EXA

[Roland Scheidegger]

Lukas Hejtmanek wrote:
> On Tue, Aug 07, 2007 at 11:30:46AM +0200, Michel Dänzer wrote:
>>> Well, the performance difference between XAA and EXA seems to be 
>>> the following: XAA uses always pixman_blt_mmx while EXA uses 
>>> combination of accelerator CopyArea and memcpy.
>> The attached patch seems to help here, how does it work for you?
> 
> Yes! It definitely helps, EXA now runs at the *same* speed as XAA 
> (i.e., 1600FPS), hope it will be included in mainline. Thanks a lot!
> 
>>> Another point is why exaPrepareAccess calls exaWaitSync, is it 
>>> really necessary?
> 
>> Yes, for serializing intermixed GPU and CPU operations.
> 
> So the i915 chip does not support something like pipelining? And is 
> it theoretically possible to change the I830WaitLpRing function not 
> to use busy waiting?
If you write to some area with the chip and then you're going to
(potentially) access the same area with direct cpu access, there is no 
way around waiting for the gpu to finish. This is not chip specific.
I can't see a reason though why you'd need to busy-wait.

Roland