[FFmpeg-devel] Bug on Bayer conversions

[Anton Khirnov]

Quoting Chema Gonzalez (2022-09-25 17:54:16)
> Hi,
> 
> I found an issue while playing with Bayer pixel format conversions.
> 
> ```
> $ echo -ne '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff'
> > image.raw
> $ xxd image.raw
> 00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000010: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000020: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> 00000030: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> ```
> 
> And then:
> ```
> $ ffmpeg -y -f rawvideo -pixel_format bayer_bggr8 -s 8x8 -i image.raw
> -f rawvideo -pix_fmt rgb24 -video_size 8x8 image.raw.rgb
> ...
> Assertion srcSliceH > 1 failed at libswscale/swscale_unscaled.c:1310
> Aborted (core dumped)ated 2 times
> ```
> 
> The issue relates to the ffmpeg parallelization.
> ```
> $ ffmpeg -y -filter_threads 1 -f rawvideo -pixel_format bayer_bggr8 -s
> 8x8 -i image.raw -f rawvideo -pix_fmt rgb24 -video_size 8x8
> image.raw.rgb
> ...
> frame=    1 fps=0.0 q=-0.0 Lsize=       0kB time=00:00:00.00
> bitrate=N/A speed=   0x    eed=N/A
> video:0kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB
> muxing overhead: 0.000000%
> $ xxd image.raw.rgb
> 00000000: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000010: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000030: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000040: 0000 0000 0000 0000 0000 0000 0000 0000  ................
> 00000050: 7f00 3f7f 0000 7f00 3f7f 0000 0000 0000  ..?.....?.......
> 00000060: ffff ffff ffff 7fbf ff7f ffff 7fbf ff7f  ................
> 00000070: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> 00000080: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> 00000090: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> 000000a0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> 000000b0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
> ```
> 
> FYI:
> ```
> $ grep processor /proc/cpuinfo  |wc -l
> 64
> ```
> 
> Problem seems to be that `ff_sws_slice_worker()`
> [libswscale/swscale.c:1222] tries to slice the input to parallelize
> the scaling task, in my case in 16 different jobs (gdb'ing the process
> shows `nb_threads == nb_jobs == 16`). The 8x8 input is therefore
> divided in eight 8x1 slices (1-pixel height), which eventually breaks
> in `bayer_to_rgb24_wrapper()` as it asserts `srcSliceH > 1`. The
> problem is the same in the 3 Bayer conversion functions
> (`bayer_to_rgb24_wrapper()`, `bayer_to_rgb48_wrapper()`, and
> `bayer_to_yv12_wrapper()`.
> 
> Wondering about the right solution: We could just enforce `nb_threads
> = nb_jobs = 1` for all Bayer inputs. That may be the simplest
> solution. Or we could make sure that `nb_threads` (and `nb_jobs`) are
> capped at `input_height / 2` (to ensure at least 2 pixels per thread).
> 
> Any suggestions?

There already is code that constrains minimum slice size, see
dst_slice_align in utils.c. Would extending that code to handle bayer
input fix the problem?

-- 
Anton Khirnov