I can see you

- Add visibility buffer fbo attachment
- Write instance id to visbuffer
- Move instance id in/out from common to impl shaders

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

@@ -49,6 +49,7 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 	private final MatrixBuffer matrixBuffer;
 
 	private final DepthPyramid depthPyramid;
+	private final VisibilityBuffer visibilityBuffer;
 
 	private boolean needsBarrier = false;
 
@@ -64,6 +65,7 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 		matrixBuffer = new MatrixBuffer();
 
 		depthPyramid = new DepthPyramid(programs.getDepthReduceProgram());
+		visibilityBuffer = new VisibilityBuffer();
 	}
 
 	@Override
@@ -100,6 +102,8 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 			matrixBuffer.bind();
 			Uniforms.bindAll();
 
+			visibilityBuffer.attach();
+
 			if (needsBarrier) {
 				glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 				needsBarrier = false;
@@ -111,6 +115,8 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 
 			MaterialRenderState.reset();
 			TextureBinder.resetLightAndOverlay();
+
+			visibilityBuffer.detach();
 		}
 	}
 
@@ -185,6 +191,10 @@ public class IndirectDrawManager extends DrawManager<IndirectInstancer<?>> {
 		crumblingDrawBuffer.delete();
 
 		programs.release();
+
+		depthPyramid.delete();
+
+		visibilityBuffer.delete();
 	}
 
 	public void renderCrumbling(List<Engine.CrumblingBlock> crumblingBlocks) {

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

@@ -0,0 +1,87 @@
+package dev.engine_room.flywheel.backend.engine.indirect;
+
+import org.lwjgl.opengl.GL30;
+import org.lwjgl.opengl.GL32;
+import org.lwjgl.opengl.GL46;
+
+import com.mojang.blaze3d.platform.GlStateManager;
+
+import dev.engine_room.flywheel.backend.FlwBackend;
+import dev.engine_room.flywheel.backend.gl.GlTextureUnit;
+import it.unimi.dsi.fastutil.ints.IntArraySet;
+import it.unimi.dsi.fastutil.ints.IntSet;
+import net.minecraft.client.Minecraft;
+
+public class VisibilityBuffer {
+	private static final int ATTACHMENT = GL30.GL_COLOR_ATTACHMENT1;
+
+	private final int textureId;
+
+	private int lastWidth = -1;
+	private int lastHeight = -1;
+
+	private final IntSet attached = new IntArraySet();
+
+	public VisibilityBuffer() {
+		textureId = GL32.glGenTextures();
+
+		GlStateManager._bindTexture(textureId);
+		GlStateManager._texParameter(GL32.GL_TEXTURE_2D, GL32.GL_TEXTURE_MIN_FILTER, GL32.GL_NEAREST);
+		GlStateManager._texParameter(GL32.GL_TEXTURE_2D, GL32.GL_TEXTURE_MAG_FILTER, GL32.GL_NEAREST);
+		GlStateManager._texParameter(GL32.GL_TEXTURE_2D, GL32.GL_TEXTURE_WRAP_S, GL32.GL_CLAMP_TO_EDGE);
+		GlStateManager._texParameter(GL32.GL_TEXTURE_2D, GL32.GL_TEXTURE_WRAP_T, GL32.GL_CLAMP_TO_EDGE);
+	}
+
+	public void attach() {
+		// TODO: clear the vis buffer. maybe do this when we read it?
+
+		var mainRenderTarget = Minecraft.getInstance()
+				.getMainRenderTarget();
+
+		setupTexture(mainRenderTarget.width, mainRenderTarget.height);
+
+		if (attached.add(mainRenderTarget.frameBufferId)) {
+			GL46.glNamedFramebufferTexture(mainRenderTarget.frameBufferId, ATTACHMENT, textureId, 0);
+
+			try {
+				mainRenderTarget.checkStatus();
+			} catch (Exception e) {
+				FlwBackend.LOGGER.error("Error attaching visbuffer", e);
+			}
+		}
+
+		// Enable writes
+		GL46.glNamedFramebufferDrawBuffers(mainRenderTarget.frameBufferId, new int[] { GL30.GL_COLOR_ATTACHMENT0, ATTACHMENT });
+	}
+
+	public void detach() {
+		var mainRenderTarget = Minecraft.getInstance()
+				.getMainRenderTarget();
+
+		// Disable writes
+		GL46.glNamedFramebufferDrawBuffers(mainRenderTarget.frameBufferId, new int[] { GL30.GL_COLOR_ATTACHMENT0 });
+	}
+
+	public void delete() {
+		GL32.glDeleteTextures(textureId);
+	}
+
+	private void setupTexture(int width, int height) {
+		if (lastWidth == width && lastHeight == height) {
+			return;
+		}
+
+		// Need to rebind to all fbos because an attachment becomes incomplete when it's resized
+		attached.clear();
+
+		lastWidth = width;
+		lastHeight = height;
+
+		GlTextureUnit.T0.makeActive();
+		GlStateManager._bindTexture(textureId);
+
+		// TODO: DSA texture storage?
+		GL32.glTexImage2D(GL32.GL_TEXTURE_2D, 0, GL32.GL_R32UI, width, height, 0, GL32.GL_RED_INTEGER, GL32.GL_UNSIGNED_INT, 0);
+		GlStateManager._bindTexture(0);
+	}
+}

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

@@ -13,9 +13,7 @@ uniform sampler2D _flw_crumblingTex;
 in vec2 _flw_crumblingTexCoord;
 #endif
 
