Recherche avancée

Sur d’autres sites (9634)

• arm : vp9 : Add NEON loop filters

  10 octobre 2016, par Martin Storsjö

  arm : vp9 : Add NEON loop filters
  This work is sponsored by, and copyright, Google.
  The implementation tries to have smart handling of cases
  
  where no pixels need the full filtering for the 8/16 width
  
  filters, skipping both calculation and writeback of the
  
  unmodified pixels in those cases. The actual effect of this
  
  is hard to test with checkasm though, since it tests the
  
  full filtering, and the benefit depends on how many filtered
  
  blocks use the shortcut.
  Examples of relative speedup compared to the C version, from checkasm :
  
                            Cortex       A7     A8     A9    A53
  
  vp9_loop_filter_h_4_8_neon :          2.72   2.68   1.78   3.15
  
  vp9_loop_filter_h_8_8_neon :          2.36   2.38   1.70   2.91
  
  vp9_loop_filter_h_16_8_neon :         1.80   1.89   1.45   2.01
  
  vp9_loop_filter_h_16_16_neon :        2.81   2.78   2.18   3.16
  
  vp9_loop_filter_mix2_h_44_16_neon :   2.65   2.67   1.93   3.05
  
  vp9_loop_filter_mix2_h_48_16_neon :   2.46   2.38   1.81   2.85
  
  vp9_loop_filter_mix2_h_84_16_neon :   2.50   2.41   1.73   2.85
  
  vp9_loop_filter_mix2_h_88_16_neon :   2.77   2.66   1.96   3.23
  
  vp9_loop_filter_mix2_v_44_16_neon :   4.28   4.46   3.22   5.70
  
  vp9_loop_filter_mix2_v_48_16_neon :   3.92   4.00   3.03   5.19
  
  vp9_loop_filter_mix2_v_84_16_neon :   3.97   4.31   2.98   5.33
  
  vp9_loop_filter_mix2_v_88_16_neon :   3.91   4.19   3.06   5.18
  
  vp9_loop_filter_v_4_8_neon :          4.53   4.47   3.31   6.05
  
  vp9_loop_filter_v_8_8_neon :          3.58   3.99   2.92   5.17
  
  vp9_loop_filter_v_16_8_neon :         3.40   3.50   2.81   4.68
  
  vp9_loop_filter_v_16_16_neon :        4.66   4.41   3.74   6.02
  The speedup vs C code is around 2-6x. The numbers are quite
  
  inconclusive though, since the checkasm test runs multiple filterings
  
  on top of each other, so later rounds might end up with different
  
  codepaths (different decisions on which filter to apply, based
  
  on input pixel differences). Disabling the early-exit in the asm
  
  doesn’t give a fair comparison either though, since the C code
  
  only does the necessary calcuations for each row.
  Based on START_TIMER/STOP_TIMER wrapping around a few individual
  
  functions, the speedup vs C code is around 4-9x.
  This is pretty similar in runtime to the corresponding routines
  
  in libvpx. (This is comparing vpx_lpf_vertical_16_neon,
  
  vpx_lpf_horizontal_edge_8_neon and vpx_lpf_horizontal_edge_16_neon
  
  to vp9_loop_filter_h_16_8_neon, vp9_loop_filter_v_16_8_neon
  
  and vp9_loop_filter_v_16_16_neon - note that the naming of horizonal
  
  and vertical is flipped between the libraries.)
  In order to have stable, comparable numbers, the early exits in both
  
  asm versions were disabled, forcing the full filtering codepath.
                             Cortex           A7      A8      A9     A53
  
  vp9_loop_filter_h_16_8_neon :             597.2   472.0   482.4   415.0
  
  libvpx vpx_lpf_vertical_16_neon :         626.0   464.5   470.7   445.0
  
  vp9_loop_filter_v_16_8_neon :             500.2   422.5   429.7   295.0
  
  libvpx vpx_lpf_horizontal_edge_8_neon :   586.5   414.5   415.6   383.2
  
  vp9_loop_filter_v_16_16_neon :            905.0   784.7   791.5   546.0
  
  libvpx vpx_lpf_horizontal_edge_16_neon : 1060.2   751.7   743.5   685.2
  Our version is consistently faster on on A7 and A53, marginally slower on
  
