From: Andreas Beck <becka@rz.uni-duesseldorf.de>
  To  : ggi-develop@eskimo.com
  Date: Thu, 1 Jul 1999 14:40:58 +0200

Re: Accelleration (Was: Re: TNT benchmark)

> >         You do realize that (in principle anyway) there is no need for one
> > single universal method of implementing acceleration in a KGI driver?  The
> > existing Dali kgicommand API explicitly defines a set of 2D accels
> > (ACCEL_DRAWBOX, etc) which for the most part works OK,
> Note that using one ioctl() per accel operation is definitely not
> recommended for anything else than initial testing, even for 2D.
> Using ioctl()s are almost one order of magnitude slower than using 
> direct access.

IIRC Jon was replying to the maintainer of the S3 virge/trio cards ...
You don't want to enable direct access on those. IIRC they both have bugs
that will cause a PCI bus lock if you access them at the wrong time.

We should define a batched request scheme for IOCTL, though, as well as what
was once called PingPong buffers. However I don't know, if it is worth the
bother, as the typical cards of that type can't do much more that lines and
boxes, which are rarely used in sequence, except in a few applications.

CU, Andy

-- 
= Andreas Beck                    |  Email :  <andreas.beck@ggi-project.org> =