MeshEXT modern, slightly fixed

src/main/java/me/cortex/voxy/client/core/VoxyRenderSystem.java

@@ -20,6 +20,7 @@ import me.cortex.voxy.client.core.rendering.hierachical.HierarchicalOcclusionTra
 import me.cortex.voxy.client.core.rendering.hierachical.NodeCleaner;
 import me.cortex.voxy.client.core.rendering.section.AbstractSectionRenderer;
 import me.cortex.voxy.client.core.rendering.section.MDICSectionRenderer;
+import me.cortex.voxy.client.core.rendering.section.MeshEXTSectionRenderer;
 import me.cortex.voxy.client.core.rendering.section.geometry.BasicSectionGeometryData;
 import me.cortex.voxy.client.core.rendering.section.geometry.IGeometryData;
 import me.cortex.voxy.client.core.rendering.util.DownloadStream;
@@ -69,7 +70,8 @@ public class VoxyRenderSystem {
 
     private static AbstractSectionRenderer<?,?> createSectionRenderer(AbstractRenderPipeline pipeline, ModelStore modelStore, IGeometryData geometryData) {
         //TODO: need todo a thing where selects optimal section render based on if supports the pipeline and geometry data type
-        return new MDICSectionRenderer(pipeline, modelStore, (BasicSectionGeometryData) geometryData);//We only have MDIC backend... for now
+        //return new MDICSectionRenderer(pipeline, modelStore, (BasicSectionGeometryData) geometryData);//We only have MDIC backend... for now
+        return new MeshEXTSectionRenderer(pipeline, modelStore, (BasicSectionGeometryData) geometryData);
     }
 
     public VoxyRenderSystem(WorldEngine world, ServiceThreadPool threadPool) {

src/main/java/me/cortex/voxy/client/core/gl/EXTMeshShader.java

@@ -0,0 +1,25 @@
+package me.cortex.voxy.client.core.gl;
+
+import org.lwjgl.opengl.GL;
+import org.lwjgl.system.JNI;
+
+public class EXTMeshShader {
+    public static final int
+            GL_MESH_SHADER_EXT = 0x9559,
+            GL_TASK_SHADER_EXT = 0x955A;
+
+    private static final long glDrawMeshTasksIndirectEXT_ptr;
+    static {
+        if (GL.getFunctionProvider() == null) {
+            throw new IllegalStateException("Class must be initalized after gl context has been created");
+        }
+        glDrawMeshTasksIndirectEXT_ptr = GL.getFunctionProvider().getFunctionAddress("glDrawMeshTasksIndirectEXT");
+    }
+
+    public static void glDrawMeshTasksIndirectEXT(long indirect) {
+        if (glDrawMeshTasksIndirectEXT_ptr == 0) {
+            throw new IllegalStateException("glDrawMeshTasksIndirectEXT not supported");
+        }
+        JNI.callPV(indirect, glDrawMeshTasksIndirectEXT_ptr);
+    }
+}

src/main/java/me/cortex/voxy/client/core/rendering/section/MeshEXTSectionRenderer.java

