Vad kan man göra på nyårsafton om han dryck inte alkohol och han är FFmpeg developer också? (Translation: what can man do on evening before New Year if he does not drink alcohol and he is FFmpeg developer?).
Naturally, I spent it hacking at some codec. One of the most annoying issues with it (at least for some people like me) is that it does not support some features, like interlaced VC-1 and decodes only I-frames in old version of WMV3. So I’ve tried to fix the latter.
There is a flag called RES_RTM which tells whether it’s final bitstream format (a.k.a. “release to manufacturer”, hence the name) or not. I’ve tried tracing its effect through binary decoder and it turned out that it only alters transform pattern decoding.
Here’s a brief outline of transform concept: WMV2, WMV3 and VC-1 may partition P or B blocks into 4×8, 8×4 or 4×4 subblocks; each subblock, obviously, may be coded or not, so decoder also needs to extract subblock coding pattern unless transform type specifies it (like “8×4 transform, left half coded”).
It turns out that older version uses different condition for subblock coding pattern reading (for 8×4 and 4×8 transforms only), without checking whether this is transform specified only for current block or for whole macroblock (i.e. group of 6 blocks). Quick hacking has not made FFmpeg decode those old file correctly though, so there is still some work left to figure out correct condition.
Would rather that you’d worked on interlaced WVC1, but oh well 😉