[FFmpeg-user] FFMPEG cuda hwaccell error

[Dennis Mungai]

On Tue, 18 Jun 2019 at 02:06, Nishant Trivedi <nishant at ever.ai> wrote:

> That worked! Thanks Dennis!
>
> On Mon, Jun 17, 2019 at 6:37 PM Dennis Mungai <dmngaie at gmail.com> wrote:
>
> > On Tue, 18 Jun 2019 at 01:20, Nishant Trivedi <nishant at ever.ai> wrote:
> >
> > > Hello,
> > >
> > > I'm trying to decode a video stream for one of my projects using the
> > > following command:
> > >
> > > *ffmpeg -hide_banner -hwaccel cuda -nostats -i
> > > rtmp://<ip>:<port>/myapp/mystream -filter_complex [0]fps=fps=1[s0] -map
> > > [s0] -f rawvideo -pix_fmt rgb24 pipe:*
> > >
> > > I'm getting the following error:
> > >
> > > *[h264 @ 0x55e36ef5ca00] Error creating a NVDEC decoder: 1*
> > > *[h264 @ 0x55e36ef5ca00] Using more than 32 (37) decode surfaces might
> > > cause nvdec to fail.*
> > > *[h264 @ 0x55e36ef5ca00] Try lowering the amount of threads. Using 16
> > right
> > > now.*
> > > *[h264 @ 0x55e36ef5ca00] Failed setup for format cuda: hwaccel
> > > initialisation returned error.*
> > >
> > > I traced this error down to this
> > > <https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/nvdec.c#L339>
> > > line
> > > so clearly there's something wrong with my setup but being a newbie in
> > > FFMPEG I don't know what it is. I'd very much appreciate if someone can
> > > help me out with this issue. Full log output follows:
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > *Input #0, flv, from 'rrtmp://<ip>:<port>/myapp/mystream:  Metadata:
> > > major_brand     : qt    minor_version   : 512    compatible_brands: qt
> > > encoder         : Lavf58.20.100    server          :
> > > SRS/2.0.263(ZhouGuowen)    srs_primary     : SRS/2.0release
> > srs_authors
> > >     : winlin,wenjie.zhao    server_version  : 2.0.263  Duration: N/A,
> > > start: 0.010000, bitrate: 2791 kb/s    Stream #0:0: Video: h264 (High),
> > > yuv420p(progressive), 1920x1080, 2791 kb/s, 30 fps, 29.97 tbr, 1k
> > tbnStream
> > > mapping:  Stream #0:0 (h264) -> fps  fps -> Stream #0:0 (rawvideo)Press
> > [q]
> > > to stop, [?] for help[h264 @ 0x55e36ef5ca00] Error creating a NVDEC
> > > decoder: 1[h264 @ 0x55e36ef5ca00] Using more than 32 (37) decode
> surfaces
> > > might cause nvdec to fail.[h264 @ 0x55e36ef5ca00] Try lowering the
> amount
> > > of threads. Using 16 right now.[h264 @ 0x55e36ef5ca00] Failed setup for
> > > format cuda: hwaccel initialisation returned error.Output #0, rawvideo,
> > to
> > > 'pipe:':  Metadata:    major_brand     : qt    minor_version   : 512
> > > compatible_brands: qt    server_version  : 2.0.263    server          :
> > > SRS/2.0.263(ZhouGuowen)    srs_primary     : SRS/2.0release
> > srs_authors
> > >     : winlin,wenjie.zhao    encoder         : Lavf58.20.100    Stream
> > #0:0:
> > > Video: rawvideo (RGB[24] / 0x18424752), rgb24, 1920x1080, q=2-31, 49766
> > > kb/s, 1 fps, 1 tbn, 1 tbc    Metadata:      encoder         :
> > Lavc58.35.100
> > > rawvideo*
> > >
> > > Best,
> > > --
> > > Nishant.
> > >
> > >
> > Hello there,
> >
> > Note that cuda hwaccel is an alias to the newer nvdec implementation, and
> > to my knowledge, has no way of setting up decoder surface limits.
> >
> > Here's a workaround you can try:
> >
> > *ffmpeg -threads 4 -hide_banner -c:v h264_cuvid -surfaces 8 -nostats -i
> > rtmp://<ip>:<port>/myapp/mystream -filter_complex [0]fps=fps=1[s0] -map
> > [s0] -f rawvideo -pix_fmt rgb24 pipe:*
> >
> > Two changes:
> >
> > 1. Call up the h264_cuvid decoder directly, and apply a surface limit
> count
> > of 8. This should be adequate.
> > 2. Limit the thread count.
> > _______________________________________________
> > ffmpeg-user mailing list
> > ffmpeg-user at ffmpeg.org
> > https://ffmpeg.org/mailman/listinfo/ffmpeg-user
> >
> > To unsubscribe, visit link above, or email
> > ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
>
>
>
> --
> --
> Nishant.
>
>
You're welcome :-)

Do reach out if you need more assistance, thanks.