A strange issue of convertor in MMAPI

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hi, I encountered a strange issue of convertor in MMAPI.

When i modify the output data of convertor, it will affect the future output.
I have modifed “00_video_decode” to reproduce this issue.
In “conv0_capture_dqbuf_thread_callback” function, my modification as follows:

static bool
conv0_capture_dqbuf_thread_callback(struct v4l2_buffer *v4l2_buf,
                                    NvBuffer * buffer, NvBuffer * shared_buffer,
                                    void *arg)
{
    context_t *ctx = (context_t *) arg;

    if (!v4l2_buf)
    {
        cerr << "Error while dequeueing conv capture plane buffer" << endl;
        abort(ctx);
        return false;
    }

    if (v4l2_buf->m.planes[0].bytesused == 0)
    {
        return false;
    }

    // Write raw video frame to file and return the buffer to converter
    // capture plane
    if (!ctx->stats && ctx->out_file)
    {
        write_video_frame(ctx->out_file, *buffer);
    }

	//TEST: modify the output data
	{
		int i, j;
		
		for(i=0; i<120; i++)
		{
			for(j=0; j<120; j++)
			{	
				buffer->planes[0].data[i*buffer->planes[0].fmt.stride + j] = 0;
			}
		}		
	}

    if (!ctx->stats && !ctx->disable_rendering)
    {
        ctx->renderer->render(buffer->planes[0].fd);
    }

    if (ctx->conv->capture_plane.qBuffer(*v4l2_buf, NULL) < 0)
    {
        return false;
    }
    return true;
}

In this function, i modify the output data after write yuv data in file. But I find the yuv data in file is incorrect.

Is there any advice?

2

Can NVIDIA answer this issue? If you cannot reproduce it, please tell me.
Thanks!

3

I have tested this issue on L4T R28.1, but it always occur.

I’m waiting the answer, please give me some advices.

4

Hi zcs, can you share more detail on the issue?
Looks like you modify the buffer after YUV is dumped.
if (!ctx->stats && ctx->out_file)
{
write_video_frame(ctx->out_file, *buffer);
}

//TEST: modify the output data
{
	int i, j;
	
	for(i=0; i<120; i++)
	{
		for(j=0; j<120; j++)
		{	
			buffer->planes[0].data[i*buffer->planes[0].fmt.stride + j] = 0;
		}
	}		
}
5

Hi DaneLLL,

The modified area looks unusual, have some line or point. But I use the cuda to modify YUV data, that is OK, I don’t know why.

6

Please refer to [url]https://devtalk.nvidia.com/default/topic/1019543/jetson-tx2/fail-to-mmap-the-capture_plane-of-decoder-in-mmapi/post/5194847/#5194847[/url]