Solidly lit

- Upload a bitset for each section indicating if blocks are solid
- When interpolating light, count the number of transparent blocks and
  divide to avoid the incorrect "AO" effect near the ground

common/src/backend/java/dev/engine_room/flywheel/backend/engine/LightStorage.java

@@ -8,6 +8,7 @@ import org.lwjgl.system.MemoryUtil;
 import dev.engine_room.flywheel.api.task.Plan;
 import dev.engine_room.flywheel.backend.engine.indirect.StagingBuffer;
 import dev.engine_room.flywheel.backend.gl.buffer.GlBuffer;
+import dev.engine_room.flywheel.lib.math.MoreMath;
 import dev.engine_room.flywheel.lib.task.SimplePlan;
 import it.unimi.dsi.fastutil.ints.IntArrayList;
 import it.unimi.dsi.fastutil.longs.Long2IntMap;
@@ -36,7 +37,10 @@ import net.minecraft.world.level.lighting.LayerLightEventListener;
  * <p>Thus, each section occupies 5832 bytes.
  */
 public class LightStorage {
-	public static final long SECTION_SIZE_BYTES = 9 * 9 * 9 * 8;
+	public static final int BLOCKS_PER_SECTION = 18 * 18 * 18;
+	public static final int LIGHT_SIZE_BYTES = BLOCKS_PER_SECTION;
+	public static final int SOLID_SIZE_BYTES = MoreMath.ceilingDiv(BLOCKS_PER_SECTION, Integer.SIZE) * Integer.BYTES;
+	public static final int SECTION_SIZE_BYTES = SOLID_SIZE_BYTES + LIGHT_SIZE_BYTES;
 	private static final int DEFAULT_ARENA_CAPACITY_SECTIONS = 64;
 	private static final int INVALID_SECTION = -1;
 
@@ -158,6 +162,8 @@ public class LightStorage {
 		// Zero it out first. This is basically free and makes it easier to handle missing sections later.
 		MemoryUtil.memSet(ptr, 0, SECTION_SIZE_BYTES);
 
+		collectSolidData(ptr, section);
+
 		collectCenter(blockLight, skyLight, ptr, section);
 
 		for (SectionEdge i : SectionEdge.values()) {
@@ -175,6 +181,53 @@ public class LightStorage {
 		collectCorners(blockLight, skyLight, ptr, section);
 	}
 
+	private void collectSolidData(long ptr, long section) {
+		var blockPos = new BlockPos.MutableBlockPos();
+		int xMin = SectionPos.sectionToBlockCoord(SectionPos.x(section));
+		int yMin = SectionPos.sectionToBlockCoord(SectionPos.y(section));
+		int zMin = SectionPos.sectionToBlockCoord(SectionPos.z(section));
+
+		var bitSet = new BitSet(BLOCKS_PER_SECTION);
+		int index = 0;
+		for (int y = -1; y < 17; y++) {
+			for (int z = -1; z < 17; z++) {
+				for (int x = -1; x < 17; x++) {
+					blockPos.set(xMin + x, yMin + y, zMin + z);
+
+					boolean isFullBlock = level.getBlockState(blockPos)
+							.isCollisionShapeFullBlock(level, blockPos);
+
+					if (isFullBlock) {
+						bitSet.set(index);
+					}
+
+					index++;
+				}
+			}
+		}
+
+		var longArray = bitSet.toLongArray();
+		for (long l : longArray) {
+			MemoryUtil.memPutLong(ptr, l);
+			ptr += Long.BYTES;
+		}
+	}
+
+	private void writeSolid(long ptr, int index, boolean blockValid) {
+		if (!blockValid) {
+			return;
+		}
+		int intIndex = index / Integer.SIZE;
+		int bitIndex = index % Integer.SIZE;
+
+		long offset = intIndex * Integer.BYTES;
+
+		int bitField = MemoryUtil.memGetInt(ptr + offset);
+		bitField |= 1 << bitIndex;
+
+		MemoryUtil.memPutInt(ptr + offset, bitField);
+	}
+
 	private void collectXStrip(LayerLightEventListener blockLight, LayerLightEventListener skyLight, long ptr, long section, SectionEdge y, SectionEdge z) {
 		var pos = SectionPos.of(section);
 		var blockData = blockLight.getDataLayerData(pos);
@@ -303,7 +356,7 @@ public class LightStorage {
 
 		long packedByte = (block & 0xF) | ((sky & 0xF) << 4);
 
-		MemoryUtil.memPutByte(ptr + offset, (byte) packedByte);
+		MemoryUtil.memPutByte(ptr + SOLID_SIZE_BYTES + offset, (byte) packedByte);
 	}
 
 	/**

common/src/backend/resources/assets/flywheel/flywheel/internal/light_lut.glsl

@@ -1,6 +1,16 @@
-const uint _FLW_LIGHT_SECTION_SIZE_BYTES = 18 * 18 * 18;
+const uint _FLW_BLOCKS_PER_SECTION = 18 * 18 * 18;
+const uint _FLW_LIGHT_SIZE_BYTES = _FLW_BLOCKS_PER_SECTION;
+const uint _FLW_SOLID_SIZE_BYTES = ((_FLW_BLOCKS_PER_SECTION + 31) / 32) * 4;
+const uint _FLW_LIGHT_START_BYTES = _FLW_SOLID_SIZE_BYTES;
+const uint _FLW_LIGHT_SECTION_SIZE_BYTES = _FLW_SOLID_SIZE_BYTES + _FLW_LIGHT_SIZE_BYTES;
+
+const uint _FLW_SOLID_START_INTS = 0;
+const uint _FLW_LIGHT_START_INTS = _FLW_SOLID_SIZE_BYTES / 4;
 const uint _FLW_LIGHT_SECTION_SIZE_INTS = _FLW_LIGHT_SECTION_SIZE_BYTES / 4;
 
+const uint _FLW_COMPLETELY_SOLID = 0x7FFFFFFu;
+const float _FLW_EPSILON = 1e-5;
+
 uint _flw_indexLut(uint index);
 
 uint _flw_indexLight(uint index);
@@ -51,17 +61,28 @@ bool _flw_chunkCoordToSectionIndex(ivec3 sectionPos, out uint index) {
     return false;
 }
 
-vec2 _flw_lightAt(uint sectionOffset, uvec3 blockInSectionPos) {
+uvec2 _flw_lightAt(uint sectionOffset, uvec3 blockInSectionPos) {
     uint byteOffset = blockInSectionPos.x + blockInSectionPos.z * 18u + blockInSectionPos.y * 18u * 18u;
 
     uint uintOffset = byteOffset >> 2u;
     uint bitOffset = (byteOffset & 3u) << 3;
 
-    uint raw = _flw_indexLight(sectionOffset + uintOffset);
+    uint raw = _flw_indexLight(sectionOffset + _FLW_LIGHT_START_INTS + uintOffset);
     uint block = (raw >> bitOffset) & 0xFu;
     uint sky = (raw >> (bitOffset + 4u)) & 0xFu;
 
-    return vec2(block, sky);
+    return uvec2(block, sky);
+}
+
+bool _flw_isSolid(uint sectionOffset, uvec3 blockInSectionPos) {
+    uint bitOffset = blockInSectionPos.x + blockInSectionPos.z * 18u + blockInSectionPos.y * 18u * 18u;
+
+    uint uintOffset = bitOffset / 32u;
+    uint bitInWordOffset = bitOffset % 32u;
+
+    uint word = _flw_indexLight(sectionOffset + _FLW_SOLID_START_INTS + uintOffset);
+
+    return (word & (1u << bitInWordOffset)) != 0;
 }
 
 bool flw_lightFetch(ivec3 blockPos, out vec2 lightCoord) {
@@ -74,7 +95,7 @@ bool flw_lightFetch(ivec3 blockPos, out vec2 lightCoord) {
 
     uvec3 blockInSectionPos = (blockPos & 0xF) + 1;
 
-    lightCoord = _flw_lightAt(sectionOffset, blockInSectionPos) / 15.;
+    lightCoord = vec2(_flw_lightAt(sectionOffset, blockInSectionPos)) / 15.;
     return true;
 }
 
@@ -104,14 +125,14 @@ bool flw_light(vec3 worldPos, out vec2 lightCoord) {
     // Fetch everything for trilinear interpolation
     // Hypothetically we could re-order these and do some calculations in-between fetches
     // to help with latency hiding, but the compiler should be able to do that for us.
-    vec2 light000 = _flw_lightAt(sectionOffset, lowestCorner);
+    vec2 light000 = vec2(_flw_lightAt(sectionOffset, lowestCorner));
-    vec2 light001 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 0, 1));
+    vec2 light001 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 0, 1)));
-    vec2 light010 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 0));
+    vec2 light010 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 0)));
-    vec2 light011 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 1));
+    vec2 light011 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 1)));
-    vec2 light100 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 0));
+    vec2 light100 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 0)));
-    vec2 light101 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 1));
+    vec2 light101 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 1)));
-    vec2 light110 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 0));
+    vec2 light110 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 0)));
-    vec2 light111 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 1));
+    vec2 light111 = vec2(_flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 1)));
 
     vec2 light00 = mix(light000, light001, interpolant.z);
     vec2 light01 = mix(light010, light011, interpolant.z);
@@ -131,13 +152,19 @@ uint _flw_lightIndex(in uvec3 p) {
 
 /// Premtively collect all light in a 3x3x3 area centered on our block.
 /// Depending on the normal, we won't use all the data, but fetching on demand will have many duplicated fetches.
-vec2[27] _flw_lightFetch3x3x3(uint sectionOffset, ivec3 blockInSectionPos) {
+uvec3[27] _flw_fetchLight3x3x3(uint sectionOffset, ivec3 blockInSectionPos, uint solid) {
-    vec2[27] lights;
+    uvec3[27] lights;
 
+    uint index = 0u;
+    uint mask = 1u;
     for (int y = -1; y <= 1; y++) {
         for (int z = -1; z <= 1; z++) {
             for (int x = -1; x <= 1; x++) {
-                lights[_flw_lightIndex(uvec3(x + 1, y + 1, z + 1))] = _flw_lightAt(sectionOffset, uvec3(blockInSectionPos + ivec3(x, y, z)));
+                // 0 if the block is solid, 1 if it's not.
+                uint flag = uint((solid & mask) == 0u);
+                lights[index] = uvec3(_flw_lightAt(sectionOffset, uvec3(blockInSectionPos + ivec3(x, y, z))), flag);
+                index++;
+                mask <<= 1;
             }
         }
     }
@@ -145,6 +172,24 @@ vec2[27] _flw_lightFetch3x3x3(uint sectionOffset, ivec3 blockInSectionPos) {
     return lights;
 }
 
+uint _flw_fetchSolid3x3x3(uint sectionOffset, ivec3 blockInSectionPos) {
+    uint ret = 0;
+
+    uint index = 0;
+    for (int y = -1; y <= 1; y++) {
+        for (int z = -1; z <= 1; z++) {
+            for (int x = -1; x <= 1; x++) {
+                bool flag = _flw_isSolid(sectionOffset, uvec3(blockInSectionPos + ivec3(x, y, z)));
+                ret |= uint(flag) << index;
+
+                index++;
+            }
+        }
+    }
+
+    return ret;
+}
+
 /// Calculate the light for a direction by averaging the light at the corners of the block.
 ///
 /// To make this reusable across directions, c00..c11 choose what values relative to each corner to use.
@@ -155,16 +200,26 @@ vec2[27] _flw_lightFetch3x3x3(uint sectionOffset, ivec3 blockInSectionPos) {
 /// @param lights The light data for the 3x3x3 area.
 /// @param interpolant The position within the center block.
 /// @param c00..c11 4 offsets to determine which "direction" we are averaging.
-vec2 _flw_lightForDirection(in vec2[27] lights, in vec3 interpolant, in uvec3 c00, in uvec3 c01, in uvec3 c10, in uvec3 c11) {
+vec2 _flw_lightForDirection(in uvec3[27] lights, in vec3 interpolant, in uvec3 c00, in uvec3 c01, in uvec3 c10, in uvec3 c11) {
 
-    vec2 light000 = lights[_flw_lightIndex(c00 + uvec3(0u, 0u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 0u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 0u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 0u, 0u))];
+    uvec3 i000 = lights[_flw_lightIndex(c00 + uvec3(0u, 0u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 0u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 0u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 0u, 0u))];
-    vec2 light001 = lights[_flw_lightIndex(c00 + uvec3(0u, 0u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 0u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 0u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 0u, 1u))];
+    uvec3 i001 = lights[_flw_lightIndex(c00 + uvec3(0u, 0u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 0u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 0u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 0u, 1u))];
-    vec2 light010 = lights[_flw_lightIndex(c00 + uvec3(0u, 1u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 1u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 1u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 1u, 0u))];
+    uvec3 i010 = lights[_flw_lightIndex(c00 + uvec3(0u, 1u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 1u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 1u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 1u, 0u))];
-    vec2 light011 = lights[_flw_lightIndex(c00 + uvec3(0u, 1u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 1u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 1u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 1u, 1u))];
+    uvec3 i011 = lights[_flw_lightIndex(c00 + uvec3(0u, 1u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(0u, 1u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(0u, 1u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(0u, 1u, 1u))];
-    vec2 light100 = lights[_flw_lightIndex(c00 + uvec3(1u, 0u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 0u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 0u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 0u, 0u))];
+    uvec3 i100 = lights[_flw_lightIndex(c00 + uvec3(1u, 0u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 0u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 0u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 0u, 0u))];
-    vec2 light101 = lights[_flw_lightIndex(c00 + uvec3(1u, 0u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 0u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 0u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 0u, 1u))];
+    uvec3 i101 = lights[_flw_lightIndex(c00 + uvec3(1u, 0u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 0u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 0u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 0u, 1u))];
-    vec2 light110 = lights[_flw_lightIndex(c00 + uvec3(1u, 1u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 1u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 1u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 1u, 0u))];
+    uvec3 i110 = lights[_flw_lightIndex(c00 + uvec3(1u, 1u, 0u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 1u, 0u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 1u, 0u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 1u, 0u))];
-    vec2 light111 = lights[_flw_lightIndex(c00 + uvec3(1u, 1u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 1u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 1u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 1u, 1u))];
+    uvec3 i111 = lights[_flw_lightIndex(c00 + uvec3(1u, 1u, 1u))] + lights[_flw_lightIndex(c01 + uvec3(1u, 1u, 1u))] + lights[_flw_lightIndex(c10 + uvec3(1u, 1u, 1u))] + lights[_flw_lightIndex(c11 + uvec3(1u, 1u, 1u))];
+
+    // Divide by the number of light transmitting blocks to get the average.
+    vec2 light000 = i000.z == 0 ? vec2(0) : vec2(i000.xy) / float(i000.z);
+    vec2 light001 = i001.z == 0 ? vec2(0) : vec2(i001.xy) / float(i001.z);
+    vec2 light010 = i010.z == 0 ? vec2(0) : vec2(i010.xy) / float(i010.z);
+    vec2 light011 = i011.z == 0 ? vec2(0) : vec2(i011.xy) / float(i011.z);
+    vec2 light100 = i100.z == 0 ? vec2(0) : vec2(i100.xy) / float(i100.z);
+    vec2 light101 = i101.z == 0 ? vec2(0) : vec2(i101.xy) / float(i101.z);
+    vec2 light110 = i110.z == 0 ? vec2(0) : vec2(i110.xy) / float(i110.z);
+    vec2 light111 = i111.z == 0 ? vec2(0) : vec2(i111.xy) / float(i111.z);
 
     vec2 light00 = mix(light000, light001, interpolant.z);
     vec2 light01 = mix(light010, light011, interpolant.z);
@@ -174,8 +229,7 @@ vec2 _flw_lightForDirection(in vec2[27] lights, in vec3 interpolant, in uvec3 c0
     vec2 light0 = mix(light00, light01, interpolant.y);
     vec2 light1 = mix(light10, light11, interpolant.y);
 
-    // Divide by 60 (15 * 4) to normalize.
+    return mix(light0, light1, interpolant.x) / 15.;
-    return mix(light0, light1, interpolant.x) / 63.;
 }
 
 bool flw_light(vec3 worldPos, vec3 normal, out vec2 lightCoord) {
@@ -194,24 +248,31 @@ bool flw_light(vec3 worldPos, vec3 normal, out vec2 lightCoord) {
     // The block's position in the section adjusted into 18x18x18 space
     ivec3 blockInSectionPos = (blockPos & 0xF) + 1;
 
+    uint solid = _flw_fetchSolid3x3x3(sectionOffset, blockInSectionPos);
+
+    if (solid == _FLW_COMPLETELY_SOLID) {
+        lightCoord = vec2(0.);
+        return true;
+    }
+
     // Fetch everything in a 3x3x3 area centered around the block.
-    vec2[27] lights = _flw_lightFetch3x3x3(sectionOffset, blockInSectionPos);
+    uvec3[27] lights = _flw_fetchLight3x3x3(sectionOffset, blockInSectionPos, solid);
 
     vec3 interpolant = fract(worldPos);
 
     vec2 lightX;
-    if (normal.x > 0) {
+    if (normal.x > _FLW_EPSILON) {
         lightX = _flw_lightForDirection(lights, interpolant, uvec3(1u, 0u, 0u), uvec3(1u, 0u, 1u), uvec3(1u, 1u, 0u), uvec3(1u, 1u, 1u));
-    } else if (normal.x < 0) {
+    } else if (normal.x < -_FLW_EPSILON) {
         lightX = _flw_lightForDirection(lights, interpolant, uvec3(0u, 0u, 0u), uvec3(0u, 0u, 1u), uvec3(0u, 1u, 0u), uvec3(0u, 1u, 1u));
     } else {
         lightX = vec2(0.);
     }
 
     vec2 lightZ;
-    if (normal.z > 0) {
+    if (normal.z > _FLW_EPSILON) {
         lightZ = _flw_lightForDirection(lights, interpolant, uvec3(0u, 0u, 1u), uvec3(0u, 1u, 1u), uvec3(1u, 0u, 1u), uvec3(1u, 1u, 1u));
-    } else if (normal.z < 0) {
+    } else if (normal.z < -_FLW_EPSILON) {
         lightZ = _flw_lightForDirection(lights, interpolant, uvec3(0u, 0u, 0u), uvec3(0u, 1u, 0u), uvec3(1u, 0u, 0u), uvec3(1u, 1u, 0u));
     } else {
         lightZ = vec2(0.);
@@ -219,9 +280,9 @@ bool flw_light(vec3 worldPos, vec3 normal, out vec2 lightCoord) {
 
     vec2 lightY;
     // Average the light in relevant directions at each corner.
-    if (normal.y > 0.) {
+    if (normal.y > _FLW_EPSILON) {
         lightY = _flw_lightForDirection(lights, interpolant, uvec3(0u, 1u, 0u), uvec3(0u, 1u, 1u), uvec3(1u, 1u, 0u), uvec3(1u, 1u, 1u));
-    } else if (normal.y < 0.) {
+    } else if (normal.y < -_FLW_EPSILON) {
         lightY = _flw_lightForDirection(lights, interpolant, uvec3(0u, 0u, 0u), uvec3(0u, 0u, 1u), uvec3(1u, 0u, 0u), uvec3(1u, 0u, 1u));
     } else {
         lightY = vec2(0.);

common/src/lib/java/dev/engine_room/flywheel/lib/math/MoreMath.java

@@ -9,6 +9,10 @@ public final class MoreMath {
 	 */
 	public static final float SQRT_3_OVER_2 = (float) (Math.sqrt(3.0) / 2.0);
 
+	public static int align32(int size) {
+		return (size + 31) & ~31;
+	}
+
 	public static int align16(int size) {
 		return (size + 15) & ~15;
 	}