Recherche avancée

Sur d’autres sites (6042)

• How does ffmpeg read stdin pipe ?

  17 novembre 2017, par ciclopez

  I’m working on a Python script that generates an image sequence based on user real time remote interaction, and uses ffmpeg to compress and stream this image sequence over the network for the user to watch it.
  As soon as an image is generated the script writes it to ffmpeg’s stdin and this action is repeated inside a loop to form a video.

  I was wondering what happens when ffmpeg is busy processing the previous image when I write to stdin.

  • Does the .stdin.write() command block execution until ffmpeg finishes processing the previous image ? (Consequence of the pipe buffer filling up)
  • What’s the buffer size of subprocess.PIPE ?
  • Is it editable ?
  • Is it possible to overwrite this buffer if it’s full, instead of blocking execution ?
  • Does ffmpeg continually read it’s input in a parallel process and buffers it until it’s free to take care of it ?
  • If that’s the case, what’s the size of ffmpeg’s input buffer and how can I change it ?

  I’m asking this because I want to fully understand how the data travels between Python and ffmpeg to reduce the latency to its minimum.

  Here is the part of the code I’m using to start ffmpeg :

  cmd = ['ffmpeg',
  
      '-f', 'rawvideo',
  
      '-s', '%dx%d'%(width, height),
  
      '-pix_fmt', 'rgba',
  
      '-r', '%d'%fps,
  
      '-i', '-',
  
      '-an',
  
      '-vcodec', 'libx264',
  
      '-tune', 'zerolatency',
  
      '-preset', 'ultrafast',
  
      '-bf', '0',
  
      '-f', 'mpegts',
  
      'udp://xxx.xxx.xxx.xxx:xxxxx']
  
  
  
  proc = subprocess.Popen(cmd, stdin=subprocess.PIPE)

  And this is what I call inside a loop when I want to pass an image to ffmpeg :

  proc.stdin.write(image)

• ffmpeg libx264 neon optimization breaks execution

  27 janvier 2014, par nmxprime

  Hi use libx264 source obtained from x264-snapshot-20140122-2245 and compiling using below script

  NDK=~/Android/android-ndk-r7c
  
  PLATFORM=$NDK/platforms/android-9/arch-arm/
  
  PREBUILT=$NDK/toolchains/arm-linux-androideabi-4.4.3/prebuilt/linux-x86
  
  function build_one
  
  {
  
  ./configure --prefix=$PREFIX \
  
  --sysroot=$PLATFORM \
  
  --disable-avs \
  
  --disable-lavf \
  
  --disable-ffms \
  
  --disable-gpac \
  
  --cross-prefix=$PREBUILT/bin/arm-linux-androideabi- \
  
  --host=arm-linux \
  
  --enable-static \
  
  --libdir=$PLATFORM/usr/lib \
  
  --includedir=$PLATFORM/usr/include \
  
  --extra-cflags="-march=armv7-a -mfloat-abi=softfp -mfpu=neon -mvectorize-with-neon-quad" \
  
  --extra-ldflags="-Wl,--fix-cortex-a8" \
  
  --enable-debug

  The config log is :

  platform:      ARM
  
  system:        LINUX
  
  cli:           yes
  
  libx264:       internal
  
  shared:        no
  
  static:        yes
  
  asm:           yes
  
  interlaced:    yes
  
  avs:           no
  
  lavf:          no
  
  ffms:          no
  
  mp4:           no
  
  gpl:           yes
  
  thread:        no
  
  opencl:        yes
  
  filters:       crop select_every
  
  debug:         yes
  
  gprof:         no
  
  strip:         no
  
  PIC:           no
  
  bit depth:     8
  
  chroma format: all
  
  
  
  You can run 'make' or 'make fprofiled' now.

  I hope the above code compiles and optimizes for NEON execution

  Doubts :

  Why Threads is no, because i didn't specified --disable-thread

  What is cli and it's significance here, also significance of opencl, such that libx264 uses opencl features ?

