marfrit/marfrit-packages
0
0
Fork 1
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

ffmpeg-v4l2-request-fourier: per-MB inspection callback for H.264 (0016) #106

Merged
marfrit merged 1 commits from claude-noether/marfrit-packages:noether/h264-mb-inspect-callback into main 2026-05-26 04:05:58 +00:00
Conversation 0 Commits 1 Files Changed 4
+237 -4
marfrit commented 2026-05-26 03:58:12 +00:00
Owner
Copy Link
Copy Source

Adds 0016-h264-mb-inspect-callback.patch to the FFmpeg fork. Foundation for option A test harness — pairs with daedalus-decoder PR-A1b (the CLI binary).

API

typedef void (*ff_h264_mb_inspect_cb)(void *opaque,
                                       const struct H264Context *h,
                                       int mb_x, int mb_y);

void ff_h264_set_mb_inspect_cb(AVCodecContext *avctx,
                                ff_h264_mb_inspect_cb cb, void *opaque);

What it does

Opt-in callback fired by ff_h264_hl_decode_mb after the existing pixel work, for every MB in coded order. Two fields appended to H264Context (internal struct in h264dec.h, no ABI surface to non-libavcodec callers). Callback receives const H264Context* so it can inspect any state — slice context via h->slice_ctx, reconstructed pixels via h->cur_pic.f->data[plane], coeffs via sl->mb, etc.

Default (cb == NULL): zero behaviour change, one load + one branch per MB.

Shape — why this is NOT the dejavu trap

The 0003-0014 patches that PR #105 reverted were per-kernel function-pointer hijacks producing one Vulkan submit per block per kernel. Per-block synchronous dispatch from libavcodec is architecturally non-competitive (~22× slower than CPU NEON on the corrected daedalus-fourier PR #37 bench).

This patch is structurally different:

  • ONE inspection point per MB (not per kernel).
  • CPU-side only — the callback just observes; no GPU work fires from inside ff_h264_hl_decode_mb.
  • Consumer (daedalus-decoder in the follow-up PR) accumulates per-MB state in CPU-side flat buffers and dispatches ONCE per frame at flush_frame.

Frame-major UMA dispatch verdict (memory: dejavu) — same architecture as the daedalus-decoder Stage 1/2 work already merged.

Used by

  • daedalus-decoder/tools/daedalus_decode_h264 (PR-A1b, follow-up) — CLI test harness that wraps libavcodec + daedalus-decoder and bit-exact-compares on real H.264 streams.
  • Future daedalus-v4l2 daemon refactor — replaces the daemon's avcodec_send_packet/receive_frame per-MB path with the inspection-callback-driven version.

Wiring

  • arch PKGBUILD: source[] + prepare().
  • debian build-deb.sh: patch sequence.
  • Both pkgrel: 12 → 13.

Refs reauktion/daedalus-decoder!12 (Stage 2 PR-b).

