Glamo Speed Improvement(s)

[Thomas White]

On Sun, 15 Nov 2009 22:23:28 +0100
mobi phil <mobi at mobiphil.com> wrote:

> Maybe I am missing some details of X11, but I think some GPU commands
> could be made async? or?
> 
> probably on normal pc desktop it is not an issue "the cpu has to wait
> for the gpu", but what you are describing should not be a generic
> solution for the dri architecture?

In an ideal world, all graphics operations would be sent to the GPU in
a "fire and forget" manner as you describe.  In practice, things aren't
so nice.  Even "proper" graphics cards need this kind of
synchronisation quite often because they can't accelerate every
single operation required.  In fact, EXA doesn't present a very large
set of operations to the driver for acceleration.

At this low a level, it becomes quite hard to make things generic.  The
implementation for Glamo is actually based on the method used in the
latest version of the Radeon driver.  It's every bit as much of a
concern for the AMD, Intel and Nouveau people...

Tom

-- 
Thomas White <taw at bitwiz.org.uk>