Created attachment 15639 [details] [review]
Patch with above described fixes

I'll attach a patch here that fixes 3 closely related issues with
acceleration of repeating sources/masks on the R300

 - The code had an incorrect check for pitch (it was using a 
   value that already had been munged to register format)
 - The code used only enabled the use of the TXPITCH register
   for non-repeating sources but it works as well for repeating
   in my testing. (This allows acceleration of width 1/2/4 
   repeating sources which are power-of-two but padded out.)
 - In the rare case that a repeating source passed the 
   incorrect pitch check, the code didn't actually turn 
   repeat on in the accelerator.

I believe this fix is R300 specific and the R100/R200 code paths
are right.