Re: Why X11 via unaccelerated driver is faster than via frame buffer driver?

[Michael <macallan%netbsd.org@localhost>]

Hello,

On Wed, 13 Apr 2016 18:34:35 +0300
Alex <lausgans%gmail.com@localhost> wrote:

> Does anybody knows why on (some?) old platforms, an unaccelerated*
> dedicated graphics driver (for example xf86-video-suncg6) is faster
> than non-shadowfb backed frame buffer device driver**?
> And also still faster than shadowfb backed wsfb on some operations
> (like 10x10 scroll / copy).
> [Actually, on CG6 / TurboGX with wsfb+shadowfb I see a slowing sliding
> effect on some x11perf tests, so I don't believe the numbers here,
> perhaps the sync issue]

That's weird, I wouldn't expect much of a difference there, both merely
point Xorg's fb module at the mmap()ed video memory and that's pretty
much it. We're not playing any prefetching or write combining games
either ( on SPARC that is, I wish we could... ).
Shadowfb is bound to be slower on small operations - it's still more
work and the speedup from not reading framebuffer memory isn't big
enough to outweight it.

> * NoAccel True
> ** wsfb, ShadowFB False. A trivial fix like
> http://pastebin.com/8BqjHWys was done in order to make wsfb work over
> cgsix console

I'll commit that fix.

have fun
Michael