Shader optimizations

Assets/Assets/Shader/VoxelRaycast.compute

@@ -52,6 +52,8 @@ struct StackEntry
     int nodeIndex;
     float3 center;
     float halfSize;
+    float entry;
+    float exit;
 };
 
 // Buffers
@@ -167,6 +169,9 @@ void CSMain(uint3 id : SV_DispatchThreadID)
     root.nodeIndex = rootIndex;
     root.center = rootCenter;
     root.halfSize = rootHalfSize;
+
+    if (IntersectAABB_fast(rootCenter, rootHalfSize, b.origin, r.direction, invDir, root.entry, root.exit))
+    {
         stack[sp++] = root;
 
         bool hasHit = false;
@@ -180,17 +185,14 @@ void CSMain(uint3 id : SV_DispatchThreadID)
     
             LinearNode n = nodes[e.nodeIndex];
     
-        float tEntry, tExit;
-        if (!IntersectAABB_fast(e.center, e.halfSize, b.origin, r.direction, invDir, tEntry, tExit)) continue;
-
             // prune with current best
-        if (tEntry >= outHit.maxDistance) continue;
+            if (e.entry >= outHit.maxDistance) continue;
     
             if (n.isLeaf == 1u)
             {
                 if (n.isOccupied == 1u)
                 {
-                float tHit = max(tEntry, 0.0);
+                    float tHit = max(e.entry, 0.0);
                     if (tHit < outHit.maxDistance)
                     {
                         // found a closer hit — commit minimal info, defer heavy ops
@@ -202,12 +204,13 @@ void CSMain(uint3 id : SV_DispatchThreadID)
                 continue;
             }
     
-        // Non-leaf: gather children that intersect and their tEntry (small array)
+            // Non-leaf: gather children that intersect and their entry (small array)
             uint childMask = n.childMask;
             // small local arrays
-        float childT[8];
             int childIdx[8];
             float3 childCenter[8];
+            float childEntry[8];
+            float childExit[8];
     
             int childCount = 0;
     
@@ -228,14 +231,14 @@ void CSMain(uint3 id : SV_DispatchThreadID)
                 float3 cCenter = e.center + offsetVec;
     
                 // pretest intersection with child AABB to get tEntry
-            float ctEntry, ctExit;
+                if (!IntersectAABB_fast(cCenter, childHalf, b.origin, r.direction, invDir, e.entry, e.exit)) continue;
-            if (!IntersectAABB_fast(cCenter, childHalf, b.origin, r.direction, invDir, ctEntry, ctExit)) continue;
+                if (e.entry >= outHit.maxDistance) continue; // prune child if already farther than best hit
-            if (ctEntry >= outHit.maxDistance) continue; // prune child if already farther than best hit
     
                 // store for near-first push
-            childT[childCount] = ctEntry;
                 childIdx[childCount] = cIndex;
                 childCenter[childCount] = cCenter;
+                childEntry[childCount] = e.entry;
+                childExit[childCount] = e.exit;
                 // temporarily store center and half? we recompute on push
                 childCount++;
             }
@@ -243,30 +246,36 @@ void CSMain(uint3 id : SV_DispatchThreadID)
             // sort children by childT ascending (insertion sort on at most 8 elements)
             for (int a = 1; a < childCount; ++a)
             {
-            float keyT = childT[a];
                 int keyIdx = childIdx[a];
                 float3 keyCenter = childCenter[a];
+                float keyEntry = childEntry[a];
+                float keyExit = childExit[a];
+
                 int j = a - 1;
-            while (j >= 0 && childT[j] > keyT) {
+                while (j >= 0 && childEntry[j] > keyEntry) {
-                childT[j+1] = childT[j];
                     childIdx[j+1] = childIdx[j];
                     childCenter[j+1] = childCenter[j];
+                    childEntry[j+1] = childEntry[j];
+                    childExit[j+1] = childExit[j];
                     j--;
                 }
-            childT[j+1] = keyT;
                 childIdx[j+1] = keyIdx;
                 childCenter[j+1] = keyCenter;
+                childEntry[j+1] = keyEntry;
+                childExit[j+1] = keyExit;
             }
     
-        StackEntry childEntry;
+            StackEntry nextChildEntry;
             // push children in reverse order (so the nearest is popped first) if stack has room
             for (int c = childCount - 1; c >= 0; --c)
             {
                 // push
-            childEntry.nodeIndex = childIdx[c];
+                nextChildEntry.nodeIndex = childIdx[c];
-            childEntry.center = childCenter[c];
+                nextChildEntry.center = childCenter[c];
-            childEntry.halfSize = childHalf;
+                nextChildEntry.halfSize = childHalf;
-            stack[sp++] = childEntry;
+                nextChildEntry.entry = childEntry[c];
+                nextChildEntry.exit = childExit[c];
+                stack[sp++] = nextChildEntry;
             }
         }
     
@@ -288,4 +297,5 @@ void CSMain(uint3 id : SV_DispatchThreadID)
                 hits.Append(outHit);
             }
         }
+    }
 }

Assets/Scripts/Debug/Player.cs

@@ -24,7 +24,7 @@ public class Player : MonoBehaviour
 
         VoxelRaycastGPU.Ray[] rays = new VoxelRaycastGPU.Ray[rayCount];
         FillRaysArray(rays);
-        voxelManager.gpuRayCaster.Init(rayCount, rays, 3);
+        voxelManager.gpuRayCaster.Init(rayCount, rays, 5);
     }
 
     void Cast( ref VoxelRaycastGPU.BatchData[] batchData, int batchCount, int iIteration )

Assets/Scripts/VoxelOctree/GPU/VoxelRaycastGpuManager.cs

@@ -95,6 +95,7 @@ public class VoxelRaycastGpuManager
             countBuffer.GetData(countArr);
             currentCount = countArr[0];
 
+            /**
                 sw.Stop();
     
                 VoxelRaycastGPU.BatchData[] hits = new VoxelRaycastGPU.BatchData[currentCount];
@@ -107,6 +108,7 @@ public class VoxelRaycastGpuManager
                 }
     
                 sw.Start();
+            */
 
             iteration++;