Invalid argument after calling gpuConvertYUYVtoRGB_kernel in 18_v4l2_camera_cuda_rgb

Autonomous Machines Jetson & Embedded Systems Jetson Orin NX
1

HI, I am using Jetson Orin Nx 8G with Jetpack 5.1.2 to test convert USB camera YUYV frame to RGB format. The application is Jetson_multimeida_api/samples/18_v4l2_camera_cuda_rgb.

I checked the return from cuda function gpuConvertYUYVtoRGB_kernel like:

cudaError_t ret = cudaGetLastError();
if ( ret != cudaSuccess){
printf(“%s \n”,cudaGetErrorString(ret));
}
when the command I used was like “./v4l2_camera_cuda_rgb -d /dev/video2 -f YUYV -o out.ppm -s 1920x1080 -m”, the “invalid argument” has been reported. But if I choose user pointer with zero copy “-u -z”, the function return cudaSuccess.

Can someone figure it out? thanks

3

Hi,
Please try the sample and see if you can capture frames successfully:

/usr/src/jetson_multimedia_api/samples/12_v4l2_camera_cuda/

It is the other sample for capturing frame data through v4l2. Would like to make sure this works first.

And please ensure the samples are installed through SDKManager or sudo apt-get install command. We have seen issues that version of the samples does not match release version. In the condition, header files may be mismatched and trigger some issues.

4

Thanks DaneLLL, yes, 12_v4l2_camera_cuda is able to capture frame and render. I installed samples through SDKManager.

What I want to test is to convert YUYV to RGB via cuda. But the cuda function --gpuConvertYUYVtoRGB_kernel return error like “invalid argument” if I choose memory copy mode while it work properly if the mode is user pointer. So I think it would be something wrong happened in pinned memory shared with CPU and GPU.

So could you have a look at?Thanks

5

Hi,
Looks like the issue occurs when -m is set. We will check if this mode can be supported.

6

I really appreciate if you could figure it out. Looking forward further feedback