On Sun, 10 Aug 2003 05:59:54 +0100 Alan Hourihane <[EMAIL PROTECTED]>
babbled:
> On Sun, Aug 10, 2003 at 02:27:13PM +1000, Carsten Haitzler wrote:
> > Would I be correct in the assumption that the only accelerated path for
> > xrender is the identity transform (1:1 scale)? all other transforms are done
> > in software? (my initial tests here with xfree86 4.3.0 & nvidia's latest
> > drivers(as of about a month ago) seem to indicate as much...) (and yes... my
> > drivers are using acceleration... GL definitely is). ?
>
> No. About 99% of the drivers don't have any xrender acceleration. I think
> only the mga driver does. Although looking furthur the sis has some, but
> it seems disabled, and the vmware driver has it too. But that's it.
>
> I guess nvidia do some acceleration in their binary drivers though, as
> you've probably found. But it's bad to assume other drivers have xrender
> acceleration.
>
> I think the thing that's holding other drivers up in getting furthur xrender
> acceleration is that there's no test suite to check that the driver is
> doing the right thing. I think Keith Packard mentioned he had intern's
> working on a test suite a while ago, but nothing has materialized as far
> as I know.
hmmm - well i could write a performance suite here... if that helps. so far as
in my other e-mail, the "hardware acceleration" provides by the nvidia drivers
so far manages to be 1/35th the speed of my own mmx asm blending routines...
this is blending with 1:1 scaling (no transform).
i just tested a quick scaling test and the software routines i have are 41 times
faster than xrender... there is something suspicious here...
a quick rundown:
my software routines are in imlib2.
machine is amd 1.7ghz
gfx is nvidia gf4 4400ti 128mb
NVAGP is set to 0 cause it manages to lock up my kernel regularly :(
this is xfree 4.3.0
nvidia drivers:
NVIDIA_kernel-1.0-4191
NVIDIA_GLX-1.0-4191
destination is 32bpp pixmap (picture) i get similar speed results with
destination being 24bpp window so that doesn't make a difference. source image
being blended is 100x100 32bpp. i just simply look 4096 times and blend it at a
random co-ord with in a 320x320 destination.
if you want i can either attach the code & data files direct, or if you give me
a little i can clean it up int something approximating a performance test suite
for xrender compositing / some basic transforms.
somehow i'd expect it to at least EQUAL software :) well there goes my plans for
a project to do an xrender display engine... opengl still is the big winner :)
--
--------------- Codito, ergo sum - "I code, therefore I am" --------------------
The Rasterman (Carsten Haitzler) [EMAIL PROTECTED]
熊耳 - 車君 [EMAIL PROTECTED]
Mobile Phone: +61 (0)413 451 899 Home Phone: 02 9698 8615
_______________________________________________
Devel mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/devel
