fast section sorting to fix remaining inter-section intra-region sorting issues stemming from bfs traversal with frustum not producing fully correct results

Author: douira

common/src/main/java/net/caffeinemc/mods/sodium/client/render/chunk/lists/ChunkRenderList.java

@@ -1,11 +1,14 @@
 package net.caffeinemc.mods.sodium.client.render.chunk.lists;
 
+import net.caffeinemc.mods.sodium.client.render.chunk.LocalSectionIndex;
 import net.caffeinemc.mods.sodium.client.render.chunk.RenderSection;
 import net.caffeinemc.mods.sodium.client.render.chunk.RenderSectionFlags;
+import net.caffeinemc.mods.sodium.client.render.chunk.region.RenderRegion;
+import net.caffeinemc.mods.sodium.client.render.viewport.CameraTransform;
+import net.caffeinemc.mods.sodium.client.util.iterator.ByteArrayIterator;
 import net.caffeinemc.mods.sodium.client.util.iterator.ByteIterator;
 import net.caffeinemc.mods.sodium.client.util.iterator.ReversibleByteArrayIterator;
-import net.caffeinemc.mods.sodium.client.util.iterator.ByteArrayIterator;
-import net.caffeinemc.mods.sodium.client.render.chunk.region.RenderRegion;
+import net.minecraft.util.Mth;
 import org.jetbrains.annotations.Nullable;
 
 public class ChunkRenderList {
@@ -37,6 +40,39 @@ public void reset(int frame) {
         this.lastVisibleFrame = frame;
     }
 
+    // clamping the relative camera position to the region bounds means there can only be very few different distances
+    private static final int SORTING_HISTOGRAM_SIZE = RenderRegion.REGION_WIDTH + RenderRegion.REGION_HEIGHT + RenderRegion.REGION_LENGTH - 2;
+
+    public void sortSections(CameraTransform transform, int[] sortItems) {
+        var cameraX = Mth.clamp((transform.intX >> 4) - this.region.getChunkX(), 0, RenderRegion.REGION_WIDTH - 1);
+        var cameraY = Mth.clamp((transform.intY >> 4) - this.region.getChunkY(), 0, RenderRegion.REGION_HEIGHT - 1);
+        var cameraZ = Mth.clamp((transform.intZ >> 4) - this.region.getChunkZ(), 0, RenderRegion.REGION_LENGTH - 1);
+
+        int[] histogram = new int[SORTING_HISTOGRAM_SIZE];
+
+        for (int i = 0; i < this.sectionsWithGeometryCount; i++) {
+            var index = this.sectionsWithGeometry[i] & 0xFF; // makes sure the byte -> int conversion is unsigned
+            var x = Math.abs(LocalSectionIndex.unpackX(index) - cameraX);
+            var y = Math.abs(LocalSectionIndex.unpackY(index) - cameraY);
+            var z = Math.abs(LocalSectionIndex.unpackZ(index) - cameraZ);
+
+            var distance = x + y + z;
+            histogram[distance]++;
+            sortItems[i] = distance << 8 | index;
+        }
+
+        // prefix sum to calculate indexes
+        for (int i = 1; i < SORTING_HISTOGRAM_SIZE; i++) {
+            histogram[i] += histogram[i - 1];
+        }
+
+        for (int i = 0; i < this.sectionsWithGeometryCount; i++) {
+            var item = sortItems[i];
+            var distance = item >>> 8;
+            this.sectionsWithGeometry[--histogram[distance]] = (byte) item;
+        }
+    }
+
     public void add(RenderSection render) {
         if (this.size >= RenderRegion.REGION_SIZE) {
             throw new ArrayIndexOutOfBoundsException("Render list is full");

common/src/main/java/net/caffeinemc/mods/sodium/client/render/chunk/lists/VisibleChunkCollector.java

@@ -1,5 +1,6 @@
 package net.caffeinemc.mods.sodium.client.render.chunk.lists;
 
+import it.unimi.dsi.fastutil.ints.IntArrays;
 import it.unimi.dsi.fastutil.objects.ObjectArrayList;
 import net.caffeinemc.mods.sodium.client.render.chunk.ChunkUpdateType;
 import net.caffeinemc.mods.sodium.client.render.chunk.RenderSection;
@@ -62,6 +63,8 @@ private void addToRebuildLists(RenderSection section) {
         }
     }
 
+    private static int[] sortItems = new int[RenderRegion.REGION_SIZE];
+
     public SortedRenderLists createRenderLists(Viewport viewport) {
         // sort the regions by distance to fix rare region ordering bugs
         var transform = viewport.getTransform();
@@ -70,20 +73,29 @@ public SortedRenderLists createRenderLists(Viewport viewport) {
         var cameraZ = transform.intZ >> (4 + RenderRegion.REGION_LENGTH_SH);
         var size = this.sortedRenderLists.size();
 
-        var items = new int[size];
+        if (sortItems.length < size) {
+            sortItems = new int[size];
+        }
+
         for (var i = 0; i < size; i++) {
             var region = this.sortedRenderLists.get(i).getRegion();
             var x = Math.abs(region.getX() - cameraX);
             var y = Math.abs(region.getY() - cameraY);
             var z = Math.abs(region.getZ() - cameraZ);
-            items[i] = (x + y + z) << 16 | i;
+            sortItems[i] = (x + y + z) << 16 | i;
         }
 
-        Arrays.sort(items);
+        IntArrays.unstableSort(sortItems, 0, size);
 
         var sorted = new ObjectArrayList<ChunkRenderList>(size);
-        for (var key : items) {
-            sorted.add(this.sortedRenderLists.get(key & 0xFFFF));
+        for (var i = 0; i < size; i++) {
+            var key = sortItems[i];
+            var renderList = this.sortedRenderLists.get(key & 0xFFFF);
+            sorted.add(renderList);
+        }
+
+        for (var list : sorted) {
+            list.sortSections(transform, sortItems);
         }
 
         return new SortedRenderLists(sorted);