Directly visible

- Don't actually need a framebuffer attachment for visibility
- Instead, process everything in pass 2 and write out the visibility
  bitset directly
- Persist visibility bits between frames for use in pass 1
- No need for indirect dispatch!
- Also saves some ssbo bindings
- Do frustum culling in both passes

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

@@ -109,6 +109,8 @@ private static CompilationHarness<InstanceType<?>> createCullingCompiler(ShaderS
 						.nameMapper(instanceType -> name + "/" + ResourceUtil.toDebugFileNameNoExtension(instanceType.cullShader()))
 						.requireExtensions(COMPUTE_EXTENSIONS)
 						.define("_FLW_SUBGROUP_SIZE", GlCompat.SUBGROUP_SIZE)
+						.enableExtension("GL_KHR_shader_subgroup_basic")
+						.enableExtension("GL_KHR_shader_subgroup_ballot")
 						.withResource(CULL_SHADER_API_IMPL)
 						.withComponent(InstanceStructComponent::new)
 						.withResource(InstanceType::cullShader)

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

@@ -1,18 +1,16 @@
 package dev.engine_room.flywheel.backend.engine.indirect;
 
 public final class BufferBindings {
-	public static final int PASS_TWO_DISPATCH = 0;
-	public static final int PASS_TWO_INSTANCE_INDEX = 1;
-	public static final int PAGE_FRAME_DESCRIPTOR = 2;
-	public static final int INSTANCE = 3;
-	public static final int DRAW_INSTANCE_INDEX = 4;
-	public static final int MODEL = 5;
-	public static final int DRAW = 6;
+	public static final int LAST_FRAME_VISIBILITY = 0;
+	public static final int PAGE_FRAME_DESCRIPTOR = 1;
+	public static final int INSTANCE = 2;
+	public static final int DRAW_INSTANCE_INDEX = 3;
+	public static final int MODEL = 4;
+	public static final int DRAW = 5;
 
-	public static final int LIGHT_LUT = 7;
-	public static final int LIGHT_SECTION = 8;
-	public static final int MATRICES = 9;
-	public static final int LAST_FRAME_VISIBILITY = 10;
+	public static final int LIGHT_LUT = 6;
+	public static final int LIGHT_SECTION = 7;
+	public static final int MATRICES = 8;
 
 	private BufferBindings() {
 	}

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

@@ -7,11 +7,12 @@
 import org.lwjgl.system.Pointer;
 
 import dev.engine_room.flywheel.backend.gl.buffer.GlBufferType;
+import dev.engine_room.flywheel.lib.math.MoreMath;
 import dev.engine_room.flywheel.lib.memory.MemoryBlock;
 
 public class IndirectBuffers {
 	// Number of vbos created.
-	public static final int BUFFER_COUNT = 7;
+	public static final int BUFFER_COUNT = 6;
 
 	public static final long INT_SIZE = Integer.BYTES;
 	public static final long PTR_SIZE = Pointer.POINTER_SIZE;
@@ -30,17 +31,15 @@ public class IndirectBuffers {
 	private static final long BUFFERS_SIZE_BYTES = SIZE_OFFSET + BUFFER_COUNT * PTR_SIZE;
 
 	// Offsets to the vbos
-	private static final long PASS_TWO_DISPATCH_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.PASS_TWO_DISPATCH * INT_SIZE;
-	private static final long PASS_TWO_INSTANCE_INDEX_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.PASS_TWO_INSTANCE_INDEX * INT_SIZE;
+	private static final long LAST_FRAME_VISIBILITY_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.LAST_FRAME_VISIBILITY * INT_SIZE;
 	private static final long PAGE_FRAME_DESCRIPTOR_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.PAGE_FRAME_DESCRIPTOR * INT_SIZE;
 	private static final long INSTANCE_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.INSTANCE * INT_SIZE;
 	private static final long DRAW_INSTANCE_INDEX_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.DRAW_INSTANCE_INDEX * INT_SIZE;
 	private static final long MODEL_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.MODEL * INT_SIZE;
 	private static final long DRAW_HANDLE_OFFSET = HANDLE_OFFSET + BufferBindings.DRAW * INT_SIZE;
 
