On Thu, 2003-08-28 at 16:51, Thomas Winischhofer wrote: 
> Michel Dänzer wrote:
> [stripped]
> >   (  3180.0/sec): 500x500 rectangle
> >  
> >   (  1920.0/sec): 500x500 tiled rectangle (17x15 tile)
> >  
> >   (   139.0/sec): ShmPutImage 500x500 square
> 
> Here (P4 Celeron, 2.0Ghz, SiSM650, DDR266) it's
> 
> 2500/sec for rect500
> 1100/sec for oddtilerect500
> 717/sec for shmput500
> 
> Seems your engine is a bit faster (rect500, oddtilerect). VideoRAM 
> access seems slower, though ?! Huh?

Well, 32 bit PowerPC CPUs aren't exactly famous for memory throughput.
:( This is even accelerated via AGP by the chip's DMA engine, otherwise
I only get around 50.

> Why the h*ck is render so fast on your machine?

Beats me.

> One possibility is that doing it unaccelerated requires the CPU to read 
> from video RAM, 

Indeed, it does AFAIK.

> which is terribly slow on my shared-memory architecture machine. dga 
> (by pressing b) reports about 500MB/sec for writing, and only 37MB/sec 
> for reading.
> 
> Perhaps the 2D engine suffers from the same bottle-neck when doing the 
> alpha blending...

Sounds plausible.


-- 
Earthling Michel Dänzer   \  Debian (powerpc), XFree86 and DRI developer
Software libre enthusiast  \     http://svcs.affero.net/rm.php?r=daenzer

_______________________________________________
Devel mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/devel

Reply via email to