First Wayland Release: Wayland 0.85 With Weston

[phoronix]

Phoronix: First Wayland Release: Wayland 0.85 With Weston

Kristian H?gsberg has just announced Wayland 0.85 and Weston 0.85, which mark the first official releases of Wayland and its reference compositor, respectively...

http://www.phoronix.com/vr.php?view=MTA1NTY

[liam]

Phoronix: First Wayland Release: Wayland 0.85 With Weston

Kristian H?gsberg has just announced Wayland 0.85 and Weston 0.85, which mark the first official releases of Wayland and its reference compositor, respectively...

http://www.phoronix.com/vr.php?view=MTA1NTY

Sounds like its time for some benchmarks!
Its too bad Owen's xcomp becnhmark is, well, X specific. That is designed to stress compositors (I'm also guessing its the reason Mutter has improved so much).

[cl333r]

Are the Wayland graphics CPU emulated?
I'm asking cause Wayland runs on top of EGL/ES2.0 which (at least with the nouveau drivers) runs on the CPU. Hence the whole shebang might be software rasterized.


Also, I hope new Wayland releases won't happen too seldom and with huge changes (like it took X.org once like 1.5 years to ship a new version but which after a few years went back to normal release schedules).

[pq__]

Are the Wayland graphics CPU emulated?
I'm asking cause Wayland runs on top of EGL/ES2.0 which (at least with the nouveau drivers) runs on the CPU. Hence the whole shebang might be software rasterized.

First, let's talk about Weston, not Wayland, to avoid confusion. Wayland does not require anything from graphics, not GL, not EGL, not even a GPU. Weston has the requirements and supporting features you are interested in.

I think you are perhaps doing something wrong with Nouveau, it should offer hardware accelerated GL flavours via EGL just as it does for desktop GL via glX. People have reported getting Weston to run on Nouveau, also without X.

Weston does not affect the graphics driver capabilities in any way, nor force software rendering. All the hardware specific graphics acceleration magic is in the various parts of Mesa. If you can hardware-accelerate anything that uses EGL, also Weston is accelerated, regardless of the Weston backend in use (X, DRM, Wayland, ...).

And if you really want to, you can also configure Mesa to use software rendering, say, llvmpipe. But all the hardware drivers should be usable. If they for any reason are not, that is a problem in Mesa.

---

Liam, to benchmark anything, you first need to come up with a relevant benchmark program, that measures an interesting thing. That may be hard, because Wayland is designed to be, and Weston is on DRM, inherently synchronized to monitor vblank. You simply cannot update the screen faster than your monitor shows it.

[89c51]

anyone knows whats the status of important apps on Wayland (ie Firefox Chrome etc) ??

we know the toolkits work

[giselher]

anyone knows whats the status of important apps on Wayland (ie Firefox Chrome etc) ??

we know the toolkits work

Firefox and Chrome have their own Toolkits. But the Chrome team is already working on Wayland support afaik.

[liam]

First, let's talk about Weston, not Wayland, to avoid confusion. Wayland does not require anything from graphics, not GL, not EGL, not even a GPU. Weston has the requirements and supporting features you are interested in.

I think you are perhaps doing something wrong with Nouveau, it should offer hardware accelerated GL flavours via EGL just as it does for desktop GL via glX. People have reported getting Weston to run on Nouveau, also without X.

Weston does not affect the graphics driver capabilities in any way, nor force software rendering. All the hardware specific graphics acceleration magic is in the various parts of Mesa. If you can hardware-accelerate anything that uses EGL, also Weston is accelerated, regardless of the Weston backend in use (X, DRM, Wayland, ...).

And if you really want to, you can also configure Mesa to use software rendering, say, llvmpipe. But all the hardware drivers should be usable. If they for any reason are not, that is a problem in Mesa.

---

Liam, to benchmark anything, you first need to come up with a relevant benchmark program, that measures an interesting thing. That may be hard, because Wayland is designed to be, and Weston is on DRM, inherently synchronized to monitor vblank. You simply cannot update the screen faster than your monitor shows it.

pq,

That's why I mentioned Owen Taylor's xcomp benchmark as it is designed to stress compositors (well, it works for non-compositors, as well as pure offscreen, but that was the use case).
It uses two parameters: 1)create a number of xterms (called depth, IIRC), 2)set number of surfaces to alpha blend together (called load, again, IIRC).
That covers exactly what a pure compositor does. The only problem with it is that it requires (required? perhaps it has been updated) X.
The fact that weston uses vsync is fine, since that is a pretty common feature of modern compositors. The interesting thing to measure is seeing at what point the compositor can no longer supply the front buffer (I assume weston is at least supporting double buffering) in time for a screen update. Owen's stress test lets us do that by adding stuff to draw.
Right now, of course, we only have the weston compositor for wayland, but before long mutter, kwin, xfwm, enlightment's wm, and I would GUESS compiz (though it wouldn't surprise we to see compiz dropped instead of updated to be wayland compliant). Still, if we ported xcomp over to wayland we could get a rough idea of wayland's ability to scale (assuming weston isn't a hacked together piece of proof of concept, but I don't think that is the case).

[pq__]

pq,

That's why I mentioned Owen Taylor's xcomp benchmark as it is designed to stress compositors (well, it works for non-compositors, as well as pure offscreen, but that was the use case).

Thanks for explaining it, it does give some ideas on what could be benchmarked.

Weston was just a proof-of-concept, but it is now developing into a reference implementation.

[liam]

Thanks for explaining it, it does give some ideas on what could be benchmarked.

Weston was just a proof-of-concept, but it is now developing into a reference implementation.

I thought it had been described as the reference? Regardless it is good that it is now being treated as such.