Alot of gpu work

build.gradle

@@ -56,14 +56,23 @@ dependencies {
     modCompileOnly "maven.modrinth:nvidium:0.2.8-beta"
 
     modImplementation("maven.modrinth:cloth-config:13.0.121+fabric")
+
     modImplementation("maven.modrinth:modmenu:11.0.1")
+
     modCompileOnly("maven.modrinth:iris:1.7.3+1.21")
+
     //modRuntimeOnly("maven.modrinth:iris:1.6.17+1.20.4")
+
     modCompileOnly("maven.modrinth:starlight:1.1.3+1.20.4")
+
     //modCompileOnly("maven.modrinth:immersiveportals:v5.1.7-mc1.20.4")
+
     modCompileOnly("maven.modrinth:vivecraft:1.20.4-1.1.6-fabric")
+
     modCompileOnly("maven.modrinth:chunky:1.3.138")
     modRuntimeOnly("maven.modrinth:chunky:1.3.138")
+
+    modImplementation('io.github.douira:glsl-transformer:2.0.1')
 }
 
 

src/main/java/me/cortex/voxy/common/world/service/ServiceThreadPool.java

@@ -0,0 +1,72 @@
+package me.cortex.voxy.common.world.service;
+
+import me.cortex.voxy.common.world.WorldEngine;
+import me.cortex.voxy.common.world.WorldSection;
+import net.minecraft.client.MinecraftClient;
+import net.minecraft.text.Text;
+
+import java.util.concurrent.ConcurrentLinkedDeque;
+import java.util.concurrent.Semaphore;
+
+//TODO:
+//FIXME:
+// FINISHME:
+// Use this instead of seperate thread pools, use a single shared pool where tasks are submitted to and worked on
+
+public class ServiceThreadPool {
+    private volatile boolean running = true;
+    private final Thread[] workers;
+    private final Semaphore jobCounter = new Semaphore(0);
+    //TODO: have a wrapper to specify extra information about the job for debugging
+    private final ConcurrentLinkedDeque<Runnable> jobQueue = new ConcurrentLinkedDeque<>();
+
+
+    public ServiceThreadPool(int workers) {
+        this.workers = new Thread[workers];
+        for (int i = 0; i < workers; i++) {
+            var worker = new Thread(this::worker);
+            worker.setDaemon(false);
+            worker.setName("Service worker #" + i);
+            worker.start();
+            this.workers[i]  = worker;
+        }
+    }
+
+    private void worker() {
+        while (true) {
+            this.jobCounter.acquireUninterruptibly();
+            if (!this.running) {
+                break;
+            }
+            var job = this.jobQueue.pop();
+            try {
+                job.run();
+            } catch (Exception e) {
+                System.err.println(e);
+                MinecraftClient.getInstance().executeSync(()->
+                        MinecraftClient.getInstance().player.sendMessage(
+                                Text.literal(
+                                        "Voxy ingester had an exception while executing service job please check logs and report error")));
+            }
+        }
+    }
+
+
+    public void shutdown() {
+        //Wait for the tasks to finish
+        while (this.jobCounter.availablePermits() != 0) {
+            Thread.onSpinWait();
+        }
+
+        //Shutdown
+        this.running = false;
+        this.jobCounter.release(1000);
+
+        //Wait for thread to join
+        try {
+            for (var worker : this.workers) {
+                worker.join();
+            }
+        } catch (InterruptedException e) {throw new RuntimeException(e);}
+    }
+}

