dubiousconst281

Back a while I tried to implement a MP4 demuxer, and I can kind of relate to that. The mdat box is sometimes an opaque blob and you need to parse the codec framing to split packets (fMP4 helps with this a bit), each codec has its own set of boxes, and the specs for each of them are paywalled...

Matroska/WebM is so much simpler and easier to parse, you can essentially abstract it away in a JSON-like DOM (obviously without loading 1GB of data into memory) and just get what you want, it's great.

One fast way to do palette reduction is by using a 16-bit RGB565 lookup table filled with all the closest colors relative to each entry index, then you can replace each pixel in the image using just a few bitwise ops and one array access.

It seems that the exact opposite problem (phi elimination / translating out of SSA form) is rarely discussed in depth, especially in introductory material.

From what I've seen, most compilers rely on the register allocator to deal with this problem (which makes total sense), or just don't bother at all and instead keep SSA in "conventional" form.

But the problem isn't really that difficult. In Sreedhar/Boissinot et al.'s paper (which btw has a typo in the copy sequentialization algorithm), the idea is quite simple: create copies for each phi argument, plus an additional copy for the phi result, then coalesce those copies to the same variable if their live-ranges don't intersect. The devil is on the details of course, and it just takes a lot more effort to implement compared to the text book construction algorithm.

There's also the chroma subsampling issue. With the standard 4:2:0 ratios, you'll get half the resolution for the two chroma channels, and if I'm not mistaken, they are more aggressively quantized.

It would be better to use YUV/YCbCr directly instead of RGB.

You can get libx264 to encode 1080p in under 10ms if you spend enough time tuning it, altough I never compared the quality to JPEG.

Widevine is a joke, L3 at least. You can easily dump all video and audio frames by hooking into Chromium's CDM wrapper.

I'm guessing the reason is because of code gen issues, there's a lot of workarounds throughout the BCL code base.