Functional frustum filter

- Manually fix alignment issue
 - Align UBO size to 16
 - Manual frustum ubo upload

src/main/java/com/jozufozu/flywheel/api/uniform/UniformProvider.java

@@ -9,7 +9,7 @@ public abstract class UniformProvider {
 	protected ByteBuffer buffer;
 	protected Notifier notifier;
 
-	public abstract int getSize();
+	public abstract int getActualByteSize();
 
 	public void updatePtr(ByteBuffer backing, Notifier notifier) {
 		this.buffer = backing;

src/main/java/com/jozufozu/flywheel/backend/instancing/indirect/FrustumUBO.java

@@ -0,0 +1,31 @@
+package com.jozufozu.flywheel.backend.instancing.indirect;
+
+import static org.lwjgl.opengl.GL46.*;
+
+import org.lwjgl.system.MemoryUtil;
+
+import com.jozufozu.flywheel.util.joml.FrustumIntersection;
+
+// This should be a push constant :whywheel:
+public class FrustumUBO {
+
+	public static final int BUFFER_SIZE = 96;
+
+	private final int ubo;
+	private final long clientStorage;
+
+	FrustumUBO() {
+		ubo = glCreateBuffers();
+		glNamedBufferStorage(ubo, BUFFER_SIZE, GL_DYNAMIC_STORAGE_BIT);
+		clientStorage = MemoryUtil.nmemAlloc(BUFFER_SIZE);
+	}
+
+	public void update(FrustumIntersection frustum) {
+		frustum.getJozuPackedPlanes(clientStorage);
+		nglNamedBufferSubData(ubo, 0, BUFFER_SIZE, clientStorage);
+	}
+
+	public void bind() {
+		glBindBufferBase(GL_UNIFORM_BUFFER, 0, ubo);
+	}
+}

src/main/java/com/jozufozu/flywheel/backend/instancing/indirect/IndirectEngine.java

@@ -17,7 +17,6 @@ import com.jozufozu.flywheel.backend.gl.GlTextureUnit;
 import com.jozufozu.flywheel.backend.instancing.Engine;
 import com.jozufozu.flywheel.backend.instancing.InstanceManager;
 import com.jozufozu.flywheel.backend.instancing.TaskEngine;
-import com.jozufozu.flywheel.backend.instancing.instancing.MeshPool;
 import com.jozufozu.flywheel.core.RenderContext;
 import com.jozufozu.flywheel.api.context.ContextShader;
 import com.jozufozu.flywheel.backend.instancing.instancing.InstancedArraysCompiler;
@@ -46,6 +45,8 @@ public class IndirectEngine implements Engine {
 	protected final List<InstancedModel<?>> uninitializedModels = new ArrayList<>();
 	protected final RenderLists renderLists = new RenderLists();
 
+	private FrustumUBO frustumUBO;
+
 	/**
 	 * The set of instance managers that are attached to this engine.
 	 */
@@ -76,7 +77,7 @@ public class IndirectEngine implements Engine {
 		setup();
 
 		for (var group : groups) {
-			group.submit();
+			group.submit(frustumUBO);
 		}
 
 	}
@@ -145,10 +146,26 @@ public class IndirectEngine implements Engine {
 
 	@Override
 	public void beginFrame(TaskEngine taskEngine, RenderContext context) {
+		if (frustumUBO == null) {
+			frustumUBO = new FrustumUBO();
+		}
+
 		for (var model : uninitializedModels) {
 			model.init(renderLists);
 		}
 		uninitializedModels.clear();
+
+		Vec3 camera = context.camera()
+				.getPosition();
+
+		var camX = (float) (camera.x - originCoordinate.getX());
+		var camY = (float) (camera.y - originCoordinate.getY());
+		var camZ = (float) (camera.z - originCoordinate.getZ());
+
+		var culler = RenderContext.createCuller(context.viewProjection(), -camX, -camY, -camZ);
+
+		frustumUBO.update(culler);
+
 	}
 
 	private void shiftListeners(int cX, int cY, int cZ) {

src/main/java/com/jozufozu/flywheel/backend/instancing/indirect/IndirectList.java

@@ -2,18 +2,15 @@ package com.jozufozu.flywheel.backend.instancing.indirect;
 
 import static org.lwjgl.opengl.GL46.*;
 
-import java.text.Format;
 import java.util.ArrayList;
 import java.util.List;
 
