[FFmpeg-devel] [PATCH 1/3] avfilter/vf_framerate: factorize SAD functions which compute SAD for a whole frame

[James Almer]

On 11/4/2018 9:06 AM, Marton Balint wrote:
> Also add SIMD which works on lines because it is faster then calculating it on
> 8x8 blocks using pixelutils.
> 
> Signed-off-by: Marton Balint <cus at passwd.hu>
> ---
>  configure                        |  3 +-
>  libavfilter/Makefile             |  1 +
>  libavfilter/framerate.h          |  4 +--
>  libavfilter/scene_sad.c          | 72 ++++++++++++++++++++++++++++++++++++++
>  libavfilter/scene_sad.h          | 44 ++++++++++++++++++++++++
>  libavfilter/vf_framerate.c       | 61 ++++-----------------------------
>  libavfilter/x86/Makefile         |  4 +++
>  libavfilter/x86/scene_sad.asm    | 74 ++++++++++++++++++++++++++++++++++++++++
>  libavfilter/x86/scene_sad_init.c | 52 ++++++++++++++++++++++++++++
>  9 files changed, 257 insertions(+), 58 deletions(-)
>  create mode 100644 libavfilter/scene_sad.c
>  create mode 100644 libavfilter/scene_sad.h
>  create mode 100644 libavfilter/x86/scene_sad.asm
>  create mode 100644 libavfilter/x86/scene_sad_init.c
> 
> diff --git a/configure b/configure
> index 2606b885b0..f5bec9fd62 100755
> --- a/configure
> +++ b/configure
> @@ -2335,6 +2335,7 @@ CONFIG_EXTRA="
>      rtpdec
>      rtpenc_chain
>      rv34dsp
> +    scene_sad
>      sinewin
>      snappy
>      srtp
> @@ -3395,7 +3396,7 @@ find_rect_filter_deps="avcodec avformat gpl"
>  firequalizer_filter_deps="avcodec"
>  firequalizer_filter_select="rdft"
>  flite_filter_deps="libflite"
> -framerate_filter_select="pixelutils"
> +framerate_filter_select="scene_sad"
>  frei0r_filter_deps="frei0r libdl"
>  frei0r_src_filter_deps="frei0r libdl"
>  fspp_filter_deps="gpl"
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index 51e48efc2e..390c2b7997 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -28,6 +28,7 @@ OBJS-$(HAVE_THREADS)                         += pthread.o
>  OBJS-$(CONFIG_QSVVPP)                        += qsvvpp.o
>  DNN-OBJS-$(CONFIG_LIBTENSORFLOW)             += dnn_backend_tf.o
>  OBJS-$(CONFIG_DNN)                           += dnn_interface.o dnn_backend_native.o $(DNN-OBJS-yes)
> +OBJS-$(CONFIG_SCENE_SAD)                     += scene_sad.o
>  
>  # audio filters
>  OBJS-$(CONFIG_ABENCH_FILTER)                 += f_bench.o

Can't you expand pixeutils instead? Adding a new set of functions that
work on lines rather than blocks.