 	// Offsets to the sizes
-	private static final long PASS_TWO_DISPATCH_SIZE_OFFSET = SIZE_OFFSET + BufferBindings.PASS_TWO_DISPATCH * PTR_SIZE;
-	private static final long PASS_TWO_INSTANCE_INDEX_SIZE_OFFSET = SIZE_OFFSET + BufferBindings.PASS_TWO_INSTANCE_INDEX * PTR_SIZE;
+	private static final long LAST_FRAME_VISIBILITY_SIZE_OFFSET = SIZE_OFFSET + BufferBindings.LAST_FRAME_VISIBILITY * PTR_SIZE;
 	private static final long PAGE_FRAME_DESCRIPTOR_SIZE_OFFSET = SIZE_OFFSET + BufferBindings.PAGE_FRAME_DESCRIPTOR * PTR_SIZE;
 	private static final long INSTANCE_SIZE_OFFSET = SIZE_OFFSET + BufferBindings.INSTANCE * PTR_SIZE;
 	private static final long DRAW_INSTANCE_INDEX_SIZE_OFFSET = SIZE_OFFSET + BufferBindings.DRAW_INSTANCE_INDEX * PTR_SIZE;
@@ -66,8 +65,7 @@ public class IndirectBuffers {
 	 */
 	private final MemoryBlock multiBindBlock;
 
-	public final ResizableStorageBuffer passTwoDispatch;
-	public final ResizableStorageArray passTwoInstanceIndex;
+	public final ResizableStorageArray lastFrameVisibility;
 	public final ObjectStorage objectStorage;
 	public final ResizableStorageArray drawInstanceIndex;
 	public final ResizableStorageArray model;
@@ -76,34 +74,29 @@ public class IndirectBuffers {
 	IndirectBuffers(long instanceStride) {
 		this.multiBindBlock = MemoryBlock.calloc(BUFFERS_SIZE_BYTES, 1);
 
-		passTwoDispatch = new ResizableStorageBuffer();
-		passTwoInstanceIndex = new ResizableStorageArray(INT_SIZE, INSTANCE_GROWTH_FACTOR);
+		lastFrameVisibility = new ResizableStorageArray(INT_SIZE, INSTANCE_GROWTH_FACTOR);
 		objectStorage = new ObjectStorage(instanceStride);
 		drawInstanceIndex = new ResizableStorageArray(INT_SIZE, INSTANCE_GROWTH_FACTOR);
 		model = new ResizableStorageArray(MODEL_STRIDE, MODEL_GROWTH_FACTOR);
 		draw = new ResizableStorageArray(DRAW_COMMAND_STRIDE, DRAW_GROWTH_FACTOR);
-
-		passTwoDispatch.ensureCapacity(INT_SIZE * 4);
 	}
 
 	void updateCounts(int instanceCount, int modelCount, int drawCount) {
 		drawInstanceIndex.ensureCapacity(instanceCount);
-		passTwoInstanceIndex.ensureCapacity(instanceCount);
+		lastFrameVisibility.ensureCapacity(MoreMath.ceilingDiv(instanceCount, 32));
 		model.ensureCapacity(modelCount);
 		draw.ensureCapacity(drawCount);
 
 		final long ptr = multiBindBlock.ptr();
 
-		MemoryUtil.memPutInt(ptr + PASS_TWO_DISPATCH_HANDLE_OFFSET, passTwoDispatch.handle());
-		MemoryUtil.memPutInt(ptr + PASS_TWO_INSTANCE_INDEX_HANDLE_OFFSET, passTwoInstanceIndex.handle());
+		MemoryUtil.memPutInt(ptr + LAST_FRAME_VISIBILITY_HANDLE_OFFSET, lastFrameVisibility.handle());
 		MemoryUtil.memPutInt(ptr + PAGE_FRAME_DESCRIPTOR_HANDLE_OFFSET, objectStorage.frameDescriptorBuffer.handle());
 		MemoryUtil.memPutInt(ptr + INSTANCE_HANDLE_OFFSET, objectStorage.objectBuffer.handle());
 		MemoryUtil.memPutInt(ptr + DRAW_INSTANCE_INDEX_HANDLE_OFFSET, drawInstanceIndex.handle());
 		MemoryUtil.memPutInt(ptr + MODEL_HANDLE_OFFSET, model.handle());
 		MemoryUtil.memPutInt(ptr + DRAW_HANDLE_OFFSET, draw.handle());
 
