Add GTAO shader support

[show50726]

Note: It has not been tested and benchmark on real devices yet.

• Create gtao.mat and gtaoImpl.fs
• Add an option to AmbientOcclusionOptions to select AO type
• Add an additional GTAO struct to AmbientOcclusionOptions for GTAO specific parameters
• Update PostProcessManager::screenSpaceAmbientOcclusion() to select different material based on the option and set the parameters
• Update ViewerGui for GTAO parameters

Screenshots

I adjust the settings to make both SAO and GTAO using the similar sample count and take some screenshots.
Noted that SAO did some adjustments to the intensity and power params before passing into the material, so the effect looks stronger than GTAO under the same params.

SAO (original)

GTAO

[show50726]

Will create another PR to update the remote webpage after this PR is in.

[pixelflinger]

I suspect that there will be issues with highp / mediump so, you definitely need to test this on real mobile device.

[show50726]

Here is the on device profiling result:

• Device: Pixel 7 Pro
• Backend: OpenGL ES
• App: sample-gltf-viewer.apk
• Model: sponza.glb
• GTAO settings:
  • Slice Count = 4
  • Steps Per Slice = 3
  • Low Pass Filter = Quality::HIGH
• SAO settings:
  • All using default values, except for Quality

No Bent Normals

Method	SAO (Quality::MEDIUM)	SAO (Quality::HIGH)	SAO (Quality::ULTRA)	GTAO
SSAO Pass	5.865ms	8.074ms	14.383ms	11.579ms
Blur Pass	1.592ms	1.464ms	1.410ms	2.815ms
Total GPU Time	7.457ms	9.538ms	15.793ms	14.394ms

Bent Normals

Method	SAO (Quality::MEDIUM)	SAO (Quality::HIGH)	SAO (Quality::ULTRA)	GTAO
SSAO Pass	7.943ms	10.753ms	19.382ms	12.638ms
Blur Pass	2.40ms	2.402ms	1.584ms	4.911ms
Total GPU Time	10.343ms	13.155ms	20.966ms	17.549ms

So given this set of default parameters, I think the performance of GTAO is between SAO with Quality::HIGH and Quality::ULTRA. Also I did not see obvious artifact that is not shown on pc build.

[pixelflinger]

@show50726 how do you explain the performance difference of the blur pass? Shouldn't it be exactly the same?

[show50726]

@show50726 how do you explain the performance difference of the blur pass? Shouldn't it be exactly the same?

Do you refer to the difference between SAO and GTAO? This is because I set Low Pass Filter = Quality::HIGH in GTAO.
The reason I do this is because I think GTAO should have this settings by default to have a better visual result or the artifacts would be more obvious.
According to the original paper, they do both spatial and temporal filtering to reduce the artifacts, but I think temporal filtering might be too expensive for us, so the alternative I do is to set the Low Pass Filter = Quality::HIGH. (I also mention this recommendation in the AmbientOcclusionOptions' comment )

[pixelflinger]

@show50726 how do you explain the performance difference of the blur pass? Shouldn't it be exactly the same?

Do you refer to the difference between SAO and GTAO? This is because I set Low Pass Filter = Quality::HIGH in GTAO.
The reason I do this is because I think GTAO should have this settings by default to have a better visual result or the artifacts would be more obvious.
According to the original paper, they do both spatial and temporal filtering to reduce the artifacts, but I think temporal filtering might be too expensive for us, so the alternative I do is to set the Low Pass Filter = Quality::HIGH. (I also mention this recommendation in the AmbientOcclusionOptions' comment )

I understand the reasoning, but I think it hurts the comparison. I think that's okay to leave the default to HIGH for GTAO, but In the table above I think they should both use HIGH then.

[show50726]

@show50726 how do you explain the performance difference of the blur pass? Shouldn't it be exactly the same?

Do you refer to the difference between SAO and GTAO? This is because I set Low Pass Filter = Quality::HIGH in GTAO.
The reason I do this is because I think GTAO should have this settings by default to have a better visual result or the artifacts would be more obvious.
According to the original paper, they do both spatial and temporal filtering to reduce the artifacts, but I think temporal filtering might be too expensive for us, so the alternative I do is to set the Low Pass Filter = Quality::HIGH. (I also mention this recommendation in the AmbientOcclusionOptions' comment )

I understand the reasoning, but I think it hurts the comparison. I think that's okay to leave the default to HIGH for GTAO, but In the table above I think they should both use HIGH then.

I agree what you say. I will do another profiling to make them all use Quality::HIGH.
I will hold this PR until I get the results.