May I suggest this thread at bottom of post for more detail information on HDR
Originally Posted by Geoff D
In lots of words: The DV master is used to create a HDR10 grade. The HDR10 grade is then analysed against the DV master to determine the data that will go into the DV enhancement layer on the disc. I don't just mean metadata (which is a separate process than this) but the things like extra highlight/lowlight detail that are 'exclusive' to the DV master. This "difference data" is then compressed using Dolby's special sauce into a 1920x1080 DV enhancement stream (don't think of it as carrying a readable video signal, the resolution isn't important as it's just carrying data packets), which is encoded onto disc alongside the 3840x2160 HDR10 base layer.
What the player does at the other end is decompress the difference data and combine it with the base layer for a full 12-bit Dolby Vision output, this is why DV needs so much processing power: not so much for 12-bit this and that, but for that actual decompression and recombining process and being able to do it in real time. If it sounds like voodoo then it's not, it's much the same line of thinking as with the core + extension system for lossless audio like DTS-HD.
However, if when the HDR10 grade is analysed against the full DV master and no differences are found then the DV layer is comprised of the dynamic metadata only, there's no extra information regarding the extension of bit depth or dynamic range. This usually happens on discs that already have 4000-nit mastered HDR10 layers e.g. Jumanji 2 or IT, or when the HDR grade is so SDR-like that it all fits into 10-bit 1000-nit HDR10 without difficulty e.g. The Last Jedi.
Now, all this shouldn't affect the compression of the HDR10 layer at all (and doesn't on most DV discs!) but alas, they are still intrinsically linked so if someone isn't paying attention during the authoring process then we end up with the sort of crapola artefacting as seen above. The crazy thing is that the reconstitution of the DV signal really does greatly enhance the compression, I can actually watch SPR without wincing (well, not at the encoding anyway) and Deer Hunter was nicely improved also, so if DV can improve the worst moments of The Fog and make the lesser moments (they're not all as bad as that ^ photo) transparent then I'll be happy enough with that.
Originally Posted by Geoff D
...the DV enhancement layer can and often does include physical picture information that is not part of the HDR10 base layer and the two are COMBINED during playback which is what makes DV so processor-intensive. It also includes the dynamic metadata as well of course. And their ICtCp processing space is much better suited to HDR whereas YCbCr can run into problems with higher brightness as it was not designed for the PQ EOTF.