XFree86-DGA

[Albert Cahalan]

On 9/26/07, Jordan Crouse <jordan.crouse at amd.com> wrote:
> On 26/09/07 11:56 -0400, Albert Cahalan wrote:

> > Isn't it rather important to support XFree86-DGA?
> > Currently it is scheduled to be disabled on the XO.
>
> No - system wide double buffering would not further increase
> performance over the current 2D acceleration, and it would consume
> (1200x900x2=2Mb) of memory that we could otherwise use for
> offscreen memory.  Plus, I'm not sure how this would interface with
> the shadow frambuffer mechanisms we use for RandR.
>
> If an application needs to be double buffered - then it
> can implement it iself with simple Xlib calls which are backed by
> accelerated hooks.

This is for one demanding app, not system-wide, and
the acceleration doesn't help for this kind of thing.

Well the exact API doesn't matter. Desired behavior:

loop{
1. app writes to inactive video memory
2. app asks X to make that memory active
}

AFAIK, XFree86-DGA provides this.
One could also skip the double buffering,
and just write to active video memory.

The simple Xlib calls would be:

loop{
1. app writes to inactive private buffer
2. app asks X to copy that buffer to video memory
}

That's a whole worthless copy.