@@ -0,0 +1,231 @@
+package me.cortex.voxy.client.core.rendering.section;
+
+
+import me.cortex.voxy.client.RenderStatistics;
+import me.cortex.voxy.client.core.AbstractRenderPipeline;
+import me.cortex.voxy.client.core.gl.Capabilities;
+import me.cortex.voxy.client.core.gl.GlBuffer;
+import me.cortex.voxy.client.core.gl.GlTexture;
+import me.cortex.voxy.client.core.gl.GlVertexArray;
+import me.cortex.voxy.client.core.gl.shader.Shader;
+import me.cortex.voxy.client.core.gl.shader.ShaderType;
+import me.cortex.voxy.client.core.model.ModelStore;
+import me.cortex.voxy.client.core.rendering.section.geometry.BasicSectionGeometryData;
+import me.cortex.voxy.client.core.rendering.util.DownloadStream;
+import me.cortex.voxy.client.core.rendering.util.LightMapHelper;
+import me.cortex.voxy.client.core.rendering.util.SharedIndexBuffer;
+import me.cortex.voxy.client.core.rendering.util.UploadStream;
+import me.cortex.voxy.common.Logger;
+import me.cortex.voxy.common.world.WorldEngine;
+import org.joml.Matrix4f;
+import org.lwjgl.system.MemoryUtil;
+
+import java.util.List;
+
+import static me.cortex.voxy.client.core.gl.EXTMeshShader.glDrawMeshTasksIndirectEXT;
+import static org.lwjgl.opengl.GL11.*;
+import static org.lwjgl.opengl.GL11C.GL_UNSIGNED_INT;
+import static org.lwjgl.opengl.GL15.GL_ELEMENT_ARRAY_BUFFER;
+import static org.lwjgl.opengl.GL15.glBindBuffer;
+import static org.lwjgl.opengl.GL30.glBindBufferBase;
+import static org.lwjgl.opengl.GL30.glBindVertexArray;
+import static org.lwjgl.opengl.GL30C.GL_R32UI;
+import static org.lwjgl.opengl.GL30C.GL_RED_INTEGER;
+import static org.lwjgl.opengl.GL31.GL_UNIFORM_BUFFER;
+import static org.lwjgl.opengl.GL33.glBindSampler;
+import static org.lwjgl.opengl.GL40C.GL_DRAW_INDIRECT_BUFFER;
+import static org.lwjgl.opengl.GL42.glMemoryBarrier;
+import static org.lwjgl.opengl.GL43.*;
+import static org.lwjgl.opengl.GL45.*;
+import static org.lwjgl.opengl.NVRepresentativeFragmentTest.GL_REPRESENTATIVE_FRAGMENT_TEST_NV;
+
+//Uses MDIC to render the sections
+public class MeshEXTSectionRenderer extends AbstractSectionRenderer<MeshViewport, BasicSectionGeometryData> {
+    private static final int STATISTICS_BUFFER_BINDING = 8;
+    private final Shader terrainShader = Shader.make()
+            .define("MESH_SIZE", 32)//16
+
+            .defineIf("HAS_STATISTICS", RenderStatistics.enabled)
+            .defineIf("STATISTICS_BUFFER_BINDING", RenderStatistics.enabled, STATISTICS_BUFFER_BINDING)
+
+            .add(ShaderType.TASK, "voxy:lod/meshext/task.glsl")
+            .add(ShaderType.MESH, "voxy:lod/meshext/mesh.glsl")
+            .add(ShaderType.FRAGMENT, "voxy:lod/meshext/frag.glsl")
+            .compile();
+
+    private final Shader cullShader = Shader.make()
+            .add(ShaderType.VERTEX, "voxy:lod/gl46/cull/raster.vert")
+            .add(ShaderType.FRAGMENT, "voxy:lod/gl46/cull/raster.frag")
+            .compile();
+
+    private final GlBuffer uniform = new GlBuffer(1024).zero();
+    private final GlBuffer cullAndMeshDrawCommand = new GlBuffer(8*4).zero();//TODO: this needs tobe in the viewport
+
+    //Statistics
+    private final GlBuffer statisticsBuffer = new GlBuffer(1024).zero();
+
+    private final AbstractRenderPipeline pipeline;
+    public MeshEXTSectionRenderer(AbstractRenderPipeline pipeline, ModelStore modelStore, BasicSectionGeometryData geometryData) {
+        super(modelStore, geometryData);
+        this.pipeline = pipeline;
+        glClearNamedBufferSubData(this.cullAndMeshDrawCommand.id, GL_R32UI,0, 4, GL_RED_INTEGER, GL_UNSIGNED_INT, new int[]{6*2*3});//count
+        glClearNamedBufferSubData(this.cullAndMeshDrawCommand.id, GL_R32UI,8, 4, GL_RED_INTEGER, GL_UNSIGNED_INT, new int[]{(1<<16)*6*2});//firstIndex
+        glClearNamedBufferSubData(this.cullAndMeshDrawCommand.id, GL_R32UI,5*4+4, 4, GL_RED_INTEGER, GL_UNSIGNED_INT, new int[]{1});//y
+        glClearNamedBufferSubData(this.cullAndMeshDrawCommand.id, GL_R32UI,5*4+8, 4, GL_RED_INTEGER, GL_UNSIGNED_INT, new int[]{1});//z
+    }
+
+    private void uploadUniformBuffer(MeshViewport viewport) {
+        long ptr = UploadStream.INSTANCE.upload(this.uniform, 0, 1024);
+        
+        var mat = new Matrix4f(viewport.MVP);
+        mat.translate(-viewport.innerTranslation.x, -viewport.innerTranslation.y, -viewport.innerTranslation.z);
+        mat.getToAddress(ptr); ptr += 4*4*4;
+
+        viewport.section.getToAddress(ptr); ptr += 4*3;
+
+        if (viewport.frameId<0) {
+            Logger.error("Frame ID negative, this will cause things to break, wrapping around");
+            viewport.frameId &= 0x7fffffff;
+        }
+        MemoryUtil.memPutInt(ptr, viewport.frameId&0x7fffffff); ptr += 4;
+        viewport.innerTranslation.getToAddress(ptr); ptr += 4*3;
+
+        ptr += 4;// padd
+
+        MemoryUtil.memPutFloat(ptr, viewport.width); ptr += 4;
+        MemoryUtil.memPutFloat(ptr, viewport.height); ptr += 4;
+
+        UploadStream.INSTANCE.commit();
+    }
+
+
+    private void bindRenderingBuffers(MeshViewport viewport) {
+        glBindBufferBase(GL_UNIFORM_BUFFER, 0, this.uniform.id);
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, viewport.getRenderList().id);