-import org.lwjgl.opengl.GL46C;
 import org.lwjgl.system.MemoryStack;
 import org.lwjgl.system.MemoryUtil;
 
 import com.jozufozu.flywheel.api.instancer.InstancedPart;
 import com.jozufozu.flywheel.api.struct.StorageBufferWriter;
 import com.jozufozu.flywheel.api.struct.StructType;
-import com.jozufozu.flywheel.backend.gl.GlNumericType;
 import com.jozufozu.flywheel.backend.gl.shader.GlProgram;
 import com.jozufozu.flywheel.core.Components;
 import com.jozufozu.flywheel.core.Materials;
@@ -24,7 +21,7 @@ import com.jozufozu.flywheel.core.vertex.Formats;
 
 public class IndirectList<T extends InstancedPart> {
 
-	private static final int DRAW_COMMAND_STRIDE = 20;
+	private static final long DRAW_COMMAND_STRIDE = 20;
 
 	final StorageBufferWriter<T> storageBufferWriter;
 	final GlProgram compute;
@@ -51,12 +48,13 @@ public class IndirectList<T extends InstancedPart> {
 	 * Stores drawIndirect structs.
 	 */
 	int drawBuffer;
+	int debugBuffer;
 
 	final IndirectMeshPool meshPool;
 
 	int vertexArray;
 
-	final int[] shaderStorageBuffers = new int[4];
+	final int[] shaderStorageBuffers = new int[5];
 
 	final List<Batch<T>> batches = new ArrayList<>();
 
@@ -73,10 +71,11 @@ public class IndirectList<T extends InstancedPart> {
 		targetBuffer = shaderStorageBuffers[1];
 		boundingSphereBuffer = shaderStorageBuffers[2];
 		drawBuffer = shaderStorageBuffers[3];
+		debugBuffer = shaderStorageBuffers[4];
 		meshPool = new IndirectMeshPool(Formats.BLOCK, 1024);
 
 		// FIXME: Resizable buffers
-		maxObjectCount = 1024L * 100L;
+		maxObjectCount = 1024L;
 		maxBatchCount = 64;
 
 		// +4 for the batch id
@@ -85,6 +84,7 @@ public class IndirectList<T extends InstancedPart> {
 		glNamedBufferStorage(targetBuffer, 4 * maxObjectCount, GL_DYNAMIC_STORAGE_BIT);
 		glNamedBufferStorage(boundingSphereBuffer, 16 * maxBatchCount, GL_DYNAMIC_STORAGE_BIT);
 		glNamedBufferStorage(drawBuffer, DRAW_COMMAND_STRIDE * maxBatchCount, GL_DYNAMIC_STORAGE_BIT);
+		glNamedBufferStorage(debugBuffer, 4 * maxObjectCount, GL_DYNAMIC_STORAGE_BIT);
 
 		objectClientStorage = MemoryUtil.nmemAlloc(objectStride * maxObjectCount);
 
@@ -115,17 +115,13 @@ public class IndirectList<T extends InstancedPart> {
 			offset += attribute
 					.getByteWidth();
 		}
-
-		glEnableVertexArrayAttrib(vertexArray, meshAttribs);
-		glVertexArrayVertexBuffer(vertexArray, meshAttribs, targetBuffer, 0, 4);
-		glVertexArrayAttribIFormat(vertexArray, meshAttribs, 1, GlNumericType.UINT.getGlEnum(), 0);
 	}
 
 	public void add(Mesh mesh, IndirectInstancer<T> instancer) {
 		batches.add(new Batch<>(instancer, meshPool.alloc(mesh)));
 	}
 
-	void submit() {
+	void submit(FrustumUBO frustumUBO) {
 		int instanceCountThisFrame = calculateTotalInstanceCount();
 
 		if (instanceCountThisFrame == 0) {
@@ -137,14 +133,15 @@ public class IndirectList<T extends InstancedPart> {
 		uploadBoundingSpheres();
 		uploadIndirectCommands();
 
-		UniformBuffer.getInstance().sync();
+		frustumUBO.bind();
-
 		dispatchCompute(instanceCountThisFrame);
 		issueMemoryBarrier();
 		dispatchDraw();
 	}
 
 	private void dispatchDraw() {
+		UniformBuffer.getInstance().sync();
+
 		draw.bind();
 		Materials.BELL.setup();
 		glVertexArrayElementBuffer(vertexArray, elementBuffer);
@@ -170,13 +167,17 @@ public class IndirectList<T extends InstancedPart> {
 				ptr += 16;
 			}
 
