Hol up

- Use way fewer memory barriers
- I didn't realize that GL_SHADER_STORAGE_BARRIER_BIT was global instead
  of operating only on the currently bound buffers. Oh, well
- Move apply program binding to IndirectDrawManager
- Fix embedded instances flickering when first loading a world. Need to
  actually bind the matrix buffer for the cull shader. Not sure how it
  worked at all before
- Minor styling/cleanup

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

@@ -51,4 +51,8 @@ public class Arena {
 	public int capacity() {
 		return top;
 	}
+
+	public long byteCapacity() {
+		return memoryBlock.size();
+	}
 }

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

@@ -16,7 +16,8 @@ public class EnvironmentStorage {
 	public final Arena arena = new Arena(MATRIX_SIZE_BYTES, 32);
 
 	{
-		arena.alloc(); // Reserve the identity matrix.
+		// Reserve the identity matrix. Burns a few bytes but oh well.
+		arena.alloc();
 	}
 
 	public void track(EmbeddedEnvironment environment) {

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

@@ -5,7 +5,6 @@ import static org.lwjgl.opengl.GL11.GL_UNSIGNED_INT;
 import static org.lwjgl.opengl.GL30.glUniform1ui;
 import static org.lwjgl.opengl.GL42.GL_COMMAND_BARRIER_BIT;
 import static org.lwjgl.opengl.GL42.glMemoryBarrier;
-import static org.lwjgl.opengl.GL43.GL_SHADER_STORAGE_BARRIER_BIT;
 import static org.lwjgl.opengl.GL43.glDispatchCompute;
 
 import java.util.ArrayList;
@@ -37,8 +36,6 @@ public class IndirectCullingGroup<I extends Instance> {
 			.thenComparing(IndirectDraw::indexOfMeshInModel)
 			.thenComparing(IndirectDraw::material, MaterialRenderState.COMPARATOR);
 
-	private static final int DRAW_BARRIER_BITS = GL_SHADER_STORAGE_BARRIER_BIT | GL_COMMAND_BARRIER_BIT;
-
 	private final InstanceType<I> instanceType;
 	private final long instanceStride;
 	private final IndirectBuffers buffers;
@@ -48,7 +45,6 @@ public class IndirectCullingGroup<I extends Instance> {
 
 	private final IndirectPrograms programs;
 	private final GlProgram cullProgram;
-	private final GlProgram applyProgram;
 
 	private boolean needsDrawBarrier;
 	private boolean needsDrawSort;
@@ -62,7 +58,6 @@ public class IndirectCullingGroup<I extends Instance> {
 
 		this.programs = programs;
 		cullProgram = programs.getCullingProgram(instanceType);
-		applyProgram = programs.getApplyProgram();
 	}
 
 	public void flushInstancers() {
@@ -123,7 +118,6 @@ public class IndirectCullingGroup<I extends Instance> {
 		cullProgram.bind();
 
 		buffers.bindForCompute();
-		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 		glDispatchCompute(GlCompat.getComputeGroupCount(instanceCountThisFrame), 1, 1);
 	}
 
@@ -132,9 +126,7 @@ public class IndirectCullingGroup<I extends Instance> {
 			return;
 		}
 
-		applyProgram.bind();
 		buffers.bindForCompute();
-		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 		glDispatchCompute(GlCompat.getComputeGroupCount(indirectDraws.size()), 1, 1);
 	}
 
@@ -241,7 +233,9 @@ public class IndirectCullingGroup<I extends Instance> {
 
 	private void drawBarrier() {
 		if (needsDrawBarrier) {
-			glMemoryBarrier(DRAW_BARRIER_BITS);
+			// In theory all command buffer writes will be protected by
+			// the shader storage barrier bit, but better safe than sorry.
+			glMemoryBarrier(GL_COMMAND_BARRIER_BIT);
 			needsDrawBarrier = false;
 		}
 	}

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

@@ -4,6 +4,8 @@ import static org.lwjgl.opengl.GL11.GL_TRIANGLES;
 import static org.lwjgl.opengl.GL11.GL_UNSIGNED_INT;
 import static org.lwjgl.opengl.GL30.glBindBufferRange;
 import static org.lwjgl.opengl.GL40.glDrawElementsIndirect;
+import static org.lwjgl.opengl.GL42.glMemoryBarrier;
+import static org.lwjgl.opengl.GL43.GL_SHADER_STORAGE_BARRIER_BIT;
 import static org.lwjgl.opengl.GL43.GL_SHADER_STORAGE_BUFFER;
 
 import java.util.HashMap;
@@ -44,6 +46,8 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 	private final LightBuffers lightBuffers;
 	private final MatrixBuffer matrixBuffer;
 
+	private boolean needsBarrier = false;
+
 	public IndirectDrawManager(IndirectPrograms programs) {
 		this.programs = programs;
 		programs.acquire();
@@ -90,6 +94,11 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 			matrixBuffer.bind();
 			Uniforms.bindAll();
 
+			if (needsBarrier) {
+				glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
+				needsBarrier = false;
+			}
+
 			for (var group : cullingGroups.values()) {
 				group.submit(visualType);
 			}
@@ -127,13 +136,27 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 
 		stagingBuffer.flush();
 
+		// We could probably save some driver calls here when there are
+		// actually zero instances, but that feels like a very rare case
+
+		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
+
+		matrixBuffer.bind();
+
 		for (var group : cullingGroups.values()) {
 			group.dispatchCull();
 		}
 
+		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
+
+		programs.getApplyProgram()
+				.bind();
+
 		for (var group : cullingGroups.values()) {
 			group.dispatchApply();
 		}
+
+		needsBarrier = true;
 	}
 
 	@Override

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

@@ -18,8 +18,8 @@ public class MatrixBuffer {
 
 		matrices.ensureCapacity(capacity);
 
-		stagingBuffer.enqueueCopy((long) arena.capacity() * EnvironmentStorage.MATRIX_SIZE_BYTES, matrices.handle(), 0, ptr -> {
+		stagingBuffer.enqueueCopy(arena.byteCapacity(), matrices.handle(), 0, ptr -> {
-			MemoryUtil.memCopy(arena.indexToPointer(0), ptr, (long) arena.capacity() * EnvironmentStorage.MATRIX_SIZE_BYTES);
+			MemoryUtil.memCopy(arena.indexToPointer(0), ptr, arena.byteCapacity());
 		});
 	}
 

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

@@ -36,8 +36,8 @@ bool _flw_testSphere(vec3 center, float radius) {
 }
 
 bool _flw_isVisible(uint instanceIndex, uint modelIndex) {
-    BoundingSphere sphere = _flw_models[modelIndex].boundingSphere;
     uint matrixIndex = _flw_models[modelIndex].matrixIndex;
+    BoundingSphere sphere = _flw_models[modelIndex].boundingSphere;
 
     vec3 center;
     float radius;