[Libav-user] How to use an existing CUDA context for decoding

Oscar Amoros Huguet oamoros at mediapro.tv
Wed May 23 23:10:56 EEST 2018


It is possible to do what you want to do. We do that, and more. Have a look at the ffmpeg-dev forums. There are some pretty interesting discussions about how to use your own cuda context, and even your own cuda stream.

Good luck!


El 23 maig 2018, a les 8:24, joep <joep at groovytunes.nl<mailto:joep at groovytunes.nl>> va escriure:


I am trying to create a program that does the following:
1. decode frames with cuvid
2. run CUDA kernels on the decoded frames
3. encode result with nvenc

In order to do step 2, I need the CUDA kernels to access the decoded frame which should be stored in the GPU memory.
For this I think I need to provide the CUDA CUcontext to ffmpeg and give the pointer to the decoded frame to the kernel.

I could not find any documentation about this topic.
In the ffmpeg docs I see the AVCodecContext has hw_device_ctx which is a AVHWDeviceContext.
AVHWDeviceContext has a hwctx field which can be a AVCUDADeviceContext.
AVCUDADeviceContext has a cuda_ctx field which is a CUcontext.

So I think I need to create a AVHWDeviceContext and set it on the AVCodecContext before calling avcodec_open2.
Are my assumptions right?
And how can I create a AVHWDeviceContext with the correct hwctx?

Libav-user mailing list
Libav-user at ffmpeg.org<mailto:Libav-user at ffmpeg.org>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://ffmpeg.org/pipermail/libav-user/attachments/20180523/a162cfc1/attachment.html>

More information about the Libav-user mailing list