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, this.geometryManager.getMetadataBuffer().id);
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, viewport.visibilityBuffer.id);
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 4, this.geometryManager.getGeometryBuffer().id);
+        this.modelStore.bind(5, 6, 0);
+        LightMapHelper.bind(1);
+        glBindTextureUnit(2, viewport.depthBoundingBuffer.getDepthTex().id);
+
+        glBindBuffer(GL_DRAW_INDIRECT_BUFFER, this.cullAndMeshDrawCommand.id);
+
+        if (RenderStatistics.enabled) {
+            glBindBufferBase(GL_SHADER_STORAGE_BUFFER, STATISTICS_BUFFER_BINDING, this.statisticsBuffer.id);
+        }
+    }
+
+    private void renderTerrain(MeshViewport viewport) {
+        //RenderLayer.getCutoutMipped().startDrawing();
+        glDisable(GL_CULL_FACE);
+        glEnable(GL_DEPTH_TEST);
+        this.terrainShader.bind();
+        this.pipeline.setupAndBindOpaque(viewport);
+        this.bindRenderingBuffers(viewport);
+
+        glMemoryBarrier(GL_COMMAND_BARRIER_BIT|GL_SHADER_STORAGE_BARRIER_BIT);//Barrier everything is needed
+
+        glDrawMeshTasksIndirectEXT(20);
+
+        glEnable(GL_CULL_FACE);
+        glBindSampler(0, 0);
+        glBindTextureUnit(0, 0);
+        glBindSampler(1, 0);
+        glBindTextureUnit(1, 0);
+
+        //RenderLayer.getCutoutMipped().endDrawing();
+    }
+
+    @Override
+    public void renderOpaque(MeshViewport viewport) {
+        if (this.geometryManager.getSectionCount() == 0) return;
+
+        this.uploadUniformBuffer(viewport);
+
+        this.renderTerrain(viewport);
+
+        //We need todo the statistics here as rastering is part of them, download then clear
+        if (RenderStatistics.enabled) {
+            DownloadStream.INSTANCE.download(this.statisticsBuffer, down->{
+                final int LAYERS = WorldEngine.MAX_LOD_LAYER+1;
+                for (int i = 0; i < LAYERS; i++) {
+                    RenderStatistics.visibleSections[i] = MemoryUtil.memGetInt(down.address+i*4L);
+                }
+
+                for (int i = 0; i < LAYERS; i++) {
+                    RenderStatistics.quadCount[i] = MemoryUtil.memGetInt(down.address+LAYERS*4L+i*4L);
+                }
+            });
+
+            this.statisticsBuffer.zero();
+        }
+    }
+
+    @Override
+    public void renderTranslucent(MeshViewport viewport) {
+        return;
+    }
+
+    @Override
+    public void buildDrawCalls(MeshViewport viewport) {
+        if (this.geometryManager.getSectionCount() == 0) return;
+        this.uploadUniformBuffer(viewport);
+        //Can do a sneeky trick, since the sectionRenderList is a list to things to render, it invokes the culler
+        // which only marks visible sections
+
+
+        {//Test occlusion
+            glCopyNamedBufferSubData(viewport.getRenderList().id, this.cullAndMeshDrawCommand.id, 0, 4, 4);//Copy counts to the draw buffer
+            glCopyNamedBufferSubData(viewport.getRenderList().id, this.cullAndMeshDrawCommand.id, 0, 20, 4);//Copy counts to the draw buffer
+
+            this.cullShader.bind();
+            if (Capabilities.INSTANCE.repFragTest) {
+                glEnable(GL_REPRESENTATIVE_FRAGMENT_TEST_NV);
+            }
+            glBindVertexArray(GlVertexArray.STATIC_VAO);
+            glBindBufferBase(GL_UNIFORM_BUFFER, 0, this.uniform.id);
+            glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, this.geometryManager.getMetadataBuffer().id);
+            glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, viewport.visibilityBuffer.id);
+            glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, viewport.getRenderList().id);
+            glBindBuffer(GL_DRAW_INDIRECT_BUFFER, this.cullAndMeshDrawCommand.id);
+            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, SharedIndexBuffer.INSTANCE.id());
+            glEnable(GL_DEPTH_TEST);
+            glColorMask(false, false, false, false);
+            glDepthMask(false);
+            glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT|GL_COMMAND_BARRIER_BIT);
+            glDrawElementsIndirect(GL_TRIANGLES, GL_UNSIGNED_BYTE, 0);
+            glDepthMask(true);
+            glColorMask(true, true, true, true);
+            glDisable(GL_DEPTH_TEST);
+            if (Capabilities.INSTANCE.repFragTest) {
+                glDisable(GL_REPRESENTATIVE_FRAGMENT_TEST_NV);
+            }
+        }
+    }
+
+    @Override
+    public void renderTemporal(MeshViewport viewport) {
+        return;
+    }
+
+    @Override
+    public void addDebug(List<String> lines) {
+        super.addDebug(lines);
+        //lines.add("SC/GS: " + this.geometryManager.getSectionCount() + "/" + (this.geometryManager.getGeometryUsed()/(1024*1024)));//section count/geometry size (MB)
+    }
+
+    @Override
+    public MeshViewport createViewport() {
+        return new MeshViewport(this.geometryManager.getMaxSectionCount());
+    }
+
+    @Override
+    public void free() {
+        this.cullAndMeshDrawCommand.free();
+        this.uniform.free();
+        this.terrainShader.free();
+        this.cullShader.free();
+        this.statisticsBuffer.free();
+    }
+}

