On Sun, Nov 12, 2017 at 10:30 AM, Hendrik Leppkes <h.lepp...@gmail.com> wrote: > --- > libavcodec/nvenc.c | 106 > ++++++++++++++++++++++++++++++++++++++++++----------- > libavcodec/nvenc.h | 11 +++++- > 2 files changed, 95 insertions(+), 22 deletions(-) >
Some details, if anyone cares (tested with 1080p) On an idle system, performance of d3d11 decode, hwdownload, nvenc is largely similar to a d3d11 -> nvenc direct connection, however once the GPU is busy (I was testing with a 3D rendering going on last night), hwdownload quickly degrades in performance, while the direct connection didn't suffer at all. Additionally, a sharp drop in CPU usage can also be observed. In case anyone is interested, I plan on working on a D3D11 VPP filter in the future to allow scaling and deinterlacing in this particular workflow, using D3D11 Video Processor APIs. - Hendrik _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
