Does AGX Xavier's vi5 driver have a low latency mode?

In the R32.4.3 BSP kernel sources , I see that vi4 has a low-latency mode, which should improve the real-time performance of the system. Does vi5 have a corresponding mode? It is measured that there is a delay of about 100ms when using vi5 to transport images.
here is the low-latency mode code in vi4_fops.c:

    chan->sequence = 0;
     chan->timeout = msecs_to_jiffies(200);
    if (!chan->low_latency)///////*CONFIG LOW LATENCY*

INIT_WORK(&chan->error_work, tegra_channel_error_worker);
INIT_WORK(&chan->status_work, tegra_channel_status_worker);

if (chan->low_latency) {///////*CONFIG LOW LATENCY*
	/* Start thread to release buffers */
	chan->kthread_release = kthread_run(
				chan, chan->video->name);
	if (IS_ERR(chan->kthread_release)) {
			"failed to run kthread for release\n");
		ret = PTR_ERR(chan->kthread_release);
		goto error_capture_setup;

and here:

  static int tegra_channel_capture_frame(struct tegra_channel *chan,
			struct tegra_channel_buffer *buf)
     int ret = 0;

     if (chan->low_latency)///////*CONFIG LOW LATENCY*
	ret = tegra_channel_capture_frame_multi_thread(chan, buf);
	ret = tegra_channel_capture_frame_single_thread(chan, buf);

      return ret;

Looks like vi5 didn’t implement it. Need to confirm it internally.

hello BugaoxingIguana,

as you can see,
low latency mode is using two threads approach to gather frames. one thread for the frame-start, the other for the frame-end.
Xavier’s VI-5 driver already implemented as the two threads approach. the idea is similar to VI-4,
it’s using kthread_capture_enqueue(), and kthread_capture_dequeue() for gathering capture frames and populate to user-space.