working z prepass shit

Cargo.toml

@@ -16,6 +16,7 @@ glam = "0.33.5"
 image = "0.25.10"
 indicatif = "0.18.6"
 itertools = "0.15.0"
+ordered-float = "5.5.0"
 pollster = "1.0.1"
 rand = "0.10.2"
 rayon = "1.12.0"

shaders/voxel.wgsl

@@ -2,11 +2,12 @@ struct VertexOutput
 {
     @builtin(position) postion: vec4<f32>,
     @location(0) @interpolate(flat) chunk_index: u32,
-    @location(1) color: vec4<f32>,
-    @location(2) cam_pos: vec3<f32>,
-    @location(3) world_pos: vec3<f32>,
-    @location(4) @interpolate(flat) structure_id: u32,
-    @location(5) chunk_position: vec3<f32>
+    @location(1) ndc: vec4<f32>,
+    @location(2) color: vec4<f32>,
+    @location(3) cam_pos: vec3<f32>,
+    @location(4) world_pos: vec3<f32>,
+    @location(5) @interpolate(flat) structure_id: u32,
+    @location(6) chunk_position: vec3<f32>,
 }
 
 struct ChunkImmediate
@@ -14,6 +15,8 @@ struct ChunkImmediate
     view_proj: mat4x4<f32>,
     cam_pos: vec3<f32>,
     frame_timestamp: u32,
+    width: u32,
+    downsampling_factor: u32,
 }
 
 var<immediate> constants: ChunkImmediate;
@@ -24,7 +27,7 @@ struct CacheChunkObject
     transform: mat4x4<f32>,
     color: vec4<f32>,
     id: u32,
-    pointer: u32
+    pointer: u32,
 }
 
 
@@ -73,10 +76,7 @@ fn unpack_color(color: u32) -> vec4<f32>
 @group(0) @binding(5) var<storage, read_write> structure_table_pointer: array<u32>;
 @group(0) @binding(6) var<storage, read_write> structure_table_request_buffer: array<atomic<u32>>;
 
-struct FragmentOutput {
-    @location(0) color: vec4<f32>,
-    @builtin(frag_depth) depth: f32, // Equivalent to gl_FragDepth
-}
+@group(1) @binding(0) var prepass_depth: texture_2d<f32>;
 
 @vertex
 fn chunk(@builtin(vertex_index) index: u32, @location(0) position: vec3<f32>, @location(1) id: u32) -> VertexOutput
@@ -119,6 +119,7 @@ fn chunk(@builtin(vertex_index) index: u32, @location(0) position: vec3<f32>, @l
 
     var output: VertexOutput;
     output.postion = output_vertex;
+    output.ndc = output_vertex / output_vertex.w;
     output.color = vec4(1.);
     output.chunk_index = 0;
     output.cam_pos = constants.cam_pos;
@@ -216,15 +217,15 @@ fn new_traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_of
 {
     let max_depth = 5;
     let dist_offset_voxel = dist_offset * f32(1 << u32(max_depth * 2));
-    let fovy_deg = 100. / 1920.;
+    let fovy_deg = 100. / f32(constants.width);
     let fovy_rad = (fovy_deg * 3.14) / 180.;
-    let cone_factor = tan(fovy_rad / 2.) * 2.;
+    let cone_factor = 1.414 * f32(constants.downsampling_factor) * (tan(fovy_rad / 2.) * 2.); // How many pixels per distance a voxel takes
 
     let st_pointer = structure_table_pointer[root_id];
-
     
     if (!node_subdivided(st_pointer))
     {
+        discard;
         var result: HitResult;
         result.color = vec4(0., 1., 0., 1.);
         result.hit_pos = ray_origin;
@@ -236,6 +237,7 @@ fn new_traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_of
         // Send request on structure table
         atomicAdd(&structure_table_request_buffer[root_id], 1); 
 
+        discard;
         var result: HitResult;
         result.color = vec4(0., 1., 0., 1.);
         result.hit_pos = ray_origin;
@@ -262,6 +264,7 @@ fn new_traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_of
     var inv_ray_dir = 1. / ray_dir;
     var ray_positive = ray_dir > vec3(0.);
     var step_dir = select(vec3(-1), vec3(1), ray_positive);
+    var min_child_size = cone_factor * dist_offset_voxel;
 
     for(var iter = 0; iter < 400; iter ++)
     {
@@ -270,17 +273,19 @@ fn new_traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_of
         child_size = 1 << u32(node_shift - 2);
 
         var child_pos = (voxel >> vec3(u32(node_shift - 2))) & vec3(3);
-        var pointer = structure_pool[dfs_stack[current_depth]].pointers[child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4];
+        var child_index = child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4;
+        var current_node = dfs_stack[current_depth];
+        var pointer = structure_pool[current_node].pointers[child_index];
         
-        let min_child_size = (length(vec3<f32>(voxel) - pos_origin) + dist_offset_voxel) * cone_factor;
+        min_child_size = (length(vec3<f32>(voxel) - pos_origin) + dist_offset_voxel) * cone_factor;
         while(node_subdivided(pointer) && 
-            f32(child_size / 4) >= min_child_size
+            f32(child_size) / 4 > min_child_size
             )
         {
             if(!node_pointer_valid(pointer) && node_subdivided(pointer))
             {
                 // Record request
-                atomicAdd(&request_buffer[dfs_stack[current_depth]].requests[child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4], 1);
+                atomicAdd(&request_buffer[current_node].requests[child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4], 1);
                 break;
             }
 
@@ -290,22 +295,26 @@ fn new_traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_of
             node_shift = (max_depth - current_depth) * 2;
             child_size = 1 << u32(node_shift - 2);
             child_pos = (voxel >> vec3(u32(node_shift - 2))) & vec3(3);
-            dfs_stack[current_depth] = node_pointer(pointer);
+            child_index = child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4;
+            current_node = node_pointer(pointer);
+            dfs_stack[current_depth] = current_node;
 
-            pointer = structure_pool[dfs_stack[current_depth]].pointers[child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4];
+            pointer = structure_pool[current_node].pointers[child_index];
 
             // Record usage
-            usage_buffer[dfs_stack[current_depth]] = constants.frame_timestamp;
+            usage_buffer[current_node] = constants.frame_timestamp;
         }
 
 
         // Check color
-        let color = color_pool[dfs_stack[current_depth]].colors[child_pos.x + child_pos.y * 4 + child_pos.z * 4 * 4];
+        let color = color_pool[current_node].colors[child_index];
         if(((color >> 24) & 0xFF) != 0)
         {
             var result: HitResult;
             result.color = unpack_color(color);
-            result.hit_pos = (far_t / f32(1 << u32(max_depth * 2))) * ray_dir + ray_origin;
+            //result.color = vec4<f32>(f32(iter) / 400.);
+            //result.hit_pos = (far_t / f32(1 << u32(max_depth * 2))) * ray_dir + ray_origin;
+            result.hit_pos = (far_t * ray_dir + pos_origin) / f32(1 << u32(max_depth * 2));
             return result;
         }
 
@@ -350,17 +359,143 @@ fn new_traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_of
     return result;
 }
 
+struct FragmentOutput {
+    @location(0) color: vec4<f32>,
+    @builtin(frag_depth) depth: f32, // Equivalent to gl_FragDepth
+}
+
+struct FragmentPrepassOutput {
+    @location(0) depth_prepass: f32, // Equivalent to gl_FragDepth
+    //@builtin(frag_depth) depth: f32, // Equivalent to gl_FragDepth
+}
+
+
+//fn fragment_prepass(in: VertexOutput) -> @location(0) vec4<f32>
+
+//@early_depth_test(less_equal)
+@fragment
+fn fragment_prepass(in: VertexOutput) -> FragmentPrepassOutput
+{
+    let ray_dir = normalize(in.world_pos - in.cam_pos);
+    let interp = box_inter(in.cam_pos, ray_dir, in.chunk_position + vec3(0.), in.chunk_position + vec3(1));
+    let ray_origin = in.cam_pos + ray_dir * max(interp.x, 0.) - in.chunk_position;
+
+    let result = new_traverse(ray_dir, ray_origin, in.structure_id, length(in.cam_pos - (ray_origin + in.chunk_position)));
+    let clip_pos = constants.view_proj * vec4(result.hit_pos + in.chunk_position, 1.);
+    let depth = clip_pos.z / clip_pos.w;
+    var frag_out: FragmentPrepassOutput;
+
+    //frag_out.color = result.color;
+    //frag_out.depth_prepass = in.postion.z;
+    frag_out.depth_prepass = length(ray_origin + in.chunk_position - in.cam_pos);
+    //frag_out.depth_prepass = interp.x;
+    //frag_out.depth_prepass = 100.;
+    //frag_out.depth = depth;
+
+    //frag_out.depth = depth;
+    //frag_out.depth_prepass = depth;
+    return frag_out;
+}
+
+@early_depth_test(less_equal)
+@fragment
+fn fragment(in: VertexOutput) -> FragmentOutput
+{
+    let surface_depth = in.postion.z;
+    let prepass_depth_sample = textureLoad(prepass_depth, vec2<i32>(in.postion.xy), 0).x;
+    let lin_depth = (100. * 0.01) / (100. - surface_depth * (100. - 0.01));
+
+    if lin_depth < prepass_depth_sample || prepass_depth_sample == -1.
+    {
+        discard;
+    }
+
+    //frag_out.color = vec4<f32>(2 * 0.01 / (100. + 0.01 - depth * (100. - 0.01)));
+    let ray_dir = normalize(in.world_pos - in.cam_pos);
+    let prepass_origin = in.cam_pos + ray_dir * max(prepass_depth_sample - 0.01, 0.);
+    let interp = box_inter(prepass_origin, ray_dir, in.chunk_position + vec3(0.), in.chunk_position + vec3(1));
+    let ray_origin = prepass_origin + ray_dir * max(interp.x, 0.) - in.chunk_position;
+    //let ray_origin = in.cam_pos + ray_dir * (max(0., lin_depth - 0.1)) - in.chunk_position;
+    //let ray_origin = in.cam_pos + ray_dir * max(interp.x, 0.) - in.chunk_position;
+    //let ray_origin = in.cam_pos + ray_dir * max(prepass_depth, 0.) - in.chunk_position;
+    //let space_ro = in.cam_pos + ray_dir * lin_depth;
+    //let ray_origin = space_ro - in.chunk_position;
+
+
+    let result = new_traverse(ray_dir, ray_origin, in.structure_id, length(in.cam_pos - (ray_origin + in.chunk_position)));
+    let clip_pos = constants.view_proj * vec4(result.hit_pos + in.chunk_position, 1.);
+    let depth = clip_pos.z / clip_pos.w;
+    var frag_out: FragmentOutput;
+    frag_out.color = result.color;
+    //frag_out.color = vec4<f32>(vec3<f32>(prepass_depth_sample) / 100., 1.);
+    //frag_out.color = vec4<f32>(smpl);
+    //frag_out.color = vec4<f32>(ray_origin, 1.);
+    //frag_out.color = result.color;
+    //frag_out.color = vec4<f32>(ray_origin, 1.);
+    //frag_out.color = vec4<f32>(vec3<f32>(lin_depth) / 100., 1.);
+    frag_out.depth = depth;
+    return frag_out;
+
+    //return vec4<f32>(ray_origin, 1.);
+    //return frag_out;
+    //return vec4(interp.y / 10.);
+}
+
+@fragment
+fn _fragment(in: VertexOutput) -> FragmentOutput
+{
+    //frag_out.color = vec4<f32>(2 * 0.01 / (100. + 0.01 - depth * (100. - 0.01)));
+    let ray_dir = normalize(in.world_pos - in.cam_pos);
+    let interp = box_inter(in.cam_pos - in.chunk_position, ray_dir, vec3(0.), vec3(1));
+    let ray_origin = (in.cam_pos - in.chunk_position) + ray_dir * (max(0., interp.x));
+
+    let result = new_traverse(ray_dir, ray_origin, in.structure_id, length(in.cam_pos - (ray_origin + in.chunk_position)));
+    let clip_pos = constants.view_proj * vec4(result.hit_pos + in.chunk_position, 1.);
+    let depth = clip_pos.z / clip_pos.w;
+    var frag_out: FragmentOutput;
+    //frag_out.color = result.color;
+    frag_out.color = result.color;
+    //frag_out.color = vec4<f32>(smpl);
+    //frag_out.color = vec4<f32>((2. * 0.01 * 100.) / (0.01 + 100. - prepass_depth * (100. - 0.01)));
+    //frag_out.depth = depth;
+    return frag_out;
+
+    //return vec4<f32>(ray_origin, 1.);
+    //return frag_out;
+    //return vec4(interp.y / 10.);
+}
+
+/*
+@fragment
+fn fragment(in: VertexOutput) -> @location(0) vec4<f32>
+{
+    let st = structure_table_pointer[0];
+    let subdivided = ((st >> 31) & 1) != 0;
+    let pointer_valid = ((st >> 30) & 1) != 0;
+    // Request stuff
+    atomicAdd(&structure_table_request_buffer[0], 1); 
+    if(subdivided && !pointer_valid)
+    {
+        return vec4(0., 1., 0., 1.);
+    }
+    return vec4(1., 0., 0., 1.);
+}
+*/
+
+/*
 fn traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_offset: f32) -> vec4<f32>
 {
     let st_pointer = structure_table_pointer[root_id];
 
     if (!node_subdivided(st_pointer))
     {
+        discard;
         return vec4(0., 1., 0., 1.);
     }
     if(!node_pointer_valid(st_pointer))
     {
         atomicAdd(&structure_table_request_buffer[root_id], 1); 
+        discard;
         return vec4(0., 1., 0., 1.);
     }
 
@@ -547,45 +682,4 @@ fn traverse(ray_dir: vec3<f32>, ray_origin: vec3<f32>, root_id: u32, dist_offset
     return vec4<f32>(1., 0., 1., 1.);
 
