Vendor dependencies for 0.3.0 release

vendor/metal/examples/raytracing/README.md

@@ -0,0 +1,11 @@
+## Raytracing
+
+A good showcase of Metal 3 raytracing features. 
+
+![Screenshot of the final render](./screenshot.png)
+
+## To Run
+
+```
+cargo run --example raytracing
+```

vendor/metal/examples/raytracing/camera.rs

@@ -0,0 +1,20 @@
+use glam::f32::Vec4;
+
+#[repr(C)]
+pub struct Camera {
+    pub position: Vec4,
+    pub right: Vec4,
+    pub up: Vec4,
+    pub forward: Vec4,
+}
+
+impl Camera {
+    pub fn new() -> Self {
+        Self {
+            position: Vec4::new(0.0, 3.0, 10.0, 0.0),
+            right: Vec4::new(1.0, 0.0, 0.0, 0.0),
+            up: Vec4::new(0.0, 1.0, 0.0, 0.0),
+            forward: Vec4::new(0.0, 0.0, -1.0, 0.0),
+        }
+    }
+}

vendor/metal/examples/raytracing/geometry.rs

@@ -0,0 +1,448 @@
+use std::{
+    mem::{size_of, transmute},
+    sync::Arc,
+};
+
+use glam::{
+    f32::{Mat4, Vec3, Vec4},
+    Vec4Swizzles,
+};
+
+use metal::*;
+
+pub const GEOMETRY_MASK_TRIANGLE: u32 = 1;
+pub const GEOMETRY_MASK_SPHERE: u32 = 2;
+pub const GEOMETRY_MASK_LIGHT: u32 = 4;
+
+pub const FACE_MASK_NONE: u16 = 0;
+pub const FACE_MASK_NEGATIVE_X: u16 = 1 << 0;
+pub const FACE_MASK_POSITIVE_X: u16 = 1 << 1;
+pub const FACE_MASK_NEGATIVE_Y: u16 = 1 << 2;
+pub const FACE_MASK_POSITIVE_Y: u16 = 1 << 3;
+pub const FACE_MASK_NEGATIVE_Z: u16 = 1 << 4;
+pub const FACE_MASK_POSITIVE_Z: u16 = 1 << 5;
+pub const FACE_MASK_ALL: u16 = (1 << 6) - 1;
+
+pub trait Geometry {
+    fn upload_to_buffers(&mut self) {
+        todo!()
+    }
+    fn clear(&mut self) {
+        todo!()
+    }
+    fn get_geometry_descriptor(&self) -> AccelerationStructureGeometryDescriptor {
+        todo!()
+    }
+    fn get_resources(&self) -> Vec<Resource> {
+        todo!()
+    }
+    fn get_intersection_function_name(&self) -> Option<&str> {
+        None
+    }
+}
+
+pub fn compute_triangle_normal(v0: &Vec3, v1: &Vec3, v2: &Vec3) -> Vec3 {
+    let e1 = Vec3::normalize(*v1 - *v0);
+    let e2 = Vec3::normalize(*v2 - *v0);
+    return Vec3::cross(e1, e2);
+}
+
+#[derive(Default)]
+#[repr(C)]
+pub struct Triangle {
+    pub normals: [Vec4; 3],
+    pub colours: [Vec4; 3],
+}
+
+pub fn get_managed_buffer_storage_mode() -> MTLResourceOptions {
+    return MTLResourceOptions::StorageModeManaged;
+}
+
+pub struct TriangleGeometry {
+    pub device: Device,
+    pub name: String,
+    pub index_buffer: Option<Buffer>,
+    pub vertex_position_buffer: Option<Buffer>,
+    pub vertex_normal_buffer: Option<Buffer>,
+    pub vertex_colour_buffer: Option<Buffer>,
+    pub per_primitive_data_buffer: Option<Buffer>,
+    pub indices: Vec<u16>,
+    pub vertices: Vec<Vec4>,
+    pub normals: Vec<Vec4>,
+    pub colours: Vec<Vec4>,
+    pub triangles: Vec<Triangle>,
+}
+
+impl TriangleGeometry {
+    pub fn new(device: Device, name: String) -> Self {
+        Self {
+            device,
+            name,
+            index_buffer: None,
+            vertex_position_buffer: None,
+            vertex_normal_buffer: None,
+            vertex_colour_buffer: None,
+            per_primitive_data_buffer: None,
+            indices: Vec::new(),
+            vertices: Vec::new(),
+            normals: Vec::new(),
+            colours: Vec::new(),
+            triangles: Vec::new(),
+        }
+    }
+
+    pub fn add_cube_face_with_cube_vertices(
+        &mut self,
+        cube_vertices: &[Vec3],
+        colour: Vec3,
+        i0: u16,
+        i1: u16,
+        i2: u16,
+        i3: u16,
+        inward_normals: bool,
+    ) {
+        let v0 = cube_vertices[i0 as usize];
+        let v1 = cube_vertices[i1 as usize];
+        let v2 = cube_vertices[i2 as usize];
+        let v3 = cube_vertices[i3 as usize];
+
+        let n0 = compute_triangle_normal(&v0, &v1, &v2) * if inward_normals { -1f32 } else { 1f32 };
+        let n1 = compute_triangle_normal(&v0, &v2, &v3) * if inward_normals { -1f32 } else { 1f32 };
