EXT: Re: [PATCH weston 0/4] Optimize pixman renderer

[Pekka Paalanen]

On Tue, 22 May 2018 05:42:15 +0000
"Ray, Ian (GE Healthcare)" <ian.ray at ge.com> wrote:

> > On 21 May 2018, at 16.29, Pekka Paalanen <ppaalanen at gmail.com> wrote:
> > 
> > Hi,
> > 
> > this series was originally written by me as you can see, and Fabien
> > lifted the patches, reviewed, tested and submitted them. If no-one
> > objects, I would like to land these on Thursday.  
> 
> Reviewed-by: Ian Ray <ian.ray at ge.com>
> 
> (Nit: patch 3, typo “shadoe”.)

Nit fixed and all pushed:
   944fae88..e0e39b66  master -> master


Thanks,
pq


> > On Mon, 23 Apr 2018 11:44:55 +0200
> > Fabien Lahoudere <fabien.lahoudere at collabora.com> wrote:
> >   
> >> Optimizes pixman renderer by:
> >> - optimizing compositing damage in DRM/pixman
> >> 
> >>  Reduce Weston's CPU usage by avoiding unnecessary compositing when updating 
> >>  the shadow buffer in pixman-renderer, under the DRM backend. The test was a 
> >>  proprietary graphical X11 application in demo mode. The effect is a drop of
> >>  total system CPU usage from 0.41 to 0.33.
> >> 
> >> - optimizing shadow buffer usage
> >> 
> >>  The shadow framebuffer is an intermediate buffer where the scene is composited
> >>  and then copied from the shadow to the actual hardware buffer. This extra step
> >>  costs memory bandwidth compared to compositing directly into a hardware buffer.
> >> 
> >>  Weston's DRM-backend with the Pixman-renderer uses a shadow framebuffer by
> >>  default. Especially on systems with dedicated VRAM, read-modify-write cycles
> >>  (a.k.a blending) into the scanout-capable buffer can be very slow. Also the
> >>  scanout pixel format may not be optimal for compositing. Therefore Weston takes
> >>  the safe default to always use a shadow framebuffer.
> >> 
> >>  However, in our use case, the hardware does not have dedicated VRAM behind a
> >>  relatively slow bus, and the graphical load has practically no blending. We 
> >>  can reduce Weston's CPU usage quite a lot by not using the shadow frambuffer.
> >> 
> >>  We test on proprietary graphical X11 application in demo mode. Using perf we
> >>  measure that this change improve weston CPU usage by 13% with dual display and
> >>  11% with clone mode.
> >> 
> >> Pekka Paalanen (4):
> >>  pixman,drm: do not composite previous damage
> >>  pixman: make shadow buffer optional
> >>  compositor-drm: expose global shadow flag for pixman
> >>  main: add setting for DRM/pixman shadow framebuffer
> >> 
> >> compositor/main.c               |  3 ++
> >> libweston/compositor-drm.c      | 30 +++++++------
> >> libweston/compositor-drm.h      |  3 ++
> >> libweston/compositor-fbdev.c    |  3 +-
> >> libweston/compositor-headless.c |  3 +-
> >> libweston/compositor-rdp.c      |  5 ++-
> >> libweston/compositor-wayland.c  |  3 +-
> >> libweston/compositor-x11.c      |  6 ++-
> >> libweston/pixman-renderer.c     | 99 ++++++++++++++++++++++++++++-------------
> >> libweston/pixman-renderer.h     | 13 +++++-
> >> man/weston-drm.man              |  4 ++
> >> 11 files changed, 118 insertions(+), 54 deletions(-)
> >>   
> > 
> > _______________________________________________
> > wayland-devel mailing list
> > wayland-devel at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/wayland-devel  
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/wayland-devel/attachments/20180524/839c6102/attachment.sig>