-			GL46C.nglNamedBufferSubData(boundingSphereBuffer, 0, size, basePtr);
+			nglNamedBufferSubData(boundingSphereBuffer, 0, size, basePtr);
 		}
 	}
 
 	private void dispatchCompute(int instanceCount) {
 		compute.bind();
-		glBindBuffersBase(GL_SHADER_STORAGE_BUFFER, 0, shaderStorageBuffers);
+		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, objectBuffer, 0, instanceCount * objectStride);
+		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 1, targetBuffer, 0, instanceCount * 4L);
+		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 2, boundingSphereBuffer, 0, batches.size() * 16L);
+		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 3, drawBuffer, 0, batches.size() * DRAW_COMMAND_STRIDE);
+		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 4, debugBuffer, 0, instanceCount * 4L);
 
 		var groupCount = (instanceCount + 31) >> 5; // ceil(totalInstanceCount / 32)
 		glDispatchCompute(groupCount, 1, 1);
@@ -197,20 +198,20 @@ public class IndirectList<T extends InstancedPart> {
 			batchID++;
 		}
 
-		GL46C.nglNamedBufferSubData(objectBuffer, 0, ptr - objectClientStorage, objectClientStorage);
+		nglNamedBufferSubData(objectBuffer, 0, ptr - objectClientStorage, objectClientStorage);
 
 	}
 
 	private void uploadIndirectCommands() {
 		try (var stack = MemoryStack.stackPush()) {
-			var size = batches.size() * DRAW_COMMAND_STRIDE;
+			long size = batches.size() * DRAW_COMMAND_STRIDE;
-			long basePtr = stack.nmalloc(size);
+			long basePtr = stack.nmalloc((int) size);
 			long writePtr = basePtr;
 			for (Batch<T> batch : batches) {
 				batch.writeIndirectCommand(writePtr);
 				writePtr += DRAW_COMMAND_STRIDE;
 			}
-			GL46C.nglNamedBufferSubData(drawBuffer, 0, size, basePtr);
+			nglNamedBufferSubData(drawBuffer, 0, size, basePtr);
 		}
 	}
 

src/main/java/com/jozufozu/flywheel/backend/instancing/indirect/RenderLists.java

@@ -1,5 +1,6 @@
 package com.jozufozu.flywheel.backend.instancing.indirect;
 
+import java.util.ArrayList;
 import java.util.Collection;
 import java.util.Collections;
 import java.util.EnumMap;

src/main/java/com/jozufozu/flywheel/core/RenderContext.java

@@ -23,13 +23,10 @@ public record RenderContext(LevelRenderer renderer, ClientLevel level, PoseStack
 		return viewProjection;
 	}
 
-	public static FrustumIntersection createCuller(Camera camera, Matrix4f viewProjection) {
+	public static FrustumIntersection createCuller(Matrix4f viewProjection, float camX, float camY, float camZ) {
 		com.jozufozu.flywheel.util.joml.Matrix4f proj = Matrix4fExtension.clone(viewProjection);
 
-		Vec3 cam = camera
+		proj.translate(camX, camY, camZ);
-				.getPosition();
-
-		proj.translate((float) -cam.x, (float) -cam.y, (float) -cam.z);
 
 		return new FrustumIntersection(proj);
 	}

src/main/java/com/jozufozu/flywheel/core/structs/oriented/OrientedStorageWriter.java

@@ -23,23 +23,23 @@ public class OrientedStorageWriter implements StorageBufferWriter<OrientedPart>
 		MemoryUtil.memPutFloat(ptr + 20, d.posY);
 		MemoryUtil.memPutFloat(ptr + 24, d.posZ);
 
-		MemoryUtil.memPutFloat(ptr + 28, d.pivotX);
+		MemoryUtil.memPutFloat(ptr + 32, d.pivotX);
-		MemoryUtil.memPutFloat(ptr + 32, d.pivotY);
+		MemoryUtil.memPutFloat(ptr + 36, d.pivotY);
-		MemoryUtil.memPutFloat(ptr + 36, d.pivotZ);
+		MemoryUtil.memPutFloat(ptr + 40, d.pivotZ);
 
-		MemoryUtil.memPutShort(ptr + 40, d.blockLight);
+		MemoryUtil.memPutShort(ptr + 44, d.blockLight);
-		MemoryUtil.memPutShort(ptr + 42, d.skyLight);
+		MemoryUtil.memPutShort(ptr + 46, d.skyLight);
 
-		MemoryUtil.memPutByte(ptr + 44, d.r);
+		MemoryUtil.memPutByte(ptr + 48, d.r);
-		MemoryUtil.memPutByte(ptr + 45, d.g);
+		MemoryUtil.memPutByte(ptr + 49, d.g);
-		MemoryUtil.memPutByte(ptr + 46, d.b);
+		MemoryUtil.memPutByte(ptr + 50, d.b);
-		MemoryUtil.memPutByte(ptr + 47, d.a);
+		MemoryUtil.memPutByte(ptr + 51, d.a);
 