 }
-
-@early_depth_test(less_equal)
-@fragment
-fn fragment(in: VertexOutput) -> FragmentOutput
-{
-    //frag_out.color = vec4<f32>(2 * 0.01 / (100. + 0.01 - depth * (100. - 0.01)));
-    let ray_dir = normalize(in.world_pos - in.cam_pos);
-    let interp = box_inter(in.cam_pos - in.chunk_position, ray_dir, vec3(0.), vec3(1));
-    let ray_origin = (in.cam_pos - in.chunk_position) + ray_dir * (max(0., interp.x));
-
-
-    let result = new_traverse(ray_dir, ray_origin, in.structure_id, length(in.cam_pos - (ray_origin + in.chunk_position)));
-    let clip_pos = constants.view_proj * vec4(result.hit_pos + in.chunk_position, 1.);
-    let depth = clip_pos.z / clip_pos.w;
-    var frag_out: FragmentOutput;
-    //frag_out.color = result.color;
-    frag_out.color = result.color;
-    frag_out.depth = depth;
-    return frag_out;
-
-    //return vec4<f32>(ray_origin, 1.);
-    //return frag_out;
-    //return vec4(interp.y / 10.);
-}
-
-/*
-@fragment
-fn fragment(in: VertexOutput) -> @location(0) vec4<f32>
-{
-    let st = structure_table_pointer[0];
-    let subdivided = ((st >> 31) & 1) != 0;
-    let pointer_valid = ((st >> 30) & 1) != 0;
-    // Request stuff
-    atomicAdd(&structure_table_request_buffer[0], 1); 
-    if(subdivided && !pointer_valid)
-    {
-        return vec4(0., 1., 0., 1.);
-    }
-    return vec4(1., 0., 0., 1.);
-}
 */
-

src/host_production.rs

@@ -0,0 +1,15 @@
+// Contains useful facilities to render data streamed in from the host
+
+// Can produce a voxel given
+// - Its depth
+// - Its position within the chunk
+// - The chunks position
+pub trait ChunkVoxelProducer
+{
+    fn produce_voxel(
+        &mut self,
+        depth: usize,
+        chunk_position: (usize, usize, usize),
+        voxel_position: (usize, usize, usize),
+    );
+}

src/main.rs

@@ -1,65 +1,49 @@
 #![feature(generic_const_exprs)]
 #![feature(float_algebraic)]
 
-use core::sync;
-use std::cell::RefCell;
 use std::collections::HashMap;
-use std::fs::File;
-use std::hash::Hash;
-use std::rc::Rc;
 use std::sync::Arc;
-use std::sync::atomic::AtomicBool;
 use std::sync::mpsc::sync_channel;
 
 use bytemuck::Pod;
 use bytemuck::Zeroable;
 use bytemuck::cast_slice;
 use crevice::std140::AsStd140;
-use crevice::std430::AsStd430;
 use egui::Color32;
-use egui::Label;
-use egui::emath::fast_midpoint;
 use egui::mutex::Mutex;
 use egui_plot::BarChart;
 use glam::Mat4;
 use glam::Vec3;
 use glam::Vec4;
 use itertools::Itertools;
-use rand::random;
+use ordered_float::OrderedFloat;
 use rayon::iter::IndexedParallelIterator;
 use rayon::iter::IntoParallelRefIterator;
 use rayon::iter::ParallelIterator;
-use wgpu::BindGroup;
-use wgpu::BindGroupEntry;
-use wgpu::BindGroupLayoutDescriptor;
-use wgpu::BindGroupLayoutEntry;
+use wgpu::BindGroupLayout;
+use wgpu::BlendState;
 use wgpu::Buffer;
 use wgpu::BufferUsages;
-use wgpu::DepthBiasState;
+use wgpu::Color;
+use wgpu::ComputePipeline;
 use wgpu::Device;
 use wgpu::Extent3d;
 use wgpu::Features;
-use wgpu::FragmentState;
 use wgpu::InstanceDescriptor;
 use wgpu::InstanceFlags;
 use wgpu::MemoryBudgetThresholds;
-use wgpu::NoopBackendOptions;
 use wgpu::Operations;
-use wgpu::PrimitiveState;
-use wgpu::RenderPassDepthStencilAttachment;
+use wgpu::Origin3d;
+use wgpu::PipelineCompilationOptions;
 use wgpu::RenderPipeline;
-use wgpu::RenderPipelineDescriptor;
-use wgpu::ShaderModuleDescriptor;
 use wgpu::ShaderStages;
-use wgpu::StencilState;
 use wgpu::Texture;
+use wgpu::TextureFormat;
 use wgpu::TextureUsages;
 use wgpu::TextureView;
-use wgpu::VertexState;
 use wgpu::util::BufferInitDescriptor;
 use wgpu::util::DeviceExt;
 use wgpu::util::DownloadBuffer;
-use wgpu::util::StagingBelt;
 use winit::application::ApplicationHandler;
 use winit::event::DeviceEvent;
 use winit::event::MouseScrollDelta;
@@ -75,24 +59,25 @@ use winit::window::WindowId;
 
 use crate::camera::Camera;
 use crate::egui_renderer::EguiRenderer;
-use crate::producers::BallGenerator;
 use crate::producers::ChunkedProducer;
-use crate::producers::Producer;
 use crate::producers::SineGenerator;
 use crate::producers::TerrainGenerator;
-use crate::voxel_cache::cache::CacheNodeRequest;
-use crate::voxel_cache::cache::CacheResponse;
-use crate::voxel_cache::cache::ColorBytes;
-use crate::voxel_cache::cache::DestinationElement;
-use crate::voxel_cache::cache::LocationPoolElement;
-use crate::voxel_cache::cache::VoxelCache;
-use crate::voxel_cache::sparse::NTreeNodeLocator;
+use crate::sparse_tree::NTreeNodeLocator;
+use crate::voxel_cache::VoxelCache;
+use crate::voxel_cache::data::CacheNodeRequest;
+use crate::voxel_cache::data::CacheResponse;
+use crate::voxel_cache::data::ColorBytes;
+use crate::voxel_cache::data::DestinationElement;
+use crate::voxel_cache::data::LocationPoolElement;
+use crate::voxel_cache::producer_interface::CacheProducerInterface;
+use crate::voxel_cache::producer_interface::CacheRequest;
 
 mod camera;
 mod egui_renderer;
+mod host_production;
 mod producers;
+mod sparse_tree;
 mod voxel_cache;
-//mod tree;
 //
 
 struct State
@@ -104,11 +89,20 @@ struct State
     size: winit::dpi::PhysicalSize<u32>,
     surface: wgpu::Surface<'static>,
     depth_buffer: (wgpu::Texture, wgpu::TextureView),
+    prepass_depth_buffer: (wgpu::Texture, wgpu::TextureView),
+    prepass_depth: (wgpu::Texture, wgpu::TextureView),
+    upsampled_prepass_depth: (wgpu::Texture, wgpu::TextureView),
+    prepass_downsampling: u32,
+    upsample_pipeline: ComputePipeline,
     surface_format: wgpu::TextureFormat,
     egui_renderer: EguiRenderer,
 
     pipeline: RenderPipeline,
+    prepass_pipeline: RenderPipeline,
+    prepass_upsample_bg_layout: BindGroupLayout,
+    prepass_depth_bind_group_layout: BindGroupLayout,
     voxel_cache: Arc<Mutex<VoxelCache<4>>>,
+    cache_interface: Arc<CacheProducerInterface<4>>,
     terrain_generator: Arc<TerrainGenerator<4>>,
     chunk_pos_map: Arc<HashMap<u32, (usize, usize, usize)>>,
     instance_buffer: Buffer,
@@ -132,6 +126,8 @@ struct Immediates
     view_proj: Mat4,
     cam_pos: Vec3,
     frame_timestamp: u32,
+    width: u32,
+    downsample_factor: u32,
 }
 
 #[derive(Debug, Clone, Copy, Zeroable, Pod)]
@@ -183,8 +179,10 @@ impl State
 
         let egui_renderer = EguiRenderer::new(&device, surface_format, &window);
 
-        let mut voxel_cache = VoxelCache::<4>::new(200_000, device.clone(), queue.clone());
+        let mut voxel_cache = VoxelCache::<4>::new(100_000, device.clone(), queue.clone());
+        let cache_interface = CacheProducerInterface::new(1024, &device);
 
+        let terrain_generator = TerrainGenerator::<4>::new(5, "vxls_height.tif", 0.2, "img.jpg");
         //let terrain_generator = TerrainGenerator::<4>::new(5, "vxls_height.tif", 0.2, "img.jpg");
         // let terrain_generator = TerrainGenerator::<4>::new(
         //     5,
@@ -192,12 +190,12 @@ impl State
         //     0.2,
         //     "./pointe_percee/ortho.jpg",
         // );
-        let terrain_generator = TerrainGenerator::<4>::new(
-            5,
-            "/home/albin/Documents/vxls_maps/lapiz/height.tif",
-            0.2,
-            "/home/albin/Documents/vxls_maps/lapiz/ortho.jpg",
-        );
+        // let terrain_generator = TerrainGenerator::<4>::new(
+        //     5,
+        //     "/home/albin/Documents/vxls_maps/orgere/height.tif",
+        //     0.2,
+        //     "/home/albin/Documents/vxls_maps/orgere/color.jpg",
+        // );
 
         let mut chunk_pos_map = HashMap::new();
         let chunk_instances = (0..terrain_generator.chunk_width)
@@ -233,16 +231,98 @@ impl State
             ),
         });
 
-        let pipeline_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
-            label: Some("Voxel pipeline layout"),
+        let prepass_pipeline_layout =
+            device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+                label: Some("Voxel pipeline layout"),
 
-            bind_group_layouts: &[Some(&voxel_cache.bind_group_layout())],
-            immediate_size: size_of::<Immediates>() as u32,
+                bind_group_layouts: &[Some(&voxel_cache.bind_group_layout())],
+                immediate_size: size_of::<Immediates>() as u32,
+            });
+
+        let prepass_pipeline = device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
+            label: Some("Render pipeline"),
+            layout: Some(&prepass_pipeline_layout),
+            vertex: wgpu::VertexState {
+                module: &shader_module,
+                entry_point: Some("chunk"),
+                compilation_options: Default::default(),
+                buffers: &[Some(wgpu::VertexBufferLayout {
+                    array_stride: (size_of::<f32>() * 3 + size_of::<u32>()) as u64,
+                    step_mode: wgpu::VertexStepMode::Instance,
+                    attributes: &[
+                        wgpu::VertexAttribute {
+                            format: wgpu::VertexFormat::Float32x3,
+                            offset: 0,
+                            shader_location: 0,
+                        },
+                        wgpu::VertexAttribute {
+                            format: wgpu::VertexFormat::Uint32,
+                            offset: (size_of::<f32>() * 3) as u64,
+                            shader_location: 1,
+                        },
+                    ],
+                })],
+            },
+            primitive: wgpu::PrimitiveState {
+                topology: wgpu::PrimitiveTopology::TriangleList,
+                strip_index_format: None,
+                front_face: wgpu::FrontFace::Ccw,
+                cull_mode: Some(wgpu::Face::Front),
+                unclipped_depth: false,
+                polygon_mode: wgpu::PolygonMode::Fill,
+                conservative: false,
+            },
+            depth_stencil: Some(wgpu::DepthStencilState {
+                format: wgpu::TextureFormat::Depth24PlusStencil8,
+                depth_write_enabled: Some(true),
+                depth_compare: Some(wgpu::CompareFunction::Less),
+                stencil: wgpu::StencilState::default(),
+                bias: wgpu::DepthBiasState::default(),
+            }),
+            multisample: wgpu::MultisampleState::default(),
+            fragment: Some(wgpu::FragmentState {
+                module: &shader_module,
+                entry_point: Some("fragment_prepass"),
+                compilation_options: wgpu::PipelineCompilationOptions::default(),
+                targets: &[Some(wgpu::ColorTargetState {
+                    format: TextureFormat::R32Float,
+                    blend: None,
+                    write_mask: wgpu::ColorWrites::all(),
+                })],
+            }),
+            multiview_mask: None,
+            cache: None,
         });
 
+        let prepass_depth_bind_group_layout =
+            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+                label: Some("prepass_upsample_bind_group_layout "),
+                entries: &[wgpu::BindGroupLayoutEntry {
+                    binding: 0,
+                    visibility: ShaderStages::FRAGMENT,
+                    ty: wgpu::BindingType::Texture {
+                        sample_type: wgpu::TextureSampleType::Float { filterable: false },
+                        view_dimension: wgpu::TextureViewDimension::D2,
+                        multisampled: false,
+                    },
+                    count: None,
+                }],
+            });
+
+        let chunk_pipeline_layout =
+            device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+                label: Some("Voxel pipeline layout"),
+
+                bind_group_layouts: &[
+                    Some(&voxel_cache.bind_group_layout()),
+                    Some(&prepass_depth_bind_group_layout),
+                ],
+                immediate_size: size_of::<Immediates>() as u32,
+            });
+
         let chunk_pipeline = device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
             label: Some("Render pipeline"),
-            layout: Some(&pipeline_layout),
+            layout: Some(&chunk_pipeline_layout),
             vertex: wgpu::VertexState {
                 module: &shader_module,
                 entry_point: Some("chunk"),
@@ -295,6 +375,85 @@ impl State
             cache: None,
         });
 
+        let prepass_upsample_bg_layout =
+            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+                label: Some("prepass_upsample_bg"),
+                entries: &[
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 0,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::StorageTexture {
+                            access: wgpu::StorageTextureAccess::ReadOnly,
+                            format: wgpu::TextureFormat::R32Float,
+                            view_dimension: wgpu::TextureViewDimension::D2,
+                        },
+                        count: None,
+                    },
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 1,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::StorageTexture {
+                            access: wgpu::StorageTextureAccess::WriteOnly,
+                            format: wgpu::TextureFormat::R32Float,
+                            view_dimension: wgpu::TextureViewDimension::D2,
+                        },
+                        count: None,
+                    },
+                ],
+            });
+
+        let prepass_upsample_layout =
+            device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+                label: Some("prepass_upsample_layout"),
+                bind_group_layouts: &[Some(&prepass_upsample_bg_layout)],
+                immediate_size: 0,
+            });
+
+        let prepass_downsampling = 4;
+        let prepass_upsample = device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+            label: Some("prepass_upsample"),
+            layout: Some(&prepass_upsample_layout),
+            module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                label: Some("prepass_upsample_module"),
+                source: wgpu::ShaderSource::Wgsl(
+                    format!(
+                        "
+                        @group(0) @binding(0) var input_tex: texture_storage_2d<r32float, read>;
+                        @group(0) @binding(1) var output_tex: texture_storage_2d<r32float, write>;
+                        @compute @workgroup_size(8, 8)
+                        fn main(@builtin(global_invocation_id) id: vec3<u32>) 
+                        {{
+                            let source_pos = vec2<i32>(id.xy);
+                            var depth = textureLoad(input_tex, source_pos + vec2<i32>(0, 0)).x;
+                            depth = min(depth, textureLoad(input_tex, source_pos + vec2<i32>(1, 0)).x);
+                            depth = min(depth, textureLoad(input_tex, source_pos + vec2<i32>(0, 1)).x);
+                            depth = min(depth, textureLoad(input_tex, source_pos + vec2<i32>(1, 1)).x);
+
+                            let dest_pos = source_pos * {prepass_downsampling};
+
+                            for(var ox = 0; ox < {prepass_downsampling}; ox ++)
+                            {{
+                                for(var oy = 0; oy < {prepass_downsampling}; oy ++)
+                                {{
+                                     textureStore(
+                                         output_tex,
+                                         dest_pos + vec2<i32>(ox, oy),
+                                         vec4<f32>(depth)
+                                     );
+                                }}
+                            }}
+
+                        }}
+                    "
+                    )
+                    .into(),
+                ),
+            }),
+            entry_point: Some("main"),
+            compilation_options: PipelineCompilationOptions::default(),
+            cache: None,
+        });
+
         let state = State {
             instance,
             window,
@@ -303,12 +462,33 @@ impl State
             surface_format,
             egui_renderer,
             depth_buffer: Self::create_depth_buffer(&device, size.width, size.height),
+            prepass_depth_buffer: Self::create_depth_buffer(
+                &device,
+                size.width / prepass_downsampling,
+                size.height / prepass_downsampling,
+            ),
+            prepass_depth: Self::create_prepass_depth_buffer(
+                &device,
+                size.width / prepass_downsampling,
+                size.height / prepass_downsampling,
+            ),
+            upsampled_prepass_depth: Self::create_prepass_depth_buffer(
+                &device,
+                size.width,
+                size.height,
+            ),
+            prepass_downsampling,
+            upsample_pipeline: prepass_upsample,
             queue,
             device,
             usage_vec: Arc::new(Mutex::new(vec![])),
             pipeline: chunk_pipeline,
+            prepass_pipeline,
+            prepass_upsample_bg_layout,
+            prepass_depth_bind_group_layout,
             voxel_cache: Arc::new(Mutex::new(voxel_cache)),
-            insertion_debounce: false,
+            cache_interface: cache_interface.into(),
+            insertion_debounce: true,
             camera: Default::default(),
             instance_buffer,
             instance_count,
@@ -330,6 +510,19 @@ impl State
 
     fn handle_event(&mut self, event: &WindowEvent)
     {
+        if let WindowEvent::KeyboardInput { event, .. } = event
+        {
+            match (event.state, event.physical_key)
+            {
+                (
+                    winit::event::ElementState::Pressed,
+                    winit::keyboard::PhysicalKey::Code(winit::keyboard::KeyCode::KeyF),
+                ) => self.insertion_debounce = !self.insertion_debounce,
+                _ =>
+                {}
+            }
+        }
+
         self.egui_renderer.handle_input(&self.window, event);
         self.camera.handle_input(event);
     }
@@ -349,6 +542,35 @@ impl State
         }
     }
 
