suggestions about 2d accelerating X/Xfbdev

[Alex]

Hello,

I'm not a Xorg hacker and I would like to have some suggestion before start 
working on something wrong or in the wrong way.

I'm using Xorg 7.5 with an embedded mips over a standard linux fbdev, X and 
Xfbdev are working correctly, but the main cpu is a bit slow.
The hw support graphic acceleration using a coprocessor, through a user 
library (proprietary closed source), it can reduce main cpu load to 0% and 
speedup solid fill to 200% (guess will gain more on blended/scaled).
The library support acceleration on copy/move/fill rects, also using blend 
and/or scale, but have some limitation:
surfaces should be created&initialized using the library (optionally 
locked&mmapped for sw access, they reside on non-system memory), also sw 
operations afer hw should be sync'ed using the library.

Currently I'm planning to accelerate Xfbdev using KdCardFuncs/initAccel family 
, something like ephyr code does where fake exa.
Instead, if I've seen correctly, with X should be used DGAFunctionRec on a new 
driver.

Something wrong?

Are these the interfaces supposed to be used or I miss something?
Some limitation of the library can't be satisfied with these interfaces?
Do the accelerated code reside on server only or it is linked in clients also?
There is some reference about adding 2d blit/rect accelerations? 
(source/patches also, I read somewhere ATI drv is the best/complete, it is?) 

Thanks

Regards