src/main/java/me/cortex/voxy/client/core/rendering/section/MeshViewport.java

@@ -0,0 +1,26 @@
+package me.cortex.voxy.client.core.rendering.section;
+
+import me.cortex.voxy.client.core.gl.GlBuffer;
+import me.cortex.voxy.client.core.rendering.Viewport;
+import me.cortex.voxy.client.core.rendering.hierachical.HierarchicalOcclusionTraverser;
+
+public class MeshViewport extends Viewport<MeshViewport> {
+    public final GlBuffer indirectLookupBuffer = new GlBuffer(HierarchicalOcclusionTraverser.MAX_QUEUE_SIZE *4+4);
+    public final GlBuffer visibilityBuffer;
+
+    public MeshViewport(int maxSectionCount) {
+        this.visibilityBuffer = new GlBuffer(maxSectionCount*4L);
+    }
+
+    @Override
+    protected void delete0() {
+        super.delete0();
+        this.visibilityBuffer.free();
+        this.indirectLookupBuffer.free();
+    }
+
+    @Override
+    public GlBuffer getRenderList() {
+        return this.indirectLookupBuffer;
+    }
+}

src/main/resources/assets/voxy/shaders/lod/meshext/frag.glsl

@@ -0,0 +1,90 @@
+#version 460 core
+
+#extension GL_EXT_mesh_shader : require
+
+layout(binding = 0) uniform sampler2D blockModelAtlas;
+layout(binding = 2) uniform sampler2D depthTex;
+
+layout(location=1) perprimitiveEXT in PerPrimData {
+    uvec4 data;
+} primIn;
+
+layout(location = 0) out vec4 outColour;
+
+vec4 uint2vec4RGBA(uint colour) {
+    return vec4((uvec4(colour)>>uvec4(24,16,8,0))&uvec4(0xFF))/255.0;
+}
+
+bool useMipmaps() {
+    return (primIn.data.x&2u)==0u;
+}
+
+bool useTinting() {
+    return (primIn.data.x&4u)!=0u;
+}
+
+bool useCutout() {
+    return (primIn.data.x&1u)==1u;
+}
+
+vec4 computeColour(vec4 colour) {
+    //Conditional tinting, TODO: FIXME: REPLACE WITH MASK OR SOMETHING, like encode data into the top bit of alpha
+    if (useTinting() && abs(colour.r-colour.g) < 0.02f && abs(colour.g-colour.b) < 0.02f) {
+        colour *= uint2vec4RGBA(primIn.data.z).yzwx;
+    }
+    return (colour * uint2vec4RGBA(primIn.data.y)) + vec4(0,0,0,float(primIn.data.w&0xFFu)/255);
+}
+
+
+uint getFace() {
+    return (primIn.data.w>>8)&7u;
+}
+
+vec2 getBaseUV() {
+    uint face = getFace();
+    uint modelId = primIn.data.x>>16;
+    vec2 modelUV = vec2(modelId&0xFFu, (modelId>>8)&0xFFu)*(1.0/(256.0));
+    return modelUV + (vec2(face>>1, face&1u) * (1.0/(vec2(3.0, 2.0)*256.0)));
+}
+
+
+void main() {
+    vec2 uv = vec2(0);
+
+    //Tile is the tile we are in
+    vec2 tile;
+    vec2 uv2 = modf(uv, tile)*(1.0/(vec2(3.0,2.0)*256.0));
+    vec4 colour;
+    vec2 texPos = uv2 + getBaseUV();
+    if (useMipmaps()) {
+        vec2 uvSmol = uv*(1.0/(vec2(3.0,2.0)*256.0));
+        vec2 dx = dFdx(uvSmol);//vec2(lDx, dDx);
+        vec2 dy = dFdy(uvSmol);//vec2(lDy, dDy);
+        colour = textureGrad(blockModelAtlas, texPos, dx, dy);
+    } else {
+        colour = textureLod(blockModelAtlas, texPos, 0);
+    }
+
+    if (any(notEqual(clamp(tile, vec2(0), vec2((primIn.data.x>>8)&0xFu, (primIn.data.x>>12)&0xFu)), tile))) {
+        discard;
+    }
+
+    //Check the minimum bounding texture and ensure we are greater than it
+    if (gl_FragCoord.z < texelFetch(depthTex, ivec2(gl_FragCoord.xy), 0).r) {
+        discard;
+    }
+
+
+    //Also, small quad is really fking over the mipping level somehow
+    if (useCutout() && (textureLod(blockModelAtlas, texPos, 0).a <= 0.1f)) {
+        //This is stupidly stupidly bad for divergence
+        //TODO: FIXME, basicly what this do is sample the exact pixel (no lod) for discarding, this stops mipmapping fucking it over
+        #ifndef DEBUG_RENDER
+        discard;
+        #endif
+    }
+
+    colour = computeColour(colour);
+
+    outColour = colour;
+}