+    fn create_prepass_depth_buffer(
+        device: &Device,
+        width: u32,
+        height: u32,
+    ) -> (Texture, TextureView)
+    {
+        let texture = device.create_texture(&wgpu::wgt::TextureDescriptor {
+            label: Some("Prepass Depth buffer"),
+            size: Extent3d {
+                width,
+                height,
+                depth_or_array_layers: 1,
+            },
+            mip_level_count: 1,
+            sample_count: 1,
+            dimension: wgpu::TextureDimension::D2,
+            format: wgpu::TextureFormat::R32Float,
+            usage: TextureUsages::RENDER_ATTACHMENT
+                | TextureUsages::TEXTURE_BINDING
+                | TextureUsages::STORAGE_BINDING,
+            view_formats: &[wgpu::TextureFormat::R32Float],
+        });
+        let texture_view = texture.create_view(&wgpu::wgt::TextureViewDescriptor {
+            label: Some("prepass depth view"),
+            ..Default::default()
+        });
+        (texture, texture_view)
+    }
+
     fn create_depth_buffer(device: &Device, width: u32, height: u32) -> (Texture, TextureView)
     {
         let texture = device.create_texture(&wgpu::wgt::TextureDescriptor {
@@ -398,6 +620,18 @@ impl State
         self.configure_surface();
         self.depth_buffer =
             Self::create_depth_buffer(&self.device, new_size.width, new_size.height);
+        self.prepass_depth_buffer = Self::create_depth_buffer(
+            &self.device,
+            new_size.width / self.prepass_downsampling,
+            new_size.height / self.prepass_downsampling,
+        );
+        self.prepass_depth = Self::create_prepass_depth_buffer(
+            &self.device,
+            new_size.width / self.prepass_downsampling,
+            new_size.height / self.prepass_downsampling,
+        );
+        self.upsampled_prepass_depth =
+            Self::create_prepass_depth_buffer(&self.device, new_size.width, new_size.height);
     }
 
     fn render(&mut self)
@@ -405,9 +639,47 @@ impl State
         self.camera.update();
         self.voxel_cache.lock().next_frame();
 
+        // Build sorted buffer
+        let mut chunks = self.chunk_pos_map.iter().collect::<Vec<_>>();
+
+        chunks.sort_by_key(|(_, (x, y, z))| {
+            OrderedFloat(
+                (Vec3::new(*x as f32, *y as f32, *z as f32) - self.camera.position).length(),
+            )
+        });
+
+        let instance_buffer = self
+            .device
+            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
+                label: Some("Instance buffer"),
+                contents: bytemuck::cast_slice(
+                    chunks
+                        .iter()
+                        .map(|(structure_id, (x, y, z))| InstanceAttribute {
+                            x: *x as f32,
+                            y: *y as f32,
+                            z: *z as f32,
+                            id: **structure_id,
+                        })
+                        .collect::<Vec<_>>()
+                        .as_slice(),
+                ),
+                usage: BufferUsages::COPY_DST | BufferUsages::VERTEX,
+            });
+
+        let prepass_depth_bind_group = self.device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("prepass_depth_bind_group"),
+            layout: &self.prepass_depth_bind_group_layout,
+            entries: &[wgpu::BindGroupEntry {
+                binding: 0,
+                resource: wgpu::BindingResource::TextureView(&self.upsampled_prepass_depth.1),
+            }],
+        });
+
         // Create texture view.
         // NOTE: We must handle Timeout because the surface may be unavailable
         // (e.g., when the window is occluded on macOS).
+        // ~~ Texture view creation ~~
         let surface_texture = match self.surface.get_current_texture()
         {
             wgpu::CurrentSurfaceTexture::Success(texture) => texture,
@@ -435,6 +707,16 @@ impl State
             }
         };
 
+        let texture_view = surface_texture
+            .texture
+            .create_view(&wgpu::TextureViewDescriptor {
+                // Without add_srgb_suffix() the image we will be working with
+                // might not be "gamma correct".
+                format: Some(self.surface_format.add_srgb_suffix()),
+                ..Default::default()
+            });
+
+        // ~~ Ray-marching timestamp query setup ~~
         let timestamp_query = self.device.create_query_set(&wgpu::QuerySetDescriptor {
             label: Some("timestamp_query_set"),
             ty: wgpu::QueryType::Timestamp,
@@ -448,18 +730,88 @@ impl State
             mapped_at_creation: false,
         });
 
-        let texture_view = surface_texture
-            .texture
-            .create_view(&wgpu::TextureViewDescriptor {
-                // Without add_srgb_suffix() the image we will be working with
-                // might not be "gamma correct".
-                format: Some(self.surface_format.add_srgb_suffix()),
-                ..Default::default()
+        // ~~ Prepass upsample bind group ~~
+        let prepass_bg = self.device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("prepass_bind_group"),
+            layout: &self.prepass_upsample_bg_layout,
+            entries: &[
+                wgpu::BindGroupEntry {
+                    binding: 0,
+                    resource: wgpu::BindingResource::TextureView(&self.prepass_depth.1),
+                },
+                wgpu::BindGroupEntry {
+                    binding: 1,
+                    resource: wgpu::BindingResource::TextureView(&self.upsampled_prepass_depth.1),
+                },
+            ],
+        });
+
+        let mut encoder = self.device.create_command_encoder(&Default::default());
+        // ~~ Prepass ~~
+        {
+            let mut renderpass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
+                label: None,
+                color_attachments: &[Some(wgpu::RenderPassColorAttachment {
+                    view: &self.prepass_depth.1,
+                    depth_slice: None,
+                    resolve_target: None,
+                    ops: Operations {
+                        load: wgpu::LoadOp::Clear(Color {
+                            r: -1.,
+                            g: 0.,
+                            b: 0.,
+                            a: 0.,
+                        }),
+                        store: wgpu::StoreOp::Store,
+                    },
+                })],
+                depth_stencil_attachment: Some(wgpu::RenderPassDepthStencilAttachment {
+                    view: &self.prepass_depth_buffer.1,
+                    depth_ops: Some(Operations {
+                        load: wgpu::LoadOp::Clear(1.),
+                        store: wgpu::StoreOp::Discard,
+                    }),
+                    stencil_ops: None,
+                }),
+                timestamp_writes: None,
+                occlusion_query_set: None,
+                multiview_mask: None,
+            });
+            renderpass.set_vertex_buffer(0, instance_buffer.slice(..));
+            renderpass.set_bind_group(0, Some(&self.voxel_cache.lock().bind_group()), &[]);
+            let imm = [Immediates {
+                view_proj: self.camera.view_proj(),
+                cam_pos: self.camera.position,
+                frame_timestamp: self.voxel_cache.lock().current_timestamp(),
+                width: self.size.width / self.prepass_downsampling,
+                downsample_factor: self.prepass_downsampling,
+            }];
+
+            renderpass.set_pipeline(&self.prepass_pipeline);
+            renderpass.set_immediates(0, unsafe { as_raw_bytes(&imm) });
+            renderpass.draw(0..36, 0..(self.instance_count as u32));
+
+            // End the renderpass.
+            drop(renderpass);
+        }
+
+        // ~~ Upsample pass ~~
+        {
+            let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+                label: Some("upsample_compute_pass"),
+                timestamp_writes: None,
             });
 
-        // Renders a GREEN screen
-        let mut encoder = self.device.create_command_encoder(&Default::default());
+            compute_pass.set_bind_group(0, Some(&prepass_bg), &[]);
+            compute_pass.set_pipeline(&self.upsample_pipeline);
+            compute_pass.dispatch_workgroups(
+                (self.size.width / self.prepass_downsampling).div_ceil(8),
+                (self.size.height / self.prepass_downsampling).div_ceil(8),
+                1,
+            );
+        }
 
+        // ~~ Main render pass ~~
         {
             let mut renderpass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
                 label: None,
@@ -468,7 +820,12 @@ impl State
                     depth_slice: None,
                     resolve_target: None,
                     ops: wgpu::Operations {
-                        load: wgpu::LoadOp::Clear(wgpu::Color::BLACK),
+                        load: wgpu::LoadOp::Clear(wgpu::Color {
+                            r: 0.,
+                            g: 2. / 255.,
+                            b: 15. / 255.,
+                            a: 1.,
+                        }),
                         store: wgpu::StoreOp::Store,
                     },
                 })],
@@ -488,15 +845,18 @@ impl State
                 occlusion_query_set: None,
                 multiview_mask: None,
             });
-
-            renderpass.set_pipeline(&self.pipeline);
-            renderpass.set_vertex_buffer(0, self.instance_buffer.slice(..));
+            renderpass.set_vertex_buffer(0, instance_buffer.slice(..));
             renderpass.set_bind_group(0, Some(&self.voxel_cache.lock().bind_group()), &[]);
+            renderpass.set_bind_group(1, Some(&prepass_depth_bind_group), &[]);
             let imm = [Immediates {
                 view_proj: self.camera.view_proj(),
                 cam_pos: self.camera.position,
                 frame_timestamp: self.voxel_cache.lock().current_timestamp(),
+                width: self.size.width,
+                downsample_factor: 1,
             }];
+
+            renderpass.set_pipeline(&self.pipeline);
             renderpass.set_immediates(0, unsafe { as_raw_bytes(&imm) });
             renderpass.draw(0..36, 0..(self.instance_count as u32));
 
@@ -506,33 +866,13 @@ impl State
             encoder.resolve_query_set(&timestamp_query, 0..2, &timestamp_buffer, 0);
         }
 
-        let requests = self.device.create_buffer(&wgpu::BufferDescriptor {
-            label: Some("dummy_dumb_dinky_aaaahhh_buffer"),
-            size: 16 * 2000,
-            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
-            mapped_at_creation: false,
-        });
-
-        if !self
-            .camera
-            .pressed_keyset
-            .contains(&winit::keyboard::KeyCode::KeyF)
-        {
-            self.voxel_cache
-                .lock()
-                .cache_post_render(&mut encoder, requests.clone());
-        }
-
-        // If you wanted to call any drawing commands, they would go here.
+        // ~~ EGUI Render pass ~~
         {
             self.egui_renderer.begin_frame(&self.window);
-            egui::Window::new("Window ! ").resizable(true).show(
+            egui::Window::new("Window ! ").resizable(false).show(
                 self.egui_renderer.context(),
                 |ui| {
-                    if self
-                        .camera
-                        .pressed_keyset
-                        .contains(&winit::keyboard::KeyCode::KeyF)
+                    if !self.insertion_debounce
                     {
                         ui.label(
                             egui::RichText::new("Cache paused")
@@ -575,7 +915,7 @@ impl State
             );
         }
 
-        // Report usage
+        // ~~ Build usage buffer histogram
         let time_stamp = self.voxel_cache.lock().current_timestamp();
         let cloned_usage_histogram = self.usage_vec.clone();
         DownloadBuffer::read_buffer(
@@ -595,27 +935,24 @@ impl State
             },
         );
 
-        // Submit the command in the queue to execute
+        // ~~ Do cache managment
+        // if !self
+        //     .camera
+        //     .pressed_keyset
+        //     .contains(&winit::keyboard::KeyCode::KeyF)
+        if self.insertion_debounce
+        {
+            self.voxel_cache
+                .lock()
+                .cache_post_render(&mut encoder, &self.cache_interface);
+        }
+
+        // ~~ Submit command buffer ~~
         self.queue.submit([encoder.finish()]);
         self.window.pre_present_notify();
         self.queue.present(surface_texture);
-        if !self
-            .camera
-            .pressed_keyset
-            .contains(&winit::keyboard::KeyCode::KeyF)
-            && self.insertion_debounce
-        {
-            self.insertion_debounce = false;
-        }
 
-        // if (self
-        //     .camera
-        //     .pressed_keyset
-        //     .contains(&winit::keyboard::KeyCode::KeyF))
-        //     && !self.insertion_debounce
-        // {
-
-        // Report frame time
+        // ~~ Get Ray-marching timestamps, report time ~~
         let cloned_rm_time = self.rm_time.clone();
         let cloned_queue = self.queue.clone();
         DownloadBuffer::read_buffer(
@@ -630,30 +967,36 @@ impl State
             },
         );
 
-        if !self
-            .camera
-            .pressed_keyset
-            .contains(&winit::keyboard::KeyCode::KeyF)
+        // ~~ Do cache managment
+        // if !self
+        //     .camera
+        //     .pressed_keyset
+        //     .contains(&winit::keyboard::KeyCode::KeyF)
+        if self.insertion_debounce
         {
-            self.insertion_debounce = true;
-            let request_count = self.voxel_cache.lock().total_request_count();
+            let request_count = self
+                .cache_interface
+                .total_request_count(&self.device, &self.queue);
             let cloned_cache = self.voxel_cache.clone();
             let cloned_device = self.device.clone();
             let cloned_queue = self.queue.clone();
             let cloned_generator = self.terrain_generator.clone();
             let cloned_map = self.chunk_pos_map.clone();
+            let cloned_cache_interface = self.cache_interface.clone();
             let (tx, rx) = sync_channel(1);
+
+            // ~~ Download request buffer, fullfill requests, writeback to cache ~~
             DownloadBuffer::read_buffer(
                 &self.device.clone(),
                 &self.queue.clone(),
-                &requests.slice(0..),
+                &self.cache_interface.requests_buffer().slice(..),
                 move |buffer| {
                     if tx.try_send(()).is_err()
                     {
                         return;
                     }
 
