- add page flipping to the mga_vid driver
- Tweak the slice at a time code to do half a slice as a smaller
	chunk will fit much better in the puny 16 kb L1 of the P6 core.
	This will require draw_slice to take a destination rectangle.
- Do idct after motion_comp and add the motion comp block where
  appropriate. The idea here is to get rid of the intermediate
  s16 write for every idct block. Convert this to mmx when it's working. -ah
- Separate the display_* stuff into libvo (vo for video-out)
- Tweak the mc mmx code to use the previous bottom line as the next top
  line in the y interp case. (partially done) -ah
- try and see if interlacing U and V is better -ml
- implement the missing mpeg features... DMV motion compensation, field
  pictures, etc... and eventually try to use the mpeg conformance
  verification bitstreams
- DGA display support
- Fix the mlib/sparc stuff once the interfaces have settled down
  a bit.
- Change y_blocks,cr_blocks,and cb_blocks in macroblock_t  and just do
  blocks[]. I think this could potentially make a lot of code simpler.

DONE
- rewrite the library interface to take a chunk of data at a time
  and return something useful if we finished a frame. -ah
- rewrite yuv2rgb to use lut for intermediate results - ml DONE
- process frames one slice at a time and write the results immediately
  to the output. (this reduces the cache footprint significantly).
  this actually ties in with (1) -ah