  ---

  Then i built ffmpeg 1.2.5with following script

  ./configure --target-os=linux \
  
  --prefix=$PREFIX \
  
  --enable-cross-compile \
  
  --extra-libs="-lgcc" \
  
  --arch=arm \
  
  --cc=$PREBUILT/bin/arm-linux-androideabi-gcc \
  
  --cross-prefix=$PREBUILT/bin/arm-linux-androideabi- \
  
  --nm=$PREBUILT/bin/arm-linux-androideabi-nm \
  
  --sysroot=$PLATFORM \
  
  --extra-cflags=" -O3 -fpic -DANDROID -DHAVE_SYS_UIO_H=1 -Dipv6mr_interface=ipv6mr_ifindex -fasm -Wno-psabi -fno-short-enums -fno-strict-aliasing -finline-limit=300 $OPTIMIZE_CFLAGS " \
  
  --disable-shared \
  
  --enable-static \
  
  --extra-ldflags="-Wl,-rpath-link=$PLATFORM/usr/lib -L$PLATFORM/usr/lib -nostdlib -lc -lm -ldl -llog -lx264 $EXTRA_LD_FLAG" \
  
  --disable-ffplay \
  
  --disable-everything \
  
  --enable-avformat \
  
  --enable-avcodec \
  
  --enable-libx264 \
  
  --enable-gpl \
  
  --enable-encoder=libx264 \
  
  --enable-encoder=libx264rgb \
  
  --enable-decoder=h264 \
  
  --disable-network \
  
  --disable-avfilter \
  
  --disable-avdevice \
  
  --enable-debug=3 \
  
  $ADDITIONAL_CONFIGURE_FLAG

  where

  ADDITIONAL_CONFIGURE_FLAG = --enable-debug=3
  
  OPTIMIZE_CFLAGS="-mfloat-abi=softfp -mfpu=neon -marm -march=$CPU -mvectorize-with-neon-quad"

  The log shows NEON supported.

  When i run the code (called in a while loop),

  ret = avcodec_encode_video2(c, &pkt, picture, &got_output);//avcodec_encode_video(c, finalout, outbuf_size, picture);
  
  
  
  fprintf(stderr,"ret = %d, got-out = %d \n",ret,got_output);
  
  if (ret < 0) 
  
          fprintf(stderr, "error encoding frame\n");
  
  
  
      if (got_output) 
  
          fprintf(stderr,"encoding frame %3d (size=%5d): (ret=%d)\n", 1, pkt.size,ret);

  it runs for 2 or 3 time(during which if(got_output) is not true), then i get SIGSEGV. Tried addr2line and ndk-stack but no use[Though i enabled debug info, ndk-stack cannot find routine info].

  I edited libx264's encoder.c code with some fprintf statements.
  Posting snippet of code

     if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
  
      {
  
          /* Nothing yet to encode, waiting for filling of buffers */
  
          pic_out->i_type = X264_TYPE_AUTO;
  
  fprintf(stderr,"EditLog:: Returns as waiting for filling \n"); //edit
  
          return 0;
  
      }
  
  }
  
  
  
  else
  
  {
  
      /* signal kills for lookahead thread */
  
      x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
  
      h->lookahead->b_exit_thread = 1;
  
      x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
  
      x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
  
  }
  
  fprintf(stderr,"After wait for fill \n");
  
  fprintf(stderr,"h: %p \n",h); //edit
  
  fprintf(stderr,"h->i_frame = %p \n",&h->i_frame); //edit
  
  h->i_frame++;
  
  fprintf(stderr,"after i_frame++");

  in log, i don't see after i_frame++, here occurs (may be) the SIGSEGV.

  Please help in solving it. The same works without neon optimization !!

• ffmpeg libx264 neon optimization breaks execution

  27 janvier 2014, par nmxprime

  Hi use libx264 source obtained from x264-snapshot-20140122-2245 and compiling using below script

  NDK=~/Android/android-ndk-r7c
  
  PLATFORM=$NDK/platforms/android-9/arch-arm/
  
  PREBUILT=$NDK/toolchains/arm-linux-androideabi-4.4.3/prebuilt/linux-x86
  
  function build_one
  
  {
  
  ./configure --prefix=$PREFIX \
  
  --sysroot=$PLATFORM \
  
  --disable-avs \
  
  --disable-lavf \
  
  --disable-ffms \
  
  --disable-gpac \
  
  --cross-prefix=$PREBUILT/bin/arm-linux-androideabi- \
  
  --host=arm-linux \
  
  --enable-static \
  
  --libdir=$PLATFORM/usr/lib \
  
  --includedir=$PLATFORM/usr/include \
  
  --extra-cflags="-march=armv7-a -mfloat-abi=softfp -mfpu=neon -mvectorize-with-neon-quad" \
  
  --extra-ldflags="-Wl,--fix-cortex-a8" \
  
  --enable-debug

  The config log is :

  platform:      ARM
  
  system:        LINUX
  
  cli:           yes
  
  libx264:       internal
  
  shared:        no
  
  static:        yes
  
  asm:           yes
  
  interlaced:    yes
  
  avs:           no
  
  lavf:          no
  
  ffms:          no
  
  mp4:           no
  
  gpl:           yes
  
  thread:        no
  
  opencl:        yes
  
  filters:       crop select_every
  
  debug:         yes
  
  gprof:         no
  
  strip:         no
  
  PIC:           no
  
  bit depth:     8
  
  chroma format: all
  
  
  
  You can run 'make' or 'make fprofiled' now.