-                    let cache_node_requests: Vec<CacheNodeRequest> =
+                    let cache_node_requests: Vec<CacheRequest> =
                         bytemuck::pod_collect_to_vec(&buffer.unwrap());
 
                     let generator = cloned_generator;
@@ -662,7 +1005,6 @@ impl State
                     let mut structure_nodes = vec![];
                     let mut color_nodes: Vec<[ColorBytes; 64]> = vec![];
                     let mut location_nodes = vec![];
-                    let mut destinations = vec![];
 
                     cache_node_requests
                         .par_iter()
@@ -674,7 +1016,7 @@ impl State
 
                             let location;
                             let node;
-                            if request.child_index == u32::MAX
+                            if request.locator == 0
                             {
                                 // Produce root node
                                 node =
@@ -688,9 +1030,8 @@ impl State
                             else
                             {
                                 // Figure out depth of request
-                                let locator = NTreeNodeLocator::<4>::from_usize(
-                                    request.structure_locator as usize,
-                                );
+                                let locator =
+                                    NTreeNodeLocator::<4>::from_usize(request.locator as usize);
                                 let depth = locator.depth();
                                 //let (x, y, z) = locator.node_location();
 
@@ -720,15 +1061,7 @@ impl State
                                 };
                             }
 
-                            (
-                                node.structure,
-                                node.colors,
-                                location,
-                                DestinationElement {
-                                    node: request.node_index,
-                                    child: request.child_index,
-                                },
-                            )
+                            (node.structure, node.colors, location)
                         })
                         .collect::<Vec<_>>()
                         .into_iter()
@@ -736,46 +1069,30 @@ impl State
                             structure_nodes.push(data.0);
                             color_nodes.push(std::array::from_fn(|i| data.1[i].into()));
                             location_nodes.push(data.2);
-                            destinations.push(data.3);
                         });
 
                     if structure_nodes.len() != 0
                     {
-                        let structre_node_buffer =
-                            cloned_device.create_buffer_init(&BufferInitDescriptor {
-                                label: None,
-                                contents: unsafe { as_raw_bytes(structure_nodes.as_slice()) },
-                                usage: BufferUsages::STORAGE,
-                            });
-                        let color_node_buffer =
-                            cloned_device.create_buffer_init(&BufferInitDescriptor {
-                                label: None,
-                                contents: unsafe { as_raw_bytes(color_nodes.as_slice()) },
-                                usage: BufferUsages::STORAGE,
-                            });
-                        let location_node_buffer =
-                            cloned_device.create_buffer_init(&BufferInitDescriptor {
-                                label: None,
-                                contents: unsafe { as_raw_bytes(location_nodes.as_slice()) },
-                                usage: BufferUsages::STORAGE,
-                            });
-                        let destinations_buffer =
-                            cloned_device.create_buffer_init(&BufferInitDescriptor {
-                                label: None,
-                                contents: unsafe { as_raw_bytes(destinations.as_slice()) },
-                                usage: BufferUsages::STORAGE,
-                            });
+                        cloned_queue.write_buffer(
+                            cloned_cache_interface.structure_nodes_buffer(),
+                            0,
+                            unsafe { as_raw_bytes(structure_nodes.as_slice()) },
+                        );
+                        cloned_queue.write_buffer(
+                            cloned_cache_interface.color_nodes_buffer(),
+                            0,
+                            unsafe { as_raw_bytes(&color_nodes.as_slice()) },
+                        );
+                        cloned_queue.write_buffer(
+                            cloned_cache_interface.location_nodes_buffer(),
+                            0,
+                            unsafe { as_raw_bytes(&&location_nodes.as_slice()) },
+                        );
 
                         let mut encoder = cloned_device.create_command_encoder(&Default::default());
-                        cloned_cache.lock().cache_insert(
-                            &mut encoder,
-                            &CacheResponse {
-                                structure_nodes: structre_node_buffer,
-                                color_nodes: color_node_buffer,
-                                locations: location_node_buffer,
-                                parents: destinations_buffer,
-                            },
-                        );
+                        cloned_cache
+                            .lock()
+                            .cache_insert(&mut encoder, &cloned_cache_interface);
                         cloned_queue.submit([encoder.finish()]);
                     }
                 },

src/producers.rs

@@ -2,12 +2,13 @@ use std::fs::File;
 use std::path::Path;
 
 use glam::Vec3;
-use indicatif::ProgressIterator;
 use itertools::Itertools;
+use rayon::iter::IntoParallelRefMutIterator;
+use rayon::iter::ParallelIterator;
 
-use crate::voxel_cache::gpu::ExplicitNTreeNode;
-use crate::voxel_cache::gpu::StructurePointer;
-use crate::voxel_cache::sparse::Color;
+use crate::sparse_tree::Color;
+use crate::voxel_cache::data::ExplicitNTreeNode;
+use crate::voxel_cache::data::StructurePointer;
 
 pub struct BallGenerator<const N: usize>
 {
@@ -311,7 +312,7 @@ where
         let heightmap_max = heightmap.iter().copied().reduce(f32::max).unwrap();
 
         heightmap
-            .iter_mut()
+            .par_iter_mut()
             .filter(|x| **x == -9999.)
             .for_each(|x| *x = heightmap_min);
 

src/sparse_tree.rs

@@ -5,8 +5,8 @@ use bytemuck::Pod;
 use bytemuck::Zeroable;
 use itertools::Itertools;
 
-use crate::voxel_cache::gpu::ExplicitNTreeNode;
-use crate::voxel_cache::gpu::StructurePointer;
+use crate::voxel_cache::data::ExplicitNTreeNode;
+use crate::voxel_cache::data::StructurePointer;
 
 #[derive(Debug, Clone, Copy, PartialEq, Pod, Zeroable)]
 #[repr(C)]

