How to know if FFmpegFrameGrabber uses GPU for decoding mp4

[thhart]

I have looked around and tried to include almost everything related to cuda, I can see cuda libraries are loaded, however still I am not able to know if FFmpegFrameGrabber uses GPU or not. Is there a simple approach to check this?

I have included this at least:

   <dependency>
        <groupId>org.bytedeco</groupId>
        <artifactId>cuda-platform-redist</artifactId>
        <version>11.8-8.6-1.5.8</version>
    </dependency>

Code looks like this:

   FFmpegFrameGrabber frameGrabber = new FFmpegFrameGrabber(fileInView.getAbsolutePath());
		try {
			frameGrabber.setOption("hwaccel", "cuvid");
			frameGrabber.setOption("hwaccel_device", "0");

[saudet]

Please make sure that FFmpegLogCallback.set() has been called.

[saudet]

Doesn't contain anything related to CUDA, so I guess not. Usually we need to set the codec manually with setVideoCodecName().

[thhart]

There is some interesting behaviour, when do the following:

recorder.setVideoCodecName("h264");

I get:

org.bytedeco.javacv.FFmpegFrameGrabber$Exception: avcodec_open2() error -22: Could not open video codec. (For more details, make sure FFmpegLogCallback.set() has been called.)

There is full log set, but there is nothing more seen than the standard output.

When I put some nonsense there:

recorder.setVideoCodecName("foo");

I get no error whatsoever, decoding works but still I don't know if Cuda/GPU is used or not.

[saudet]

Try with the names given here:
https://trac.ffmpeg.org/wiki/HWAccelIntro#CUDANVENCNVDEC

For "hwaccel", I'd try setVideoOption(), not setOption().

[thhart]

Hi Samuel, I appreciate your input but trying things here and there is a bit frustrating. In the meantime I monitored the GPU (nvtop) while transcoding with JavaCV and it looks like it is working with the GPU. However a better way of control would be nice as well a better feedback which options are working in the underlying ffmpeg structures. I hope this can be seen as enhancement request, I really appreciate all the work in the whole framework as such and what is in there to have this level of performance.

[saudet]

Sure thing! Contributions are welcome