AM62A7: Request if there's GstPool realization within gst_wrapper

Hi TI experts,

I'm recently trying to understand the project of edgeai-gst-apps and to do some custom modification of it. Then I have tried to setup a pipeline of 

src_pipe_1 = gst_parse_launch("v4l2src device=/dev/video3 io-mode=5 ! video/x-bayer, width=2592, height=1944, format=bggi10 ! appsink name=sen_1 max-buffers=2 drop=true", NULL);

sen_1 = gst_bin_get_by_name(GST_BIN(src_pipe_1), "sen_1");

and then to use 

m_pipe.getBuffer("sen_1",m_buf_1,false,false,m_pipe.sen_1);

to pull raw image into gst_wrapper.c . But the trying failed because gst_wrapper doesn't support io-mode=dmabuf-import for now. But if I remove "io-mode=5" from the pipeline, the frame rate dropped from 30 to 8.

I have also noticed you left the annotation in the code says that GstPool could improve performance. 

int32_t GstPipe::allocBuffer(GstWrapperBuffer  &buf,
                             uint32_t           width,
                             uint32_t           height,
                             const string       format)
{
    GstSample  *sample = nullptr;
    GstBuffer  *buffer = nullptr;
    GstCaps    *caps = nullptr;
    int32_t     size;
    int32_t     ret;
    int32_t     status = 0;

    freeBuffer(buf);
    
    size = width * height;

    if (format == "RGB")
    {
        /* 3 * size * width. */
        size *= 3;
    }
    else if (format == "NV12")
    {
        /* 1.5 * size * width. */
        size += size/2;
    }
    else if (format == "UYVY")
    {
        /* 2 * size * width. */
        size *= 2;
    }

    /* TODO: When optimizing performance, this should be allocated from the
     * GstPool passed by appsrc's peer element so that kmssink can allocate
     * physically contiguous buffer so as to avoid doig a buffer copy
     */
    buffer = gst_buffer_new_allocate(NULL, size, NULL);

I wonder if you have some sample code of the GstPool realization of here. 

Sincerely,

Yizhe Fan