New volumetric fog implementation

[Try]

I'm testing new stochastic fog right now. While quality is definitely better, performance cost is quite noticeable.
Please test on you hardware how much slower this fog is and post fps before-vs-after on same scene.
Also fps with Radial fog disabled is interesting.

UPD: Merged

Performance

Here is frame comparative chart. New rendering is 6.5% slower:

Core part of algorithm is to use path tracing with 32 ray steps, + random offset for each pixel.
You may notice that L1 cache (Yellow) is completely trashed, probably to large distance span between shadow-map samples, on each sample.
Tested with low-res (512x256) shadows - helps a little, but only a little.

Testing

Here build to test with new shading:
https://ci.appveyor.com/project/Try/opengothic/builds/46140513/job/5j0a451cbalra790/artifacts

Please report, if it works well on your machine.

[Try]

UPD:
Want to encourage testings, before voting. 6.5% performance may not be representation to your GPU. In principle RTX measurements are hardly representative.

[thokkat]

Tested on a GTX 1060 in 2560x1440. Everything looked good so far.

Actually this would be the perfect scenario for an ingame benchmark to get comparable numbers.

Fps table:

	scene 1	scene 2	scene 3	scene 4	scene 5	scene 6
off	107	122	133	139	139	129
new	103	121	133	129	130	131
old	118	136	153	161	159	145

old/new fog screenshots

scene 1 old

scene 1 new

scene 2 old

scene 2 new

scene 3 old

scene 3 new

scene 4 old

scene 4 new

scene 5 old

scene 5 new

scene 6 old

scene 6 new

[Try]

Hm, interesting that off is slower on your machine than old. Practically current off suppose to have no changes in compare to previous off.

[thokkat]

There used to be a 20 % performance penalty when activating fog. At some time off got lower fps, maybe when the sun got that large halo.

I compared with initial implementation (13th September build). Different scenes than in previous table due to save game incompatibility.

	scene 1	scene 2	scene 3
off initial	141	135	159
off current	120	115	126
new	123	108	127
old initial	120	111	129
old current	133	128	142

[Try]

After 3.5 long days of work:

0.80ms - occlusion lut
0.14ms - draw-pass