Around 14 o'clock on Aug 21, Peter Kaczowka wrote:

> I guess that's true for reads, but my experience is that quadword writes
> are faster to some PCI devices.

Yeah, if you have 64-bit PCI.  With write-combining turned on, a 32-bit 
PCI or AGP bus is best written 32 bits at a time.  The current frame 
buffer code has a compile-time switch to use 64-bit accesses, and it's 
faster running in 32-bit mode than 64-bit to the screen.

Write-combining does help quite a bit when moving data from the CPU to the
card, but even that isn't as good as DMA.  The problem here is that
write-combining mode in the CPU bypasses the usual memory pathways and runs
through a separate buffer (which performs the write combining).  That
buffer is not deep enough to cover the AGP latency, so the CPU can only 
use something like 1/2 of the available AGP bandwidth.

> Who would have thought that X would live on as XFree86?  I guess Keith
> among others did; congratulations on keeping X alive.

It wasn't me -- I thought X was dead in 1995 or so.  We are all fortunate
that the original XFree86 developers (David Dawes and many others) kept
working through the dark ages.

Keith Packard        XFree86 Core Team        HP Cambridge Research Lab


_______________________________________________
Render mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/render

Reply via email to