src/voxel_cache.rs

@@ -1,4 +1,900 @@
-pub mod cache;
-pub mod gpu;
-pub mod pipeline;
-pub mod sparse;
+use bytemuck::Zeroable;
+use wgpu::{BindGroup, BindGroupLayout, Buffer, BufferUsages, CommandEncoder, ComputePipeline, Device, Queue, ShaderStages};
+
+use crate::voxel_cache::{data::{CacheNodeRequest, CacheResponse, ColorPoolElement, LocationPoolElement, StructurePoolElement}, producer_interface::CacheProducerInterface, request_buffer::RequestBuffer, structure_table::StructureTable, usage_buffer::UsageBuffer};
+
+
+pub mod request_buffer;
+pub mod structure_table;
+pub mod usage_buffer;
+pub mod producer_interface;
+
+pub mod data;
+
+pub struct VoxelCache<const N: usize>
+{
+    size: usize,
+    device: Device,
+    queue: Queue,
+    structure_pool: Buffer,
+    color_pool: Buffer,
+    location_pool: Buffer,
+
+    pub structure_table: StructureTable,
+
+    request_buffer: RequestBuffer<N>,
+
+    voxel_cache_bind_group: BindGroup,
+    structure_table_bind_group_layout: BindGroupLayout,
+    voxel_cache_render_bind_group_layout: BindGroupLayout,
+
+    // Writing requests
+    request_write_pipeline: ComputePipeline,
+
+    // User response -> caching
+    caching_pipeline: ComputePipeline,
+
+    // Invalidation
+    invalidation_pipeline: ComputePipeline,
+
+    pub usage_buffer: UsageBuffer,
+}
+
+
+#[derive(Zeroable, bytemuck::Pod, Clone, Copy)]
+#[repr(C)]
+struct CachingPipelineImmediates
+{
+    frame_timestamp: u32,
+    write_pointers: u32, // Boolean
+}
+
+unsafe fn as_raw_bytes<T: Sized>(slice: &[T]) -> &[u8]
+{
+    let size = slice.len() * size_of::<T>();
+    unsafe { std::slice::from_raw_parts(slice.as_ptr() as *const u8, size) }
+}
+
+impl<const N: usize> VoxelCache<N>
+where
+    [(); N * N * N]:,
+{
+    pub fn new(cache_size: usize, device: Device, queue: Queue) -> Self
+    {
+        let structure_pool = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("Structure pool"),
+            size: (size_of::<StructurePoolElement<N>>() * cache_size) as u64,
+            usage: BufferUsages::STORAGE,
+            mapped_at_creation: false,
+        });
+
+        let location_pool = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("Location pool"),
+            size: (size_of::<LocationPoolElement>() * cache_size) as u64,
+            usage: BufferUsages::STORAGE,
+            mapped_at_creation: false,
+        });
+
+        let color_pool = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("Color pool"),
+            size: (size_of::<ColorPoolElement<N>>() * cache_size) as u64,
+            usage: BufferUsages::STORAGE,
+            mapped_at_creation: false,
+        });
+
+        let request_buffer = RequestBuffer::new(cache_size, device.clone());
+        let usage_buffer = UsageBuffer::new(cache_size, device.clone());
+
+        let request_interface_bind_group_layout = CacheProducerInterface::<N>::request_side_bind_group_layout(&device);
+
+        // Rendering bing group layouts
+        let voxel_cache_render_bind_group_layout =
+            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+                label: Some("voxel_cache_render_bind_group_layout"),
+                entries: &[
+                    // Structure pool
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 0,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Color pool
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 1,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Location pool
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 2,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Request buffer
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 3,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Usage buffer
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 4,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+
+                    // Structure table stuff
+                    // Structure table pointers
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 5,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+
+                    // Structure table request buffer
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 6,
+                        visibility: ShaderStages::FRAGMENT,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                ],
+            });
+
+        let voxel_cache_bind_group_layout =
+            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+                label: Some("voxel_cache_bind_group_layout"),
+                entries: &[
+                    // Structure pool
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 0,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Color pool
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 1,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Location pool
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 2,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Sorted requests
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 3,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: true },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // LRU List
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 4,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: true },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Usage buffer
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 5,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // pools request count
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 6,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: true },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                ],
+            });
+
+        let voxel_cache_bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("voxel_cache_bind_group"),
+            layout: &voxel_cache_bind_group_layout,
+            entries: &[
+                // Structure pool
+                wgpu::BindGroupEntry {
+                    binding: 0,
+                    resource: structure_pool.as_entire_binding(),
+                },
+                // Color pool
+                wgpu::BindGroupEntry {
+                    binding: 1,
+                    resource: color_pool.as_entire_binding(),
+                },
+                // Location pool
+                wgpu::BindGroupEntry {
+                    binding: 2,
+                    resource: location_pool.as_entire_binding(),
+                },
+                // Sorted requests
+                wgpu::BindGroupEntry {
+                    binding: 3,
+                    resource: request_buffer.sort_buffer().as_entire_binding(),
+                },
+                // LRU list
+                wgpu::BindGroupEntry {
+                    binding: 4,
+                    resource: usage_buffer.sort_buffer().as_entire_binding(),
+                },
+                // Usage buffer
+                wgpu::BindGroupEntry {
+                    binding: 5,
+                    resource: usage_buffer.usage_buffer().as_entire_binding(),
+                },
+                // Pools request count
+                wgpu::BindGroupEntry {
+                    binding: 6,
+                    resource: request_buffer.request_count_buffer().as_entire_binding(),
+                },
+            ],
+        });
+
+        let structure_table_bind_group_layout =
+            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+                label: Some("structure_table_bind_group_layout"),
+                entries: &[
+                    // Pointer table
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 0,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Sorted requests
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 1,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Requests count
+                    wgpu::BindGroupLayoutEntry {
+                        binding: 2,
+                        visibility: ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                ],
+            });
+
+        let children_count = N*N*N;
+        let wgsl_bindings = 
+            format!("
+                struct StructurePoolElement
+                {{
+                     pointers: array<u32, {children_count}>
+                }}
+
+                struct ColorPoolElement
+                {{
+                      colors: array<u32, {children_count}>
+                }}
+
+                struct LocationPoolElement
+                {{
+                     structure_id: u32,
+                     structure_locator: u32
+                }}
+                
+                struct SortedRequestsElement
+                {{
+                     node: u32,
+                     child: u32
+                }}
+
+                @group(0) @binding(0) var<storage, read_write> structure_pool: array<StructurePoolElement>;
+                @group(0) @binding(1) var<storage, read_write> color_pool: array<ColorPoolElement>;
+                @group(0) @binding(2) var<storage, read_write> location_pool: array<LocationPoolElement>;
+                @group(0) @binding(3) var<storage, read> sorted_requests: array<SortedRequestsElement>;
+                @group(0) @binding(4) var<storage, read> lru_list: array<u32>;
+                @group(0) @binding(5) var<storage, read_write> usage_buffer: array<u32>;
+                @group(0) @binding(6) var<storage, read> pools_request_count: u32;
+
+                @group(1) @binding(0) var<storage, read_write> structure_table_pointers: array<u32>;
+                @group(1) @binding(1) var<storage, read_write> structure_table_sorted_requests: array<SortedRequestsElement>;
+                @group(1) @binding(2) var<storage, read_write> structure_table_request_count: u32;
+            ");
+
+        let write_requests_pipeline_layout =
+            device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+                label: Some("write_requests_pipeline_layout"),
+                bind_group_layouts: &[
+                    Some(&voxel_cache_bind_group_layout),
+                    Some(&structure_table_bind_group_layout),
+                    Some(&request_interface_bind_group_layout),
+                ],
+                immediate_size: 0,
+            });
+
+        // The user bindgroup will be created on the fly
+        let write_requests =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("write_requests_pipeline"),
+                layout: Some(&write_requests_pipeline_layout),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("write_requests_pipeline_shader_module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+                            {wgsl_bindings}
+
+                            struct CacheInterfaceRequest
+                            {{
+                                structure_id: u32,
+                                locator: u32,
+                                child_index: u32
+                            }}
+
+                            struct CacheInterfaceRequestWb
+                            {{
+                                node_index: u32,
+                                child_index: u32
+                            }}
+
+                            @group(2) @binding(0) var<storage, read_write> requests: array<CacheInterfaceRequest>; 
+                            @group(2) @binding(1) var<storage, read_write> requests_wb: array<CacheInterfaceRequestWb>; 
+                            @group(2) @binding(2) var<storage, read_write> request_count: u32; 
+
+                            @group(2) @binding(3) var<storage, read> structure_nodes: array<StructurePoolElement>; 
+                            @group(2) @binding(4) var<storage, read> color_nodes: array<ColorPoolElement>; 
+                            @group(2) @binding(5) var<storage, read> locations: array<LocationPoolElement>; 
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+                                 // One shader invocation per invocation on both structure table domain
+                                 // and cache domain
+                                 var index = global_invocation_id.x;
+                                 let max_requests_count = min(arrayLength(&requests), arrayLength(&lru_list));
+                                 request_count = min(max_requests_count, pools_request_count + structure_table_request_count);
+                                 if(index >= request_count)
+                                 {{
+                                      return;
+                                 }}
+
+                                 if(index < structure_table_request_count)
+                                 {{
+                                     var request: CacheInterfaceRequest;
+                                     // Index in structure table IS structure id
+                                     request.structure_id = structure_table_sorted_requests[index].node;
+                                     request.locator = 0; // Root request -> locator 0
+                                     request.child_index = 0xFFFFFFFF; // child index u32::MAX ->
+
+                                     var request_wb: CacheInterfaceRequestWb;
+                                     request_wb.node_index = request.structure_id;
+                                     request_wb.child_index = 0xFFFFFFFF; // child index u32::MAX ->
+                                                                       // Structure table request
+                                     // Write request
+                                     requests[index] = request;
+                                     requests_wb[index] = request_wb;
+                                     return;
+                                 }}
+
+                                 let pool_index = index - structure_table_request_count;
+                                     
+                                 // Request location comes from location pool
+                                 var request: CacheInterfaceRequest;
+                                 request.structure_id = location_pool[sorted_requests[pool_index].node].structure_id;
+                                 request.locator = location_pool[sorted_requests[pool_index].node].structure_locator;
+                                 request.child_index = sorted_requests[pool_index].child;
+
+                                 var request_wb: CacheInterfaceRequestWb;
+                                 request_wb.node_index = sorted_requests[pool_index].node;
+                                 request_wb.child_index = sorted_requests[pool_index].child;
+
+                                 // Write request
+                                 requests[index] = request;
+                                 requests_wb[index] = request_wb;
+
+                            }}
+                        "
+                        )
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        // ~~~ Caching pipeline ~~~
+        // Pipeline that takes user fulling ~some~ requests
+        // and writes them to the cache based on the eviction list
+
+        let caching_pipeline_layout =
+            device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+                label: Some("caching_pipelien_layout"),
+                bind_group_layouts: &[
+                    Some(&voxel_cache_bind_group_layout),
+                    Some(&structure_table_bind_group_layout),
+                    Some(&request_interface_bind_group_layout),
+                ],
+                immediate_size: size_of::<CachingPipelineImmediates>() as u32, // Current frame timestamp
+            });
+
+        let children_count = N * N * N;
+        let caching_pipeline =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("caching_pipeline"),
+                layout: Some(&caching_pipeline_layout),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("caching_pipeline_shader_module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+
+                            struct DestinationElement
+                            {{
+                                node: u32,
+                                child: u32
+                            }}
+
+                            struct CachingPipelineImmediates
+                            {{
+                                frame_timestamp: u32,
+                                write_pointers: u32
+                            }}
+
+                            {wgsl_bindings}
+                            
+                            var<immediate> parameters: CachingPipelineImmediates;
+
+                            struct CacheInterfaceRequest
+                            {{
+                                structure_id: u32,
+                                locator: u32,
+                                child_index: u32
+                            }}
+
+                            struct CacheInterfaceRequestWb
+                            {{
+                                node_index: u32,
+                                child_index: u32
+                            }}
+
+                            @group(2) @binding(0) var<storage, read_write> requests: array<CacheInterfaceRequest>; 
+                            @group(2) @binding(1) var<storage, read_write> requests_wb: array<CacheInterfaceRequestWb>; 
+                            @group(2) @binding(2) var<storage, read_write> request_count: u32; 
+
+                            @group(2) @binding(3) var<storage, read_write> structure_nodes: array<StructurePoolElement>; 
+                            @group(2) @binding(4) var<storage, read_write> color_nodes: array<ColorPoolElement>; 
+                            @group(2) @binding(5) var<storage, read_write> locations: array<LocationPoolElement>; 
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+                                    // Copy with indirection
+                                    let index = global_invocation_id.x;
+                                    let total = arrayLength(&structure_nodes);
+                                    let total_cache = arrayLength(&lru_list);
+                                    if(index >= total || index >= total_cache)
+                                    {{
+                                        return;
+                                    }}
+
+                                    let overwritten_element = lru_list[index];
+                                    if(parameters.write_pointers == 0 && usage_buffer[overwritten_element] != parameters.frame_timestamp)
+                                    {{
+                                        // Phase 1
+                                        // Copy into cache page
+                                        for(var i = 0; i < {children_count}; i++)
+                                        {{
+                                            structure_pool[overwritten_element].pointers[i] = select(u32(0), u32(1)<<31, structure_nodes[index].pointers[i] != 0);
+                                        }}
+                                        color_pool[overwritten_element] = color_nodes[index];
+                                        location_pool[overwritten_element] = locations[index];
+
+                                        // Mark dirty/correct timestamp
+                                        usage_buffer[overwritten_element] = parameters.frame_timestamp + 1;
+                                    }}
+
+                                    if(parameters.write_pointers != 0 && usage_buffer[overwritten_element] != parameters.frame_timestamp)
+                                    {{
+                                        // Phase 2
+                                        
+                                        // Point parent to new page
+                                        let new_pointer = (1 << 31) | (1 << 30) | overwritten_element;
+                                        if(requests_wb[index].child_index == 0xFFFFFFFF)
+                                        {{
+                                            structure_table_pointers[requests_wb[index].node_index] = new_pointer;
+                                        }}else if usage_buffer[requests_wb[index].node_index] != parameters.frame_timestamp + 1
+                                        {{
+                                            structure_pool[requests_wb[index].node_index].pointers[requests_wb[index].child_index] = new_pointer;
+                                        }}
+                                    }}
+
+                            }}
+                        ")
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        // ~~~ Invalidation pipeline ~~~
+        // Pipeline invalidates old nodes that points to ones
+        // that have been replaces base on timestamp information
+        let invalidation_pipeline_layout =
+            device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+                label: Some("invalidation_pipeline_layout"),
+                bind_group_layouts: &[
+                    Some(&voxel_cache_bind_group_layout),
+                    Some(&structure_table_bind_group_layout),
+                ],
+                immediate_size: size_of::<u32>() as u32, // frame_timestamp
+            });
+
+        let invalidation_pipeline =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("invalidation_pipeline"),
+                layout: Some(&invalidation_pipeline_layout),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("invalidation_pipeline_shader_module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+                            {wgsl_bindings}
+                            
+                            var<immediate> frame_timestamp: u32;
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+                                    var index = global_invocation_id.x;
+                                    let total_pool = arrayLength(&structure_pool);
+                                    let total_table = arrayLength(&structure_table_pointers);
+
+                                    if index < total_pool
+                                    {{
+                                         for(var i = 0; i < {children_count}; i += 1)
+                                         {{
+                                            let structure_pointer = structure_pool[index].pointers[i];
+                                            let pointer = structure_pointer & 0x3FFFFFFF;
+                                            let pointer_subdiv = ((structure_pointer >> 31) & 1) != 0;
+                                            let pointed_timestamp = usage_buffer[pointer];
+                                            if(pointed_timestamp == frame_timestamp + 1) // Future
+                                                                                         // timestamp
+                                                                                         // -> new page
+                                            {{
+                                                 // Invalidate
+                                                 structure_pool[index].pointers[i] = select(u32(0), u32(1), pointer_subdiv) << 31;
+                                            }}
+                                         }}
+                                         return;
+                                    }} 
+                                    index -= total_pool;
+                                    if index < total_table
+                                    {{ 
+                                            let structure_pointer = structure_table_pointers[index];
+                                            // Is pointer pointing to something valid
+                                            let pointer_valid = ((structure_pointer >> 30) & 1) != 0;
+                                            let pointer_subdiv = ((structure_pointer >> 31) & 1) != 0;
+                                            let pointer = structure_pointer & 0x3FFFFFFF;
+                                            let pointed_timestamp = usage_buffer[pointer];
+                                            if(pointed_timestamp == frame_timestamp + 1 && pointer_valid && pointer_subdiv) // Future
+                                                                                         // timestamp
+                                                                                         // -> new page
+                                            {{
+                                                 // Invalidate
+                                                 structure_table_pointers[index] &= select(u32(0), u32(1), pointer_subdiv) << 31;
+                                            }}
+                                    }}
+
+                            }}
+                        ")
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        Self {
+            size: cache_size,
+            structure_table: StructureTable::new(device.clone(), queue.clone()),
+            queue,
+            device,
+            structure_pool,
+            color_pool,
+            location_pool,
+            request_buffer,
+            usage_buffer,
+
+            voxel_cache_bind_group,
+            structure_table_bind_group_layout,
+            voxel_cache_render_bind_group_layout,
+
+            // Write requests stage
+            request_write_pipeline: write_requests,
+
+            // Caching operation
+            caching_pipeline,
+
+            // Invalidation
+            invalidation_pipeline,
+        }
+    }
+
+    pub fn bind_group_layout(&self) -> BindGroupLayout
+    {
+        self.voxel_cache_render_bind_group_layout.clone()
+    }
+
+    pub fn bind_group(&self) -> BindGroup
+    {
+        self.device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("voxel_cache_render_bind_group"),
+            layout: &self.voxel_cache_render_bind_group_layout,
+            entries: &[
+                // Structure pool
+                wgpu::BindGroupEntry {
+                    binding: 0,
+                    resource: self.structure_pool.as_entire_binding(),
+                },
+                // Color pool
+                wgpu::BindGroupEntry {
+                    binding: 1,
+                    resource: self.color_pool.as_entire_binding(),
+                },
+                // Location pool
+                wgpu::BindGroupEntry {
+                    binding: 2,
+                    resource: self.location_pool.as_entire_binding(),
+                },
+                // Request buffer 
+                wgpu::BindGroupEntry {
+                    binding: 3,
+                    resource: self.request_buffer.request_buffer().as_entire_binding(),
+                },
+                // Usage buffer
+                wgpu::BindGroupEntry {
+                    binding: 4,
+                    resource: self.usage_buffer.usage_buffer().as_entire_binding(),
+                },
+                // Structure table
+                // Structure table pointers
+                wgpu::BindGroupEntry {
+                    binding: 5,
+                    resource: self.structure_table.pointer_table.as_entire_binding(),
+                },
+                // Pools request count
+                wgpu::BindGroupEntry {
+                    binding: 6,
+                    resource: self.structure_table.request_buffer.request_buffer().as_entire_binding(),
+                },
+            ],
+        })
+    }
+
+    fn structure_table_bind_group(&self) -> BindGroup
+    {
+        self.device.create_bind_group(&wgpu::BindGroupDescriptor
+            {
+                label: Some("structure_table_bind_group"),
+                layout: &self.structure_table_bind_group_layout,
+                entries: 
+                &[
+                    wgpu::BindGroupEntry
+                    {
+                        binding: 0,
+                        resource: self.structure_table.pointer_table.as_entire_binding(),
+                    },
+                    wgpu::BindGroupEntry
+                    {
+                        binding: 1,
+                        resource: self.structure_table.request_buffer.sort_buffer().as_entire_binding(),
+                    },
+                    wgpu::BindGroupEntry
+                    {
+                        binding: 2,
+                        resource: self.structure_table.request_buffer.request_count_buffer().as_entire_binding(),
+                    },
+                ],
+            })
+    }
+
+    pub fn cache_post_render(&mut self, encoder: &mut CommandEncoder, request_interface: &CacheProducerInterface<N>)
+    {
+        // Sort requests, reset request buffers to count requests
+        self.usage_buffer.sort_usage(encoder);
+        self.request_buffer.sort_requests(encoder);
+
+        self.structure_table
+            .request_buffer_mut()
+            .sort_requests(encoder);
+
+        // Sorted requests are now in the group
+        let mut write_requests_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+            label: Some("write_requests_pass"),
+            timestamp_writes: None,
+        });
+
+        write_requests_pass.set_bind_group(0, Some(&self.voxel_cache_bind_group), &[]);
+        write_requests_pass.set_bind_group(1, Some(&self.structure_table_bind_group()), &[]);
+        write_requests_pass.set_bind_group(2, Some(&request_interface.request_side_bind_group), &[]);
+        write_requests_pass.set_pipeline(&self.request_write_pipeline);
+
+        // Compute necessary shader invocations
+        let shader_invocation_count = request_interface.size;
+        let workgroup_invocation_count = shader_invocation_count.div_ceil(64);
+        write_requests_pass.dispatch_workgroups(workgroup_invocation_count as u32, 1, 1);
+        drop(write_requests_pass);
+
+        self.request_buffer.reset_requests(encoder);
+        self.structure_table
+            .request_buffer_mut()
+            .reset_requests(encoder);
+    }
+
+    pub fn current_timestamp(&self) -> u32
+    {
+        self.usage_buffer.timestamp()
+    }
+
+    pub fn cache_insert(&mut self, encoder: &mut CommandEncoder, cache_interface: &CacheProducerInterface<N>)
+    {
+        // Each of the entry of the cache insertion, is matched with the usage list to evict
+        let mut cache_insertion_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+            label: Some("cache_insertion_pass"),
+            timestamp_writes: None,
+        });
+
+        // Copy insertions to evicted lru
+        // Point parents to children
+        let structure_table_bind_group = self.structure_table_bind_group();
+
+        // ~~~ Write data + Dirty flagging/timestamp update ~~~
+        {
+            cache_insertion_pass.set_pipeline(&self.caching_pipeline);
+            cache_insertion_pass.set_bind_group(0, Some(&self.voxel_cache_bind_group), &[]);
+            cache_insertion_pass.set_bind_group(1, Some(&structure_table_bind_group), &[]);
+            cache_insertion_pass.set_bind_group(2, Some(&cache_interface.request_side_bind_group), &[]);
+            cache_insertion_pass.set_immediates(
+                0,
+                bytemuck::bytes_of(&CachingPipelineImmediates {
+                    frame_timestamp: self.usage_buffer.timestamp(),
+                    write_pointers: 0, // false
+                }),
+            );
+
+            // Compute dispatch amounts
+            let shader_invocation_count = cache_interface.size;
+            let workgroup_invocations = shader_invocation_count.div_ceil(64);
+
+            cache_insertion_pass.dispatch_workgroups(workgroup_invocations as u32, 1, 1);
+        }
+
+        // ~~~ Invalidate pointers ~~~
+        {
+            cache_insertion_pass.set_pipeline(&self.invalidation_pipeline);
+            cache_insertion_pass.set_bind_group(2, None, &[]);
+            cache_insertion_pass
+                .set_immediates(0, bytemuck::bytes_of(&self.usage_buffer.timestamp()));
+
+            // Compute dispatch amounts
+            let shader_invocation_count = self.size + self.structure_table.allocation_table.len();
+            let workgroup_invocations = shader_invocation_count.div_ceil(64);
+
+            cache_insertion_pass.dispatch_workgroups(workgroup_invocations as u32, 1, 1);
+        }
+
+        // ~~~ Write new pointers ~~~
+        {
+            cache_insertion_pass.set_pipeline(&self.caching_pipeline);
+            cache_insertion_pass.set_bind_group(2, Some(&cache_interface.request_side_bind_group), &[]);
+            cache_insertion_pass.set_immediates(
+                0,
+                bytemuck::bytes_of(&CachingPipelineImmediates {
+                    frame_timestamp: self.usage_buffer.timestamp(),
+                    write_pointers: 1, // true
+                }),
+            );
+
+            // Compute dispatch amounts
+            let shader_invocation_count = cache_interface.size;
+            let workgroup_invocations = shader_invocation_count.div_ceil(64);
+
+            cache_insertion_pass.dispatch_workgroups(workgroup_invocations as u32, 1, 1);
+        }
+
+    }
+
+    pub fn next_frame(&mut self)
+    {
+        self.usage_buffer.next_frame();
+    }
+
+}

