Toru Nishimura <nisimura@itc.aist-nara.ac.jp> wrote:

> > You'll be pleased to know that it'll drive PixelStamp adapters. I have a
> > test server running 32-bit color on my PXG at home. It's faster than the
> > `real' frame-buffers for most things, but painfully slow for anything
> > involving big pixmaps (so no Gimp or colourful root window tiles).
> 
> Doh, poor performance of pixmap transfer is _the one_ blamed by ULTRIX
> FAQ.
> 
> The speed enhancement of root window titling is very visible to users.
> I have an idea to accelerate titling.  Build 32bit aligned block of
> title in PXG memory first, then replicate it double in size, then
> double it, double it...  With just a few iteration you will have a
> large image, then copy and title the resulting to the target area.

For `colorful' tiles, that's an idea (I've also considered (when possible)
devoting the back buffer(s) to pixmaps in general, but that's some way
away).

For tiling window backgrounds (and rectangles), it's smarter than that in
some cases though - e.g., if the tile is <= 16x16, both dimensions are a
power of two, and it *uses* only two colours, then the root window is
stippled with two (or more) rectangles (each using an XYMASK). That turns
out to be very fast (and suits the default tile).

Andrew