how to enable the vsync in opengl application?

Autonomous Machines Jetson & Embedded Systems Jetson TX1
1

when display the image by direct fb, there is screen tearing, could i eliminate it by open the vsync, open it will call the code below code in kernel ?
int tegra_dc_wait_for_vsync(struct tegra_dc dc)
{
unsigned long timeout_ms;
unsigned long refresh; / in 1000th Hz */
int ret;

mutex_lock(&dc->lp_lock);
mutex_lock(&dc->lock);
if (!dc->enabled) {
	ret = -ENOTTY;
	goto out;
}
refresh = tegra_dc_calc_refresh(&dc->mode);
/* time out if waiting took more than 2 frames */
timeout_ms = DIV_ROUND_UP(2 * 1000000, refresh);
_tegra_dc_user_vsync_enable(dc, true);
mutex_unlock(&dc->lock);
ret = wait_for_completion_interruptible_timeout(
	&dc->out->user_vblank_comp, msecs_to_jiffies(timeout_ms));
mutex_lock(&dc->lock);
_tegra_dc_user_vsync_enable(dc, false);

out:
mutex_unlock(&dc->lock);
mutex_unlock(&dc->lp_lock);
return ret;
}

2

Hi Chenghaibo,

Have you tried https://devtalk.nvidia.com/default/topic/932150/enabling-disabling-vsync-jetson-tx1-/ ?

3

Hi Chenghaibo,

I use a simple gl application with vsync enable and the kernel code above is not called.

You can enable/disable vsync by export __GL_SYNC_TO_VBLANK=1/0.

4

You can take a look at /kernel/drivers/video/tegra/dc/window.c: _tegra_dc_program_windows(…)

This function is the one called when vsync enable by GL API.

5

thank you very much, It looks like, the vsync might be enabled by default, under the X11 environment, the tegra_dc_ext_flip was called by 60HZ, and the image looks like perfect with out any screen tearing, but under the direct fb, for purpose of eliminate the tear, i call the tegra_dc_wait_for_vsync,but that is useless, should i use the tegra_fb_blank??

6

Only fb.c and dev.c can receive the ioctl calls from user space.

I am still looking into it, but you can try tegra_fb_blank first.

7

Hi Chenghaibo,

How is your app going?

8

I’m sorry for communicate with you so late, in company, if i submit my reply, my account will be logout.

during present, i can’t find out which code relate to the screen tear.

below is my guess

the cea-861-d notice the 1920x1080 at 59.94 Hz
total H line:2200;
total v line:1125;
vblank is 45 line;

so, if the dma can’t transfer 1920x1080x4 byte per (45 /(59.94 x 1125) ) seconds that equal to
12429MB / S, it is a huge number, so the data should be transferred during the active period.

i hope my guess could help you

9

Hi Chenghaibo,

Could you print the return value of ioctl FBIO_WAITFORVSYNC ?

I have tried to flip framebuffer with FBIOPAN_DISPLAY. If WAITFORVSYNC is called, I can see smooth result.

However, I didn’t test it with any cuda kernel. Only used cpu to keep drawing framebuffer.