The dehazing effect of 9-16bit is not as good as 8bit

[Kiyamou]

I think I have replaced all numbers related to 8bit with variable “peak” and "bits". But the dehazing effect of 9-16bit is not as good as 8bit.

Maybe the empirical parameters in algorithm are related to 8bit?

(From left to right: origin, 8bit, 16bit (parameters same with 8bit), 16bit (adjust parameters))
I use a wrong sample.

single picture:
* origin: ~~
~~ https://i.loli.net/2020/06/27/z68WVxSCJDLFj1r.png
* 8bit: ~~
~~ https://i.loli.net/2020/06/27/PUYp6Ma5AKz3Vbt.png
* 16bit: ~~
~~ https://i.loli.net/2020/06/27/Y3RU18jF6rXuNbt.png
* 16bit adjust: ~~
~~ https://i.loli.net/2020/06/27/uciV8v6atedHQr7.png

Update: Actually, not all of 9-16bit have this problem. The output of 10bit is almost the same as 8bit.

[Kiyamou]

At that moment (c1083e6), there are indeed problems in code (data overflow when high bit depth). But the picture for test has problem. The "origin" has been dehazed. So the problems caused by overflow are more obvious. (>__<)
All in all, I think I have found the major cause and fix it latter.

Update: I think the problem has been fixed in fbe0502. Maybe there is still slight color problem (more dark) for high bit depth. Related to gamma correction probably?

Update2: 16bit output (for some picture is 15-16bit output) still has some difference from low bit depth. Although the difference isn't always bad, yet there is still someting to fix.