There's processing precision (normally full floating point precision, running it at half precision doesn't give any advantage on desktop gpus and should not affect compatibility) and texture (that hold intermediate result) precision (less precision = less bandwidth = faster). Rendering works this way: input -> processing -> texture -> processing -> texture. It uses 16 bits and has at least 11 bits of precision in range.Īnother option is 8/10/16/32-bit (normalized, means maps to ) integer textures.įor example madVR uses 16-bit integer textures, but has 10-bit option (in trade quality for performance). So the problem is that something breaks.ītw, what does full precision really do, i get that it works in 32bit or something, but what does it normally work at?The other floating point precision is half-float (). Hmm, about my GPU not supporting it, that kinda sucks, feels weird as it's fairly new (R9 380):SNah, I was thinking about something ancient, not modern gpus (AFAIK DX10 requires full 32-bit floating point support). to create a fallback to the internal subtitle renderer but this need to be developped.Īnyway, thank you very much for the huge amount of work on MPCHC! And the option is already defined in xySubfilter. Does this "link" is already planned or could be added?įurthermore, I can see that xySubfilter is not loaded when no subtitle is detected at the creation of the graph, meaning that the loading of an external subtitle could be problematic afterward. My expectation is to have the CTR+L function linked to the "open" function of xySubfilter. Indeed, I used this function a lot to add on the fly external subtitles to my physical media (Bluray, DVD.). This is a great addition! My only issue is related to the "Load Subtitle CTR+L" function that does not work with xySubfilter (and maybe standard version of vsfilter). I recently updated my version of MPCHC and seen the better integration of xySubfilter as subtitle renderer.
0 kommentar(er)