src/main/resources/assets/voxy/shaders/lod/hierarchial/node.glsl

@@ -0,0 +1,85 @@
+
+layout(binding = NODE_DATA_INDEX, std430) restrict buffer NodeData {
+//Needs to be read and writeable for marking data,
+//(could do an evil violation, make this readonly, then have a writeonly varient, which means that writing might not be visible but will show up by the next frame)
+//Nodes are 16 bytes big (or 32 cant decide, 16 might _just_ be enough)
+    ivec4[] nodes;
+};
+
+//First 2 are joined to be the position
+
+
+
+
+//All node access and setup into global variables
+//TODO: maybe make it global vars
+struct UnpackedNode {
+    uint nodeId;
+
+    ivec3 pos;
+    uint lodLevel;
+
+    uint flags;
+
+    uint meshPtr;
+    uint childPtr;
+};
+
+#define NULL_NODE ((1<<25)-1)
+#define NULL_MESH ((1<<24)-1)
+
+void unpackNode(inout UnpackedNode node, uint nodeId) {
+    node.nodeId = nodeId;
+    uvec4 compactedNode = nodes[nodeId];
+    node.lodLevel = compactedNode.x >> 28;
+
+    {
+        int y = ((int(compactedNode.x)<<4)>>24);
+        int x = (int(compactedNode.y)<<4)>>8;
+        int z = int((int(compactedNode.x)&((1<<20)-1))<<4);
+        z |= int(compactedNode.y>>28);
+        z <<= 8;
+        z >>= 8;
+
+        node.pos = ivec3(x, y, z);
+    }
+
+    node.meshPtr = compactedNode.z&0xFFFFFFu;
+    node.childPtr = compactedNode.w&0x1FFFFFFu;
+    node.flags = (compactedNode.z>>24) | ((compactedNode.w>>23)<<8);
+}
+
+bool hasMesh(in UnpackedNode node) {
+    return node.meshPtr != NULL_MESH;
+}
+
+bool hasChildren(in UnpackedNode node) {
+    return node.childPtr != NULL_NODE;
+}
+
+bool isEmpty(in UnpackedNode node) {
+    return (node.flags&2u) != 0;
+}
+
+bool hasRequested(in UnpackedNode node) {
+    return (node.flags&1u) != 0u;
+}
+
+uint getMesh(in UnpackedNode node) {
+    return node.meshPtr;
+}
+
+uint getId(in UnpackedNode node) {
+    return node.nodeId;
+}
+
+uint getChild(in UnpackedNode node) {
+    return node.flags >> 2;
+}
+
+//-----------------------------------
+
+void markRequested(inout UnpackedNode node) {
+    node.flags |= 1u;
+    nodes[node.nodeId].z |= 1u<<24;
+}

src/main/resources/assets/voxy/shaders/lod/hierarchial/screenspace.glsl

@@ -12,22 +12,55 @@
 // substantually for performance (for both persistent threads and incremental)
 
 
+layout(binding = HIZ_BINDING_INDEX) uniform sampler2DShadow hizDepthSampler;
 
+//TODO: maybe do spher bounds aswell? cause they have different accuracies but are both over estimates (liberals (non conservative xD))
+// so can do &&
 
+vec3 minBB;
+vec3 maxBB;
+vec2 size;
 
-
+vec3 projPoint(mat4 mat, vec3 pos) {
+    vec4 t = mat * vec4(vec3(pos),1);
+    return t.xyz/t.w;
+}
 
 //Sets up screenspace with the given node id, returns true on success false on failure/should not continue
 //Accesses data that is setup in the main traversal and is just shared to here
-bool setupScreenspace() {
+void setupScreenspace(in UnpackedNode node) {
+    //TODO: Need to do aabb size for the nodes, it must be an overesimate of all the children
 
+    mat4 mvp;
+
+    vec3 basePos;
+    vec3 minSize;
+    vec3 maxSize;
+
+
+    vec3 minPos = minSize + basePos;
+    vec3 maxPos = maxSize + basePos;
+
+    minBB = projPoint(mvp, minPos);
+    maxBB = minBB;
+
+    for (int i = 1; i < 8; i++) {
+        vec3 point = projPoint(mvp, mix(minPos, maxPos, bvec3((i&1)!=0,(i&2)!=0,(i&4)!=0)));
+        minBB = min(minBB, point);
+        maxBB = max(maxBB, point);
+    }
+
+    size = maxBB.xy - minBB.xy;
 }
 
 bool isCulledByHiz() {
-
+    vec2 ssize = size.xy * vec2(ivec2(screensize));
+    float miplevel = ceil(log2(max(max(ssize.x, ssize.y),1)));
+    vec2 midpoint = (maxBB.xy + minBB.xy)*0.5;
+    return textureLod(hizDepthSampler, vec3(midpoint, minBB.z), miplevel) > 0.0001;
 }
 
 //Returns if we should decend into its children or not
 bool shouldDecend() {
-
+    return (size.x*size.y) > (64*64F);
 }

src/main/resources/assets/voxy/shaders/lod/hierarchial/traversal.glsl

@@ -12,15 +12,10 @@ layout(local_size_x=1, local_size_y=1) in;
 // then increment head strictly _AFTER_ writing to the queue, this ensures that the data is always written and avaible in the queue
 
 layout(binding = 0, std140) uniform SceneUniform {
-
+    uint a;
-};
-
-layout(binding = 1, std430) restrict buffer NodeData {//Needs to be read and writeable for marking data,
-    //(could do an evil violation, make this readonly, then have a writeonly varient, which means that writing might not be visible but will show up by the next frame)
-    //Nodes are 16 bytes big (or 32 cant decide, 16 might _just_ be enough)
-    ivec4[] nodes;
 };
 
