From: Brian S. Julin <bri@tull.umassp.edu>
  To  : Andreas Beck <becka@rz.uni-duesseldorf.de>
  Date: Wed, 18 Aug 1999 15:38:57 -0400 (EDT)

Re: Ping-pong buffers on KGIcon are here!

On Wed, 18 Aug 1999, Andreas Beck wrote:
> > LibGGI would kind
> > of suck if you had to call Flush between each primitive because some of
> > them write to the MMAP and some use the accel engine :).
> 
> Right. That has to happen automatically. And IMHO it should do the same for
> DB vs. primitive, but I'll wait for Marcus' answer on that, as he might have
> data on the 3dfx stuff or something that says that would really suck badly.

Hmm... sucks any way you look at it, really.  In the very worst case (broken
hardware) any single non-accel operation must flush the ping pongs, put
a pixel or several, and then resume queueing accels.  My hunch is eventually
this will get bothersome enough to put fb-byte-poke "kgicommands" in the
accelqueue when there are only a few between accel calls and let the kernel
do the poking.  Hopefully that practice won't bloat things too much.

--
Brian