src/voxel_cache/data.rs

@@ -0,0 +1,102 @@
+use bytemuck::Pod;
+use bytemuck::Zeroable;
+use wgpu::Buffer;
+
+use crate::sparse_tree::Color;
+
+#[derive(Clone, Copy, Zeroable, Pod, Debug)]
+#[repr(C)]
+pub struct CacheNodeRequest
+{
+    // Requested ressource
+    pub structure_id: u32,
+    pub structure_locator: u32,
+
+    // Write back info
+    pub node_index: u32,
+    pub child_index: u32,
+}
+
+pub struct RequestBufferElement<const N: usize>
+where
+    [(); N * N * N]:,
+{
+    request_count: [u32; N * N * N],
+}
+
+pub struct StructurePoolElement<const N: usize>
+where
+    [(); N * N * N]:,
+{
+    pointers: [StructurePointer; N * N * N],
+}
+
+pub struct DestinationElement
+{
+    pub node: u32,
+    pub child: u32,
+}
+
+pub struct ColorBytes(pub u8, pub u8, pub u8, pub u8);
+
+impl From<Color> for ColorBytes
+{
+    fn from(value: Color) -> Self
+    {
+        Self(
+            (value.0 * 255.) as u8,
+            (value.1 * 255.) as u8,
+            (value.2 * 255.) as u8,
+            (value.3 * 255.) as u8,
+        )
+    }
+}
+
+pub struct ColorPoolElement<const N: usize>
+where
+    [(); N * N * N]:,
+{
+    colors: [ColorBytes; N * N * N],
+}
+
+pub struct LocationPoolElement
+{
+    pub structure_id: u32,
+    pub structure_locator: u32,
+}
+
+pub struct CacheResponse
+{
+    // Each buffer contains the same amount of elements (structure of arrays style)
+
+    // Cache data to bring in
+    pub structure_nodes: Buffer,
+    pub color_nodes: Buffer,
+    pub locations: Buffer,
+
+    // Which nodes this extends : node_index + child_index
+    pub parents: Buffer,
+}
+
+#[derive(Clone, Copy, Pod, Zeroable)]
+#[repr(transparent)]
+pub struct StructurePointer(pub u32);
+
+impl StructurePointer
+{
+    pub fn new(subdivided: bool, pointer_valid: bool, pointer: u32) -> Self
+    {
+        assert!(pointer >> 30 == 0);
+        StructurePointer((subdivided as u32) << 31 | (pointer_valid as u32) << 30 | pointer)
+    }
+}
+
+#[derive(Clone, Copy, Zeroable)]
+#[repr(C)]
+pub struct ExplicitNTreeNode<const N: usize>
+where
+    [(); N * N * N]:,
+{
+    pub structure: [StructurePointer; N * N * N],
+    pub colors: [Color; N * N * N],
+}

src/voxel_cache/gpu.rs

@@ -1,27 +0,0 @@
-use bytemuck::Pod;
-use bytemuck::Zeroable;
-
-use crate::voxel_cache::sparse::Color;
-
-#[derive(Clone, Copy, Pod, Zeroable)]
-#[repr(transparent)]
-pub struct StructurePointer(pub u32);
-
-impl StructurePointer
-{
-    pub fn new(subdivided: bool, pointer_valid: bool, pointer: u32) -> Self
-    {
-        assert!(pointer >> 30 == 0);
-        StructurePointer((subdivided as u32) << 31 | (pointer_valid as u32) << 30 | pointer)
-    }
-}
-
-#[derive(Clone, Copy, Zeroable)]
-#[repr(C)]
-pub struct ExplicitNTreeNode<const N: usize>
-where
-    [(); N * N * N]:,
-{
-    pub structure: [StructurePointer; N * N * N],
-    pub colors: [Color; N * N * N],
-}

src/voxel_cache/pipeline.rs

@@ -1,68 +0,0 @@
-use std::num::NonZero;
-
-use bytemuck::cast_slice;
-use crevice::std140::AsStd140;
-use glam::Mat4;
-use wgpu::BindGroup;
-use wgpu::BindGroupDescriptor;
-use wgpu::BindGroupEntry;
-use wgpu::BindGroupLayout;
-use wgpu::Buffer;
-use wgpu::BufferUsages;
-use wgpu::CommandEncoder;
-use wgpu::CommandEncoderDescriptor;
-use wgpu::ComputePass;
-use wgpu::ComputePassDescriptor;
-use wgpu::ComputePipeline;
-use wgpu::Device;
-use wgpu::Operations;
-use wgpu::Queue;
-use wgpu::RenderPipeline;
-use wgpu::ShaderModuleDescriptor;
-use wgpu::ShaderStages;
-use wgpu::TextureFormat;
-use wgpu::TextureView;
-use wgpu::VertexBufferLayout;
-use wgpu::util::StagingBelt;
-
-use crate::as_raw_bytes;
-use crate::camera::Camera;
-use crate::voxel_cache::cache::ColorPoolElement;
-use crate::voxel_cache::cache::LocationPoolElement;
-use crate::voxel_cache::cache::RequestBufferElement;
-use crate::voxel_cache::cache::StructurePoolElement;
-use crate::voxel_cache::gpu::StructurePointer;
-use crate::voxel_cache::sparse::Color;
-
-// Represents a chunk to be rendered by the voxel pipeline
-#[derive(Clone, Copy)]
-#[repr(C)]
-pub struct ChunkObject
-{
-    // Chunk object transform
-    pub transform: Mat4,
-
-    // Chunk data
-    pub color: Color,
-    pub subdivided: bool,
-
-    // Producer specific data
-    pub id: u32,
-}
-
-#[derive(Clone, Copy)]
-#[repr(C)]
-pub struct CacheChunkObject
-{
-    // Chunk object transform
-    transform: Mat4,
-
-    // Chunk data
-    color: Color,
-
-    // Producer specific data
-    id: u32,
-
-    // Pointer into cache
-    pointer: StructurePointer,
-}

src/voxel_cache/producer_interface.rs

@@ -0,0 +1,369 @@
+// The cache emits requests, and receives answer to these requests
+// The cache producer interface provides the necessary buffers to contains
+// these, so that use applications can fullfill the requests
+
+use bytemuck::Pod;
+use bytemuck::Zeroable;
+use wgpu::BindGroup;
+use wgpu::BindGroupEntry;
+use wgpu::BindGroupLayout;
+use wgpu::BindGroupLayoutEntry;
+use wgpu::Buffer;
+use wgpu::BufferUsages;
+use wgpu::CommandEncoder;
+use wgpu::Device;
+use wgpu::Queue;
+use wgpu::util::DownloadBuffer;
+
+use crate::voxel_cache::data::ColorPoolElement;
+use crate::voxel_cache::data::LocationPoolElement;
+use crate::voxel_cache::data::StructurePoolElement;
+
+#[derive(Clone, Copy, Zeroable, Pod)]
+#[repr(C)]
+pub struct CacheRequest
+{
+    pub structure_id: u32,
+    pub locator: u32,
+    pub child_index: u32,
+}
+
+pub(super) struct CacheRequestWbInfo
+{
+    node_index: u32,
+    child_index: u32,
+}
+
+pub struct CacheProducerInterface<const N: usize>
+where
+    [(); N * N * N]:,
+{
+    // Request emissions
+    pub(super) size: usize,
+
+    // Contains the specific node being requested
+    pub(super) requests: Buffer,
+
+    // Contains how many request have effectively been emitted
+    pub(super) request_count: Buffer,
+
+    // Contains on which node the request was done
+    pub(super) requests_wb_info: Buffer,
+    // Request fullfillment
+    pub(super) structure_nodes: Buffer,
+    pub(super) color_nodes: Buffer,
+    pub(super) location_nodes: Buffer,
+
+    pub(super) request_side_bind_group_layout: BindGroupLayout,
+    pub(super) request_side_bind_group: BindGroup,
+
+    pub(super) producer_side_bind_group_layout: BindGroupLayout,
+    pub(super) producer_side_bind_group: BindGroup,
+}
+impl<const N: usize> CacheProducerInterface<N>
+where
+    [(); N * N * N]:,
+{
+    pub fn new(size: usize, device: &Device) -> Self
+    {
+        let requests = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("cache_interface_request_buffer"),
+            size: (size_of::<CacheRequest>() * size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            mapped_at_creation: false,
+        });
+
+        let requests_wb_info = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("cache_interface_writeback_info"),
+            size: (size_of::<CacheRequestWbInfo>() * size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            mapped_at_creation: false,
+        });
+
+        let structure_nodes = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("cache_interface_structure_nodes"),
+            size: (size_of::<StructurePoolElement<N>>() * size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_DST,
+            mapped_at_creation: false,
+        });
+
+        let color_nodes = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("cache_interface_structure_nodes"),
+            size: (size_of::<ColorPoolElement<N>>() * size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_DST,
+            mapped_at_creation: false,
+        });
+
+        let location_nodes = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("cache_interface_structure_nodes"),
+            size: (size_of::<LocationPoolElement>() * size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_DST,
+            mapped_at_creation: false,
+        });
+
+        let request_count = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("cache_interface_request_count"),
+            size: (size_of::<u32>()) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            mapped_at_creation: false,
+        });
+
+        let request_side_bind_group_layout = Self::request_side_bind_group_layout(device);
+        let producer_side_bind_group_layout =
+            device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+                label: Some("cache_interface_producer_side_bind_group_layout"),
+                entries: &[
+                    // Reqests
+                    BindGroupLayoutEntry {
+                        binding: 0,
+                        visibility: wgpu::ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: true },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Request count
+                    BindGroupLayoutEntry {
+                        binding: 1,
+                        visibility: wgpu::ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: true },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Structure nodes
+                    BindGroupLayoutEntry {
+                        binding: 2,
+                        visibility: wgpu::ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Color nodes
+                    BindGroupLayoutEntry {
+                        binding: 3,
+                        visibility: wgpu::ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                    // Location nodes
+                    BindGroupLayoutEntry {
+                        binding: 4,
+                        visibility: wgpu::ShaderStages::COMPUTE,
+                        ty: wgpu::BindingType::Buffer {
+                            ty: wgpu::BufferBindingType::Storage { read_only: false },
+                            has_dynamic_offset: false,
+                            min_binding_size: None,
+                        },
+                        count: None,
+                    },
+                ],
+            });
+
+        // Bind groups
+        let request_side_bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("cache_interface_request_side_bind_group"),
+            layout: &request_side_bind_group_layout,
+            entries: &[
+                BindGroupEntry {
+                    binding: 0,
+                    resource: requests.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 1,
+                    resource: requests_wb_info.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 2,
+                    resource: request_count.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 3,
+                    resource: structure_nodes.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 4,
+                    resource: color_nodes.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 5,
+                    resource: location_nodes.as_entire_binding(),
+                },
+            ],
+        });
+
+        let producer_side_bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("cache_interface_producer_side_bind_group"),
+            layout: &producer_side_bind_group_layout,
+            entries: &[
+                BindGroupEntry {
+                    binding: 0,
+                    resource: requests.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 1,
+                    resource: request_count.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 2,
+                    resource: structure_nodes.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 3,
+                    resource: color_nodes.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 4,
+                    resource: location_nodes.as_entire_binding(),
+                },
+            ],
+        });
+
+        Self {
+            size,
+            requests,
+            request_count,
+            requests_wb_info,
+            structure_nodes,
+            color_nodes,
+            location_nodes,
+
+            request_side_bind_group_layout,
+            request_side_bind_group,
+            producer_side_bind_group_layout,
+            producer_side_bind_group,
+        }
+    }
+
+    pub fn requests_buffer(&self) -> &Buffer
+    {
+        &self.requests
+    }
+
+    pub fn structure_nodes_buffer(&self) -> &Buffer
+    {
+        &self.structure_nodes
+    }
+
+    pub fn color_nodes_buffer(&self) -> &Buffer
+    {
+        &self.color_nodes
+    }
+
+    pub fn location_nodes_buffer(&self) -> &Buffer
+    {
+        &self.location_nodes
+    }
+
+    pub fn total_request_count(&self, device: &Device, queue: &Queue) -> u32
+    {
+        let (tx, rx) = std::sync::mpsc::sync_channel(1);
+        wgpu::util::DownloadBuffer::read_buffer(
+            device,
+            queue,
+            &self.request_count.slice(0..),
+            move |download_buffer| {
+                let vec = download_buffer.unwrap().to_vec();
+                let value = u32::from_ne_bytes(std::array::from_fn(|i| vec[i]));
+                let _ = tx.try_send(value);
+            },
+        );
+
+        loop
+        {
+            if let Ok(value) = rx.try_recv()
+            {
+                return value;
+            }
+            device.poll(wgpu::wgt::PollType::Poll).unwrap();
+        }
+
+        //panic!("Could not retrieve total request count.");
+    }
+
+    pub fn request_side_bind_group_layout(device: &Device) -> BindGroupLayout
+    {
+        device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+            label: Some("cache_interface_request_side_bind_group_layout"),
+            entries: &[
+                // Reqests
+                BindGroupLayoutEntry {
+                    binding: 0,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                // Requests writeback
+                BindGroupLayoutEntry {
+                    binding: 1,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                // Request count
+                BindGroupLayoutEntry {
+                    binding: 2,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                // Structure nodes
+                BindGroupLayoutEntry {
+                    binding: 3,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                // Color nodes
+                BindGroupLayoutEntry {
+                    binding: 4,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                // Location nodes
+                BindGroupLayoutEntry {
+                    binding: 5,
+                    visibility: wgpu::ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+            ],
+        })
+    }
+}