-		MemoryUtil.memPutAddress(ptr + PASS_TWO_DISPATCH_SIZE_OFFSET, passTwoDispatch.capacity());
-		MemoryUtil.memPutAddress(ptr + PASS_TWO_INSTANCE_INDEX_SIZE_OFFSET, INT_SIZE * instanceCount);
+		MemoryUtil.memPutAddress(ptr + LAST_FRAME_VISIBILITY_SIZE_OFFSET, INT_SIZE * MoreMath.ceilingDiv(instanceCount, 32));
 		MemoryUtil.memPutAddress(ptr + PAGE_FRAME_DESCRIPTOR_SIZE_OFFSET, objectStorage.frameDescriptorBuffer.capacity());
 		MemoryUtil.memPutAddress(ptr + INSTANCE_SIZE_OFFSET, objectStorage.objectBuffer.capacity());
 		MemoryUtil.memPutAddress(ptr + DRAW_INSTANCE_INDEX_SIZE_OFFSET, INT_SIZE * instanceCount);
@@ -112,24 +105,23 @@ void updateCounts(int instanceCount, int modelCount, int drawCount) {
 	}
 
 	public void bindForCullPassOne() {
-		multiBind(0, 6);
+		multiBind(0, 5);
 	}
 
 	public void bindForCullPassTwo() {
-		multiBind(0, 6);
-		GlBufferType.DISPATCH_INDIRECT_BUFFER.bind(passTwoDispatch.handle());
+		multiBind(0, 5);
 	}
 
 	public void bindForApply() {
-		multiBind(5, 2);
+		multiBind(4, 2);
 	}
 
 	public void bindForModelReset() {
-		multiBind(5, 1);
+		multiBind(4, 1);
 	}
 
 	public void bindForDraw() {
-		multiBind(3, 4);
+		multiBind(2, 4);
 		GlBufferType.DRAW_INDIRECT_BUFFER.bind(draw.handle());
 	}
 
@@ -155,7 +147,6 @@ public void delete() {
 		drawInstanceIndex.delete();
 		model.delete();
 		draw.delete();
-		passTwoDispatch.delete();
-		passTwoInstanceIndex.delete();
+		lastFrameVisibility.delete();
 	}
 }

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

@@ -6,16 +6,13 @@
 import static org.lwjgl.opengl.GL42.GL_COMMAND_BARRIER_BIT;
 import static org.lwjgl.opengl.GL42.glMemoryBarrier;
 import static org.lwjgl.opengl.GL43.glDispatchCompute;
-import static org.lwjgl.opengl.GL43.glDispatchComputeIndirect;
 
 import java.util.ArrayList;
 import java.util.Comparator;
 import java.util.EnumMap;
 import java.util.List;
 import java.util.Map;
 
-import org.lwjgl.opengl.GL46;
-
 import dev.engine_room.flywheel.api.instance.Instance;
 import dev.engine_room.flywheel.api.instance.InstanceType;
 import dev.engine_room.flywheel.api.material.Material;
@@ -29,7 +26,6 @@
 import dev.engine_room.flywheel.backend.engine.uniform.Uniforms;
 import dev.engine_room.flywheel.backend.gl.GlCompat;
 import dev.engine_room.flywheel.backend.gl.shader.GlProgram;
-import dev.engine_room.flywheel.lib.material.LightShaders;
 import dev.engine_room.flywheel.lib.math.MoreMath;
 
 public class IndirectCullingGroup<I extends Instance> {
@@ -54,12 +50,6 @@ public class IndirectCullingGroup<I extends Instance> {
 	private boolean needsDrawSort;
 	public int instanceCountThisFrame;
 
-	private int pagesLastFrame = 0;
-	private int pagesThisFrame = 0;
-
-	private int visibilityWriteOffsetPages = 0;
-	private int visibilityReadOffsetPages = 0;
-
 	IndirectCullingGroup(InstanceType<I> instanceType, IndirectPrograms programs) {
 		this.instanceType = instanceType;
 		instanceStride = MoreMath.align4(instanceType.layout()
@@ -95,17 +85,6 @@ public void flushInstancers() {
 		}
 	}
 
-	public int flipVisibilityOffsets(int visibilityWriteOffsetPages) {
-		this.visibilityReadOffsetPages = this.visibilityWriteOffsetPages;
-		this.visibilityWriteOffsetPages = visibilityWriteOffsetPages;
-
-		pagesLastFrame = pagesThisFrame;
-
-		pagesThisFrame = buffers.objectStorage.capacity();
-
-		return pagesThisFrame;
-	}
-
 	public void upload(StagingBuffer stagingBuffer) {
 		if (nothingToDo()) {
 			return;
@@ -127,8 +106,6 @@ public void upload(StagingBuffer stagingBuffer) {
 		}
 
 		uploadDraws(stagingBuffer);
-
-		GL46.nglClearNamedBufferData(buffers.passTwoDispatch.handle(), GL46.GL_R32UI, GL46.GL_RED, GL46.GL_UNSIGNED_INT, 0);
 	}
 
 	public void dispatchCull() {
@@ -139,8 +116,6 @@ public void dispatchCull() {
 		Uniforms.bindAll();
 		earlyCull.bind();
 
