Recherche avancée

Sur d’autres sites (8172)

• why is ffmpeg so fast

  8 juillet 2017, par jx.xu

  I have written a ffmpeg-based C++ program about converting yuv to rgb using libswscale, similar to the official example.
  Just simple copy and modify the official example before build in my visual studio 2017 on windows 10. However, the time performance is much slower than the ffmpeg.exe executable file, as 36 seconds vs 12 seconds.
  I already know that ffmpeg uses some optimization techniques like SIMD instructions. While in my performance profiling, the bottleneck is disk I/O writing which takes at least 2/3 time.
  Then I develop a concurrent version where a dedicated thread will handle all I/O task while the situation doesn’t seem to improve. It’s worth noting that I use Boost C++ library to utilize multi-thread and asynchronous events.

  So, I just wanna know how can I modify program using the libraries of ffmpeg or the time performance gap towards ffmpeg.exe just can’ be catched up.

  As requested by friendly answers, I post my codes. Compiler is msvc in vs2017 and I turn on the full optimization /Ox .

  Supplement my main question, I make another plain disk I/O test merely copying the file of the same size. It’s surprising to find that plain sequential disk I/O costs 28 seconds while the front codes cost 36 seconds in total... Any one knows how can ffmpeg finishes the same job in only 12 seconds ? That must use some optimization techniques, like random disk I/O or memory buffer reusing ?

  #include "stdafx.h"
  
  #define __STDC_CONSTANT_MACROS
  
  extern "C" {
  
  #include <libavutil></libavutil>imgutils.h>
  
  #include <libavutil></libavutil>parseutils.h>
  
  #include <libswscale></libswscale>swscale.h>
  
  }
  
  
  
  #ifdef _WIN64
  
  #pragma comment(lib, "avformat.lib")
  
  #pragma comment(lib, "avcodec.lib")
  
  #pragma comment(lib, "avutil.lib")
  
  #pragma comment(lib, "swscale.lib")
  
  #endif
  
  
  
  #include <common></common>cite.hpp>   // just include headers of c++ STL/Boost
  
  
  
  int main(int argc, char **argv)
  
  {
  
      chrono::duration<double> period;
  
      auto pIn = fopen("G:/Panorama/UHD/originalVideos/DrivingInCountry_3840x1920_30fps_8bit_420_erp.yuv", "rb");
  
      auto time_mark = chrono::steady_clock::now();
  
  
  
      int src_w = 3840, src_h = 1920, dst_w, dst_h;
  
      enum AVPixelFormat src_pix_fmt = AV_PIX_FMT_YUV420P, dst_pix_fmt = AV_PIX_FMT_RGB24;
  
      const char *dst_filename = "G:/Panorama/UHD/originalVideos/out.rgb";
  
      const char *dst_size = "3840x1920";
  
      FILE *dst_file;
  
      int dst_bufsize;
  
      struct SwsContext *sws_ctx;
  
      int i, ret;
  
      if (av_parse_video_size(&amp;dst_w, &amp;dst_h, dst_size) &lt; 0) {
  
          fprintf(stderr,
  
              "Invalid size '%s', must be in the form WxH or a valid size abbreviation\n",
  
              dst_size);
  
          exit(1);
  
      }
  
      dst_file = fopen(dst_filename, "wb");
  
      if (!dst_file) {
  
          fprintf(stderr, "Could not open destination file %s\n", dst_filename);
  
          exit(1);
  
      }
  
      /* create scaling context */
  
      sws_ctx = sws_getContext(src_w, src_h, src_pix_fmt,
  
          dst_w, dst_h, dst_pix_fmt,
  
          SWS_BILINEAR, NULL, NULL, NULL);
  
      if (!sws_ctx) {
  
          fprintf(stderr,
  
              "Impossible to create scale context for the conversion "
  
              "fmt:%s s:%dx%d -> fmt:%s s:%dx%d\n",
  
              av_get_pix_fmt_name(src_pix_fmt), src_w, src_h,
  
              av_get_pix_fmt_name(dst_pix_fmt), dst_w, dst_h);
  
          ret = AVERROR(EINVAL);
  
          exit(1);
  
      }
  
      io_service srv;   // Boost.Asio class
  
      auto work = make_shared(srv); 
  
      thread t{ bind(&amp;io_service::run,&amp;srv) }; // I/O worker thread
  
      vector> result;
  
      /* utilize function class so that lambda can capture itself */
  
      function)> recursion;  
  
      recursion = [&amp;](int left, unique_future<bool> writable)   
  
      {
  
          if (left &lt;= 0)
  
          {
  
              writable.wait();
  
              return;
  
          }
  
          uint8_t *src_data[4], *dst_data[4];
  
          int src_linesize[4], dst_linesize[4];
  
          /* promise-future pair used for thread synchronizing so that the file part is written in the correct sequence  */
  
          promise<bool> sync; 
  
          /* allocate source and destination image buffers */
  
          if ((ret = av_image_alloc(src_data, src_linesize,
  
              src_w, src_h, src_pix_fmt, 16)) &lt; 0) {
  
              fprintf(stderr, "Could not allocate source image\n");
  
          }
  
          /* buffer is going to be written to rawvideo file, no alignment */
  
          if ((ret = av_image_alloc(dst_data, dst_linesize,
  
              dst_w, dst_h, dst_pix_fmt, 1)) &lt; 0) {
  
              fprintf(stderr, "Could not allocate destination image\n");
  
