Re: Discrepancy in byte count when reading from appsink with NV16_10LE32 and NV16 formats

Hi,

Le mercredi 14 juin 2023 à 17:05 +0200, Israel Díaz via gstreamer-devel a
écrit :
> 
> Hello GStreamer developers,
> 
> I have a question regarding reading bytes from an appsink in a GStreamer
> pipeline. I have set up a pipeline using the "videotestsrc" source with two
> different video formats: "NV16_10LE32" and "NV16". Here are the pipelines I am
> using:

> 
> 1. With NV16_10LE32 format:
> videotestsrc is-live=true ! video/x-
> raw,width=1920,height=1080,framerate=30/1,format=(string)NV16_10LE32 ! appsink
> name=videosourcesink
> 
> 
> 2. With NV16 format:
> videotestsrc is-live=true ! video/x-
> raw,width=1920,height=1080,framerate=30/1,format=(string)NV16 ! appsink
> name=videosourcesink
> 
> 
> I have implemented a callback function for the appsink, where I attempt to
> read the bytes from the received buffer. Here is the callback function
> implementation:
> 
> ```
> GstFlowReturn
> VDRE2GStreamerCaptureEncodingSourceTask::on_new_sample_from_videoSource_sink(G
> stElement *elt, ProgramData *data)
> {
>   GstSample *sample;
>   GstBuffer *app_buffer, *buffer;
>   GstElement *source;
>   GstFlowReturn ret;
>   GstMapInfo map;
> 
>   // ...
> 
>   sample = gst_app_sink_pull_sample(GST_APP_SINK(elt));
>   buffer = gst_sample_get_buffer(sample);
>   
>   app_buffer = gst_buffer_copy(buffer);
>   gst_buffer_map(app_buffer, &map, GST_MAP_WRITE);
> 
>   g_print("Array size: %ld \n", map.size);
> 
>   // ...
> 
>   gst_buffer_unmap(app_buffer, &map);
>   gst_object_unref(source);
>   gst_sample_unref(sample);
> 
>   return ret;
> }
> ```
> 
> My issue is as follows: when using the NV16_10LE32 (4:2:2 10 bits per
> component packed to 16) format, where a frame size of 8294400 bytes is
> expected (resolution 1920x1080), I am getting a size of 5529600 bytes instead.
> However, when I switch to the NV16 (4:2:2 8 bits per component) format, where
> a size of 4147200 bytes is expected, I get the correct result.

NV16_10LE32 is Xilinx format, wer 3 pixels are packed into 32bit (with 2 bit
padding).  The size seems correct to me ((width * height + 2) / 3 * 4 * 1080 *
2). I think you are confusing it Microsoft P016_LE/BE formats.

regards,
Nicolas

> 
> My question is: What could be causing this discrepancy in the byte count
> between the NV16_10LE32 and NV16 formats? Is there something wrong with my
> implementation or the pipeline configuration?
> 
> I appreciate your help and guidance in resolving this issue. If you need any
> additional information or further details, I would be happy to provide them.
> 
> Thank you,
> Israel