OK,

Back to the video card emulation vs pseudo-card+driver debate...

I want to make a call on this topic this week and see if we
can get someone working on it.

>From a performance perspective, I think the major issue is
what privilege level code is reading/writing to the video
memory.  (IO to tweak the video registers to set up the card
and put it a certain mode don't matter since that is
infrequent compared to memory accesses).

If we emulate a 2D video card with 1:1 frame buffer, and let
guest user code (ring3) run natively, direct accesses from user
code to the emulated video memory don't require much extra overhead.
(just the extra copy of modified regions to the host GUI)

Accesses from kernel space will not be from natively executing
code (it will be from emulated or DT'd code), thus there will
be quite some overhead.

In general, I'd think user space libraries would want to memory
map the video frame buffer into user space, for performance, and
read/write directly to it.  This would be good for us, if we emulate
a real video card.

So, under which OSes is this _not_ the case?

-Kevin


-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Kevin Lawton                        [EMAIL PROTECTED]
MandrakeSoft, Inc.                  Plex86 developer
http://www.linux-mandrake.com/      http://www.plex86.org/

Reply via email to