Hi !
> Yes, I've wondered about something. I measured (with ./demo) the speed
> of ggiCopyBox on my X server (XF86_SVGA running on the ViRGE) and on
> the kgicon driver. I got the following figures:
> xlib: 400 boxes/sex
O.K. - accelerated xlib drawing.
> kgicon (accelerated): 200 boxes/sec
O.K. - More overhead due to ioctl access probably.
> x: 200 boxes/sec
> kgicon (unaccelerated): 20 boxes/sec
> The only guess I can make it that the access to the card memory must
> be much slower (PCI) than that to my main RAM (100MHz SDRAM).
Yes. Bandwidth on PCI bus is much slower than to main RAM. Especially, as
the area that is written to is relatively small, so that it might fit in the
cache. The main problem with Video card RAM access is often _reading_.
The whole system is optimized to be read by the graphics controller only and
to be written to by the accelerator and mmap. Reading by mmap is often
horribly slow.
> (Or does the x target accelerate CopyBox?)
No. How should it ? O.K. - one could program an in-memory DMA (can a PC do
that ?), but :-) ...
CU, Andy
--
Andreas Beck | Email : <[EMAIL PROTECTED]>
