mach64 EXA bits

[Michael Lorenz]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On Aug 9, 2006, at 07:14, Benjamin Herrenschmidt wrote:

> Hi George !
>
> I've seen with interest your various commits to the mach64 driver.
> However, I wonder if your EXA implementation has been tested on big
> endian at all. ATI chips generally have swappers at the front of the
> framebuffer, and thus need tweaking when doing things like memcpy
> to/from it (UFS/DFS can be called for different bit depth than the 
> front
> buffer) and need proper Prepare/FinishAccess. Also, I don't know for
> sure (I don't have docs at hand) how the swappers work on mach64 but if
> you don't have surfaces like radeon, you might have to fail multiple
> calls to PrepareAccess for pixmaps of different depth (thus possibly
> forcing a full migration to memory for some composite fallbacks).

Not sure if that's what you mean but usually the framebuffer aperture 
offers two views to video memory and registers, one swapped one not. I 
seem to remember the upper half is swapped but that may be wrong. Just 
write a couple bytes into video memory and read them back as an int32 
at BAR+offset and at BAR+aperture_size/2+offset, then use whatever is 
more suitable. Registers are the upper 1kb or so of each view, newer 
mach64 have them in additional memory and/or i/o BARs.

have fun
Michael
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (Darwin)

iQEVAwUBRNosrMpnzkX8Yg2nAQIOSwgApjnrUkMPpAbCTml8UB5F0vjG80sijCuo
8achLX6SxFeFqauAcMM3UeSdwYc2Ot1sJdvtFRCO0ZOgNoF18MF6LspbPPntwVoX
2Tb1pZiUw67FpGcl6pk/yTpaaSdlII6fjVZP2XGudGllDFoZ1ILy7j4WzWKLWr+3
vCMtliJIKPloiqKixZ4tqEZPcBKjUwx5fKrec5bV/05fsdpZ/7d9BUD92wAcBhbl
5g2QpD4ppqXFVlseX4sHuanQvgMG3Og523cbykoeh0gD+LiFbl+OrMlVYZncxjB2
F3EDjK5XeMiEq3zo++a+voDLNn9WysIigmZteTSOy+qmij2KrZKW6A==
=k7q7
-----END PGP SIGNATURE-----