+
+        let first_index = self.indices.len();
+        let base_index = self.vertices.len() as u16;
+
+        self.indices.push(base_index + 0);
+        self.indices.push(base_index + 1);
+        self.indices.push(base_index + 2);
+        self.indices.push(base_index + 0);
+        self.indices.push(base_index + 2);
+        self.indices.push(base_index + 3);
+
+        self.vertices.push(From::from((v0, 0.0)));
+        self.vertices.push(From::from((v1, 0.0)));
+        self.vertices.push(From::from((v2, 0.0)));
+        self.vertices.push(From::from((v3, 0.0)));
+
+        self.normals
+            .push(From::from((Vec3::normalize(n0 + n1), 0.0)));
+        self.normals.push(From::from((n0, 0.0)));
+        self.normals
+            .push(From::from((Vec3::normalize(n0 + n1), 0.0)));
+        self.normals.push(From::from((n1, 0.0)));
+
+        for _ in 0..4 {
+            self.colours.push(From::from((colour, 0.0)));
+        }
+
+        for triangle_index in 0..2 {
+            let mut triangle = Triangle::default();
+            for i in 0..3 {
+                let index = self.indices[first_index + triangle_index * 3 + i];
+                triangle.normals[i] = self.normals[index as usize];
+                triangle.colours[i] = self.colours[index as usize];
+            }
+            self.triangles.push(triangle);
+        }
+    }
+
+    pub fn add_cube_with_faces(
+        &mut self,
+        face_mask: u16,
+        colour: Vec3,
+        transform: Mat4,
+        inward_normals: bool,
+    ) {
+        let mut cube_vertices = [
+            Vec3::new(-0.5, -0.5, -0.5),
+            Vec3::new(0.5, -0.5, -0.5),
+            Vec3::new(-0.5, 0.5, -0.5),
+            Vec3::new(0.5, 0.5, -0.5),
+            Vec3::new(-0.5, -0.5, 0.5),
+            Vec3::new(0.5, -0.5, 0.5),
+            Vec3::new(-0.5, 0.5, 0.5),
+            Vec3::new(0.5, 0.5, 0.5),
+        ];
+
+        for i in 0..8 {
+            let transformed_vertex = Vec4::from((cube_vertices[i], 1.0));
+            let transformed_vertex = transform * transformed_vertex;
+            cube_vertices[i] = transformed_vertex.xyz();
+        }
+
+        const CUBE_INDICES: [[u16; 4]; 6] = [
+            [0, 4, 6, 2],
+            [1, 3, 7, 5],
+            [0, 1, 5, 4],
+            [2, 6, 7, 3],
+            [0, 2, 3, 1],
+            [4, 5, 7, 6],
+        ];
+
+        for face in 0..6 {
+            if face_mask & (1 << face) != 0 {
+                self.add_cube_face_with_cube_vertices(
+                    &cube_vertices,
+                    colour,
+                    CUBE_INDICES[face][0],
+                    CUBE_INDICES[face][1],
+                    CUBE_INDICES[face][2],
+                    CUBE_INDICES[face][3],
+                    inward_normals,
+                );
+            }
+        }
+    }
+}
+
+impl Geometry for TriangleGeometry {
+    fn upload_to_buffers(&mut self) {
+        self.index_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(self.indices.as_ptr()),
+                (self.indices.len() * size_of::<u16>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.vertex_position_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(self.vertices.as_ptr()),
+                (self.vertices.len() * size_of::<Vec4>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.vertex_normal_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(self.normals.as_ptr()),
+                (self.normals.len() * size_of::<Vec4>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.vertex_colour_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(self.colours.as_ptr()),
+                (self.colours.len() * size_of::<Vec4>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.per_primitive_data_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(self.triangles.as_ptr()),
