How to write to depth buffer in Custom Pass?

I am creating a sandstorm effect (looks like the one in the latest Forza Horizon trailer) using the volumetric cloud approach. I used Post Process Volume at first, but ran into an issue where the sandstorm draws on top of the fog and particle effects. I looked into the diagram (HDRP-frame-graph-diagram.png (3250×1026) (unity3d.com)) and thought it might be solved by switching to full screen custom pass and use an earlier injection point so that both the fog and particle effect draw on top of the sandstorm. But I don't know how to overwrite the depth buffer so that sky and fog can properly depth test against the sandstorm. Any help would be appreciated.

 

It's also stated here Custom Pass | High Definition RP | 7.1.8 (unity3d.com) in the description of AfterOpaqueDepthAndNormal injection point: "Buffers will contain all opaque objects. Here you can modify the normal, roughness and depth buffer, it will be taken in account in the lighting and the depth pyramid. Note that normals and roughness are in the same buffer, you can use DecodeFromNormalBuffer and EncodeIntoNormalBuffer functions to read/write normal and roughness data."
So I assume there is also a way to write custom values to depth buffer like the mentioned functions for normals, but I couldn't find anywhere a function that looks like EncodeIntoDepthBuffer.

 

Thank you! I will try and see if that works in my scenario and if the fog would be applied correctly.

 

@antoinel_unity Hi Antoinel, I ran into another problem. The given example only writes 0 depth. I don't really know what to write to the SV_DEPTH. For example, if the world position of the pixel I want to overwrite is (0, 100, 1000), what should the value to be written into the depth buffer be?

 

Thank you so much!

 

I've tried this. But I'm still not getting the result I want.

In the custom pass C# script, I wrote this:

Code (CSharp):

1.         // get scene color and depth buffer
2.         RTHandle source;
3.         RTHandle depthBuffer;
4.         if(targetColorBuffer == TargetBuffer.Camera)
5.         {
6.             GetCameraBuffers(out source, out depthBuffer);
7.         }
8.         else
9.         {
10.             GetCustomBuffers(out source, out depthBuffer);
11.         }
12.  
13.         // shader pass 2, overwrite depth
14.         HDUtils.DrawFullScreen(cmd, material, _temp, depthBuffer, matProperties, 2);
15.        
16.         //test depth
17.         HDUtils.DrawFullScreen(cmd, material, source, matProperties, 3);

And in the shader, I wrote this (to visualize if the depth got overwritten correctly):

Code (CSharp):

1.  
2. // shader pass 2
3. float4 WriteToDepth(PackedVaryingsToPS packedInput, out float outputDepth : SV_Depth) : SV_Target
4. {
5. UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(packedInput);
6. FragInputs input = UnpackVaryingsMeshToFragInputs(packedInput.vmesh);
7.  
8. // input.positionSS is SV_Position
9. PositionInputs posInput = GetPositionInput(input.positionSS.xy, _ScreenSize.zw, input.positionSS.z, input.positionSS.w, input.positionRWS);
10.  
11. float3 rayPos = _WorldSpaceCameraPos;
12. float3 rayDir = normalize(posInput.positionWS);
13. float2 rayToContainerInfo = rayBoxDst(_boundsMin, _boundsMax, rayPos, 1/rayDir);
14. float dstToBox = rayToContainerInfo.x;
15. float dstInsideBox = rayToContainerInfo.y;
16. float3 entryPoint = rayPos + rayDir * dstToBox;
17.  
18. outputDepth = ComputeNormalizedDeviceCoordinatesWithZ(entryPoint, GetWorldToHClipMatrix()).z;
19.  
20. return float4(0,0,0,0);
21. }
22.  
23. // shader pass 3
24. float4 ViewDepth (Varyings varyings) : SV_Target
25. {
26.     UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(varyings);
27.    
28.     float nonlin_depth = LoadCameraDepth(varyings.positionCS.xy);
29.     float depth = Linear01Depth(nonlin_depth, _ZBufferParams);
30.     return depth;
31. }

And no matter what world space position value I give to it, the depth buffer always remains unchanged (just showing the original depth of the scene).

Is there anything wrong with the way I did it?

 

I don't know anyone else to turn to for help, so @antoinel_unity sensei, please help.

 

This is probably too complex for me . I went for another way of rendering the sandstorm. I first raymarch the sandstorm to a render texture in a compute shader, which is dispatched in a custom pass to ensure the depth is correct. This way I can control the render ratio of the sandstorm to enhance performance. And then I use a huge quad placed very far away to sample this render texture. And since I can enable "Receive fog" on this quad, fog is also applied correctly. The sandstorm rendered this way looks good enough for my purpose.

(I ran into another problem with TAA though. I will post in another thread.)