Graphics Performance

[Carsten Haitzler (The Rasterman)]

On Sat, 4 Apr 2009 15:33:26 +0200 Rask Ingemann Lambertsen <rask at sygehus.dk>
said:

> On Fri, Apr 03, 2009 at 09:14:16PM +0100, Ian Stirling wrote:
> > Iain B. FIndleton wrote:
> 
> > > I 
> > > wonder what the issue is in moving the frame buffer data? At the speeds 
> > > available on the FR, moving 640 x 480 x 2 = 614,400 bytes from memory to 
> > > a video buffer at 30 Hz needs about 18 MB/sec.
> > > 
> > > What is the bandwidth for memory moves?
> > 
> > About 6-8 or so - with 100% CPU utilisation
> 
>    There are four DMA channels in the s3c422b, so you shouldn't use the CPU
> for straight memcpy() of such large blocks. The kernel doesn't yet support
> mem->mem DMA transfers, but I'm working on that to use it in the glamo SD
> card driver.

that was tried. results were worse than with the cpu. try again - but.. been
there, done that. ask dodji skeleti. :)

1. the memory bus is locked up anyway - be it cpu or dma cycles. if dma does it
the cpu is still locked out, so it's not done async.
2. the actual performance was much worse - at best it was 1/2 the speed of
using the cpu. but it was actually overall even worse on smaller blocks.


-- 
------------- Codito, ergo sum - "I code, therefore I am" --------------
The Rasterman (Carsten Haitzler)    raster at rasterman.com