          }
  
          dst_bufsize = ret;
  
          fread(src_data[0], src_h*src_w, 1, pIn);
  
          fread(src_data[1], src_h*src_w / 4, 1, pIn);
  
          fread(src_data[2], src_h*src_w / 4, 1, pIn);
  
          result.push_back(async([&amp;] {
  
              /* convert to destination format */
  
              sws_scale(sws_ctx, (const uint8_t * const*)src_data,
  
                  src_linesize, 0, src_h, dst_data, dst_linesize);
  
              if (left>0)
  
              {
  
                  assert(writable.get() == true);
  
                  srv.post([=]
  
                  {
  
                      /* write scaled image to file */
  
                      fwrite(dst_data[0], 1, dst_bufsize, dst_file);
  
                      av_freep((void*)&amp;dst_data[0]);
  
                  });
  
              }
  
              sync.set_value(true);
  
              av_freep(&amp;src_data[0]);
  
          }));
  
          recursion(left - 1, sync.get_future());
  
      };
  
  
  
      promise<bool> root;
  
      root.set_value(true);
  
      recursion(300, root.get_future());  // .yuv file only has 300 frames
  
      wait_for_all(result.begin(), result.end()); // wait for all unique_future to callback
  
      work.reset();    // io_service::work releses
  
      srv.stop();      // io_service stops
  
      t.join();        // I/O thread joins
  
      period = steady_clock::now() - time_mark;  // calculate valid time
  
      fprintf(stderr, "Scaling succeeded. Play the output file with the command:\n"
  
          "ffplay -f rawvideo -pix_fmt %s -video_size %dx%d %s\n",
  
          av_get_pix_fmt_name(dst_pix_fmt), dst_w, dst_h, dst_filename);
  
      cout &lt;&lt; "period" &lt;&lt; et &lt;&lt; period &lt;&lt; en;
  
  end:
  
      fclose(dst_file);
  
      //  av_freep(&amp;src_data[0]);
  
      //  av_freep(&amp;dst_data[0]);
  
      sws_freeContext(sws_ctx);
  
      return ret &lt; 0;
  
  }
  
  </bool></bool></bool></double>

• avformat : add AV1 RTP depacketizer and packetizer

  26 août 2024, par Chris Hodges

  avformat : add AV1 RTP depacketizer and packetizer
  Add RTP packetizer and depacketizer according to (most)
  
  of the official AV1 RTP specification. This enables
  
  streaming via RTSP between ffmpeg and ffmpeg and has
  
  also been tested to work with AV1 RTSP streams via
  
  GStreamer.
  It also adds the required SDP attributes for AV1.
  AV1 RTP encoding is marked as experimental due to
  
  draft specification status, debug amount reduced
  
  and other changes suggested by Tristan.
  Added optional code for searching the sequence
  
  header to determine the first packet for broken
  
  AV1 encoders / parsers.
  Stops depacketizing on corruption until next keyframe,
  
  no longer prematurely issues packet on decoding if
  
  temporal unit was not complete yet.
  Change-Id : I90f5c5b9d577908a0d713606706b5654fde5f910
  
  Signed-off-by : Chris Hodges <chrishod@axis.com>
  
  Signed-off-by : Ronald S. Bultje <rsbultje@gmail.com>
  

  • [DH] libavformat/Makefile
  • [DH] libavformat/demux.c
  • [DH] libavformat/rtp_av1.h
  • [DH] libavformat/rtpdec.c
  • [DH] libavformat/rtpdec_av1.c
  • [DH] libavformat/rtpdec_formats.h
  • [DH] libavformat/rtpenc.c
  • [DH] libavformat/rtpenc.h
  • [DH] libavformat/rtpenc_av1.c
  • [DH] libavformat/sdp.c

• Developers and vendors : Want a Matomo Hoodie ? Add a tag to the Matomo Open Source Tag Manager and this could be yours !

  7 juin 2018, par Matomo Core Team — Community, Development

  The Free Open Source Tag Manager is now available as a public beta on the Matomo Marketplace. Don’t know what a Tag Manager is ? Learn more here. In Short : It lets you easily manage all your third party JavaScript and HTML snippets (analytics, ads, social media, remarketing, affiliates, etc) through a single interface.

  Over the last few months we have worked on building the core for the Matomo Tag Manager which comes with a great set of features and a large set of pre-configured triggers and variables. However, we currently lack tags.

  This is where we need your help ! Together we can build a complete and industry leading open source tag manager.

  Tag examples include Google AdWords Conversion Tracking, Facebook Buttons, Facebook Pixels, Twitter Universal Website Tags, LinkedIn Insights.

  Are you a developer who is familiar with JavaScript and keen on adding a tag ? Or are you a vendor ? Don’t be shy, we appreciate any tags, even analytics related :) We have documented how to develop a new tag here, which is quite easy and straightforward. You may also need to understand a tiny bit of PHP but you’ll likely be fine even if you don’t (here is an example PHP file and the related JS file).

  As we want to ship the Matomo Tag Manager with as many tags as possible out of the box, we appreciate any new tag additions as a pull request on https://github.com/matomo-org/tag-manager.

  We will send out “Matomo Contributor” stickers that cannot be purchased anywhere for every contributor who contributes a tag within the next 3 months. As for the top 3 contributors… you’ll receive a Matomo hoodie ! Simply send us an email at hello@matomo.org after your tag has been merged. If needed, a draw will decide who gets the hoodies.

  FYI : The Matomo Tag Manager is already prepared to be handled in different contexts and we may possibly generate containers for Android and iOS. If you are keen on building the official Matomo SDKs for any of these mobile platforms, please get in touch.