2D Rendering On X11 Remains Barely Faster Than CPU Rendering

[raster]

aaah nothing new. figured this out 15+ years ago and moved to pure "client-side" software rendering with a view to using pure opengl once it was sane/possible (which is has become). so sw rendering only... or full blown gpu. there really isn't much else except maybe trying to use hw planes as a compositor if they are there and reliable/useful for your case. that decision to go this design path has proven correct as time has gone on.

[sdack]

Which is why every sane program, WM and DE uses the supplied power-of-two icon sizes. No scaling anywhere.

You mean you need an example. Take Xfce4 and it's panels. Change the size of the panels and it will change the size of icons on it. Take it's notification area and turn its border on and off - it, too, changes the size of the icons within the notification area. Change the size of your font and all icons on the menus change in size with it.

"No scaling anywhere" you say?

[curaga]

Which is why every sane program, WM and DE uses the supplied power-of-two icon sizes. No scaling anywhere.

[sdack]

So which UI elements are actually getting excessively rescaled at runtime in your system? For example, the wallpaper picture is only rescaled once when you set it or change the screen resolution, but then cached and reused for obvious performance reasons. Please note that rescaling also introduces some image quality loss, especially when done with a fast bilinear filter. There are many reasons to avoid unnecessary rescaling, and software developers already take this into account.

And simple image scaling is reasonably fast with software rendering too. It's the arbitrary-angle rotation, which is really troublesome.

The buttons, icons and thumbnails you see just come in a few fixed sizes. These then get constantly up- or downscaled wherever these are being used and before they can get cached. It goes into the start times of applications and opening times of windows and menus. Having a bit more speed here would be nice.

Other than this will faster 2D help with HTML rendering and bring more web content. I'm guessing advertisement and flashy web pages will profit from accelerated 2D the most. If this is a good or a bad thing is another question, but shouldn't be a reason not the have faster 2D.

[ssvb]

All the time?

So which UI elements are actually getting excessively rescaled at runtime in your system? For example, the wallpaper picture is only rescaled once when you set it or change the screen resolution, but then cached and reused for obvious performance reasons. Please note that rescaling also introduces some image quality loss, especially when done with a fast bilinear filter. There are many reasons to avoid unnecessary rescaling, and software developers already take this into account.

And simple image scaling is reasonably fast with software rendering too. It's the arbitrary-angle rotation, which is really troublesome.

[xpue]

how often do you see anything that is even simply upscaled/downscaled?

All the time?

[ssvb]

As mentioned in the talk, the benchmarking methology used is flawed.

Yes, the methodology is unfortunately giving an unfair advantage to the GPU rendering. The latest additions to the cairo traces repository introduced quite a bunch of html5 based animations, which heavily use affine transformations and are a bit difficult to handle by the CPU. So unless you are into modern web based gaming, the full set of cairo traces is not very representative. Older traces represented somewhat more relevant use cases (such as a terminal emulator with text scrolling, a web browser zooming in and out a forum page with tiled background, etc.).

Generally, if you see somebody showcasing a GPU acceleration advantage, in most cases you will see an emphasis on image rotation (I mean arbitrary rotation and not the easy 90/180/270 degrees cases) or something like this But how often does you Linux desktop system display anything that is rotated? In fact, how often do you see anything that is even simply upscaled/downscaled?

[sdack]

Why the hell is the audio quality on all open source conferences so awful? I won't listen to this shat because my ears start bleeding.

Yes, it is. I've tried listening to it on YouTube and even downloaded the 160k Opus audio stream, but the quality must be getting squished by something before that. Partly is it the headset and one can hear every breath coming in and out of the mouth and nose of the speaker, but it's also been exposed to some very lossy audio compression, because one can hear the typical artefacts (chirps), too. Only thing one can do is to share a box of tampons for the ears.

[birdie]

Why the hell is the audio quality on all open source conferences so awful? I won't listen to this shat because my ears start bleeding.

[Linuxhippy]

As mentioned in the talk, the benchmarking methology used is flawed.