You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Version of mpv that introduced the problem: commit e3af545
Reproduction steps
While setting up fuzzing of mpv, I noticed that execution with benchmark options (--untimed and friends) is significantly slower in current master than release 0.37.0. git bisect located the problem at commit e3af545.
user@laptop:~/mpv-e3af545$ hyperfine './build/mpv --vo=null --untimed --ao=null --ao-null-untimed ~/9bitwhite.mkv'
Benchmark 1: ./build/mpv --vo=null --untimed --ao=null --ao-null-untimed ~/9bitwhite.mkv
Time (mean ± σ): 515.6 ms ± 5.0 ms [User: 216.0 ms, System: 51.4 ms]
Range (min … max): 509.9 ms … 527.5 ms 10 runs
user@laptop:~/mpv-e3af545$ hyperfine './build/mpv --untimed ~/9bitwhite.mkv'
Benchmark 1: ./build/mpv --untimed ~/9bitwhite.mkv
Time (mean ± σ): 1.951 s ± 0.018 s [User: 0.342 s, System: 0.107 s]
Range (min … max): 1.932 s … 1.982 s 10 runs
A 9 second video file took roughly 0.5s to play with vo=null and --untimed options. With normal video output and --untimed option only, it took roughly 1.9s to play.
user@laptop:~/mpv-7051e94$ hyperfine './build/mpv --vo=null --untimed --ao=null --ao-null-untimed ~/9bitwhite.mkv'
Benchmark 1: ./build/mpv --vo=null --untimed --ao=null --ao-null-untimed ~/9bitwhite.mkv
Time (mean ± σ): 101.7 ms ± 3.1 ms [User: 206.8 ms, System: 53.0 ms]
Range (min … max): 95.7 ms … 108.8 ms 27 runs
user@laptop:~/mpv-7051e94$ hyperfine './build/mpv --untimed ~/9bitwhite.mkv'
Benchmark 1: ./build/mpv --untimed ~/9bitwhite.mkv
Time (mean ± σ): 1.618 s ± 0.019 s [User: 0.375 s, System: 0.093 s]
Range (min … max): 1.593 s … 1.650 s 10 runs
The same 9 second video file took roughly 0.1s to play with vo=null and --untimed options. With normal video output and --untimed option only, it took roughly 1.6s to play.
There is a 5x performance regression in --untimed between these two commits.
Expected behavior
Do not sleep when outputting video frames with --untimed enabled.
Actual behavior
Additional delay other than decoding cost introduced between frames even when --untimed is enabled.
Ended up being too flawed and caused trouble in other areas. There's
other approaches to trying to solve the issue this meant to address in
the works that should be better, so let's wait on that. Fixesmpv-player#13613 and
fixesmpv-player#13622.
This reverts commit e3af545.
Important information
Version information:
Reproduction steps
While setting up fuzzing of mpv, I noticed that execution with benchmark options (
--untimed
and friends) is significantly slower in current master than release 0.37.0.git bisect
located the problem at commit e3af545.First bad commit:
--untimed
optionA 9 second video file took roughly 0.5s to play with
vo=null
and--untimed
options. With normal video output and--untimed
option only, it took roughly 1.9s to play.Last good commit:
--untimed
optionThe same 9 second video file took roughly 0.1s to play with
vo=null
and--untimed
options. With normal video output and--untimed
option only, it took roughly 1.6s to play.There is a 5x performance regression in
--untimed
between these two commits.Expected behavior
Do not sleep when outputting video frames with
--untimed
enabled.Actual behavior
Additional delay other than decoding cost introduced between frames even when
--untimed
is enabled.Sample files
The sample file used in reproduction steps: https://github.com/strongcourage/fuzzing-corpus/blob/master/mkv/9bitwhite.mkv
The issue seems to be independent of the input file.
The text was updated successfully, but these errors were encountered: