Lighting, for instance

- Sideport light lut stuffs to instancing engine
- Move actual lookup logic to light_lut.glsl, and have backend mains
  provide functions to index the backing storages for sanity's sake
- Standardize naming of lut and sections
- Pull in pepper's loom fix, so I can build :lwe:
- Allow specifying the internal format of texture buffers so light can
  be a simple uint array

common/src/backend/java/dev/engine_room/flywheel/backend/Samplers.java

@@ -8,4 +8,6 @@ public class Samplers {
 	public static final GlTextureUnit LIGHT = GlTextureUnit.T2;
 	public static final GlTextureUnit CRUMBLING = GlTextureUnit.T3;
 	public static final GlTextureUnit INSTANCE_BUFFER = GlTextureUnit.T4;
+	public static final GlTextureUnit LIGHT_LUT = GlTextureUnit.T5;
+	public static final GlTextureUnit LIGHT_SECTIONS = GlTextureUnit.T6;
 }

common/src/backend/java/dev/engine_room/flywheel/backend/compile/Pipelines.java

@@ -11,7 +11,11 @@ public final class Pipelines {
 			.vertexMain(Flywheel.rl("internal/instancing/main.vert"))
 			.fragmentMain(Flywheel.rl("internal/instancing/main.frag"))
 			.assembler(BufferTextureInstanceComponent::new)
-			.onLink(program -> program.setSamplerBinding("_flw_instances", Samplers.INSTANCE_BUFFER))
+			.onLink(program -> {
+				program.setSamplerBinding("_flw_instances", Samplers.INSTANCE_BUFFER);
+				program.setSamplerBinding("_flw_lightLut", Samplers.LIGHT_LUT);
+				program.setSamplerBinding("_flw_lightSections", Samplers.LIGHT_SECTIONS);
+			})
 			.build();
 
 	public static final Pipeline INDIRECT = Pipeline.builder()

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

@@ -8,6 +8,7 @@ import dev.engine_room.flywheel.api.event.RenderContext;
 import dev.engine_room.flywheel.api.task.Plan;
 import dev.engine_room.flywheel.backend.engine.EnvironmentStorage;
 import dev.engine_room.flywheel.backend.engine.indirect.StagingBuffer;
+import dev.engine_room.flywheel.backend.gl.buffer.GlBuffer;
 import dev.engine_room.flywheel.lib.task.SimplePlan;
 import it.unimi.dsi.fastutil.ints.IntArrayList;
 import it.unimi.dsi.fastutil.longs.Long2IntMap;
@@ -205,10 +206,6 @@ public class LightStorage {
 		arena.delete();
 	}
 
-	public boolean hasChanges() {
-		return !changed.isEmpty();
-	}
-
 	public boolean checkNeedsLutRebuildAndClear() {
 		var out = needsLutRebuild;
 		needsLutRebuild = false;
@@ -222,6 +219,15 @@ public class LightStorage {
 		changed.clear();
 	}
 
+	public void upload(GlBuffer buffer) {
+		if (changed.isEmpty()) {
+			return;
+		}
+
+		buffer.upload(arena.indexToPointer(0), arena.capacity() * SECTION_SIZE_BYTES);
+		changed.clear();
+	}
+
 	public IntArrayList createLut() {
 		// TODO: incremental lut updates
 		return LightLut.buildLut(section2ArenaIndex);

common/src/backend/java/dev/engine_room/flywheel/backend/engine/indirect/BufferBindings.java

@@ -6,8 +6,8 @@ public class BufferBindings {
 	public static final int MODEL_INDEX_BUFFER_BINDING = 2;
 	public static final int MODEL_BUFFER_BINDING = 3;
 	public static final int DRAW_BUFFER_BINDING = 4;
-	public static final int EMBEDDING_LUT_BINDING = 5;
-	public static final int EMBEDDING_LIGHT_BINDING = 6;
+	public static final int LIGHT_LUT_BINDING = 5;
+	public static final int LIGHT_SECTION_BINDING = 6;
 
 	private BufferBindings() {
 	}

common/src/backend/java/dev/engine_room/flywheel/backend/engine/indirect/LightBuffers.java

@@ -6,7 +6,7 @@ import org.lwjgl.system.MemoryUtil;
 import dev.engine_room.flywheel.backend.engine.embed.LightStorage;
 
 public class LightBuffers {
-	private final ResizableStorageArray lightArena = new ResizableStorageArray(LightStorage.SECTION_SIZE_BYTES);
+	private final ResizableStorageArray sections = new ResizableStorageArray(LightStorage.SECTION_SIZE_BYTES);
 	private final ResizableStorageArray lut = new ResizableStorageArray(4);
 
 	public LightBuffers() {
@@ -19,8 +19,8 @@ public class LightBuffers {
 			return;
 		}
 
-		lightArena.ensureCapacity(capacity);
-		light.uploadChangedSections(staging, lightArena.handle());
+		sections.ensureCapacity(capacity);
+		light.uploadChangedSections(staging, sections.handle());
 
 		if (light.checkNeedsLutRebuildAndClear()) {
 			var lut = light.createLut();
@@ -36,11 +36,11 @@ public class LightBuffers {
 	}
 
 	public void bind() {
-		if (lightArena.capacity() == 0) {
+		if (sections.capacity() == 0) {
 			return;
 		}
 
-		GL46.glBindBufferRange(GL46.GL_SHADER_STORAGE_BUFFER, BufferBindings.EMBEDDING_LUT_BINDING, lut.handle(), 0, lut.byteCapacity());
-		GL46.glBindBufferRange(GL46.GL_SHADER_STORAGE_BUFFER, BufferBindings.EMBEDDING_LIGHT_BINDING, lightArena.handle(), 0, lightArena.byteCapacity());
+		GL46.glBindBufferRange(GL46.GL_SHADER_STORAGE_BUFFER, BufferBindings.LIGHT_LUT_BINDING, lut.handle(), 0, lut.byteCapacity());
+		GL46.glBindBufferRange(GL46.GL_SHADER_STORAGE_BUFFER, BufferBindings.LIGHT_SECTION_BINDING, sections.handle(), 0, sections.byteCapacity());
 	}
 }

common/src/backend/java/dev/engine_room/flywheel/backend/engine/instancing/InstancedDrawManager.java

@@ -43,6 +43,7 @@ public class InstancedDrawManager extends DrawManager<InstancedInstancer<?>> {
 	private final MeshPool meshPool;
 	private final GlVertexArray vao;
 	private final TextureBuffer instanceTexture;
+	private final InstancedLight light;
 
 	public InstancedDrawManager(InstancingPrograms programs) {
 		programs.acquire();
@@ -51,6 +52,7 @@ public class InstancedDrawManager extends DrawManager<InstancedInstancer<?>> {
 		meshPool = new MeshPool();
 		vao = GlVertexArray.create();
 		instanceTexture = new TextureBuffer();
+		light = new InstancedLight();
 
 		meshPool.bind(vao);
 	}
@@ -78,6 +80,8 @@ public class InstancedDrawManager extends DrawManager<InstancedInstancer<?>> {
 		}
 
 		meshPool.flush();
+
+		light.flush(lightStorage);
 	}
 
 	@Override
@@ -92,6 +96,7 @@ public class InstancedDrawManager extends DrawManager<InstancedInstancer<?>> {
 			Uniforms.bindAll();
 			vao.bindForDraw();
 			TextureBinder.bindLightAndOverlay();
+			light.bind();
 
 			drawSet.draw(instanceTexture, programs);
 
@@ -114,6 +119,8 @@ public class InstancedDrawManager extends DrawManager<InstancedInstancer<?>> {
 		programs.release();
 		vao.delete();
 
+		light.delete();
+
 		super.delete();
 	}
 

common/src/backend/java/dev/engine_room/flywheel/backend/engine/instancing/InstancedLight.java

@@ -0,0 +1,62 @@
+package dev.engine_room.flywheel.backend.engine.instancing;
+
+import org.lwjgl.opengl.GL32;
+import org.lwjgl.system.MemoryUtil;
+
+import dev.engine_room.flywheel.backend.Samplers;
+import dev.engine_room.flywheel.backend.engine.embed.LightStorage;
+import dev.engine_room.flywheel.backend.gl.TextureBuffer;
+import dev.engine_room.flywheel.backend.gl.buffer.GlBuffer;
+import dev.engine_room.flywheel.lib.memory.MemoryBlock;
+
+public class InstancedLight {
+	private final GlBuffer lut;
+	private final GlBuffer sections;
+	private final TextureBuffer lutTexture;
+	private final TextureBuffer sectionsTexture;
+
+	public InstancedLight() {
+		sections = new GlBuffer();
+		lut = new GlBuffer();
+		lutTexture = new TextureBuffer(GL32.GL_R32UI);
+		sectionsTexture = new TextureBuffer(GL32.GL_R32UI);
+	}
+
+	public void bind() {
+		Samplers.LIGHT_LUT.makeActive();
+		lutTexture.bind(lut.handle());
+		Samplers.LIGHT_SECTIONS.makeActive();
+		sectionsTexture.bind(sections.handle());
+	}
+
+	public void flush(LightStorage light) {
+		if (light.capacity() == 0) {
+			return;
+		}
+
+		light.upload(sections);
+
+		if (light.checkNeedsLutRebuildAndClear()) {
+			var lut = light.createLut();
+
+			var up = MemoryBlock.malloc((long) lut.size() * Integer.BYTES);
+
+			long ptr = up.ptr();
+
+			for (int i = 0; i < lut.size(); i++) {
+				MemoryUtil.memPutInt(ptr + (long) Integer.BYTES * i, lut.getInt(i));
+			}
+
+			this.lut.upload(up);
+
+			up.free();
+		}
+	}
+
+	public void delete() {
+		sections.delete();
+		lut.delete();
+		lutTexture.delete();
+		sectionsTexture.delete();
+	}
+}

common/src/backend/java/dev/engine_room/flywheel/backend/gl/TextureBuffer.java

@@ -5,14 +5,20 @@ import org.lwjgl.opengl.GL32;
 public class TextureBuffer extends GlObject {
 	public static final int MAX_TEXELS = GL32.glGetInteger(GL32.GL_MAX_TEXTURE_BUFFER_SIZE);
 	public static final int MAX_BYTES = MAX_TEXELS * 16; // 4 channels * 4 bytes
+	private final int format;
 
 	public TextureBuffer() {
+		this(GL32.GL_RGBA32UI);
+	}
+
+	public TextureBuffer(int format) {
 		handle(GL32.glGenTextures());
+		this.format = format;
 	}
 
 	public void bind(int buffer) {
 		GL32.glBindTexture(GL32.GL_TEXTURE_BUFFER, handle());
-		GL32.glTexBuffer(GL32.GL_TEXTURE_BUFFER, GL32.GL_RGBA32UI, buffer);
+		GL32.glTexBuffer(GL32.GL_TEXTURE_BUFFER, format, buffer);
 	}
 
 	@Override

common/src/backend/resources/assets/flywheel/flywheel/internal/common.frag

@@ -13,9 +13,7 @@ uniform sampler2D _flw_crumblingTex;
 in vec2 _flw_crumblingTexCoord;
 #endif
 
-#ifdef _FLW_EMBEDDED
 bool _flw_embeddedLight(vec3 worldPos, out vec2 lightCoord);
-#endif
 
 flat in uint _flw_instanceID;
 

common/src/backend/resources/assets/flywheel/flywheel/internal/indirect/buffer_bindings.glsl

@@ -3,5 +3,5 @@
 #define _FLW_MODEL_INDEX_BUFFER_BINDING 2
 #define _FLW_MODEL_BUFFER_BINDING 3
 #define _FLW_DRAW_BUFFER_BINDING 4
-#define _FLW_EMBEDDING_LUT_BINDING 5
-#define _FLW_EMBEDDING_LIGHT_BINDING 6
+#define _FLW_LIGHT_LUT_BINDING 5
+#define _FLW_LIGHT_SECTIONS_BINDING 6

common/src/backend/resources/assets/flywheel/flywheel/internal/indirect/main.frag

@@ -1,131 +1,25 @@
 #include "flywheel:internal/common.frag"
+#include "flywheel:internal/light_lut.glsl"
 #include "flywheel:internal/indirect/buffer_bindings.glsl"
 
 flat in uvec3 _flw_packedMaterial;
 
-#ifdef _FLW_EMBEDDED
-
-layout(std430, binding = _FLW_EMBEDDING_LUT_BINDING) restrict readonly buffer EmbeddingLut {
-    uint _flw_embeddingLut[];
+layout(std430, binding = _FLW_LIGHT_LUT_BINDING) restrict readonly buffer LightLut {
+    uint _flw_lightLut[];
 };
 
-const uint _FLW_LIGHT_SECTION_SIZE_BYTES = 18 * 18 * 18;
-const uint _FLW_LIGHT_SECTION_SIZE_INTS = _FLW_LIGHT_SECTION_SIZE_BYTES / 4;
-
-layout(std430, binding = _FLW_EMBEDDING_LIGHT_BINDING) restrict readonly buffer LightSections {
+layout(std430, binding = _FLW_LIGHT_SECTIONS_BINDING) restrict readonly buffer LightSections {
     uint _flw_lightSections[];
 };
 