+#define NODE_DATA_INDEX 1
 
 layout(binding = 2, std430) restrict buffer Atomics {
     uint requestQueueIndex;
@@ -51,20 +46,12 @@ layout(binding = 2, std430) restrict buffer QueueData {
 
 #import <voxy:lod/hierarchial/node.glsl>
 
-layout(binding = 0) uniform sampler2DShadow hizDepthSampler;
+#define HIZ_BINDING_INDEX 0
-
-
-void aqcuireNewBatch() {
-
-}
 
 //Contains all the screenspace computation
 #import <voxy:lod/hierarchial/screenspace.glsl>
 
 
-
-
-
 //If a request is successfully added to the RequestQueue, must update NodeData to mark that the node has been put into the request queue
 // to prevent it from being requested every frame and blocking the queue
 
@@ -81,32 +68,37 @@ void aqcuireNewBatch() {
 // not sure
 
 
-void addRequest() {
-    if (!hasRequested()) {
-        //TODO: request this node (cpu side can figure out what it wants/needs)
 
-        //Mark node as having a request submitted to prevent duplicate submissions
+void addRequest(inout UnpackedNode node) {
+    if (!hasRequested(node)) {
+        //TODO: maybe try using only 1 variable and it being <0 being bad
+        if (atomics.requestQueueIndex < atomic.requestQueueMaxSize) {
+            //Mark node as having a request submitted to prevent duplicate submissions
+            requestQueue[atomicAdd(atomics.requestQueueIndex, 1)] = getId(node);
+            markRequested(node);
+        }
     }
 }
 
-void enqueueSelfForRender() {
+void enqueueChildren(in UnpackedNode node) {
-    //TODO: Draw mesh and stop with this node (good path)
+
 }
 
-void enqueueChildren() {
+void enqueueSelfForRender(in UnpackedNode node) {
-
+    renderQueue[atomicAdd(atomics.renderQueueIndex, 1)] = getMesh(node);
 }
 
 //TODO: need to add an empty mesh, as a parent node might not have anything to render but the children do??
 void main() {
-    uint id = 0;
+    UnpackedNode node;
 
     //Setup/unpack the node
-    unpackNode(id);
+    unpackNode(node, gl_GlobalInvocationID.x);
+
     //TODO: check the node is OK first??? maybe?
 
     //Compute screenspace
-    setupScreenspace();
+    setupScreenspace(node);
 
     if (isCulledByHiz()) {
         //We are done here, dont do any more, the issue is the shader barriers maybe
@@ -115,25 +107,25 @@ void main() {
         //It is visible, TODO: maybe do a more detailed hiz test? (or make it so that )
 
         if (shouldDecend()) {
-            if (hasChildren()) {
+            if (hasChildren(node)) {
-                enqueueChildren();
+                enqueueChildren(node);
             } else {
-                addRequest();
+                addRequest(node);
                 //TODO: use self mesh (is error state if it doesnt have one since all leaf nodes should have a mesh)
                 // Basicly guarenteed to have a mesh, if it doesnt it is very very bad and incorect since its a violation of the graph properties
                 // that all leaf nodes must contain a mesh
-                enqueueSelfForRender();
+                enqueueSelfForRender(node);
             }
         } else {
-            if (hasMesh()) {
+            if (hasMesh(node)) {
-                enqueueSelfForRender();
+                enqueueSelfForRender(node);
             } else {
                 //!! not ideal, we want to render this mesh but dont have it. If we havent sent a request
                 // then send a request for a mesh for this node.
-                addRequest();
+                addRequest(node);
 
                 //TODO: Decend into children? maybe add a bitflag saying is bad if the immediate children dont have meshes
-                enqueueChildren();
+                enqueueChildren(node);
             }
         }
     }

src/main/resources/assets/voxy/shaders/lod/hierarchical/node.glsl

@@ -1,29 +0,0 @@
-
-//All node access and setup into global variables
-uint lodLevel;
-ivec3 position;
-
-uint childrenPtr;
-uint meshPtr;
-
-uint flags;
-
-void unpackNode(uint nodeId) {
-
-}
-
-bool hasMesh() {
-
-}
-
-bool hasChildren() {
-
-}
-
-bool hasRequested() {
-
-}
-
-void markRequested() {
-
-}