src/main/resources/assets/voxy/shaders/lod/meshext/mesh.glsl

@@ -0,0 +1,327 @@
+#version 460 core
+
+#extension GL_EXT_mesh_shader : require
+
+#extension GL_ARB_gpu_shader_int64 : enable
+
+#extension GL_KHR_shader_subgroup_arithmetic: require
+#extension GL_KHR_shader_subgroup_basic : require
+#extension GL_KHR_shader_subgroup_ballot : require
+#extension GL_KHR_shader_subgroup_vote : require
+
+
+//TODO: finetune the local size and emission size
+layout(local_size_x = MESH_SIZE, local_size_y=1, local_size_z=1) in;
+layout(triangles, max_vertices=(MESH_SIZE*4), max_primitives=(MESH_SIZE*2)) out;
+
+//Tightly packed, prefix sum + offset
+//uvec4 binA;
+//uvec4 binB;
+struct Task {
+    uint bins[8];
+
+    vec3 cameraOffset;
+    uint lodLvl;
+
+    uint baseQuad;
+    uint quadCount;
+};
+
+taskPayloadSharedEXT Task task;
+
+layout(location=1) perprimitiveEXT out PerPrimData {
+    uvec4 data;
+} primOut[];
+
+
+uint getQuadId() {
+    uint mid = gl_GlobalInvocationID.x;
+    uint cv = (mid<<16)|0xFFFFu;
+    /*
+    //Funny method
+    uvec4 a = mix(uvec4(0), uvec4( 1, 2, 4,  8), lessThanEqual(uvec4(task.bins[0],task.bins[1],task.bins[2],task.bins[3]), uvec4(cv))) +
+              mix(uvec4(0), uvec4(16,32,64,128), lessThanEqual(uvec4(task.bins[4],task.bins[5],task.bins[6],task.bins[7]), uvec4(cv)));
+    uint act = a.x+a.y+a.z+a.w;
+    uint id = findLSB(act^(act>>1));
+
+    //uint point = mix(binB, binA, id<4)[id&3u];
+    uint point = task.bins[id];
+
+    return (point&0xFFFFu)+(mid-(point>>16));
+    */
+    #pragma unroll
+    for (uint i = 0; i<7; i++) {
+        uint point = task.bins[i];
+        if (point<=cv&&cv<task.bins[i+1]) {
+            return (point&0xFFFFu)+(mid-(point>>16));
+        }
+    }
+    return -1;
+
+
+
+    /*
+    for (uint i = 0; i<7; i++) {
+        uint point = task.bins[i];
+        if (point <= ((mid<<16)|0xFFFFu) && ((mid<<16)|0xFFFFu)<task.bins[i+1]) {
+            binId = i;
+            return (point&0xFFFFu)+(mid-(point>>16));
+        }
+    }
+    return -1;
+    */
+}
+
+#import <voxy:lod/quad_format.glsl>
+#import <voxy:lod/block_model.glsl>
+
+layout(binding = 0, std140) uniform SceneUniform {
+    mat4 MVP;
+    ivec3 baseSectionPos;
+    uint frameId;
+    vec3 cameraSubPos;
+    uint pad_;
+    vec2 screenSize;
+};
+
+layout(binding = 4, std430) readonly restrict buffer QuadBuffer {
+    Quad quadData[];
+};
+
+layout(binding = 5, std430) readonly restrict buffer ModelBuffer {
+    BlockModel modelData[];
+};
+
+layout(binding = 6, std430) readonly restrict buffer ModelColourBuffer {
+    uint colourData[];
+};
+
+layout(binding = 1) uniform sampler2D lightSampler;
+vec4 getLighting(uint index) {
+    int i2 = int(index);
+    return texture(lightSampler, clamp((vec2((i2>>4)&0xF, i2&0xF))/16, vec2(8.0f/255), vec2(248.0f/255)));
+}
+
+//===============
+
+
+vec3 swizzelDataAxis(uint axis, vec3 data) {
+    return mix(mix(data.zxy,data.xzy,bvec3(axis==0)),data,bvec3(axis==1));
+}
+
+vec4 getFaceSize(uint faceData) {
+    float EPSILON = 0.00005f;
+
+    vec4 faceOffsetsSizes = extractFaceSizes(faceData);
+