-/// Find the index for the next step in the LUT.
-/// @param base The base index in the LUT, should point to the start of a coordinate span.
-/// @param coord The coordinate to look for.
-/// @param next Output. The index of the next step in the LUT.
-/// @return true if the coordinate is not in the span.
-bool _flw_nextLut(uint base, int coord, out uint next) {
-    // The base coordinate.
-    int start = int(_flw_embeddingLut[base]);
-    // The width of the coordinate span.
-    uint size = _flw_embeddingLut[base + 1];
-
-    // Index of the coordinate in the span.
-    int i = coord - start;
-
-    if (i < 0 || i >= size) {
-        // We missed.
-        return true;
+uint _flw_indexLut(uint index) {
+    return _flw_lightLut[index];
 }
 
-    next = _flw_embeddingLut[base + 2 + i];
-
-    return false;
+uint _flw_indexLight(uint index) {
+    return _flw_lightSections[index];
 }
 
-bool _flw_chunkCoordToSectionIndex(ivec3 sectionPos, out uint index) {
-    uint y;
-    if (_flw_nextLut(0, sectionPos.x, y) || y == 0) {
-        return true;
-    }
-
-    uint z;
-    if (_flw_nextLut(y, sectionPos.y, z) || z == 0) {
-        return true;
-    }
-
-    uint sectionIndex;
-    if (_flw_nextLut(z, sectionPos.z, sectionIndex) || sectionIndex == 0) {
-        return true;
-    }
-
-    // The index is written as 1-based so we can properly detect missing sections.
-    index = sectionIndex - 1;
-
-    return false;
-}
-
-vec2 _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_lightSections[sectionOffset + uintOffset];
-    uint block = (raw >> bitOffset) & 0xFu;
-    uint sky = (raw >> (bitOffset + 4u)) & 0xFu;
-
-    return vec2(block, sky);
-}
-
-bool _flw_embeddedLight(vec3 worldPos, out vec2 lightCoord) {
-    // Always use the section of the block we are contained in to ensure accuracy.
-    // We don't want to interpolate between sections, but also we might not be able
-    // to rely on the existence neighboring sections, so don't do any extra rounding here.
-    ivec3 blockPos = ivec3(floor(worldPos));
-
-    uint lightSectionIndex;
-    if (_flw_chunkCoordToSectionIndex(blockPos >> 4, lightSectionIndex)) {
-        // TODO: useful debug mode for this.
-        // flw_fragOverlay = ivec2(0, 3);
-        return false;
-    }
-    // The offset of the section in the light buffer.
-    uint sectionOffset = lightSectionIndex * _FLW_LIGHT_SECTION_SIZE_INTS;
-
-    // The block's position in the section adjusted into 18x18x18 space
-    uvec3 blockInSectionPos = (blockPos & 0xF) + 1;
-
-    // The lowest corner of the 2x2x2 area we'll be trilinear interpolating.
-    // The ugly bit on the end evaluates to -1 or 0 depending on which side of 0.5 we are.
-    uvec3 lowestCorner = blockInSectionPos + ivec3(floor(fract(worldPos) - 0.5));
-
-    // The distance our fragment is from the center of the lowest corner.
-    vec3 interpolant = fract(worldPos - 0.5);
-
-    // 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 light001 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 0, 1));
-    vec2 light010 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 0));
-    vec2 light011 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 1));
-    vec2 light100 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 0));
-    vec2 light101 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 1));
-    vec2 light110 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 0));
-    vec2 light111 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 1));
-
-    vec2 light00 = mix(light000, light001, interpolant.z);
-    vec2 light01 = mix(light010, light011, interpolant.z);
-    vec2 light10 = mix(light100, light101, interpolant.z);
-    vec2 light11 = mix(light110, light111, interpolant.z);
-
-    vec2 light0 = mix(light00, light01, interpolant.y);
-    vec2 light1 = mix(light10, light11, interpolant.y);
-
-    lightCoord = mix(light0, light1, interpolant.x) / 15.;
-    return true;
-}
-
-#endif
-
 void main() {
     _flw_uberMaterialFragmentIndex = _flw_packedMaterial.x;
     _flw_unpackUint2x16(_flw_packedMaterial.y, _flw_uberCutoutIndex, _flw_uberFogIndex);

common/src/backend/resources/assets/flywheel/flywheel/internal/instancing/main.frag

@@ -1,12 +1,18 @@
 #include "flywheel:internal/common.frag"
+#include "flywheel:internal/light_lut.glsl"
 
 uniform uvec4 _flw_packedMaterial;
 
