+CPU per cell frustum culling, can render 10M grass on adreno612 now

Author: ColinLeung-NiloCat

Assets/URPMobileGrassInstancedIndirectDemo/InstancedIndirectGrass/Core/CullingCompute.compute

@@ -5,14 +5,15 @@
 //cullingComputeShader.SetMatrix("_VPMatrix", p * v); //set from C#
 float4x4 _VPMatrix;
 float _MaxDrawDistance;
-StructuredBuffer<float3> _AllInstancesTransformBuffer; //will not change until instance count change
-AppendStructuredBuffer<uint> _VisibleInstanceOnlyTransformIDBuffer; //will set counter to 0 per frame, then fill in by this compute shader
+uint _StartOffset;
+StructuredBuffer<float3> _AllInstancesPosWSBuffer; //will not change until instance count change
+AppendStructuredBuffer<uint> _VisibleInstancesOnlyPosWSIDBuffer; //will set counter to 0 per frame, then fill in by this compute shader
 
-[numthreads(1024,1,1)]
+[numthreads(64,1,1)]
 void CSMain (uint3 id : SV_DispatchThreadID)
 {
 	//posWS -> posCS
-	float4 absPosCS = abs(mul(_VPMatrix,float4(_AllInstancesTransformBuffer[id.x],1.0)));
+	float4 absPosCS = abs(mul(_VPMatrix,float4(_AllInstancesPosWSBuffer[id.x + _StartOffset],1.0)));
 
 	//do culling test in clip space, result is the same as doing test in NDC space.
 	//prefer clip space here because doing culling test in clip space is faster than doing culling test in NDC, because we can skip 1 division.
@@ -21,5 +22,5 @@ void CSMain (uint3 id : SV_DispatchThreadID)
 	//y test allow 50% more threshold (hardcode for grass)
 	//x test allow 10% more threshold (hardcode for grass)
     if (absPosCS.z <= absPosCS.w && absPosCS.y <= absPosCS.w*1.5 && absPosCS.x <= absPosCS.w*1.1 && absPosCS.w <= _MaxDrawDistance)
-    	_VisibleInstanceOnlyTransformIDBuffer.Append(id.x);
+		_VisibleInstancesOnlyPosWSIDBuffer.Append(id.x + _StartOffset);
 }

Assets/URPMobileGrassInstancedIndirectDemo/InstancedIndirectGrass/Core/InstancedIndirectGrass.mat

@@ -92,7 +92,7 @@ Material:
     - _ZWrite: 1
     m_Colors:
     - _BaseColor: {r: 0.41936636, g: 0.7169812, b: 0.29423285, a: 1}
-    - _BoundSize: {r: 55.9017, g: 55.9017, b: 0, a: 0}
+    - _BoundSize: {r: 661.4378, g: 661.4378, b: 0, a: 0}
     - _Color: {r: 1, g: 1, b: 1, a: 1}
     - _EmissionColor: {r: 0, g: 0, b: 0, a: 1}
     - _GroundColor: {r: 0.2195755, g: 0.46226418, b: 0.2581851, a: 1}

Assets/URPMobileGrassInstancedIndirectDemo/InstancedIndirectGrass/Core/InstancedIndirectGrassPosDefine.cs

@@ -0,0 +1,67 @@
+using System.Collections;
+using System.Collections.Generic;
+using UnityEngine;
+
+public class InstancedIndirectGrassPosDefine : MonoBehaviour
+{
+    [Range(1, 40000000)]
+    public int instanceCount = 1000000;
+    public float drawDistance = 125;
+
+    int cacheCount = -1;
+
+    // Start is called before the first frame update
+    void Start()
+    {
+        UpdatePosIfNeeded();
+    }
+
+    private void UpdatePosIfNeeded()
+    {
+        if (instanceCount == cacheCount)
+            return;
+
+        Debug.Log("UpdatePos (Slow)");
+
+        //same seed to keep grass visual the same
+        UnityEngine.Random.InitState(123);
+
+        //auto keep density the same
+        float scale = Mathf.Sqrt((instanceCount / 4)) / 2f;
+        transform.localScale = new Vector3(scale, transform.localScale.y, scale);
+
+        //////////////////////////////////////////////////////////////////////////
+        //can define any posWS in this section, random is just an example
+        //////////////////////////////////////////////////////////////////////////
+        List<Vector3> positions = new List<Vector3>(instanceCount);
+        for (int i = 0; i < instanceCount; i++)
+        {
+            Vector3 pos = Vector3.zero;
+
+            pos.x = UnityEngine.Random.Range(-1f, 1f) * transform.lossyScale.x;
+            pos.z = UnityEngine.Random.Range(-1f, 1f) * transform.lossyScale.z;
+
+            //transform to posWS in C#
+            pos += transform.position;
+
+            positions.Add(new Vector3(pos.x, pos.y, pos.z));
+        }
+
+        //send all posWS to renderer
+        InstancedIndirectGrassRenderer.instance.allGrassPos = positions;
+        cacheCount = positions.Count;
+    }
+    private void Update()
+    {
+        UpdatePosIfNeeded();
+    }
+    private void OnGUI()
+    {
+        GUI.Label(new Rect(300, 50, 200, 30), "Instance Count: " + instanceCount/1000000 + "Million");
+        instanceCount = Mathf.Max(1, (int)(GUI.HorizontalSlider(new Rect(300, 100, 200, 30), instanceCount / 1000000f, 1, 10)) * 1000000);
+
+        GUI.Label(new Rect(300, 150, 200, 30), "Draw Distance: " + drawDistance);
+        drawDistance = Mathf.Max(1, (int)(GUI.HorizontalSlider(new Rect(300, 200, 200, 30), drawDistance / 25f, 1, 8)) * 25);
+        InstancedIndirectGrassRenderer.instance.drawDistance = drawDistance;
+    }
+}