  I hope the above code compiles and optimizes for NEON execution

  Doubts :

  Why Threads is no, because i didn't specified --disable-thread

  What is cli and it's significance here, also significance of opencl, such that libx264 uses opencl features ?

  ---

  Then i built ffmpeg 1.2.5with following script

  ./configure --target-os=linux \
  
  --prefix=$PREFIX \
  
  --enable-cross-compile \
  
  --extra-libs="-lgcc" \
  
  --arch=arm \
  
  --cc=$PREBUILT/bin/arm-linux-androideabi-gcc \
  
  --cross-prefix=$PREBUILT/bin/arm-linux-androideabi- \
  
  --nm=$PREBUILT/bin/arm-linux-androideabi-nm \
  
  --sysroot=$PLATFORM \
  
  --extra-cflags=" -O3 -fpic -DANDROID -DHAVE_SYS_UIO_H=1 -Dipv6mr_interface=ipv6mr_ifindex -fasm -Wno-psabi -fno-short-enums -fno-strict-aliasing -finline-limit=300 $OPTIMIZE_CFLAGS " \
  
  --disable-shared \
  
  --enable-static \
  
  --extra-ldflags="-Wl,-rpath-link=$PLATFORM/usr/lib -L$PLATFORM/usr/lib -nostdlib -lc -lm -ldl -llog -lx264 $EXTRA_LD_FLAG" \
  
  --disable-ffplay \
  
  --disable-everything \
  
  --enable-avformat \
  
  --enable-avcodec \
  
  --enable-libx264 \
  
  --enable-gpl \
  
  --enable-encoder=libx264 \
  
  --enable-encoder=libx264rgb \
  
  --enable-decoder=h264 \
  
  --disable-network \
  
  --disable-avfilter \
  
  --disable-avdevice \
  
  --enable-debug=3 \
  
  $ADDITIONAL_CONFIGURE_FLAG

  where

  ADDITIONAL_CONFIGURE_FLAG = --enable-debug=3
  
  OPTIMIZE_CFLAGS="-mfloat-abi=softfp -mfpu=neon -marm -march=$CPU -mvectorize-with-neon-quad"

  The log shows NEON supported.

  When i run the code (called in a while loop),

  ret = avcodec_encode_video2(c, &pkt, picture, &got_output);//avcodec_encode_video(c, finalout, outbuf_size, picture);
  
  
  
  fprintf(stderr,"ret = %d, got-out = %d \n",ret,got_output);
  
  if (ret < 0) 
  
          fprintf(stderr, "error encoding frame\n");
  
  
  
      if (got_output) 
  
          fprintf(stderr,"encoding frame %3d (size=%5d): (ret=%d)\n", 1, pkt.size,ret);

  it runs for 2 or 3 time(during which if(got_output) is not true), then i get SIGSEGV. Tried addr2line and ndk-stack but no use[Though i enabled debug info, ndk-stack cannot find routine info].

  I edited libx264's encoder.c code with some fprintf statements.
  Posting snippet of code

     if( h->frames.i_input <= h->frames.i_delay + 1 - h->i_thread_frames )
  
      {
  
          /* Nothing yet to encode, waiting for filling of buffers */
  
          pic_out->i_type = X264_TYPE_AUTO;
  
  fprintf(stderr,"EditLog:: Returns as waiting for filling \n"); //edit
  
          return 0;
  
      }
  
  }
  
  
  
  else
  
  {
  
      /* signal kills for lookahead thread */
  
      x264_pthread_mutex_lock( &h->lookahead->ifbuf.mutex );
  
      h->lookahead->b_exit_thread = 1;
  
      x264_pthread_cond_broadcast( &h->lookahead->ifbuf.cv_fill );
  
      x264_pthread_mutex_unlock( &h->lookahead->ifbuf.mutex );
  
  }
  
  fprintf(stderr,"After wait for fill \n");
  
  fprintf(stderr,"h: %p \n",h); //edit
  
  fprintf(stderr,"h->i_frame = %p \n",&h->i_frame); //edit
  
  h->i_frame++;
  
  fprintf(stderr,"after i_frame++");

  in log, i don't see after i_frame++, here occurs (may be) the SIGSEGV.

  Please help in solving it. The same works without neon optimization !!