-flat in uint _flw_instanceID;
+layout(location = 0) out vec4 _flw_outputColor;
-
-out vec4 _flw_outputColor;
 
 float _flw_diffuseFactor() {
     if (flw_material.diffuse) {
@@ -29,7 +27,7 @@ float _flw_diffuseFactor() {
     }
 }
 
-void _flw_main() {
+void _flw_main(uint instanceID) {
     flw_sampleColor = texture(flw_diffuseTex, flw_vertexTexCoord);
     flw_fragColor = flw_vertexColor * flw_sampleColor;
     flw_fragOverlay = flw_vertexOverlay;
@@ -72,7 +70,7 @@ void _flw_main() {
         color = vec4(flw_vertexNormal * .5 + .5, 1.);
         break;
         case 2u:
-        color = _flw_id2Color(_flw_instanceID);
+        color = _flw_id2Color(instanceID);
         break;
         case 3u:
         color = vec4(vec2((flw_fragLight * 15.0 + 0.5) / 16.), 0., 1.);

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

@@ -71,9 +71,7 @@ mat4 _flw_modelMatrix;
 mat3 _flw_normalMatrix;
 #endif
 
-flat out uint _flw_instanceID;
+void _flw_main(in FlwInstance instance) {
-
-void _flw_main(in FlwInstance instance, in uint stableInstanceID) {
     _flw_layoutVertex();
     flw_instanceVertex(instance);
     flw_materialVertex();
@@ -92,6 +90,4 @@ void _flw_main(in FlwInstance instance, in uint stableInstanceID) {
     flw_distance = fogDistance(flw_vertexPos.xyz, flw_cameraPos, flw_fogShape);
 
     gl_Position = flw_viewProjection * flw_vertexPos;
-
-    _flw_instanceID = stableInstanceID;
 }

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

@@ -4,10 +4,16 @@
 
 flat in uvec3 _flw_packedMaterial;
 
+flat in uint _flw_instanceID;
+
+layout(location = 1) out uint _flw_out_instanceID;
+
 void main() {
     _flw_uberMaterialFragmentIndex = _flw_packedMaterial.x;
     _flw_unpackUint2x16(_flw_packedMaterial.y, _flw_uberFogIndex, _flw_uberCutoutIndex);
     _flw_unpackMaterialProperties(_flw_packedMaterial.z, flw_material);
 
-    _flw_main();
+    _flw_main(_flw_instanceID);
+
+    _flw_out_instanceID = _flw_instanceID;
 }

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

@@ -21,8 +21,14 @@ layout(std430, binding = _FLW_MATRIX_BUFFER_BINDING) restrict buffer MatrixBuffe
 
 uniform uint _flw_baseDraw;
 
+// We read the visibility buffer for all culling groups into a single shared buffer.
+// This offset is used to know where each culling group starts.
+uniform uint _flw_globalInstanceIdOffset = 0;
+
 flat out uvec3 _flw_packedMaterial;
 
+flat out uint _flw_instanceID;
+
 void main() {
 #if __VERSION__ < 460
     uint drawIndex = gl_DrawIDARB + _flw_baseDraw;
@@ -49,5 +55,8 @@ void main() {
 #endif
     FlwInstance instance = _flw_unpackInstance(instanceIndex);
 
-    _flw_main(instance, instanceIndex);
+    _flw_main(instance);
+
+    // Add 1 because a 0 instance id means null.
+    _flw_instanceID = _flw_globalInstanceIdOffset + instanceIndex + 1;
 }

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

@@ -0,0 +1,25 @@
+layout(local_size_x = 8, local_size_y = 8) in;
+
+layout(binding = 0) uniform usampler2D visBuffer;
+
+layout(std430) restrict buffer VisibleFlagBuffer {
+    uint _flw_visibleFlag[];
+};
+
+void main() {
+    uint instanceID = texelFetch(visBuffer, ivec2(gl_GlobalInvocationID.xy), 0).r;
+
+    // Null instance id.
+    if (instanceID == 0) {
+        return;
+    }
+
+    // Adjust for null to find the actual index.
+    instanceID = instanceID - 1;
+
+    uint index = instanceID >> 5;
+
+    uint mask = 1u << (instanceID & 31u);
+
+    atomicOr(_flw_visibleFlag[index], mask);
+}

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

@@ -3,10 +3,12 @@
 
 uniform uvec4 _flw_packedMaterial;
 
+flat in uint _flw_instanceID;
+
 void main() {
     _flw_uberMaterialFragmentIndex = _flw_packedMaterial.y;
     _flw_unpackUint2x16(_flw_packedMaterial.z, _flw_uberFogIndex, _flw_uberCutoutIndex);
     _flw_unpackMaterialProperties(_flw_packedMaterial.w, flw_material);
 
-    _flw_main();
+    _flw_main(_flw_instanceID);
 }

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

@@ -10,6 +10,8 @@ uniform mat4 _flw_modelMatrixUniform;
 uniform mat3 _flw_normalMatrixUniform;
 #endif
 
+flat out uint _flw_instanceID;
+
 void main() {
     _flw_uberMaterialVertexIndex = _flw_packedMaterial.x;
     _flw_unpackMaterialProperties(_flw_packedMaterial.w, flw_material);
@@ -21,5 +23,7 @@ void main() {
     _flw_normalMatrix = _flw_normalMatrixUniform;
     #endif
 
-    _flw_main(instance, uint(gl_InstanceID));
+    _flw_main(instance);
+
+    _flw_instanceID = gl_InstanceID;
 }