* things we dont implement yet
	* more verbose error reporting
	* export MC information (for XvMC or for error resilience)
	* export quantizer information (for postprocessing filters)
	* dont crash on bad streams, make sure we can resync after a while
		* possible chunk buffer overflow while reading bits
	* synchronization stuff (play at correct speed)
	* IDCT precision with sparse matrixes
	* sparc IDCT optimizations
	* support for still pictures (decode before receiving next startcode !)

* structural optimizations
	* do yuv per sub-slice (probably big speed boost)
	* try different memory arrangements for pictures (yuyv, stride, ...)
	* once we have sync, call draw_frame before decoding I or P not after

* local optimizations
	* put most common fields at start of decoder_t structure
	* fix code that uses multiples of the stride (use preshifted value ?)
	* avoid 8-bit accesses particularly on alpha
	* use 64-bit shift register for parsing on 64-bit arches
	* use restrict (__restrict__) pointers: int * restrict p;
	* try feig IDCT ?
	* review the use of static inline functions
	* improve MMX motion comp inner routines
	* optimize IDCT for very sparse input matrixes ?
	* optimize startcode search loop ?
	* bit parsing / DCT parsing optimizations

* clean up
	* clean up header file usage
	* clean up yuv2rgb for interlaced pictures (handling of uv)
	* clean up decoder_t structure (some variables should be local ?)
	* clean up slice_init