search

MythTV / mythtv

The official MythTV repository
https://www.mythtv.org
GNU General Public License v2.0
708 stars 346 forks source link

Regression in V4L2 M2M Prime decoding #209

Closed mark-kendall closed 3 years ago

mark-kendall commented 4 years ago

a1e378ad431c74c029fb4aba0bc34194838672e1 updated the V4L2 M2M decoder for the latest patch set.

No issues noted with RPI4 but @warpme reports a regression in decoder performance on s905 based boards.

My only suspect is the use of V4L2_MEMORY_DMABUF instead of V4L2_MEMORY_MMAP

mark-kendall commented 4 years ago

Test patch: https://gist.github.com/mark-kendall/bb9923ade6a41d3e7d901b8d8dc8b0e8

warpme commented 4 years ago

Mark - unfortunately test patch not helps. issue is still present.

warpme commented 4 years ago

Summarise my experience with current master (as of b3d42cff2):

  1. with a1e378a I have

    a\ issue of randomly skipping frames on s905 when v4l2 m2m decode is used; b\ s912/sm1 playabck hangs.

  2. reverting a1e378a fixes a\ and b\
  3. applying https://gist.github.com/mark-kendall/bb9923ade6a41d3e7d901b8d8dc8b0e8 fixes b\ but not a\
mark-kendall commented 3 years ago

@warpme I decided the easiest approach was to just to revert to the original implementation. The updated patches offered some theoretical improvements/fixes but I couldn't see any actual improvements and I simply don't have the time to dig into the regressions.

Closing this to focus on the request API.