+    //Expand the quads by a very small amount (because of the subtraction after this also becomes an implicit add)
+    faceOffsetsSizes.xz -= vec2(EPSILON);
+
+    //Make the end relative to the start
+    faceOffsetsSizes.yw -= faceOffsetsSizes.xz;
+
+    return faceOffsetsSizes;
+}
+
+vec3 faceNormal(uint face) {
+    //TODO: optimize this garbage
+    return vec3(uint((face>>1)==2), uint((face>>1)==0), uint((face>>1)==1)) * (float(int(face)&1)*2-1);
+}
+
+uint packVec4(vec4 vec) {
+    uvec4 vec_=uvec4(vec*255)<<uvec4(24,16,8,0);
+    return vec_.x|vec_.y|vec_.z|vec_.w;
+}
+
+//===============
+vec3 cornerPos;
+vec2 axisFaceSize;
+uint face;
+
+vec4 faceSize;
+
+uint modelId;
+BlockModel model;
+uint faceData;
+bool isTranslucent;
+bool hasAO;
+bool isShaded;
+
+void setup(Quad quad) {
+    face = extractFace(quad);
+    modelId = extractStateId(quad);
+    model = modelData[modelId];
+    faceData = model.faceData[face];
+    isTranslucent = modelIsTranslucent(model);
+    hasAO = modelHasMipmaps(model);//TODO: replace with per face AO flag
+    isShaded = hasAO;//TODO: make this a per face flag
+
+    ivec2 quadSize = extractSize(quad);
+
+    faceSize = getFaceSize(faceData);
+
+    cornerPos = extractPos(quad);
+    float depthOffset = extractFaceIndentation(faceData);
+    cornerPos += swizzelDataAxis(face>>1, vec3(faceSize.xz, mix(depthOffset, 1-depthOffset, float(face&1u))));
+    cornerPos *= (1<<task.lodLvl);
+    cornerPos += task.cameraOffset;
+
+    axisFaceSize = (faceSize.yw + quadSize - 1);
+
+    //uv =
+
+}
+
+vec2 getUvCorner(uint corner) {
+    return faceSize.xz + axisFaceSize*vec2((corner>>1)&1u, corner&1u);;
+}
+
+uvec4 createQuadData(Quad quad) {
+    uint flags = faceHasAlphaCuttout(faceData);
+
+    ivec2 quadSize = extractSize(quad);
+    //We need to have a conditional override based on if the model size is < a full face + quadSize > 1
+    flags |= uint(any(greaterThan(quadSize, ivec2(1)))) & faceHasAlphaCuttoutOverride(faceData);
+
+    flags |= uint(!modelHasMipmaps(model))<<1;
+
+    //Compute lighting
+    vec4 tinting = getLighting(extractLightId(quad));
+
+    //Apply model colour tinting
+    uint tintColour = model.colourTint;
+    if (modelHasBiomeLUT(model)) {
+        tintColour = colourData[tintColour + extractBiomeId(quad)];
+    }
+
+    uint conditionalTinting = 0;
+    if (tintColour != uint(-1)) {
+        flags |= 1u<<2;
+        conditionalTinting = tintColour;
+    }
+
+    uint addin = 0;
+    if (!isTranslucent) {
+        tinting.w = 0.0;
+        //Encode the face, the lod level and
+        uint encodedData = 0;
+        encodedData |= face;
+        encodedData |= (task.lodLvl<<3);
+        encodedData |= uint(hasAO)<<6;
+        addin = encodedData;
+    }
+
+    //Apply face tint
+    if (isShaded) {
+        //TODO: make branchless, infact apply ahead of time to the texture itself in ModelManager since that is
+        // per face
+        if ((face>>1) == 1) {//NORTH, SOUTH
+            tinting.xyz *= 0.8f;
+        } else if ((face>>1) == 2) {//EAST, WEST
+            tinting.xyz *= 0.6f;
+        } else if (face == 0) {//DOWN
