XAA vs. EXA in old hardware support

[Keith Packard]

On Tue, 2006-06-20 at 09:25 +0100, Alan Cox wrote:

> I'm curious why you think it would be a loss. For a normal desktop case
> it would not be much of a gain and may be performance neutral. Why do
> you expect a noticable performance loss ?

Hardware line drawing isn't always a win, that's for sure. If the
hardware isn't pipelined for 2D line drawing, you have to spin waiting
for it to complete in any case, and even still you have to occasionally
read back status information from the card, which is a performance
killer if needed after every line. Plus, you still have a bunch of data
to upload per line (XY start, length, bresenham values).

For short lines, it's generally faster to draw with software than
hardware. On the ancient hardware I measured it on (S3 Trio, PCI), the
cut-over was somewhere around 50 or 60 pixels long, so for that embedded
system, it didn't make sense to even bother with hardware line drawing.
For longer lines, the delta in performance wasn't significant (perhaps a
factor of 2), but for shorter lines, the software drawing was
dramatically faster (more than 10x for 1 pixel lines). Switching
on-the-fly isn't sensible either as you'd have to completely stop the
GPU on each transition.

Of course, trying again with current hardware would make sense; things
are different now.

-- 
keith.packard at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.x.org/archives/xorg/attachments/20060620/5570e7b4/attachment.pgp>