The depths of the rabbit hole

- Fix mip levels being half the size they should be
- Use the next lowest po2 from the main render target size for mip 0
- Map from dst texel to src texel rather than naively multiply by 2
- Clamp the estimated mip level in the cull shader
- Use texel fetches in the cull shader (not sure if necessary?)

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

@@ -35,8 +35,8 @@ public class DepthPyramid {
 		var mainRenderTarget = Minecraft.getInstance()
 				.getMainRenderTarget();
 
-		int width = mainRenderTarget.width;
+		int width = mip0Size(mainRenderTarget.width);
-		int height = mainRenderTarget.height;
+		int height = mip0Size(mainRenderTarget.height);
 
 		int mipLevels = getImageMipLevels(width, height);
 
@@ -53,15 +53,15 @@ public class DepthPyramid {
 		depthReduceProgram.bind();
 
 		for (int i = 0; i < mipLevels; i++) {
-			int mipWidth = Math.max(1, width >> i);
+			int mipWidth = mipSize(width, i);
-			int mipHeight = Math.max(1, height >> i);
+			int mipHeight = mipSize(height, i);
 
 			int srcTexture = (i == 0) ? depthBufferId : pyramidTextureId;
 			GL46.glBindTexture(GL32.GL_TEXTURE_2D, srcTexture);
 
 			GL46.glBindImageTexture(0, pyramidTextureId, i, false, 0, GL32.GL_WRITE_ONLY, GL32.GL_R32F);
 
-			depthReduceProgram.setUVec2("imageSize", mipWidth, mipHeight);
+			depthReduceProgram.setVec2("imageSize", mipWidth, mipHeight);
 			depthReduceProgram.setInt("lod", Math.max(0, i - 1));
 
 			GL46.glDispatchCompute(MoreMath.ceilingDiv(mipWidth, 8), MoreMath.ceilingDiv(mipHeight, 8), 1);
@@ -85,20 +85,28 @@ public class DepthPyramid {
 		GL32.glBindTexture(GL32.GL_TEXTURE_2D, pyramidTextureId);
 
 		for (int i = 0; i < mipLevels; i++) {
-			int mipWidth = Math.max(1, width >> (i + 1));
+			int mipWidth = mipSize(width, i);
-			int mipHeight = Math.max(1, height >> (i + 1));
+			int mipHeight = mipSize(height, i);
 
 			GL32.glTexImage2D(GL32.GL_TEXTURE_2D, i, GL32.GL_R32F, mipWidth, mipHeight, 0, GL32.GL_RED, GL32.GL_FLOAT, 0);
 		}
 	}
 
-	private static int getImageMipLevels(int width, int height) {
+	public static int mipSize(int mip0Size, int level) {
+		return Math.max(1, mip0Size >> level);
+	}
+
+	public static int mip0Size(int screenSize) {
+		return Integer.highestOneBit(screenSize);
+	}
+
+	public static int getImageMipLevels(int width, int height) {
 		int result = 1;
 
 		while (width > 2 && height > 2) {
 			result++;
-			width /= 2;
+			width >>= 1;
-			height /= 2;
+			height >>= 1;
 		}
 
 		return result;

common/src/backend/java/dev/engine_room/flywheel/backend/engine/uniform/FrameUniforms.java

@@ -8,6 +8,7 @@ import org.lwjgl.system.MemoryUtil;
 
 import dev.engine_room.flywheel.api.RenderContext;
 import dev.engine_room.flywheel.api.visualization.VisualizationManager;
+import dev.engine_room.flywheel.backend.engine.indirect.DepthPyramid;
 import dev.engine_room.flywheel.backend.mixin.LevelRendererAccessor;
 import net.minecraft.client.Camera;
 import net.minecraft.client.Minecraft;
@@ -17,7 +18,7 @@ import net.minecraft.world.level.Level;
 import net.minecraft.world.phys.Vec3;
 
 public final class FrameUniforms extends UniformWriter {
-	private static final int SIZE = 96 + 64 * 9 + 16 * 5 + 8 * 2 + 8 + 4 * 16;
+	private static final int SIZE = 96 + 64 * 9 + 16 * 5 + 8 * 2 + 8 + 4 * 17;
 	static final UniformBuffer BUFFER = new UniformBuffer(Uniforms.FRAME_INDEX, SIZE);
 
 	private static final Matrix4f VIEW = new Matrix4f();
@@ -182,12 +183,20 @@ public final class FrameUniforms extends UniformWriter {
 	}
 