+            tinting.xyz *= 0.5f;
+        }
+    }
+
+
+
+    uvec4 interData;
+
+    interData.x = (modelId<<16) | flags | (uint(quadSize.x-1)<<8) | (uint(quadSize.y-1)<<12);
+
+    interData.y = packVec4(tinting);
+    interData.z = conditionalTinting;
+    interData.w = addin|(face<<8);
+
+    return interData;
+}
+
+vec4 emitVertexPos(int corner) {
+    vec3 pointPos = swizzelDataAxis(face>>1,vec3(axisFaceSize*mix(vec2(0),vec2(1<<task.lodLvl),bvec2((corner>>1)&1, corner&1)),0))+cornerPos;
+    return MVP*vec4(pointPos, 1.0);
+}
+
+bvec2 whatRender(vec4 p1, vec4 p2, vec4 p0, vec4 p3) {
+    vec2 ssmin = ((p1.xy/p1.w)+1)*screenSize;
+    vec2 ssmax = ssmin;
+
+    vec2 point = ((p2.xy/p2.w)+1)*screenSize;
+    ssmin = min(ssmin, point);
+    ssmax = max(ssmax, point);
+
+    point = ((p0.xy/p0.w)+1)*screenSize;
+    vec2 t0min = min(ssmin, point);
+    vec2 t0max = max(ssmax, point);
+
+    point = ((p3.xy/p3.w)+1)*screenSize;
+    vec2 t1min = min(ssmin, point);
+    vec2 t1max = max(ssmax, point);
+
+    //Possibly cull the triangles if they dont cover the center of a pixel on the screen (degen)
+    float degenBias = 0.01f;
+    bool t0draw = all(notEqual(round(t0min-degenBias),round(t0max+degenBias)));
+    bool t1draw = all(notEqual(round(t1min-degenBias),round(t1max+degenBias)));
+    return bvec2(t0draw, t1draw);
+}
+
+#ifdef HAS_STATISTICS
+layout(binding = STATISTICS_BUFFER_BINDING, std430) restrict buffer statisticsBuffer {
+    uint visibleSectionCounts[5];
+    uint quadCounts[5];
+};
+#endif
+
+void main() {
+    uint qid = uint(-1);
+    Quad quad;
+    if (gl_GlobalInvocationID.x<task.quadCount) {
+        qid = getQuadId() + task.baseQuad;
+        quad = quadData[qid];
+        setup(quad);
+        bool render = dot(faceNormal(face), cornerPos-cameraSubPos) <= 0;
+        qid = render?qid:uint(-1);
+    }
+
+
+    subgroupBarrier();
+    uint triId_ = subgroupExclusiveAdd(qid!=uint(-1)?2:0);
+    uint qc = subgroupMax(triId_+(qid!=uint(-1)?2:0));
+    SetMeshOutputsEXT(qc*2, qc);
+
+    #ifdef HAS_STATISTICS
+    if (subgroupElect()) {
+        atomicAdd(quadCounts[task.lodLvl], qc);
+    }
+    #endif
+
+
+
+    if (qid != uint(-1)) {
+        uint vertId_ = triId_*2;//hack cause we are emitting full quads (2 tris) just to test
+        uint triId = triId_;
+        uint vertId = vertId_;
+
+        vec4 p1 = emitVertexPos(1);
+        vec4 p2 = emitVertexPos(2);
+        vec4 p0 = emitVertexPos(0);
+        vec4 p3 = emitVertexPos(3);
+
+        uvec4 data = createQuadData(quad);
+
+        //Emit common
+        gl_MeshVerticesEXT[vertId++].gl_Position = p1;
+        gl_MeshVerticesEXT[vertId++].gl_Position = p2;
+
+
+        //Prim 1
+            gl_PrimitiveTriangleIndicesEXT[triId] = uvec3(vertId_+0, vertId_+1, vertId);
+            gl_MeshVerticesEXT[vertId++].gl_Position = p0;
+            primOut[triId++].data = data;
+
+        //Prim 2
+            gl_PrimitiveTriangleIndicesEXT[triId] = uvec3(vertId_+0, vertId, vertId_+1);
+            gl_MeshVerticesEXT[vertId++].gl_Position = p3;
+            primOut[triId++].data = data;
+    }
+}