-		MemoryUtil.memPutInt(ptr + 48, batchID);
+		MemoryUtil.memPutInt(ptr + 52, batchID);
 	}
 
 	@Override
 	public int getAlignment() {
-		return 52;
+		return 64;
 	}
 }

src/main/java/com/jozufozu/flywheel/core/uniform/FogProvider.java

@@ -12,7 +12,7 @@ public class FogProvider extends UniformProvider {
 
 
 	@Override
-	public int getSize() {
+	public int getActualByteSize() {
 		return 16 + 8 + 4;
 	}
 

src/main/java/com/jozufozu/flywheel/core/uniform/UniformBuffer.java

@@ -47,11 +47,11 @@ public class UniformBuffer {
 		int totalBytes = 0;
 		int index = 0;
 		for (UniformProvider provider : providers) {
-			int size = provider.getSize();
+			int size = alignPo2(provider.getActualByteSize(), 16);
 
 			builder.add(new Allocated(provider, totalBytes, size, index));
 
-			totalBytes = align(totalBytes + size);
+			totalBytes = alignUniformBuffer(totalBytes + size);
 			index++;
 		}
 
@@ -82,7 +82,7 @@ public class UniformBuffer {
 	}
 
 	// https://stackoverflow.com/questions/3407012/rounding-up-to-the-nearest-multiple-of-a-number
-	private static int align(int numToRound) {
+	private static int alignUniformBuffer(int numToRound) {
 		if (PO2_ALIGNMENT) {
 			return (numToRound + OFFSET_ALIGNMENT - 1) & -OFFSET_ALIGNMENT;
 		} else {
@@ -90,6 +90,10 @@ public class UniformBuffer {
 		}
 	}
 
+	private static int alignPo2(int numToRound, int alignment) {
+		return (numToRound + alignment - 1) & -alignment;
+	}
+
 	private class Allocated implements UniformProvider.Notifier {
 		private final UniformProvider provider;
 		private final int offset;

src/main/java/com/jozufozu/flywheel/core/uniform/ViewProvider.java

@@ -26,7 +26,7 @@ public class ViewProvider extends UniformProvider {
 	}
 
 	@Override
-	public int getSize() {
+	public int getActualByteSize() {
 		return 4 * 16 + 16 + 4;
 	}
 

src/main/java/com/jozufozu/flywheel/mixin/LevelRendererMixin.java

@@ -27,6 +27,7 @@ import net.minecraft.client.renderer.GameRenderer;
 import net.minecraft.client.renderer.LevelRenderer;
 import net.minecraft.client.renderer.LightTexture;
 import net.minecraft.client.renderer.RenderBuffers;
+import net.minecraft.world.phys.Vec3;
 import net.minecraftforge.common.MinecraftForge;
 
 @Mixin(value = LevelRenderer.class, priority = 1001) // Higher priority to go after sodium
@@ -44,7 +45,8 @@ public class LevelRendererMixin {
 	@Inject(at = @At("HEAD"), method = "renderLevel")
 	private void beginRender(PoseStack pPoseStack, float pPartialTick, long pFinishNanoTime, boolean pRenderBlockOutline, Camera pCamera, GameRenderer pGameRenderer, LightTexture pLightTexture, Matrix4f pProjectionMatrix, CallbackInfo ci) {
 		var viewProjection = RenderContext.createViewProjection(pPoseStack, pProjectionMatrix);
-		var culler = RenderContext.createCuller(pCamera, viewProjection);
+		var cameraPos = pCamera.getPosition();
+		var culler = RenderContext.createCuller(viewProjection, (float) -cameraPos.x, (float) -cameraPos.y, (float) -cameraPos.z);
 		renderContext = new RenderContext((LevelRenderer) (Object) this, level, pPoseStack, viewProjection, pProjectionMatrix, renderBuffers, pCamera, culler);
 
 		try (var restoreState = GlStateTracker.getRestoreState()) {

src/main/java/com/jozufozu/flywheel/util/joml/FrustumIntersection.java

@@ -1004,11 +1004,9 @@ public class FrustumIntersection {
 	 * {@code vec2(nzZ, pzZ)}<br>
 	 * {@code vec2(nzW, pzW)}<br>
 	 *
-	 * @param buffer The buffer to write the planes to.
+	 * @param addr The buffer to write the planes to.
 	 */
-	public void getJozuPackedPlanes(ByteBuffer buffer) {
+	public void getJozuPackedPlanes(long addr) {
-		long addr = MemoryUtil.memAddress(buffer);
-
 		MemoryUtil.memPutFloat(addr, nxX);
 		MemoryUtil.memPutFloat(addr + 4, pxX);
 		MemoryUtil.memPutFloat(addr + 8, nyX);

src/main/resources/assets/flywheel/flywheel/compute/cull_instances.glsl

@@ -3,7 +3,10 @@ layout(local_size_x = FLW_SUBGROUP_SIZE) in;
 #use "flywheel:compute/objects.glsl"
 #use "flywheel:util/quaternion.glsl"
 
