[Port] [6000.4] [UUM-97985] Fix HDRP FPTL and Cluster artifacts when having many lights

Author: svc-reach-platform-support

Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/LightLoop/LightLoop.cs

@@ -860,7 +860,7 @@ void LightLoopNewFrame(CommandBuffer cmd, HDCamera hdCamera)
 
         static int NumLightIndicesPerClusteredTile()
         {
-            return ShaderConfig.FPTLMaxLightCount * (1 << k_Log2NumClusters);       // total footprint for all layers of the tile (measured in light index entries)
+            return (ShaderConfig.FPTLMaxLightCount + 1) * (1 << (k_Log2NumClusters + 1));       // total footprint for all layers of the tile (measured in light index entries)
         }
 
         void LightLoopAllocResolutionDependentBuffers(HDCamera hdCamera, int width, int height)

Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/LightLoop/lightlistbuild-clustered.compute

@@ -61,7 +61,9 @@ RWStructuredBuffer<float> g_logBaseBuffer : register( u3 );             // don't
 
 
 #define MAX_NR_COARSE_ENTRIES LIGHT_CLUSTER_MAX_COARSE_ENTRIES
+#define MAX_NR_VISIBLE_LIGHTS 826                                       // correspond to m_MaxLightsOnScreen in LightLoop.cs    
 
+groupshared unsigned int ldsTilePassList[MAX_NR_VISIBLE_LIGHTS];
 groupshared unsigned int coarseList[MAX_NR_COARSE_ENTRIES];
 groupshared unsigned int clusterIdxs[MAX_NR_COARSE_ENTRIES/2];
 groupshared float4 lightPlanes[4*6]; // Each plane is defined by a float4. 6 planes per light, 4 lights (24 planes)
@@ -241,6 +243,16 @@ void LIGHTLISTGEN(uint threadID : SV_GroupIndex, uint3 u3GroupID : SV_GroupID)
 
     uint2 tileIDX = u3GroupID.xy;
     uint t=threadID;
+    int i;
+
+    const bool lightSortRequired = g_iNrVisibLights > MAX_NR_COARSE_ENTRIES;  // Uniform runtime branch
+    if (lightSortRequired)
+    {
+        UNITY_LOOP
+        for(i=t; i<MAX_NR_VISIBLE_LIGHTS; i+=NR_THREADS)
+            if(i<MAX_NR_VISIBLE_LIGHTS)
+                ldsTilePassList[i]=0;
+    }
 
     const uint log2TileSize = firstbithigh(TILE_SIZE_CLUSTERED);
     uint nrTilesX = ((uint)g_screenSize.x +(TILE_SIZE_CLUSTERED-1))>>log2TileSize;
@@ -340,7 +352,27 @@ void LIGHTLISTGEN(uint threadID : SV_GroupIndex, uint3 u3GroupID : SV_GroupID)
             unsigned int uInc = 1;
             unsigned int uIndex;
             InterlockedAdd(lightOffs, uInc, uIndex);
-            if(uIndex<MAX_NR_COARSE_ENTRIES) coarseList[uIndex] = l;        // add to light list
+            if (lightSortRequired)
+                ldsTilePassList[l] = 1;
+            else
+                if(uIndex<MAX_NR_COARSE_ENTRIES) coarseList[uIndex] = l;        // add to light list
+        }
+    }
+
+    if (lightSortRequired)
+    {
+#if NR_THREADS > PLATFORM_LANE_COUNT
+        GroupMemoryBarrierWithGroupSync();
+#endif
+        // Ascending sort to prevent unpredictable light list when the number of visible lights is greater than the coarse entries.
+        if(t==0)
+        {
+            int c=0;
+            for(int ii=0; ii<g_iNrVisibLights && c < MAX_NR_COARSE_ENTRIES; ii++)
+            {
+                if(ldsTilePassList[ii] == 1)
+                    coarseList[c++] = ii;
+            }
         }
     }
 
@@ -399,7 +431,7 @@ void LIGHTLISTGEN(uint threadID : SV_GroupIndex, uint3 u3GroupID : SV_GroupID)
     //////////////////////////////////////////////////////////
 
     uint start = 0;
-    int i=(int) t;
+    i=(int) t;
     int iSpaceAvail = 0;
     int iSum = 0;
     if(i<nrClusters)

Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/LightLoop/lightlistbuild.compute

@@ -49,7 +49,9 @@ StructuredBuffer<uint> g_vBigTileLightList : register( t4 );        // don't sup
 RWStructuredBuffer<uint> g_vLightList : register( u0 );             // don't support RWBuffer yet in unity
 
 #define CATEGORY_LIST_SIZE          LIGHTCATEGORY_COUNT
+#define MAX_NR_VISIBLE_LIGHTS       826                             // correspond to m_MaxLightsOnScreen in LightLoop.cs
 
+groupshared unsigned int ldsTilePassList[MAX_NR_VISIBLE_LIGHTS];
 groupshared unsigned int coarseList[LIGHT_LIST_MAX_COARSE_ENTRIES];
 groupshared unsigned int prunedList[LIGHT_LIST_MAX_COARSE_ENTRIES];     // temporarily support room for all 64 while in LDS
 
@@ -155,6 +157,15 @@ void TileLightListGen(uint3 dispatchThreadId : SV_DispatchThreadID, uint threadI
         if(i<LIGHT_LIST_MAX_COARSE_ENTRIES)
             prunedList[i]=0;
 
+    const bool lightSortRequired = g_iNrVisibLights > LIGHT_LIST_MAX_COARSE_ENTRIES;  // Uniform runtime branch
+    if (lightSortRequired)
+    {
+        UNITY_LOOP
+        for(i=t; i<MAX_NR_VISIBLE_LIGHTS; i+=NR_THREADS)
+            if(i<MAX_NR_VISIBLE_LIGHTS)
+                ldsTilePassList[i]=0;
+    }
+
     uint iWidth = g_viDimensions.x;
     uint iHeight = g_viDimensions.y;
     uint nrTilesX = (iWidth+15)/16;
@@ -244,14 +255,34 @@ void TileLightListGen(uint3 dispatchThreadId : SV_DispatchThreadID, uint threadI
             unsigned int uInc = 1;
             unsigned int uIndex;
             InterlockedAdd(lightOffs, uInc, uIndex);
-            if(uIndex<LIGHT_LIST_MAX_COARSE_ENTRIES) coarseList[uIndex] = l;        // add to light list
+            if (lightSortRequired)
+                ldsTilePassList[l] = 1;
+            else
+                if(uIndex<LIGHT_LIST_MAX_COARSE_ENTRIES) coarseList[uIndex] = l;        // add to light list
         }
     }
 
 #ifdef FINE_PRUNING_ENABLED
     if(t<LIGHT_FPTL_VISIBILITY_DWORD_COUNTS) ldsDoesLightIntersect[t] = 0;
 #endif
 
+    if (lightSortRequired)
+    {
+#if NR_THREADS > PLATFORM_LANE_COUNT
+        GroupMemoryBarrierWithGroupSync();
+#endif
+        // Ascending sort to prevent unpredictable light list when the number of visible lights is greater than the coarse entries.
+        if(t==0)
+        {
+            int c=0;
+            for(int ii=0; ii<g_iNrVisibLights && c < LIGHT_LIST_MAX_COARSE_ENTRIES; ii++)
+            {
+                if(ldsTilePassList[ii] == 1)
+                    coarseList[c++] = ii;
+            }
+        }
+    }
+
 #if NR_THREADS > PLATFORM_LANE_COUNT
     GroupMemoryBarrierWithGroupSync();
 #endif
@@ -626,7 +657,7 @@ void FinePruneLights(uint threadID, int iNrCoarseLights, uint2 viTilLL, float vL
                 if (uIndex < LIGHT_LIST_MAX_COARSE_ENTRIES)
                 {
                     // InterlockedAdd ensures atomic and ordered writing to prunedList, preventing threads from overwriting each other's values.
-                    // Increment ldsNrLightsFinal (represents the total number of lights for the tile). 
+                    // Increment ldsNrLightsFinal (represents the total number of lights for the tile).
                     unsigned int uInc = 1;
                     unsigned int finalPrunedLightIndex;
                     InterlockedAdd(ldsNrLightsFinal, uInc, finalPrunedLightIndex);