src/main/resources/assets/voxy/shaders/lod/meshext/task.glsl

@@ -0,0 +1,125 @@
+#version 460 core
+
+#extension GL_EXT_mesh_shader : require
+
+layout(local_size_x=1, local_size_y=1, local_size_z=1) in;
+
+#import <voxy:lod/section.glsl>
+
+layout(binding = 0, std140) uniform SceneUniform {
+    mat4 MVP;
+    ivec3 baseSectionPos;
+    uint frameId;
+    vec3 cameraSubPos;
+    uint pad_;
+    vec2 screenSize;
+};
+
+layout(binding = 1, std430) restrict readonly buffer IndirectSectionLookupBuffer {
+    uint sectionCount;
+    uint indirectLookup[];
+};
+
+layout(binding = 2, std430) restrict readonly buffer SectionBuffer {
+    SectionMeta sectionData[];
+};
+
+layout(binding = 3, std430) restrict readonly buffer VisibilityBuffer {
+    uint visibilityData[];
+};
+
+#ifdef HAS_STATISTICS
+layout(binding = STATISTICS_BUFFER_BINDING, std430) restrict buffer statisticsBuffer {
+    uint visibleSectionCounts[5];
+    uint quadCounts[5];
+};
+#endif
+
+struct Task {
+    //Tightly packed, prefix sum + offset
+    //uvec4 binA;
+    //uvec4 binB;
+    uint bins[8];
+
+    vec3 cameraOffset;
+    uint lodLvl;
+
+    uint baseQuad;
+    uint quadCount;
+};
+taskPayloadSharedEXT Task task;
+
+#define BIN(br, cnt) if (br) { task.bins[i++] = (sum<<16)|off; sum += cnt; } off += cnt;
+//#define BIN(br, cnt) if (br) { batch[i++] = (sum<<16)|off; sum += cnt; } off += cnt;
+
+bvec3 and(bvec3 a, bvec3 b) {
+    return bvec3(a.x&&b.x, a.y&&b.y, a.z&&b.z);
+}
+uint fillBins(uvec4 counts, ivec3 relative) {//Returns quad count
+    #pragma unroll
+    for (uint i = 0; i < 8; i++) task.bins[i] = uint(-1);
+
+    uvec3 cA = counts.yzw&0xFFFFu;
+    uvec3 cB = counts.yzw>>16;
+
+    bvec3 a = and(notEqual(cA, uvec3(0)), lessThanEqual(ivec3(0), relative.yzx));
+    bvec3 b = and(notEqual(cB, uvec3(0)), lessThanEqual(relative.yzx, ivec3(0)));
+
+    uint dsc = counts.x>>16;//double sided quads
+    uint sum = 0;
+    uint off = counts.x&0xFFFFu;//translucent quads
+    uint i = 0;
+
+    //TODO: might need to move this into shared memory or somethign? so that compiler can reason about it (or make the bin an array in here and mesh)
+    //uint batch[8] = {uint(-1), uint(-1), uint(-1), uint(-1), uint(-1),uint(-1),uint(-1),uint(-1)};
+
+    BIN(dsc!=0, dsc);//Double sided quads
+
+    //TODO: compute prefix sums and then jsut batch set into the array (this is an optimization)
+
+    BIN(a.x, cA.x);//Down
+    BIN(b.x, cB.x);//Up
+    BIN(a.y, cA.y);//North
+    BIN(b.y, cB.y);//South
+    BIN(a.z, cA.z);//West
+    BIN(b.z, cB.z);//East
+
+    //task.binA = uvec4(batch[0], batch[1], batch[2], batch[3]);
+    //task.binB = uvec4(batch[4], batch[5], batch[6], batch[7]);
+    return sum;
+}
+
+
+void main() {
+    uint secId = indirectLookup[gl_WorkGroupID.x];
+    uint vis = visibilityData[secId];
+
+    bool shouldRender = (vis&0x7fffffffu) == frameId-1;//-1 since we are technically in the next frame for the primary rasterization
+    bool renderTemporally = (vis&0x80000000u)==0;
+
+    task.quadCount = 0;
+
+    if (shouldRender) {
+        SectionMeta section = sectionData[secId];
+
+        uint detail = extractDetail(section);
+        ivec3 ipos = extractPosition(section);
+
+        ivec3 relative = ipos-(baseSectionPos>>detail);
+
+        #ifdef HAS_STATISTICS
+        atomicAdd(visibleSectionCounts[detail], 1);
+        #endif
+
+        //TODO: here enqueue the id here for both translucent and temporal (if relevant) (* note technically dont need for temporal as can just check :tm: if we are in temporal render mode)
+
+        task.baseQuad  = extractQuadStart(section);
+        task.quadCount = fillBins(section.b, relative);
+
+        task.cameraOffset = vec3(((ipos<<detail) - baseSectionPos)<<5);
+        task.lodLvl = detail;
+    }
+
+    //It appears to be valid to read from taskPayloadSharedEXT
+    EmitMeshTasksEXT((task.quadCount+(MESH_SIZE-1))/MESH_SIZE, 1, 1);
+}