 	private static long writeCullData(long ptr) {
+		var mc = Minecraft.getInstance();
+		var mainRenderTarget = mc.getMainRenderTarget();
+
+		int pyramidWidth = DepthPyramid.mip0Size(mainRenderTarget.width);
+		int pyramidHeight = DepthPyramid.mip0Size(mainRenderTarget.height);
+		int pyramidDepth = DepthPyramid.getImageMipLevels(pyramidWidth, pyramidHeight);
+
 		ptr = writeFloat(ptr, 0.05F); // zNear
-		ptr = writeFloat(ptr, Minecraft.getInstance().gameRenderer.getDepthFar()); // zFar
+		ptr = writeFloat(ptr, mc.gameRenderer.getDepthFar()); // zFar
 		ptr = writeFloat(ptr, PROJECTION.m00()); // P00
 		ptr = writeFloat(ptr, PROJECTION.m11()); // P11
-		ptr = writeFloat(ptr, Minecraft.getInstance().getMainRenderTarget().width >> 1); // pyramidWidth
+		ptr = writeFloat(ptr, pyramidWidth); // pyramidWidth
-		ptr = writeFloat(ptr, Minecraft.getInstance().getMainRenderTarget().height >> 1); // pyramidHeight
+		ptr = writeFloat(ptr, pyramidHeight); // pyramidHeight
+		ptr = writeInt(ptr, pyramidDepth - 1); // pyramidLevels
 		ptr = writeInt(ptr, 0); // useMin
 
 		return ptr;

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

@@ -91,12 +91,18 @@ bool _flw_isVisible(uint instanceIndex, uint modelIndex) {
             float width = (aabb.z - aabb.x) * _flw_cullData.pyramidWidth;
             float height = (aabb.w - aabb.y) * _flw_cullData.pyramidHeight;
 
-            float level = floor(log2(max(width, height)));
+            int level = clamp(0, int(ceil(log2(max(width, height)))), _flw_cullData.pyramidLevels);
 
-            float depth01 = textureLod(_flw_depthPyramid, aabb.xw, level).r;
+            ivec2 levelSize = textureSize(_flw_depthPyramid, level);
-            float depth11 = textureLod(_flw_depthPyramid, aabb.zw, level).r;
+
-            float depth10 = textureLod(_flw_depthPyramid, aabb.zy, level).r;
+            ivec4 levelSizePair = ivec4(levelSize, levelSize);
-            float depth00 = textureLod(_flw_depthPyramid, aabb.xy, level).r;
+
+            ivec4 bounds = ivec4(aabb * vec4(levelSizePair));
+
+            float depth01 = texelFetch(_flw_depthPyramid, bounds.xw, level).r;
+            float depth11 = texelFetch(_flw_depthPyramid, bounds.zw, level).r;
+            float depth10 = texelFetch(_flw_depthPyramid, bounds.zy, level).r;
+            float depth00 = texelFetch(_flw_depthPyramid, bounds.xy, level).r;
 
             float depth;
             if (_flw_cullData.useMin == 0) {

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

@@ -3,7 +3,7 @@ layout(local_size_x = 8, local_size_y = 8) in;
 layout(binding = 0, r32f) uniform writeonly image2D outImage;
 layout(binding = 1) uniform sampler2D inImage;
 
-uniform uvec2 imageSize;
+uniform vec2 imageSize;
 uniform int lod;
 
 uniform int useMin = 0;
@@ -11,7 +11,9 @@ uniform int useMin = 0;
 void main() {
     uvec2 pos = gl_GlobalInvocationID.xy;
 
-    ivec2 samplePos = ivec2(pos) * 2;
+    // Map the output texel to an input texel. Properly do the division because generating mip0 maps from the actual
+    // full resolution depth buffer and the aspect ratio may be different from our Po2 pyramid.
+    ivec2 samplePos = ivec2(floor(vec2(pos) * vec2(textureSize(inImage, lod)) / imageSize));
 
     float depth01 = texelFetchOffset(inImage, samplePos, lod, ivec2(0, 1)).r;
     float depth11 = texelFetchOffset(inImage, samplePos, lod, ivec2(1, 1)).r;

common/src/backend/resources/assets/flywheel/flywheel/internal/uniforms/frame.glsl

@@ -16,6 +16,7 @@ struct _FlwCullData {
     float P11;
     float pyramidWidth;
     float pyramidHeight;
+    int pyramidLevels;
     uint useMin;
 };