src/voxel_cache/request_buffer.rs

@@ -0,0 +1,644 @@
+use bytemuck::bytes_of;
+use wgpu::BindGroup;
+use wgpu::BindGroupEntry;
+use wgpu::Buffer;
+use wgpu::BufferUsages;
+use wgpu::CommandEncoder;
+use wgpu::ComputePipeline;
+use wgpu::Device;
+use wgpu::ShaderStages;
+
+use crate::voxel_cache::data::RequestBufferElement;
+
+pub struct RequestBuffer<const N: usize>
+{
+    pub cache_size: usize,
+    pub request_buffer: Buffer,
+    pub request_count_buffer: Buffer,
+    pub indirect_count_storage: Buffer,
+    pub element_sort_count_buffer: Buffer,
+    pub indirect_count: Buffer,
+    pub sort_buffer: Buffer,
+    pub reset_pipeline: ComputePipeline,
+    pub sort_pipeline: ComputePipeline,
+    pub bindgroup: BindGroup,
+    pub device: Device,
+
+    pub compaction_pipeline: ComputePipeline,
+    pub running_sum_pipeline: ComputePipeline,
+}
+
+impl<const N: usize> RequestBuffer<N>
+where
+    [(); N * N * N]:,
+{
+    pub fn new(cache_size: usize, device: Device) -> Self
+    {
+        let request_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some(format!("request_buffer_{N}").as_str()),
+            size: (size_of::<RequestBufferElement<N>>() * cache_size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_DST,
+            mapped_at_creation: false,
+        });
+
+        let request_count_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("Request_count_buffer"),
+            size: size_of::<u32>() as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            mapped_at_creation: false,
+        });
+
+        let indirect_count = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("indirect_count_buffer"),
+            usage: BufferUsages::INDIRECT | BufferUsages::COPY_DST,
+            size: size_of::<wgpu::util::DispatchIndirectArgs>() as u64,
+            mapped_at_creation: false,
+        });
+
+        let indirect_count_storage = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("indirect_count_storage_buffer"),
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            size: size_of::<wgpu::util::DispatchIndirectArgs>() as u64,
+            mapped_at_creation: false,
+        });
+
+        let element_sort_count_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("indirect_count_storage_buffer"),
+            usage: BufferUsages::STORAGE,
+            size: size_of::<u32>() as u64,
+            mapped_at_creation: false,
+        });
+
+        // One element number per child entry
+        // One number to identify node, one to identify sub child
+        let sort_buffer_count = cache_size * N * N * N;
+        let sort_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some(format!("request_sort_buffer_{N}").as_str()),
+            size: (size_of::<(u32, u32)>() * sort_buffer_count) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            mapped_at_creation: true,
+        });
+
+        let init_data = (0..(cache_size as u32))
+            .flat_map(|i| (0..((N * N * N) as u32)).map(move |j| (i, j)))
+            .flat_map(|(i, j)| [i, j]) // Because bytemuck does not like tuples ...
+            .collect::<Vec<_>>();
+
+        sort_buffer
+            .get_mapped_range_mut(0..)
+            .unwrap()
+            .copy_from_slice(bytemuck::cast_slice(&init_data));
+        sort_buffer.unmap();
+
+        let bindgroup_layout = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+            label: Some("request_sort_pipeline_bing_group_layout"),
+            entries: &[
+                wgpu::BindGroupLayoutEntry {
+                    binding: 0,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                wgpu::BindGroupLayoutEntry {
+                    binding: 1,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                wgpu::BindGroupLayoutEntry {
+                    binding: 2,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                wgpu::BindGroupLayoutEntry {
+                    binding: 3,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                wgpu::BindGroupLayoutEntry {
+                    binding: 4,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+            ],
+        });
+
+        let bindgroup = device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("request_sort_pipeline_bindgroup"),
+            layout: &bindgroup_layout,
+            entries: &[
+                BindGroupEntry {
+                    binding: 0,
+                    resource: request_buffer.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 1,
+                    resource: sort_buffer.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 2,
+                    resource: request_count_buffer.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 3,
+                    resource: indirect_count_storage.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 4,
+                    resource: element_sort_count_buffer.as_entire_binding(),
+                },
+            ],
+        });
+
+        let pipeline_layouts = &device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+            label: Some("request_buffer_pipeline_layout"),
+            bind_group_layouts: &[Some(&bindgroup_layout)],
+            immediate_size: size_of::<u32>() as u32,
+        });
+
+        let children_count = N * N * N;
+        let sort_pipeline =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("Sort node request"),
+                layout: Some(pipeline_layouts),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("clear_chunk_requests shader module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+                            struct RequestElement
+                            {{
+                                children: array<u32, {children_count}> 
+                            }}
+
+                            struct SortElement
+                            {{
+                                 node: u32,
+                                 child: u32
+                            }}
+
+                            var<immediate> both_phase: u32;
+
+                            @group(0) @binding(0) var<storage, read_write> request_buffer: array<RequestElement>;
+                            @group(0) @binding(1) var<storage, read_write> sort_indirection: array<SortElement>; 
+                            @group(0) @binding(2) var<storage, read_write> request_count: u32; 
+                            @group(0) @binding(3) var<storage, read_write> indirect_count: vec3<u32>;
+                            @group(0) @binding(4) var<storage, read_write> element_sort_count: u32; 
+
+                            fn big_fusion(index: u32, phase_size: u32) -> vec2<u32>
+                            {{
+                                // Find out in which block index this invocation pertains
+                                let block_index = index / (phase_size / 2);
+                                let element_index = index % (phase_size / 2); 
+
+                                let offset = block_index * phase_size;
+                                let a = offset + element_index;
+                                let b = offset + phase_size - 1 - element_index;
+                                return vec2<u32>(a, b);
+                            }}
+
+                            fn small_fusion(index: u32, phase: u32, sub_phase: u32) -> vec2<u32>
+                            {{
+                                let phase_size = u32(1 << (phase - sub_phase + 1));
+                                
+                                let block_index = index / (phase_size / 2);
+                                let element_index = index % (phase_size / 2); 
+
+                                let offset = block_index * phase_size;
+                                let a = offset + element_index;
+                                let b = a + (phase_size / 2);
+                                return vec2<u32>(a, b);
+                            }}
+
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+
+                                 let index = global_invocation_id.x;
+                                 let total = arrayLength(&sort_indirection);
+
+                                 let sub_phase = (both_phase >> 16) & 0xFFFF;
+                                 let phase = both_phase & 0xFFFF;
+                                 let phase_total_width = u32(1 << (phase + 1));
+
+
+                                 // Check if phase is last
+                                 /*
+                                 if(phase == sub_phase)
+                                 {{
+                                      let next_phase_width = phase_total_width * 2;
+                                      if(next_phase_width >= element_sort_count && phase_total_width >= element_sort_count)
+                                      {{
+                                        // This was the final phase, stop
+                                        indirect_count = vec3<u32>(0);
+                                      }}
+                                 }}
+                                 */
+
+                                 if(phase == 0)
+                                 {{
+                                    let a = index * 2;
+                                    let b = a + 1;
+                                    sort_indirection[a].child = a % {children_count};
+                                    sort_indirection[b].child = b % {children_count};
+                                 }}
+
+                                 var swap_indices = vec2<u32>(0, 0);
+
+                                 if(sub_phase == 0)
+                                 {{
+                                    // Bitonic fusion
+                                    swap_indices = big_fusion(index, phase_total_width);
+                                 }}else
+                                 {{
+                                    swap_indices = small_fusion(index, phase, sub_phase);
+                                 }}
+
+                                 // Do swap
+
+                                 //if(swap_indices.y >= request_count_round_up)
+                                 if(swap_indices.y >= element_sort_count)
+                                 {{
+                                    // Suppose that swap_indices.y is -inf, dont swap
+                                    return;
+                                 }}
+                                 
+                                 let av = request_buffer[sort_indirection[swap_indices.x].node].children[sort_indirection[swap_indices.x].child];
+                                 let bv = request_buffer[sort_indirection[swap_indices.y].node].children[sort_indirection[swap_indices.y].child];
+
+                                 if(bv > av)
+                                 {{
+                                    let temp = sort_indirection[swap_indices.x];
+                                    sort_indirection[swap_indices.x] = sort_indirection[swap_indices.y];
+                                    sort_indirection[swap_indices.y] = temp;
+                                 }}
+                            }}
+                        ")
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        let running_sum_pipeline =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("running_sum_pipeline"),
+                layout: Some(pipeline_layouts),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("running_sum shader module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+                            struct RequestElement
+                            {{
+                                children: array<u32, {children_count}> 
+                            }}
+
+                            struct SortElement
+                            {{
+                                 node: u32,
+                                 child: u32
+                            }}
+
+                            var<immediate> phase: u32;
+
+                            @group(0) @binding(0) var<storage, read_write> request_buffer: array<RequestElement>;
+                            @group(0) @binding(1) var<storage, read_write> sort_indirection: array<SortElement>; 
+                            @group(0) @binding(2) var<storage, read_write> request_counts: atomic<u32>; 
+                            @group(0) @binding(3) var<storage, read_write> indirect_count: u32; 
+                            @group(0) @binding(4) var<storage, read_write> node_sort_count: u32; 
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+                                let index = global_invocation_id.x;
+                                let len = arrayLength(&request_buffer);
+                                if(index > len) {{ return; }}
+                                let sindex = index * {children_count};
+                                 
+                                if(phase == 0)
+                                {{
+                                    var count = 0;
+                                    for(var i = 0; i < {children_count}; i++)
+                                    {{
+                                        count += select(0, 1, request_buffer[index].children[i] != 0);
+                                    }}
+
+                                    sort_indirection[sindex].child = select(u32(0), u32(1), count != 0);
+                                    if count != 0
+                                    {{
+                                        atomicAdd(&request_counts, 1);
+                                    }}
+                                    return;
+                                }}
+
+                                if(phase == 0xFFFFFFFF)
+                                {{
+                                    // Double buffering bring back
+                                    sort_indirection[sindex].child = sort_indirection[sindex].node;
+                                }}
+
+                                // Phase is not zero, running sum part
+                                let running_sum_phase = phase - 1;
+                                let running_sum_offset = u32((1 << running_sum_phase) * {children_count});
+
+                                var add = u32(0);
+                                // Double buffering
+                                if(running_sum_phase % 2 == 0)
+                                {{
+                                     if(sindex >= running_sum_offset)
+                                     {{ 
+                                        add = sort_indirection[sindex - running_sum_offset].child;
+                                     }}
+                                     sort_indirection[sindex].node = sort_indirection[sindex].child + add;
+                                }}
+                                else
+                                {{
+                                     if(sindex >= running_sum_offset)
+                                     {{ 
+                                        add = sort_indirection[sindex - running_sum_offset].node;
+                                     }}
+                                     sort_indirection[sindex].child = sort_indirection[sindex].node + add;
+                                }}
+                            }}
+                        ")
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        let compaction_pipeline =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("compaction_pipeline"),
+                layout: Some(pipeline_layouts),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("compaction shader module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+                            struct RequestElement
+                            {{
+                                children: array<u32, {children_count}> 
+                            }}
+
+                            struct SortElement
+                            {{
+                                 node: u32,
+                                 child: u32
+                            }}
+
+                            var<immediate> phase: u32;
+
+                            @group(0) @binding(0) var<storage, read_write> request_buffer: array<RequestElement>;
+                            @group(0) @binding(1) var<storage, read_write> sort_indirection: array<SortElement>; 
+                            @group(0) @binding(2) var<storage, read_write> request_counts: atomic<u32>; 
+                            @group(0) @binding(3) var<storage, read_write> indirect_count: vec3<u32>;
+                            @group(0) @binding(4) var<storage, read_write> element_sort_count: u32; 
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+                                 let index = global_invocation_id.x;
+                                 let sindex = index * {children_count};
+                                 let len = arrayLength(&request_buffer);
+
+                                 element_sort_count = sort_indirection[(len - 1) * {children_count}].child * {children_count};
+                                 let invocation_count = (element_sort_count / 2) + select(u32(0), u32(1), element_sort_count % 2 != 0);
+                                 indirect_count = vec3(
+                                     (invocation_count / 64) + select(u32(0), u32(1), invocation_count % 64 != 0),
+                                     1, 1
+                                 );
+                                 let destination_index = sort_indirection[sindex].child - 1;
+
+                                 var count = u32(0);
+                                 for(var i = 0; i < {children_count}; i++)
+                                 {{
+                                      count += request_buffer[index].children[i];
+                                 }}
+
+                                 if(count != 0) // Keep ?
+                                 {{
+                                    for(var i = u32(0); i < u32({children_count}); i++)
+                                    {{
+                                        sort_indirection[destination_index * u32({children_count}) + i].node = index;
+                                    }}
+                                 }}
+                            }}
+                        ")
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        let reset_pipeline =
+            device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+                label: Some("reset_node_requests"),
+                layout: Some(pipeline_layouts),
+                module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                    label: Some("clear_chunk_requests shader module"),
+                    source: wgpu::ShaderSource::Wgsl(
+                        format!("
+                            struct RequestElement
+                            {{
+                                children: array<u32, {children_count}> 
+                            }}
+
+                            @group(0) @binding(0) var<storage, read_write> request_buffer: array<RequestElement>;
+                            @group(0) @binding(1) var<storage, read_write> _ignore: array<u32>; 
+                            @group(0) @binding(2) var<storage, read_write> request_count: u32;
+                            @group(0) @binding(4) var<storage, read_write> node_sort_count: u32; 
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+                                    let index = global_invocation_id.x;
+                                    let total = arrayLength(&request_buffer);
+                                    if(index >= total)
+                                    {{ return; }}
+
+                                    if(index < total)
+                                    {{
+                                        for(var i = 0; i < {children_count}; i += 1)
+                                        {{
+                                             /*
+                                            if(request_buffer[index].children[i] != 0)
+                                            {{
+                                                atomicAdd(&request_count, 1);
+                                            }}
+                                            */
+                                            request_buffer[index].children[i] = 0;
+                                        }}
+                                        request_count = 0;
+                                    }}
+                            }}
+                        ")
+                        .into(),
+                    ),
+                }),
+                entry_point: Some("main"),
+                compilation_options: Default::default(),
+                cache: None,
+            });
+
+        Self {
+            cache_size,
+            request_buffer,
+            request_count_buffer,
+            indirect_count,
+            indirect_count_storage,
+            element_sort_count_buffer,
+            running_sum_pipeline,
+            sort_buffer,
+            sort_pipeline,
+            reset_pipeline,
+            compaction_pipeline,
+            bindgroup,
+            device,
+        }
+    }
+
+    pub fn request_buffer(&self) -> &Buffer
+    {
+        &self.request_buffer
+    }
+
+    pub fn request_count_buffer(&self) -> &Buffer
+    {
+        &self.request_count_buffer
+    }
+
+    pub fn sort_buffer(&self) -> &Buffer
+    {
+        &self.sort_buffer
+    }
+
+    pub fn reset_requests(&self, encoder: &mut CommandEncoder)
+    {
+        let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+            label: Some("request_buffer_reset_compute_pass"),
+            timestamp_writes: None,
+        });
+
+        compute_pass.set_bind_group(0, Some(&self.bindgroup), &[]);
+        compute_pass.set_pipeline(&self.reset_pipeline);
+
+        let shader_invocations = self.cache_size; // one invocation per element
+        let workgroup_invocations = shader_invocations.div_ceil(64);
+        compute_pass.dispatch_workgroups(workgroup_invocations as u32, 1, 1);
+    }
+
+    pub fn sort_requests(&self, encoder: &mut CommandEncoder)
+    {
+        let mut compaction_compute_pass =
+            encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+                label: Some("request_buffer_compaction_compute_pass"),
+                timestamp_writes: None,
+            });
+
+        compaction_compute_pass.set_bind_group(0, Some(&self.bindgroup), &[]);
+        let request_element_count = self.cache_size; // Each child slot is sorted
+        let workgroups_invocations = request_element_count.div_ceil(64);
+
+        // = Perform list compaction
+
+        // == Running sum
+        compaction_compute_pass.set_pipeline(&self.running_sum_pipeline);
+
+        // Phase 0: Put ones in correct location
+        compaction_compute_pass.set_immediates(0, bytes_of(&0));
+        compaction_compute_pass.dispatch_workgroups(workgroups_invocations as u32, 1, 1);
+        // Phase _: running sum
+        // Running sum phase
+        let running_sum_steps = request_element_count.next_power_of_two().ilog2();
+        for i in 1..=running_sum_steps
+        {
+            compaction_compute_pass.set_immediates(0, bytes_of(&i));
+            compaction_compute_pass.dispatch_workgroups(workgroups_invocations as u32, 1, 1);
+        }
+
+        if !running_sum_steps.is_multiple_of(2)
+        {
+            // Bring back double buffer
+            compaction_compute_pass.set_immediates(0, bytes_of(&0xFFFFFFFF_u32));
+            compaction_compute_pass.dispatch_workgroups(workgroups_invocations as u32, 1, 1);
+        }
+
+        // == Stream compaction
+        compaction_compute_pass.set_pipeline(&self.compaction_pipeline);
+        compaction_compute_pass.dispatch_workgroups(workgroups_invocations as u32, 1, 1);
+        drop(compaction_compute_pass);
+
+        // == Copy count into indirect buffer
+
+        // = Sort
+
+        // == Bitonic sort
+        let sort_element_count = self.cache_size * N * N * N;
+        let phases_upper_bound = sort_element_count.next_power_of_two().ilog2();
+
+        // Phase 0 dispatch all
+        for i in 0..=phases_upper_bound
+        {
+            encoder.copy_buffer_to_buffer(
+                &self.indirect_count_storage,
+                0,
+                &self.indirect_count,
+                0,
+                Some(size_of::<wgpu::util::DispatchIndirectArgs>() as u64),
+            );
+            let mut sort_compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+                label: Some(format!("request_buffer_sort_compute_pass_{}", 0).as_str()),
+                timestamp_writes: None,
+            });
+            sort_compute_pass.set_bind_group(0, Some(&self.bindgroup), &[]);
+            sort_compute_pass.set_pipeline(&self.sort_pipeline);
+            for j in 0..=i
+            {
+                sort_compute_pass.set_immediates(0, bytemuck::bytes_of(&(i | (j << 16))));
+                sort_compute_pass.dispatch_workgroups_indirect(&self.indirect_count, 0);
+            }
+            drop(sort_compute_pass);
+        }
+    }
+}

