[FFmpeg-devel] [GSoC 2026] APV Hardware Acceleration, VideoToolbox & VAAPI, - Interest & Mentor Request

[Soham Kute via ffmpeg-devel <ffmpeg-devel@ffmpeg.org>]

Hello FFmpeg developers,

My name is Soham Kute. I am a pre-final year undergraduate student at
Veermata Jijabai Technological Institute (VJTI), Mumbai. I am writing
to express interest in GSoC 2026 with FFmpeg and to discuss a project
I have been studying.

I enjoy reading codec specifications, finding edge case bugs in decoders,
and understanding how bitstream parsing maps to actual implementation.
This is what drew me to FFmpeg.

Project Idea:

I would like to propose implementing hardware accelerated decoding for
the APV (Advanced Professional Video) codec, specifically targeting
Apple VideoToolbox (macOS/iOS) and Intel/AMD VAAPI (Linux).

APV's native software decoder (apv_decode.c) and liboapv based encoder
were merged in FFmpeg 8.0. However, no hwaccel path exists yet no
videotoolbox_apv.c, no vaapi_apv.c, no hwaccels.h declaration, and no
configure entry for any hardware target. Given that Google has adopted
OpenAPV for Android 16 and Samsung cameras already record in APV.

Technical Approach:

I have studied videotoolbox_av1.c and vaapi_av1.c as templates and
understand the full implementation path:

  VideoToolbox:
  - ff_videotoolbox_apvc_extradata_create() using CBS-parsed
    APVRawFrameHeader fields (profile_idc, chroma_format_idc,
    bit_depth_minus8) from the existing cbs_apv.c infrastructure.
  - Tile accumulation in decode_slice, mirroring AV1 tile_group handling.
  - CVPixelBuffer mappings for 4:2:2 and 4:4:4 at 10-bit and 12-bit.

  VAAPI:
  - Profile/entrypoint probing via vaQueryConfigProfiles().
  - VAPictureParameterBufferAPV + VAIQ atrixBufferAPV from CBS frame header.
  - Per-tile VASliceParameterBufferAPV submission loop.

Qualification Task:

I understand that completing a qualification patch is a prerequisite
for GSoC consideration. I would appreciate guidance from the mentors
on an appropriate task,  whether that is a robustness fix in the
existing APV decoder, an improvement to the CBS APV infrastructure,
or any other area the mentors feel would best demonstrate readiness
for this project.

I am ready to start immediately.

Mentor Request:

I would like to request that Mark Thompson consider mentoring this project,
given his authorship of the native APV decoder and expertise in FFmpeg
hardware acceleration. I have also CC'd James Almer (APV encoder and parser
co-author) and Dariusz Frankiewicz from Samsung Electronics.

If the scope needs adjustment or if I have misunderstood any aspect of
the APV hwaccel design, I would very much welcome feedback before the
application deadline.

I am subscribed to ffmpeg-devel and active on #ffmpeg-devel (Libera.chat)
as SohamK.

Thank you for your time.

Best regards,
Soham Kute
_______________________________________________
ffmpeg-devel mailing list -- ffmpeg-devel@ffmpeg.org
To unsubscribe send an email to ffmpeg-devel-leave@ffmpeg.org