-		earlyCull.setUInt("_flw_visibilityReadOffsetPages", visibilityReadOffsetPages);
-
 		buffers.bindForCullPassOne();
 		glDispatchCompute(buffers.objectStorage.capacity(), 1, 1);
 	}
@@ -154,7 +129,7 @@ public void dispatchCullPassTwo() {
 		lateCull.bind();
 
 		buffers.bindForCullPassTwo();
-		glDispatchComputeIndirect(0);
+		glDispatchCompute(buffers.objectStorage.capacity(), 1, 1);
 	}
 
 	public void dispatchApply() {
@@ -257,8 +232,6 @@ public void submit(VisualType visualType) {
 				// Don't need to do this unless the program changes.
 				drawProgram.bind();
 				baseDrawUniformLoc = drawProgram.getUniformLocation("_flw_baseDraw");
-
-				drawProgram.setUInt("_flw_visibilityWriteOffsetInstances", visibilityWriteOffsetPages << ObjectStorage.LOG_2_PAGE_SIZE);
 			}
 
 			glUniform1ui(baseDrawUniformLoc, multiDraw.start);

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

@@ -53,11 +53,6 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 	private final MatrixBuffer matrixBuffer;
 
 	private final DepthPyramid depthPyramid;
-	private final VisibilityBuffer visibilityBuffer;
-
-	private int totalPagesLastFrame = 0;
-
-	private boolean needsBarrier = false;
 
 	public IndirectDrawManager(IndirectPrograms programs) {
 		this.programs = programs;
@@ -73,7 +68,6 @@ public IndirectDrawManager(IndirectPrograms programs) {
 		matrixBuffer = new MatrixBuffer();
 
 		depthPyramid = new DepthPyramid(programs);
-		visibilityBuffer = new VisibilityBuffer(programs);
 	}
 
 	@Override
@@ -112,8 +106,6 @@ public void render(VisualType visualType) {
 
 		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT | GL_BUFFER_UPDATE_BARRIER_BIT);
 
-		visibilityBuffer.bind();
-
 		for (var group1 : cullingGroups.values()) {
 			group1.dispatchCull();
 		}
@@ -124,8 +116,6 @@ public void render(VisualType visualType) {
 
 		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 
-		visibilityBuffer.attach();
-
 		submitDraws();
 
 		depthPyramid.generate();
@@ -156,8 +146,6 @@ public void render(VisualType visualType) {
 
 		MaterialRenderState.reset();
 		TextureBinder.resetLightAndOverlay();
-
-		visibilityBuffer.detach();
 	}
 
 	private void dispatchApply() {
@@ -185,20 +173,12 @@ public void flush(LightStorage lightStorage, EnvironmentStorage environmentStora
 			group.flushInstancers();
 		}
 
-		visibilityBuffer.read(totalPagesLastFrame);
-		visibilityBuffer.clear();
-
 		cullingGroups.values()
 				.removeIf(IndirectCullingGroup::checkEmptyAndDelete);
 
 		instancers.values()
 				.removeIf(instancer -> instancer.instanceCount() == 0);
 
-		int totalPagesThisFrame = 0;
-		for (var group : cullingGroups.values()) {
-			totalPagesThisFrame += group.flipVisibilityOffsets(totalPagesThisFrame);
-		}
-
 		meshPool.flush();
 
 		stagingBuffer.reclaim();
@@ -215,10 +195,6 @@ public void flush(LightStorage lightStorage, EnvironmentStorage environmentStora
 
 		// We could probably save some driver calls here when there are
 		// actually zero instances, but that feels like a very rare case
-
-		needsBarrier = true;
-
-		totalPagesLastFrame = totalPagesThisFrame;
 	}
 
 	@Override
@@ -238,8 +214,6 @@ public void delete() {
 		programs.release();
 
 		depthPyramid.delete();
-
-		visibilityBuffer.delete();
 	}
 
 	public void renderCrumbling(List<Engine.CrumblingBlock> crumblingBlocks) {