264_start_code = true only for VAProfileH264* and VAProfileHEVCMain. False for MPEG2/VP8/VP9. iter1 (MPEG-2) and iter3 (VP8) had this same bug latent — they passed because their criterion-4 verification used different paths (iter1 direct readback was small enough to mask, iter3 used transitive proof not pixel comparison). The Phase 7 byte-level pixel comparison is what exposed it. Empirical proof of the fix on fresnel: - pre-fix submission FRAME control bytes 0-23: lf.flags=0x01 (only DELTA_ENABLED), base_q_idx=0x41 — bit-misaligned because parser was reading the prefix bytes. - post-fix submission FRAME control bytes 0-23 byte-match Phase 3 kernel-direct anchor: lf.flags=0x03 (ENABLED\|UPDATE), base_q_idx=0x2e (46). Transitive-proof leg 1 (backend-payload == kernel-direct-payload) satisfied for the keyframe. - s(6) bit-width fix in vp9.c (4 mag + 1 sign -> 6 mag + 1 sign per VP9 spec) was a real bug too, latent because Bug 1 (this commit's fix) prevented its code path from running. Both fixes ship together. Pixels still produce 0x4c constant pattern post-fix — that is Bug 2 (substrate-wide cap_pool readback regression on linux-fresnel-fourier 7.0-1) per phase7_iter4_verification.md. Bug 2 is out of iter4 scope per Option-A choice; transitive proof remains the criterion-4 verification path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 09:50:25 +00:00
claude-noether	406d08e122	fresnel-fourier iter4 Phase 6 commit B: NEW src/vp9.c + src/vp9.h + meson.build + context.h (vp9_lf) + surface.h (params.vp9) VP9 codec dispatcher implementing 12 contract clauses against V4L2_CID_STATELESS_VP9_FRAME (0xa40a2c) + V4L2_CID_STATELESS_VP9_COMPRESSED_HDR (0xa40a2d). 2 batched controls per frame; rkvdec on RK3399 mandatorily requires both per drivers/staging/media/rkvdec/rkvdec-vp9.c::rkvdec_vp9_run_preamble:752. Implementation: - ~80 LOC VPX range coder (vp9_rac_) — minimal port of FFmpeg vpx_rac.[ch] + vp89_rac.h. Stateless static helpers. - inv_map_table[255] + read_prob_delta — verbatim copy from v4l2_request_vp9.c:44-97. - vp9_parse_uncompressed_header_lf_quant — partial parse for the fields VAAPI doesn't expose: lf_delta_enabled / lf_delta_update / lf_ref_delta[4] / lf_mode_delta[2] / base_q_idx / delta_q_y_dc / delta_q_uv_dc / delta_q_uv_ac. ~120 LOC. - vp9_fill_compressed_hdr — port of FFmpeg fill_compressed_hdr with Phase 5 C3 out_reference_mode parameter. ~140 LOC. - vp9_set_controls — orchestrates Clauses 1+2+4+5+7+10+11+12. ~120 LOC. Phase 5 amendments incorporated in code: - C1: frame.interpolation_filter = direct from VAAPI's mcomp_filter_type (NO XOR; vaapi_vp9.c:62 already applied it before storing into VAAPI's mcomp_filter_type). - C2: persistent vp9_lf state added to object_context (in context.h). Initialized to VP9 spec defaults {1,0,-1,-1,0,0} on keyframe / intra_only / error_resilient. Updated only when parser sees lf_delta.update=1. Always copied to kernel control. - C3: vp9_fill_compressed_hdr takes uint8_t out_reference_mode; threaded through call site. allowcompinter derived from VAAPI sign-bias bits. Phase 5 S4: uv_mode memcpy from FFmpeg's fill_compressed_hdr omitted — rkvdec reads uv_mode from kernel's persistent probability_tables, NOT from prob_updates ctrl. Clause 3 compile-time _Static_assert on struct sizes (168/2040) matches Phase 3 empirical baseline; UAPI shifts will fail loudly. surface.h: extends params union with vp9 { picture, slice }. context.h: adds vp9_lf { ref_deltas[4], mode_deltas[2], initialized }. meson.build: adds vp9.c + vp9.h. Build: clean on fresnel (linux-fresnel-fourier 7.0-1, libva 1.23). Runtime: not yet wired in picture.c — next commit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-10 06:46:11 +00:00

Author

SHA1

Message

Date

claude-noether

692eaa0053

fresnel-fourier iter4 Phase 7 fix-forward: gate ANNEX-B start-code prepend on H.264/HEVC profiles

Root cause for VP9 criterion-4 failure traced via runtime
instrumentation: context.c:194 unconditionally set
context_object->h264_start_code = true for every CreateContext,
regardless of codec profile. picture.c:70 then prepends 0x00 0x00 0x01
(ANNEX-B start code) to ALL slice data including VP9 frames.

