Optimize JPEG scanline copy of yuv format

[chraac]

Summary

In our current source, when encoding JPEG with CPU we'll first convert the yuv pixel int rgb, but from jpeg spec, encoder using YCbCr which is yuv internally (see also: JPEG#Encoding), this means we can passthrough the yuv value to encoder directly without converting to RGB.

Changes

• For V4L2_PIX_FMT_YUYV and V4L2_PIX_FMT_UYVY copy the yuv value directly into scanline buffer instead of convert to rgb

Test

• Run on my RK3399 board with asan enabled, no error found

• Performance test

  • compiling flag: -O3 -march='armv8-a' -mtune='cortex-a72.cortex-a53
  • running params: ./ustreamer --device=/dev/video0 -r 1280x720 --quality=80 --encoder=CPU

  \	Before	After
  latency	0.231~0.301	0.201
  cpu usage

[argakon]

It is actually really good performance optimisation. I have Elgato Camlink 4k stick and with pixel format YUYV and 1920x1080 50fps i have more than twice cpu utilisation difference. It uses only one core right now on RK3588, was 2,5 cores.

[argakon]

I've added NV12 format support in my repo. It also should close #97 issue. But it converts frame into RGB24, i tried to pass interpolated YUV data from NV12 directly into jpeg_write_scanlines with YCbCr, but no success, could i ask you to look in it?

[chraac]

I've added NV12 format support in my repo. It also should close #97 issue. But it converts frame into RGB24, i tried to pass interpolated YUV data from NV12 directly into jpeg_write_scanlines with YCbCr, but no success, could i ask you to look in it?

Sure, leave some comment on it, let me known if have any question about

[mdevaev]

Please resolve a conflict.

[chraac]

Please resolve a conflict.

Resolved!