src/voxel_cache/structure_table.rs

@@ -0,0 +1,128 @@
+use std::num::NonZero;
+
+use wgpu::Buffer;
+use wgpu::BufferUsages;
+use wgpu::CommandEncoder;
+use wgpu::Device;
+use wgpu::Queue;
+use wgpu::util::StagingBelt;
+
+use crate::voxel_cache::data::StructurePointer;
+use crate::voxel_cache::request_buffer::RequestBuffer;
+
+// Stores root pointers to the cache
+pub struct StructureTable
+{
+    pub(crate) device: Device,
+    pub(crate) queue: Queue,
+    pub(crate) allocation_table: Vec<bool>,
+    pub(crate) available_slots: usize,
+    pub(crate) pointer_table: Buffer,
+    pub(crate) request_buffer: RequestBuffer<1>,
+    pub(crate) write_staging: StagingBelt,
+}
+
+impl StructureTable
+{
+    pub fn new(device: Device, queue: Queue) -> Self
+    {
+        let pointer_table = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("structure_table_pointer_table"),
+            size: size_of::<u32>() as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_DST | BufferUsages::COPY_SRC,
+            mapped_at_creation: false,
+        });
+
+        StructureTable {
+            request_buffer: RequestBuffer::new(1, device.clone()),
+            write_staging: StagingBelt::new(device.clone(), size_of::<u32>() as u64),
+            device,
+            queue,
+            allocation_table: vec![false],
+            available_slots: 1,
+            pointer_table,
+        }
+    }
+
+    fn double_capacity(&mut self, encoder: &mut CommandEncoder)
+    {
+        self.available_slots += self.allocation_table.len();
+        self.allocation_table
+            .extend(vec![false; self.allocation_table.len()]);
+
+        // Copy buffers into bigger buffers
+        let pointer_table = self.device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("structure_table_pointer_table"),
+            size: size_of::<u32>() as u64 * self.allocation_table.len() as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC | BufferUsages::COPY_DST,
+            mapped_at_creation: false,
+        });
+
+        encoder.copy_buffer_to_buffer(
+            &self.pointer_table,
+            0,
+            &pointer_table,
+            0,
+            self.pointer_table.size(),
+        );
+
+        self.pointer_table = pointer_table;
+        self.request_buffer = RequestBuffer::new(self.allocation_table.len(), self.device.clone());
+    }
+
+    pub fn request_buffer(&self) -> &RequestBuffer<1>
+    {
+        &self.request_buffer
+    }
+
+    pub fn request_buffer_mut(&mut self) -> &mut RequestBuffer<1>
+    {
+        &mut self.request_buffer
+    }
+
+    pub fn free_structure(&mut self, structure_id: u32)
+    {
+        self.available_slots += 1;
+        self.allocation_table[structure_id as usize] = false;
+    }
+
+    // Returns a structure Id
+    pub fn allocate_structure(&mut self, subdivided: bool) -> u32
+    {
+        let mut encoder = self
+            .device
+            .create_command_encoder(&wgpu::CommandEncoderDescriptor {
+                label: Some("structure_table_resize_encoder"),
+            });
+
+        // Get first available element
+        if self.available_slots == 0
+        {
+            self.double_capacity(&mut encoder);
+        }
+        let (first_id, _) = self
+            .allocation_table
+            .iter()
+            .enumerate()
+            .find(|(_, allocated)| !**allocated)
+            .unwrap();
+        self.allocation_table[first_id] = true;
+        self.available_slots -= 1;
+
+        // Write empty pointer to new pointer
+        self.write_staging
+            .write_buffer(
+                &mut encoder,
+                &self.pointer_table,
+                size_of::<u32>() as u64 * first_id as u64,
+                NonZero::new(size_of::<u32>() as u64).unwrap(),
+            )
+            .copy_from_slice(bytemuck::bytes_of(
+                &StructurePointer::new(subdivided, false, 0).0,
+            ));
+        self.write_staging.finish_and_recall_on_submit(&encoder);
+        self.queue.submit([encoder.finish()]);
+
+        first_id as u32
+    }
+}

src/voxel_cache/usage_buffer.rs

@@ -0,0 +1,249 @@
+use bytemuck::bytes_of;
+use wgpu::BindGroup;
+use wgpu::BindGroupEntry;
+use wgpu::Buffer;
+use wgpu::BufferUsages;
+use wgpu::CommandEncoder;
+use wgpu::ComputePipeline;
+use wgpu::Device;
+use wgpu::ShaderStages;
+
+pub struct UsageBuffer
+{
+    pub cache_size: usize,
+    pub current_timestamp: u32,
+    pub usage_buffer: Buffer,
+    pub sort_buffer: Buffer,
+    pub sort_pipeline: ComputePipeline,
+    pub bindgroup: BindGroup,
+    pub device: Device,
+}
+
+impl UsageBuffer
+{
+    pub fn new(cache_size: usize, device: Device) -> Self
+    {
+        let initial_timestamp = 0;
+        let usage_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("Usage buffer"),
+            size: (size_of::<u32>() * cache_size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_DST | BufferUsages::COPY_SRC,
+            mapped_at_creation: true,
+        });
+        let sort_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+            label: Some("usage_sort_buffer"),
+            size: (size_of::<u32>() * cache_size) as u64,
+            usage: BufferUsages::STORAGE | BufferUsages::COPY_SRC,
+            mapped_at_creation: true,
+        });
+
+        usage_buffer
+            .get_mapped_range_mut(0..)
+            .unwrap()
+            .copy_from_slice(bytemuck::cast_slice(
+                vec![initial_timestamp; cache_size].as_slice(),
+            ));
+        usage_buffer.unmap();
+
+        sort_buffer
+            .get_mapped_range_mut(0..)
+            .unwrap()
+            .copy_from_slice(bytemuck::cast_slice(
+                &(0..(cache_size as u32)).collect::<Vec<_>>(),
+            ));
+        sort_buffer.unmap();
+
+        let bindgroup_layout = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+            label: Some("usage_sort_pipeline_bing_group_layout"),
+            entries: &[
+                wgpu::BindGroupLayoutEntry {
+                    binding: 0,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+                wgpu::BindGroupLayoutEntry {
+                    binding: 1,
+                    visibility: ShaderStages::COMPUTE,
+                    ty: wgpu::BindingType::Buffer {
+                        ty: wgpu::BufferBindingType::Storage { read_only: false },
+                        has_dynamic_offset: false,
+                        min_binding_size: None,
+                    },
+                    count: None,
+                },
+            ],
+        });
+
+        let bindgroup = device.create_bind_group(&wgpu::BindGroupDescriptor {
+            label: Some("usage_sort_pipeline_bindgroup"),
+            layout: &bindgroup_layout,
+            entries: &[
+                BindGroupEntry {
+                    binding: 0,
+                    resource: usage_buffer.as_entire_binding(),
+                },
+                BindGroupEntry {
+                    binding: 1,
+                    resource: sort_buffer.as_entire_binding(),
+                },
+            ],
+        });
+
+        let pipeline_layouts = &device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+            label: Some("usage_buffer_pipeline_layout"),
+            bind_group_layouts: &[Some(&bindgroup_layout)],
+            immediate_size: size_of::<u32>() as u32,
+        });
+
+        let sort_pipeline = device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+            label: Some("Sort node request"),
+            layout: Some(pipeline_layouts),
+            module: &device.create_shader_module(wgpu::ShaderModuleDescriptor {
+                label: Some("clear_chunk_requests shader module"),
+                source: wgpu::ShaderSource::Wgsl(
+                    "
+
+                            @group(0) @binding(0) var<storage, read_write> usage_buffer: array<u32>;
+                            @group(0) @binding(1) var<storage, read_write> sort_buffer: array<u32>; 
+                            var<immediate> both_phase: u32;
+
+                            fn big_fusion(index: u32, phase_size: u32) -> vec2<u32>
+                            {{
+                                // Find out in which block index this invocation pertains
+                                let block_index = index / (phase_size / 2);
+                                let element_index = index % (phase_size / 2); 
+
+                                let offset = block_index * phase_size;
+                                let a = offset + element_index;
+                                let b = offset + phase_size - 1 - element_index;
+                                return vec2<u32>(a, b);
+                            }}
+
+                            fn small_fusion(index: u32, phase: u32, sub_phase: u32) -> vec2<u32>
+                            {{
+                                let phase_size = u32(1 << (phase - sub_phase + 1));
+                                
+                                let block_index = index / (phase_size / 2);
+                                let element_index = index % (phase_size / 2); 
+
+                                let offset = block_index * phase_size;
+                                let a = offset + element_index;
+                                let b = a + (phase_size / 2);
+                                return vec2<u32>(a, b);
+                            }}
+
+
+                            @compute
+                            @workgroup_size(64)
+                            fn main(
+                                    @builtin(global_invocation_id) global_invocation_id: vec3<u32>
+                                )
+                            {{
+
+                                 let length = arrayLength(&sort_buffer);
+                                 let index = global_invocation_id.x;
+                                 let sub_phase = (both_phase >> 16) & 0xFFFF;
+                                 let phase = both_phase & 0xFFFF;
+                                 let phase_total_width = u32(1 << (phase + 1));
+
+                                 var swap_indices = vec2<u32>(0, 0);
+
+                                 if(sub_phase == 0)
+                                 {{
+                                    // Bitonic fusion
+                                    swap_indices = big_fusion(index, phase_total_width);
+                                 }}else
+                                 {{
+                                    swap_indices = small_fusion(index, phase, sub_phase);
+                                 }}
+
+                                 // Do swap
+                                 if(swap_indices.y >= length)
+                                 {{
+                                    // Suppose that swap_indices.y is -inf, dont swap
+                                    return;
+                                 }}
+                                 
+                                 let av = usage_buffer[sort_buffer[swap_indices.x]];
+                                 let bv = usage_buffer[sort_buffer[swap_indices.y]];
+
+                                 if(bv < av)
+                                 {{
+                                    let temp = sort_buffer[swap_indices.x];
+                                    sort_buffer[swap_indices.x] = sort_buffer[swap_indices.y];
+                                    sort_buffer[swap_indices.y] = temp;
+                                 }}
+                            }}
+                        "
+                    .into(),
+                ),
+            }),
+            entry_point: Some("main"),
+            compilation_options: Default::default(),
+            cache: None,
+        });
+
+        Self {
+            cache_size,
+            current_timestamp: initial_timestamp,
+            usage_buffer,
+            sort_buffer,
+            sort_pipeline,
+            bindgroup,
+            device,
+        }
+    }
+
+    pub fn timestamp(&self) -> u32
+    {
+        self.current_timestamp
+    }
+
+    pub fn next_frame(&mut self)
+    {
+        let (new_timestamp, _) = self.current_timestamp.overflowing_add(1);
+        self.current_timestamp = new_timestamp;
+    }
+
+    pub fn usage_buffer(&self) -> &Buffer
+    {
+        &self.usage_buffer
+    }
+
+    pub fn sort_buffer(&self) -> &Buffer
+    {
+        &self.sort_buffer
+    }
+
+    pub fn sort_usage(&self, encoder: &mut CommandEncoder)
+    {
+        let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+            label: Some("usage_buffer_sort_compute_pass"),
+            timestamp_writes: None,
+        });
+
+        compute_pass.set_bind_group(0, Some(&self.bindgroup), &[]);
+        compute_pass.set_pipeline(&self.sort_pipeline);
+
+        // Compute required shader invocations
+        let sort_steps = self.cache_size.next_power_of_two().ilog2(); // Each element is sorted
+        let shader_invocations = self.cache_size.div_ceil(2); // bitonic sorting :
+        // half as many shaders
+        // per element
+        let workgroup_invocations = shader_invocations.div_ceil(64);
+
+        for i in 0..=sort_steps
+        {
+            for j in 0..=i
+            {
+                compute_pass.set_immediates(0, bytes_of(&(j << 16 | i)));
+                compute_pass.dispatch_workgroups(workgroup_invocations as u32, 1, 1);
+            }
+        }
+    }
+}