  A8, and sometimes faster, sometimes slower on A9 (marginally slower in all
  
  three tests in this particular test run).
  Signed-off-by : Martin Storsjö <martin@martin.st>
  

  • [DBH] libavcodec/arm/Makefile
  • [DBH] libavcodec/arm/vp9dsp_init_arm.c
  • [DBH] libavcodec/arm/vp9lpf_neon.S

• rtmp audio out of sync, http works fine

  21 janvier 2014, par marca

  We have encoded and distributed videos for some years now, using FFMPEG to produce h.264/mp4 files that have been working great for us. We have been using HTML mode and fall-backed to flash for browsers that does not support it natively using flowplayer.

  We use cloudfront to serve our files from a s3 bucket and have been using http progressive streaming.

  Recently we started distribute the files in flashmode over rtmp instead, using a cloudfront streaming distribution pointing to the same amazon s3 bucket.

  All good for some weeks, until yesterday when we notice a couple of files with audio sync issues in rtmp mode.
  The same file have no sync problems in flash with direct url to file.

  What can be the case ?

  Not working when streamed via RTMP, but file work with http streaming/progressive.
  You see the sync issue 15 sec's into the video.
  rtmp ://s2xe2avk54qztf.cloudfront.net:1935/cfx/st/mp4:95fvOY255bdPspO3z6tEvGi3Em7/default.mp4
  http://media.shootitlive.com/95fvOY255bdPspO3z6tEvGi3Em7/default.mp4

  Another file that have no sync issue at all.
  rtmp ://s2xe2avk54qztf.cloudfront.net:1935/cfx/st/mp4:P4EuH2TZxfV6BvpupP6dxrrs7gD/default.mp4
  http://media.shootitlive.com/P4EuH2TZxfV6BvpupP6dxrrs7gD/default.mp4

  Both files have the same format for video and audio and have been encoded the exact same way with ffmpeg. It's not player related as we see the audio sync issue on several players and when playing stream in VLC.

• AWS Lambda execution time for FFMPEG transcoding

  4 janvier 2023, par FlamingMoe

  I'm using AWS Lambda for converting files from WEBM to MP4

  


  I'm using ffmpeg version 4.3.1-static https://johnvansickle.com/ffmpeg/ (I have done the following tests also with the ffmpeg in serverless AWS ffmpeg layer (that includes de 4.1.3), but results are even worse (about 25% slower)

  


  I'm using Node 10x as container.

  


  WEBM size   Time to convert.  Memory Lambda.  Memory used (as shown in log)

80Mb             ~44s              3008            410
40Mb             ~44s              3008            375

80Mb             ~70s              1024            321
40Mb             ~70s              1024            279


  


  All videos are 80s length. So as far as I can see, it does not matter the size of the WEBM, if the length of the video is the same, it takes the same to convert. So ffmpeg takes more time if the video length is higher, not if the file size is higher ... curious ;-)

  


  But in the other hand, I'm confused with Lambda memory. I know memory and CPU comes together in Lambda ... the more memory you choose, the more CPU is assigned.

  


  But...

  


  

  1. Why ffmpeg just take about 300/400Mb if it has more to run ?

  


  2. How can I tell ffmpeg to use more memory ?

  


  3. Is there any option to accelerate the process in Lambda ?

  


  


  Btw, In all tests, all ffmpeg are the same, and

  


  cpu-used paramenter)

  


  

  • I added to ffmpeg parameters cpu-used=100, and it does not matter at all if I put cpu-used=5 ... times are the same, so I guess that parameter is useless (i don't know why)

  


  


  threads parameter)

  


  

  • Also I did some tests with "threads" parameters, but it's useless also.

  


  


  I know it's not a good comparison, but same files takes about 5 seconds to be converted in a simple dedicated server (8 vCores and 8GB RAM in OVH Centos VPS).

  


  Btw, Amazon Elastic Transcoder is not an option :
a) it's extremely more expensive
b) it has just his profiles to convert, and my ffmpeg commands are very complex (watermarks, effects, etc ...)