* things we dont implement yet
	* field pictures
	* mpeg1 - full-pixel MC, have to find test streams
	* dont crash on bad streams, make sure we can resync after a while
		* possible chunk buffer overflow (writing or reading back)
		* destination macroblock overflow
		* MC source out of bounds (maybe signal handler for that one ?)
	* make the decoder reentrant (if the speed impact is small enough)

* structural optimizations
	* integrate idct_add and idct_copy into the main idct routine
	* get rid of slice structure, put it into picture
	* do yuv per sub-slice (probably big speed boost)
	* pass pictures to libvo in decoding order not display order
	* try different memory arrangements for pictures (yuyv, pitch, ...)

* local optimizations
	* review the use of static inline functions
	* improve MMX motion comp inner routines
	* optimize DCT for very sparse input matrixes ?
	* optimize startcode search loop ?
	* bit parsing / DCT parsing optimizations

* clean up
	* clean up header file usage
	* clean up DMV code

* synchronization stuff 
