Re: RENDER ext. & hardware acceleration, libart integration

Juliusz Chroboczek <[email protected]> writes:

> >> If MMX is already being used in Render or has already tried, I'm sorry to
> >> imply that no one has thought of this before.
> 
> KP> It has been tried with no effect on performance.
> 
> Curious.  Do you understand why?
> 
> I would expect a 64-bit MMX read or a 128-bit SSE read from write-com-
> bining memory to do something smarter to the PCI bus than 2 or 4
> 32-bit reads.

While I'm no PCI bus expert, my conclusion when I fooled around
with this area some is that the video card manufacturers simply
don't spend any money/effort on the read-from-video-memory part of 
their PCI hardware. 

I seem to remember that the figures I was getting for 
read-from-video-memory were equivalent to about 1 dword every
12 pci bus cycles or something. 

Regards,
                                        Owen