VP9 has no start codes — its uncompressed_header begins with the raw
frame_marker byte (0x10 in the high 2 bits). The 3-byte prefix
shifted the rkvdec driver's bitstream-read by 24 bits, producing a
silent decode failure (frame_marker mismatch -> driver fails to
locate a valid frame -> CAPTURE slot stays at cap_pool init pattern,
the dim 0x4c green visible in Phase 7 hwdownload PNGs).

iter4 fix: switch on config_object->profile in RequestCreateContext.
Set h264_start_code = true only for VAProfileH264* and VAProfileHEVCMain.
False for MPEG2/VP8/VP9.

iter1 (MPEG-2) and iter3 (VP8) had this same bug latent — they passed
because their criterion-4 verification used different paths (iter1
direct readback was small enough to mask, iter3 used transitive proof
not pixel comparison). The Phase 7 byte-level pixel comparison is what
exposed it.

Empirical proof of the fix on fresnel:
- pre-fix submission FRAME control bytes 0-23: lf.flags=0x01 (only
  DELTA_ENABLED), base_q_idx=0x41 — bit-misaligned because parser was
  reading the prefix bytes.
- post-fix submission FRAME control bytes 0-23 byte-match Phase 3
  kernel-direct anchor: lf.flags=0x03 (ENABLED|UPDATE), base_q_idx=0x2e
  (46). Transitive-proof leg 1 (backend-payload == kernel-direct-payload)
  satisfied for the keyframe.
- s(6) bit-width fix in vp9.c (4 mag + 1 sign -> 6 mag + 1 sign per
  VP9 spec) was a real bug too, latent because Bug 1 (this commit's fix)
  prevented its code path from running. Both fixes ship together.

Pixels still produce 0x4c constant pattern post-fix — that is Bug 2
(substrate-wide cap_pool readback regression on
linux-fresnel-fourier 7.0-1) per phase7_iter4_verification.md.
Bug 2 is out of iter4 scope per Option-A choice; transitive proof
remains the criterion-4 verification path.

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

2026-05-10 09:50:25 +00:00

claude-noether

406d08e122

fresnel-fourier iter4 Phase 6 commit B: NEW src/vp9.c + src/vp9.h + meson.build + context.h (vp9_lf) + surface.h (params.vp9)

VP9 codec dispatcher implementing 12 contract clauses against
V4L2_CID_STATELESS_VP9_FRAME (0xa40a2c) +
V4L2_CID_STATELESS_VP9_COMPRESSED_HDR (0xa40a2d). 2 batched
controls per frame; rkvdec on RK3399 mandatorily requires both
per drivers/staging/media/rkvdec/rkvdec-vp9.c::rkvdec_vp9_run_preamble:752.

Implementation:
- ~80 LOC VPX range coder (vp9_rac_*) — minimal port of FFmpeg
  vpx_rac.[ch] + vp89_rac.h. Stateless static helpers.
- inv_map_table[255] + read_prob_delta — verbatim copy from
  v4l2_request_vp9.c:44-97.
- vp9_parse_uncompressed_header_lf_quant — partial parse for the
  fields VAAPI doesn't expose: lf_delta_enabled / lf_delta_update /
  lf_ref_delta[4] / lf_mode_delta[2] / base_q_idx /
  delta_q_y_dc / delta_q_uv_dc / delta_q_uv_ac. ~120 LOC.
- vp9_fill_compressed_hdr — port of FFmpeg fill_compressed_hdr
  with Phase 5 C3 out_reference_mode parameter. ~140 LOC.
- vp9_set_controls — orchestrates Clauses 1+2+4+5+7+10+11+12.
  ~120 LOC.

Phase 5 amendments incorporated in code:
- C1: frame.interpolation_filter = direct from VAAPI's
  mcomp_filter_type (NO XOR; vaapi_vp9.c:62 already applied it
  before storing into VAAPI's mcomp_filter_type).
- C2: persistent vp9_lf state added to object_context (in
  context.h). Initialized to VP9 spec defaults
  {1,0,-1,-1,0,0} on keyframe / intra_only / error_resilient.
  Updated only when parser sees lf_delta.update=1. Always
  copied to kernel control.
- C3: vp9_fill_compressed_hdr takes uint8_t *out_reference_mode;
  threaded through call site. allowcompinter derived from VAAPI
  sign-bias bits.

Phase 5 S4: uv_mode memcpy from FFmpeg's fill_compressed_hdr
omitted — rkvdec reads uv_mode from kernel's persistent
probability_tables, NOT from prob_updates ctrl.

Clause 3 compile-time _Static_assert on struct sizes (168/2040)
matches Phase 3 empirical baseline; UAPI shifts will fail loudly.

surface.h: extends params union with vp9 { picture, slice }.
context.h: adds vp9_lf { ref_deltas[4], mode_deltas[2], initialized }.
meson.build: adds vp9.c + vp9.h.

Build: clean on fresnel (linux-fresnel-fourier 7.0-1, libva 1.23).
Runtime: not yet wired in picture.c — next commit.

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

2026-05-10 06:46:11 +00:00

2 Commits