-layout(std140, binding = 3) uniform FrameData {
+uint flw_objectID;
+uint flw_batchID;
+
+layout(std140, binding = 0) uniform FrameData {
     vec4 a1; // vec4(nx.x, px.x, ny.x, py.x)
     vec4 a2; // vec4(nx.y, px.y, ny.y, py.y)
     vec4 a3; // vec4(nx.z, px.z, ny.z, py.z)
@@ -31,37 +34,55 @@ layout(std430, binding = 3) buffer DrawCommands {
     MeshDrawCommand drawCommands[];
 };
 
+layout(std430, binding = 4) writeonly buffer DebugVisibility {
+    uint objectVisibilityBits[];
+};
+
 // 83 - 27 = 56 spirv instruction results
 bool testSphere(vec3 center, float radius) {
-    return
+    bvec4 resultA = greaterThanEqual(fma(frustum.a1, center.xxxx, fma(frustum.a2, center.yyyy, fma(frustum.a3, center.zzzz, frustum.a4))), -radius.xxxx);
-    all(lessThanEqual(fma(frustum.a1, center.xxxx, fma(frustum.a2, center.yyyy, fma(frustum.a3, center.zzzz, frustum.a4))), -radius.xxxx)) &&
+    bvec2 resultB = greaterThanEqual(fma(frustum.b1, center.xx, fma(frustum.b2, center.yy, fma(frustum.b3, center.zz, frustum.b4))), -radius.xx);
-    all(lessThanEqual(fma(frustum.b1, center.xx, fma(frustum.b2, center.yy, fma(frustum.b3, center.zz, frustum.b4))), -radius.xx));
+
+    uint debug = uint(resultA.x);
+    debug |= uint(resultA.y) << 1;
+    debug |= uint(resultA.z) << 2;
+    debug |= uint(resultA.w) << 3;
+    debug |= uint(resultB.x) << 4;
+    debug |= uint(resultB.y) << 5;
+
+    objectVisibilityBits[flw_objectID] = debug;
+
+    return all(resultA) && all(resultB);
 }
 
-bool isVisible(uint objectID, uint batchID) {
+void flw_transformBoundingSphere(in Instance i, inout vec3 center, inout float radius) {
-    vec4 sphere = boundingSpheres[batchID];
+    center = rotateVertexByQuat(center - i.pivot, i.rotation) + i.pivot + i.pos;
+    radius = radius;
+}
 
-    vec3 pivot = objects[objectID].pivot;
+bool isVisible() {
-    vec3 center = rotateVertexByQuat(sphere.xyz - pivot, objects[objectID].rotation) + pivot + objects[objectID].pos;
+    vec4 sphere = boundingSpheres[flw_batchID];
+
+    vec3 center = sphere.xyz;
     float radius = sphere.r;
+    flw_transformBoundingSphere(objects[flw_objectID], center, radius);
 
-    return true; //testSphere(center, radius);
+    return testSphere(center, radius);
 }
 
 void main() {
-    uint objectID = gl_GlobalInvocationID.x;
+    flw_objectID = gl_GlobalInvocationID.x;
 
-    if (objectID >= objects.length()) {
+    if (flw_objectID >= objects.length()) {
         return;
     }
 
-    uint batchID = objects[objectID].batchID;
+    flw_batchID = objects[objectID].batchID;
-    bool visible = isVisible(objectID, batchID);
 
-    if (visible) {
+    if (isVisible()) {
-        uint batchIndex = atomicAdd(drawCommands[batchID].instanceCount, 1);
+        uint batchIndex = atomicAdd(drawCommands[flw_batchID].instanceCount, 1);
-        uint globalIndex = drawCommands[batchID].baseInstance + batchIndex;
+        uint globalIndex = drawCommands[flw_batchID].baseInstance + batchIndex;
 
-        objectIDs[globalIndex] = objectID;
+        objectIDs[globalIndex] = flw_objectID;
     }
 }