-#ifdef _FLW_EMBEDDED
-bool _flw_embeddedLight(vec3 worldPos, out vec2 lightCoord) {
-    return true;
+uniform usamplerBuffer _flw_lightLut;
+uniform usamplerBuffer _flw_lightSections;
+
+uint _flw_indexLut(uint index) {
+    return texelFetch(_flw_lightLut, int(index)).r;
+}
+
+uint _flw_indexLight(uint index) {
+    return texelFetch(_flw_lightSections, int(index)).r;
 }
-#endif
 
 void main() {
     _flw_uberMaterialFragmentIndex = _flw_packedMaterial.y;

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

@@ -0,0 +1,115 @@
+const uint _FLW_LIGHT_SECTION_SIZE_BYTES = 18 * 18 * 18;
+const uint _FLW_LIGHT_SECTION_SIZE_INTS = _FLW_LIGHT_SECTION_SIZE_BYTES / 4;
+
+uint _flw_indexLut(uint index);
+
+uint _flw_indexLight(uint index);
+
+/// Find the index for the next step in the LUT.
+/// @param base The base index in the LUT, should point to the start of a coordinate span.
+/// @param coord The coordinate to look for.
+/// @param next Output. The index of the next step in the LUT.
+/// @return true if the coordinate is not in the span.
+bool _flw_nextLut(uint base, int coord, out uint next) {
+    // The base coordinate.
+    int start = int(_flw_indexLut(base));
+    // The width of the coordinate span.
+    uint size = _flw_indexLut(base + 1);
+
+    // Index of the coordinate in the span.
+    int i = coord - start;
+
+    if (i < 0 || i >= size) {
+        // We missed.
+        return true;
+    }
+
+    next = _flw_indexLut(base + 2 + i);
+
+    return false;
+}
+
+bool _flw_chunkCoordToSectionIndex(ivec3 sectionPos, out uint index) {
+    uint y;
+    if (_flw_nextLut(0, sectionPos.x, y) || y == 0) {
+        return true;
+    }
+
+    uint z;
+    if (_flw_nextLut(y, sectionPos.y, z) || z == 0) {
+        return true;
+    }
+
+    uint sectionIndex;
+    if (_flw_nextLut(z, sectionPos.z, sectionIndex) || sectionIndex == 0) {
+        return true;
+    }
+
+    // The index is written as 1-based so we can properly detect missing sections.
+    index = sectionIndex - 1;
+
+    return false;
+}
+
+vec2 _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 block = (raw >> bitOffset) & 0xFu;
+    uint sky = (raw >> (bitOffset + 4u)) & 0xFu;
+
+    return vec2(block, sky);
+}
+
+bool _flw_embeddedLight(vec3 worldPos, out vec2 lightCoord) {
+    // Always use the section of the block we are contained in to ensure accuracy.
+    // We don't want to interpolate between sections, but also we might not be able
+    // to rely on the existence neighboring sections, so don't do any extra rounding here.
+    ivec3 blockPos = ivec3(floor(worldPos));
+
+    uint lightSectionIndex;
+    if (_flw_chunkCoordToSectionIndex(blockPos >> 4, lightSectionIndex)) {
+        // TODO: useful debug mode for this.
+        // flw_fragOverlay = ivec2(0, 3);
+        return false;
+    }
+    // The offset of the section in the light buffer.
+    uint sectionOffset = lightSectionIndex * _FLW_LIGHT_SECTION_SIZE_INTS;
+
+    // The block's position in the section adjusted into 18x18x18 space
+    uvec3 blockInSectionPos = (blockPos & 0xF) + 1;
+
+    // The lowest corner of the 2x2x2 area we'll be trilinear interpolating.
+    // The ugly bit on the end evaluates to -1 or 0 depending on which side of 0.5 we are.
+    uvec3 lowestCorner = blockInSectionPos + ivec3(floor(fract(worldPos) - 0.5));
+
+    // The distance our fragment is from the center of the lowest corner.
+    vec3 interpolant = fract(worldPos - 0.5);
+
+    // 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 light001 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 0, 1));
+    vec2 light010 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 0));
+    vec2 light011 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(0, 1, 1));
+    vec2 light100 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 0));
+    vec2 light101 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 0, 1));
+    vec2 light110 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 0));
+    vec2 light111 = _flw_lightAt(sectionOffset, lowestCorner + uvec3(1, 1, 1));
+
+    vec2 light00 = mix(light000, light001, interpolant.z);
+    vec2 light01 = mix(light010, light011, interpolant.z);
+    vec2 light10 = mix(light100, light101, interpolant.z);
+    vec2 light11 = mix(light110, light111, interpolant.z);
+
+    vec2 light0 = mix(light00, light01, interpolant.y);
+    vec2 light1 = mix(light10, light11, interpolant.y);
+
+    lightCoord = mix(light0, light1, interpolant.x) / 15.;
+    return true;
+}
+