URP Depth issue

Hi,
I want am having issue with depth intersection effect using URP.
I get the correct result in editor ( top picture ) but when built on Android I get weird banding ( bottom picture )



here's the simplified version of the code

Code (CSharp):

1. float pixelDepth = IN.screenCoord.z; // (ObjectToViewPos.z)
2. float2 screenUV = IN.screenCoord.xy / IN.screenCoord.w; // ComputeScreenPos()
3. float rawDepth = SAMPLE_TEXTURE2D(_CameraDepthTexture, sampler_ScreenTextures_linear_clamp, screenUV.xy);
4. float depth = LinearEyeDepth(rawDepth, _ZBufferParams);
5.  
6. float start = 0;
7. float end = 0.5;
8. float dist = ((depth - pixelDepth) - end) / (start - end);
9.  
10. return saturate(dist);

Am I doing something wrong?
I'm using unity 2020.3.0 and URP 10.3.2

Thanks!

 

My guess is the android versions using a 16 bit depth buffer. If you can, push out the near clip on the camera. If that’s the problem, even a small amount might help a ton.

 

Hey,
so I reduced the far clip plane and increased the near plane and it did help a lot indeed. Thanks for that!

In SRP you could declare your sampler like this sampler2D_float _CameraDepthTexture; and it would use full float precision data.

Any way to do that with the TEXTURE2D(_CameraDepthTexture); way of doing it?
I found TEXTURE2D_FLOAT however the define is just the same as regular TEXTURE2D

Or is it really just 16bits buffer and higher precision sampler will no change anything? ( in my build settings I did check the 32bits buffer box )

thanks again!

 

I honestly don't understand what the point of the

_float

suffix on the

sampler2D

definitions are for, since in OpenGLES 3.0 and desktop all

sampler2D

s are float precision.

My guess is the issue isn't anything to do with the shader code, but rather Unity is creating the camera depth texture as a half precision depth texture on Android, regardless of what the system depth buffer is using. You'd need to use a gpu profiler (one for the device you're using, not Unity's) to find out if that's what's happening, and I have no idea how you would fix that if it is the issue.

 

Looks like

sampler2D_float

adds

highp

precision modifier in front of the

sampler2D

in the glsl. So it might do something on some platforms.