[FFmpeg-user] can swscale handle AV_PIX_FMT_YUV444P input/output?

Andrew Randrianasulu randrianasulu at gmail.com
Wed Sep 4 23:27:59 EEST 2024


https://www.cinelerra-gg.org/bugtracker/view.php?id=665

we run  into strange problem:

if cinelerra-gg  uses RGBA-8 format internally  we can process yuv444p
(full chroma) test sample made with

ffmpeg -f lavfi -i yuvtestsrc -frames 1 -color_range 2
/dev/shm/yuv-test.y4m

by rendering it to y4m (yuv444p pixel format) and then comparing two files
does not show big difference.

Yet if we switch our pipeline to yuv (8 bit, 4:4:4) result is totally off
(in YUVviewer) for same y4m render.

I looked in libswscale and can't see much handling for specifically

guest at slax:~/botva/src/mplayer/ffmpeg/libswscale$ grep "AV_PIX_FMT_YUV444P"
*.c
input.c:    case AV_PIX_FMT_YUV444P9LE:
input.c:    case AV_PIX_FMT_YUV444P10LE:
input.c:    case AV_PIX_FMT_YUV444P12LE:
input.c:    case AV_PIX_FMT_YUV444P14LE:
input.c:    case AV_PIX_FMT_YUV444P16LE:
input.c:    case AV_PIX_FMT_YUV444P9BE:
input.c:    case AV_PIX_FMT_YUV444P10BE:
input.c:    case AV_PIX_FMT_YUV444P12BE:
input.c:    case AV_PIX_FMT_YUV444P14BE:
input.c:    case AV_PIX_FMT_YUV444P16BE:
input.c:    case AV_PIX_FMT_YUV444P9LE:
input.c:    case AV_PIX_FMT_YUV444P10LE:
input.c:    case AV_PIX_FMT_YUV444P12LE:
input.c:    case AV_PIX_FMT_YUV444P14LE:
input.c:    case AV_PIX_FMT_YUV444P16LE:
input.c:    case AV_PIX_FMT_YUV444P9BE:
input.c:    case AV_PIX_FMT_YUV444P10BE:
input.c:    case AV_PIX_FMT_YUV444P12BE:
input.c:    case AV_PIX_FMT_YUV444P14BE:
input.c:    case AV_PIX_FMT_YUV444P16BE:
swscale_unscaled.c:    if ((srcFormat == AV_PIX_FMT_YUV444P || srcFormat ==
AV_PIX_FMT_YUVA444P) &&
swscale_unscaled.c:    if (dstFormat == AV_PIX_FMT_YUV444P &&
swscale_unscaled.c:        IS_DIFFERENT_ENDIANESS(srcFormat, dstFormat,
AV_PIX_FMT_YUV444P9)  ||
swscale_unscaled.c:        IS_DIFFERENT_ENDIANESS(srcFormat, dstFormat,
AV_PIX_FMT_YUV444P10) ||
swscale_unscaled.c:        IS_DIFFERENT_ENDIANESS(srcFormat, dstFormat,
AV_PIX_FMT_YUV444P12) ||
swscale_unscaled.c:        IS_DIFFERENT_ENDIANESS(srcFormat, dstFormat,
AV_PIX_FMT_YUV444P14) ||
swscale_unscaled.c:        IS_DIFFERENT_ENDIANESS(srcFormat, dstFormat,
AV_PIX_FMT_YUV444P16))
utils.c:    [AV_PIX_FMT_YUV444P]     = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P9BE]  = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P9LE]  = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
utils.c:    [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
utils.c:        *format = AV_PIX_FMT_YUV444P;
utils.c:    case AV_PIX_FMT_YUVA444P:           return AV_PIX_FMT_YUV444P;
utils.c:    case AV_PIX_FMT_YUVA444P9BE:        return AV_PIX_FMT_YUV444P9;
utils.c:    case AV_PIX_FMT_YUVA444P9LE:        return AV_PIX_FMT_YUV444P9;
utils.c:    case AV_PIX_FMT_YUVA444P10BE:       return AV_PIX_FMT_YUV444P10;
utils.c:    case AV_PIX_FMT_YUVA444P10LE:       return AV_PIX_FMT_YUV444P10;
utils.c:    case AV_PIX_FMT_YUVA444P16BE:       return AV_PIX_FMT_YUV444P16;
utils.c:    case AV_PIX_FMT_YUVA444P16LE:       return AV_PIX_FMT_YUV444P16;

it seems input.c does not deal well with specifically this format? There is
no default case in switches ...

in swscale_unscaled I also mostly see specific yuv-nv12 wrappers, not
yuv444 standalone.

util.c is again declares what to do with >8 bpc yuv, mostly.

did I miss something important?

our ffmpeg mappings live at


https://git.cinelerra-gg.org/git/?p=goodguy/cinelerra.git;a=blob;f=cinelerra-5.1/cinelerra/ffmpeg.C;h=9b8832dd718c94dd8b815edc7eb5c9d732953dd6;hb=HEAD

ps: non-ffmpeg png writer seems to output nearly exactly same png if I set
cingg to yuv8 or rgba8 internal processing. So, for same input test file as
long as libswscale did its job once on input -  pipeline itself does not
ruin video, as long as there was no second swscale ? unfortunately we do
not have direct yuv writer, so at the end our yuv444 transformed into rgba
for saving into tga or png or ppm by our routines.


More information about the ffmpeg-user mailing list