Adds `0016-h264-mb-inspect-callback.patch` to the FFmpeg fork. Foundation for option A test harness — pairs with daedalus-decoder PR-A1b (the CLI binary). ## API ```c typedef void (*ff_h264_mb_inspect_cb)(void *opaque, const struct H264Context *h, int mb_x, int mb_y); void ff_h264_set_mb_inspect_cb(AVCodecContext *avctx, ff_h264_mb_inspect_cb cb, void *opaque); ``` ## What it does Opt-in callback fired by `ff_h264_hl_decode_mb` **after** the existing pixel work, for every MB in coded order. Two fields appended to `H264Context` (internal struct in `h264dec.h`, no ABI surface to non-libavcodec callers). Callback receives `const H264Context*` so it can inspect any state — slice context via `h->slice_ctx`, reconstructed pixels via `h->cur_pic.f->data[plane]`, coeffs via `sl->mb`, etc. Default (`cb == NULL`): zero behaviour change, one load + one branch per MB. ## Shape — why this is NOT the dejavu trap The `0003-0014` patches that PR #105 reverted were per-kernel function-pointer hijacks producing one Vulkan submit per block per kernel. Per-block synchronous dispatch from libavcodec is architecturally non-competitive (~22× slower than CPU NEON on the corrected daedalus-fourier PR #37 bench). This patch is structurally different: - **ONE** inspection point per MB (not per kernel). - **CPU-side only** — the callback just observes; no GPU work fires from inside `ff_h264_hl_decode_mb`. - Consumer (`daedalus-decoder` in the follow-up PR) accumulates per-MB state in CPU-side flat buffers and dispatches **ONCE per frame** at `flush_frame`. Frame-major UMA dispatch verdict (memory: `dejavu`) — same architecture as the daedalus-decoder Stage 1/2 work already merged. ## Used by - daedalus-decoder/tools/daedalus_decode_h264 (PR-A1b, follow-up) — CLI test harness that wraps libavcodec + daedalus-decoder and bit-exact-compares on real H.264 streams. - Future daedalus-v4l2 daemon refactor — replaces the daemon's `avcodec_send_packet`/`receive_frame` per-MB path with the inspection-callback-driven version. ## Wiring - arch PKGBUILD: source[] + prepare(). - debian build-deb.sh: patch sequence. - Both pkgrel: 12 → 13. Refs reauktion/daedalus-decoder!12 (Stage 2 PR-b).
marfrit added 1 commit 2026-05-26 03:58:13 +00:00
ffmpeg-v4l2-request-fourier: per-MB inspection callback for H.264 (0016) 87cbb9b70a 
Adds 0016-h264-mb-inspect-callback.patch to the FFmpeg fork.  Adds an
opt-in callback fired by ff_h264_hl_decode_mb after the existing
pixel work, for tools that need per-MB visibility into H.264 decode.

API:
  typedef void (*ff_h264_mb_inspect_cb)(void *opaque,
                                         const struct H264Context *h,
                                         int mb_x, int mb_y);
  void ff_h264_set_mb_inspect_cb(AVCodecContext *avctx,
                                  ff_h264_mb_inspect_cb cb, void *opaque);

Two new fields appended to H264Context (internal struct, declared in
h264dec.h not h264.h, no ABI surface to non-libavcodec callers).
Callback fires post-pixel-work for every MB in coded order; receives
const H264Context* so it can inspect any state (slice ctx via
h->slice_ctx, reconstructed pixels via h->cur_pic.f->data[plane],
etc.).

Default (cb==NULL): zero behaviour change, one load + one branch per
MB in the decoder hot path.

Shape distinction: per-MB observation, NOT per-kernel function-pointer
hijack (the 0003-0014 substitution-arc pattern that PR #105 reverted
+ daedalus-fourier PR #37's measurement-correction architecturally
retired).  Per-block synchronous Vulkan dispatch from libavcodec is
non-competitive; per-MB CPU-side observation feeding a per-frame
daedalus-decoder batch submit is the right shape (frame-major UMA
dispatch verdict, memory: dejavu).

Used by:
  - daedalus-decoder/tools/daedalus_decode_h264 (PR-A1b, follow-up)
  - future daedalus-v4l2 daemon refactor

Wired into arch PKGBUILD source[] + prepare() and debian build-deb.sh
patch sequence.  pkgrel bumped 12 → 13.

Refs reauktion/daedalus-decoder!12.
marfrit merged commit 59901bceca into main 2026-05-26 04:05:58 +00:00
marfrit deleted branch noether/h264-mb-inspect-callback 2026-05-26 04:05:58 +00:00
Sign in to join this conversation.
Reviewers
No Reviewers
Labels
No items
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
marfrit (Markus Fritsche)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: marfrit/marfrit-packages#106
Delete Branch "claude-noether/marfrit-packages:noether/h264-mb-inspect-callback"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?