Trying Out Fedora 25 With Wayland, Early Benchmarks Included

[Mystro256]

For anyone who is interested in a rough look of what's unfinished:

https://fedoraproject.org/wiki/Wayland_features

[cj.wijtmans]

Wanna bet EGLStreams would add a few FPS instead of gbm.

[milkylainen]

Enlighten me please.
How is gaming on a display server more than broadly relevant? I.e how badly could you screw up direct access through SDL where the display server would have to do nothing, beside some user input and other non cpu intensive tasks? (Besides background services and other crap dirtying cpu's like there is no tomorrow?).
I am obviously assuming that this type of measurements happen in a fullscreen and not windowed mode.
Would not actual timing in compositor and desktop rendering be more relevant than "more or less" passthrough rendering?

[theghost]

Compare with Ubuntu 16.10 with latest Unity 8 and Mir, that would be interesting, SDL2 runs on Mir I believe.

That would be very interesting to see how Mir compares atm.

[johanb]

Compare with Ubuntu 16.10 with latest Unity 8 and Mir, that would be interesting, SDL2 runs on Mir I believe.

SDL2 support for Wayland and Mir is pretty much equivalent right now it seems so that shouldn't be a big factor.

It was added in 2.0.2 (March 2014)

https://www.phoronix.com/scan.php?page=news_item&px=MTYyNjU

It was enabled in 2.0.4 (January 2016)

http://phoronix.com/scan.php?page=news_item&px=SDL-2.0.4-Released

[hiryu]

In the full two days of using the Fedora 25 development system, it's gone rather smoothly. My daily workflow was not interrupted, I did not run into a single crash, and at least for my usage I didn't encounter any shortcomings or areas where my Fedora experience was impaired at all. In fact, it felt just like using the traditional Fedora Workstation atop X.Org.

Michael,

When you say it feels just like using X.org, does that not mean you don't see any improvements from using Wayland?

When I've tried Rebecca Black OS, I immediately noticed a huge difference using Wayland. Things were much smoother and it seemed less laggy as well compared to X.

I think this line from https://www.phoronix.com/scan.php?pa...6-review&num=6 sums up my experiences with Wayland perfectly for me:

Using X11 you may not consciously notice moments of stutter, flickering, or tearing - but I never realised just how much my brain was suppressing these imperfections until a proper session with Wayland. The best way to describe it would be to say that Wayland removes the "mental fatigue" of constantly being exposed to many glitches.

[david_lynch]

Hmm, interesting results. But as long as wayland is a loser, performance wise, compared to X11, it's just not going to fly.

[darkbasic]

Wanna bet EGLStreams would add a few FPS instead of gbm.

LOL (stupid character limit)

[AsuMagic]

Stop judging Wayland's performance by this. It has an overhead caused by XWayland, and it can't really be dodged. I recall the last benchmarks about XWayland were much worse (or my memory is tricking with me), so good job to the guys behind those changes.

[bregma]

I think #1 is because all of those games were probably running through the XWayland wrapper, which goes:
Game > SDL (maybe) > XWayland (lighweight X wrapper) > Wayland > Display

Eventually, as games go to newer SDL versions and those SDL versions go through Wayland directly, that will change to:
Game > SDL > Wayland > Display

The overhead of wrapping the X calls to wayland probably accounts for some of the difference.

If it's not because of the lack of unredirect support in the compositor, then that's likely.

libSDL2 has full support for both Wayland and Mir, but libSDL1.2 doesn't so you need to pick your games.

At Canonical, we also found that libSDL would probe for X11 first before any other rendering back end, so if there was any hint whatsoever of an X11 server being present, it would use that preferentially. We ended up distro-patching libSDL2 to try X11 last instead so we could test Wayland and Mir more easily. If Fedora's strategy is to always have XWayland available as a backup, this could explain some performance regressions.

Another catch is that an awful lot of games and third-party libraries still have a hardcoded linux==X11 assumption even if they're using SDL. One particular set egregious offenders are the OpenGL extension wrangler libraries like GLEW. Your game can be completely innocent of all X11 calls but wants to do the right thing with respect to OpenGL and suddenly there's a crash on startup in XCB while GLEW probes for the version of GLX supported on your $DISPLAY. The X11 calls are simply hardcoded when compiled for Linux. For shame. Hopefully a modern extension wrangler built for 2016 is working its way through the queue.