Damage Rectangle Interface Proposed For Atomic DRM Drivers

[phoronix]

Phoronix: Damage Rectangle Interface Proposed For Atomic DRM Drivers

DisplayLink developers in cooperation with interest from VMware's virtual graphics driver team have sent out a draft proposal for adding a damage interface to the Direct Rendering Manager drivers...

http://www.phoronix.com/scan.php?page=news_item&px=Damage-Rectangles-DRM

[davidbepo]

the description sounds like something that could be useful for everything.
can somebody expert explain me why not?

[trek]

the description sounds like something that could be useful for everything.
can somebody expert explain me why not?

probably because there is no interest to save bandwidth on a dedicated wire to the monitor (VGA, DVI, HDMI), but it become useful on a shared bus (USB)

[TheOne]

the description sounds like something that could be useful for everything.
can somebody expert explain me why not?

Yeah, maybe this could help with tearing...

[linuxgeex]

probably because there is no interest to save bandwidth on a dedicated wire to the monitor (VGA, DVI, HDMI), but it become useful on a shared bus (USB)

It's not useful for "everything", but it's useful for software that needs to pay attention to what is happening on the screen.

Sometimes, instead of bandwidth, it's about saving local compute. There's a whole class of software that needs to monitor changes to the framebuffer. ie window manager, compositor, vidcap, desktop automation, I'm sure there's more.

[linuxgeex]

the description sounds like something that could be useful for everything.
can somebody expert explain me why not?

The main reason this won't be useful for 'everything' is that it will have some small overhead and there won't always be a way to make good use of the notifications, so then that overhead goes to waste.

Imagine a web browser sending notifications to the desktop compositor via this interface every time a rectangle is updated in the browser. It would be a huge win to update only the 32x32 rotating loading image instead of the entire browser window at every frame. But sending notifications for every image in a page with hundreds of images, or for each character of a scrolling line of text, or for each cell of a table during initial layout... those would all be fails. So it needs to be used strategically.

[oiaohm]

probably because there is no interest to save bandwidth on a dedicated wire to the monitor (VGA, DVI, HDMI), but it become useful on a shared bus (USB)

It will be interesting to watch if that remains the case. As we are getting higher and higher resolution screens and items like display-port wanting to put 3 monitors on 1 cable and DVI can in fact do 2 monitors on 1 cable..

Yeah, maybe this could help with tearing...

Adaptive Sync, G-sync(nvidia), Freesync(amd) are designed to prevent tearing by altering the frame rate up and down. So tearing is prevented as this comes into usage. The reason for looking at thisdamaged rectangle stuff is bandwidth.

Now the possible usage cases of this change long term.

1. Prime buffer sharing could possibly benefit from damaged rectangle so only what is changed has to be transferred between video card/user-space. https://www.kernel.org/doc/html/v4.1...buffer-sharing
2. VM interfaces like virtio-gpu and VMware's virtual graphics driver
3. Bandwidth restricted Items like display link evdi that is over usb. Or possible hdmi/dvi/displayport if screen resolutions keep on growing.
4. Remote Desktop

One of the reasons why Splice, RDP and VNC have be implemented at the compositor/server level this has been how to get access to damaged rectangle information that you need to make doing them perform. Now if this information is now at DRM level in kernel and standard this alters the game and may fix the issue of each wayland compositor having to implement RDP/VNC .... Instead just make them a graphics driver..

Now people have said RDP has like video acceleration and Wayland design does not allow for this. If implemented as a kernel driver set RDP screens could expose accelerated video decode.

This is another one of these areas were we need to think about have we been implementing remote desktop software in the right place.

[ssokolow]

It will be interesting to watch if that remains the case. As we are getting higher and higher resolution screens and items like display-port wanting to put 3 monitors on 1 cable and DVI can in fact do 2 monitors on 1 cable..

DVI does it by physically having two different sets of pins, resulting in "Dual-link DVI" cables being two cables in one, electrically. It's still a protocol built around fixed timing.

[Hi-Angel]

The main reason this won't be useful for 'everything' is that it will have some small overhead and there won't always be a way to make good use of the notifications, so then that overhead goes to waste.

Imagine a web browser sending notifications to the desktop compositor via this interface every time a rectangle is updated in the browser. It would be a huge win to update only the 32x32 rotating loading image instead of the entire browser window at every frame. But sending notifications for every image in a page with hundreds of images, or for each character of a scrolling line of text, or for each cell of a table during initial layout... those would all be fails. So it needs to be used strategically.

I think it's possible to make a heuristic that chooses whether it's worth sending a few damaged regions or rather whole screen.

[oiaohm]

DVI does it by physically having two different sets of pins, resulting in "Dual-link DVI" cables being two cables in one, electrically. It's still a protocol built around fixed timing.

Digital video stream:
Single link: 1920 × 1200 (WUXGA) @ 60 Hz
Dual link: 2560 × 1600 (WQXGA) @ 60 Hz
Above is what DVI specification says.
But you do find single link displays doing WQXGA (2560 × 1600) @ 30 Hz with CVT-RB blanking (132 MHz).

The reality here is yes DVI is fixed speed to a point. Add adaptive sync and only sending what has changed could be used to exceed these limits by a lot larger margins than currently done. Its not like unapproved extensions to have not turned up before like how to output HDMI audio out a DVI port. Time will tell if DVI ports add a mode to only send what has changed.

As screen size increases bandwidth limits of cables comes more of a problem even cables we would think are very fast.