+                (self.triangles.len() * size_of::<Triangle>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.index_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.index_buffer.as_ref().unwrap().length(),
+            ));
+        self.vertex_position_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.vertex_position_buffer.as_ref().unwrap().length(),
+            ));
+        self.vertex_normal_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.vertex_normal_buffer.as_ref().unwrap().length(),
+            ));
+        self.vertex_colour_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.vertex_colour_buffer.as_ref().unwrap().length(),
+            ));
+        self.per_primitive_data_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.per_primitive_data_buffer.as_ref().unwrap().length(),
+            ));
+
+        self.index_buffer
+            .as_ref()
+            .unwrap()
+            .set_label(&format!("index buffer of {}", self.name));
+        self.vertex_position_buffer
+            .as_ref()
+            .unwrap()
+            .set_label(&format!("vertex position buffer of {}", self.name));
+        self.vertex_normal_buffer
+            .as_ref()
+            .unwrap()
+            .set_label(&format!("vertex normal buffer of {}", self.name));
+        self.vertex_colour_buffer
+            .as_ref()
+            .unwrap()
+            .set_label(&format!("vertex colour buffer of {}", self.name));
+        self.per_primitive_data_buffer
+            .as_ref()
+            .unwrap()
+            .set_label(&format!("per primitive data buffer of {}", self.name));
+    }
+
+    fn clear(&mut self) {
+        self.indices.clear();
+        self.vertices.clear();
+        self.normals.clear();
+        self.colours.clear();
+        self.triangles.clear();
+    }
+
+    fn get_geometry_descriptor(&self) -> AccelerationStructureGeometryDescriptor {
+        let descriptor = AccelerationStructureTriangleGeometryDescriptor::descriptor();
+
+        descriptor.set_index_buffer(Some(self.index_buffer.as_ref().unwrap()));
+        descriptor.set_index_type(MTLIndexType::UInt16);
+        descriptor.set_vertex_buffer(Some(self.vertex_position_buffer.as_ref().unwrap()));
+        descriptor.set_vertex_stride(size_of::<Vec4>() as NSUInteger);
+        descriptor.set_triangle_count((self.indices.len() / 3) as NSUInteger);
+        descriptor
+            .set_primitive_data_buffer(Some(self.per_primitive_data_buffer.as_ref().unwrap()));
+        descriptor.set_primitive_data_stride(size_of::<Triangle>() as NSUInteger);
+        descriptor.set_primitive_data_element_size(size_of::<Triangle>() as NSUInteger);
+        From::from(descriptor)
+    }
+
+    fn get_resources(&self) -> Vec<Resource> {
+        vec![
+            From::from(self.index_buffer.as_ref().unwrap().clone()),
+            From::from(self.vertex_normal_buffer.as_ref().unwrap().clone()),
+            From::from(self.vertex_colour_buffer.as_ref().unwrap().clone()),
+        ]
+    }
+}
+
+#[repr(C)]
+pub struct BoundingBox {
+    pub min: Vec3,
+    pub max: Vec3,
+}
+
+#[repr(C)]
+pub struct Sphere {
+    pub origin_radius_squared: Vec4,
+    pub colour_radius: Vec4,
+}
+
+pub struct SphereGeometry {
+    pub device: Device,
+    pub sphere_buffer: Option<Buffer>,
+    pub bounding_box_buffer: Option<Buffer>,
+    pub per_primitive_data_buffer: Option<Buffer>,
+    pub spheres: Vec<Sphere>,
+}
+
+impl SphereGeometry {
+    pub fn new(device: Device) -> Self {
+        Self {
+            device,
+            sphere_buffer: None,
+            bounding_box_buffer: None,
+            per_primitive_data_buffer: None,
+            spheres: Vec::new(),
+        }
+    }
+
+    pub fn add_sphere_with_origin(&mut self, origin: Vec3, radius: f32, colour: Vec3) {
+        self.spheres.push(Sphere {
+            origin_radius_squared: Vec4::from((origin, radius * radius)),
+            colour_radius: Vec4::from((colour, radius)),
+        });
+    }
+}
+
+impl Geometry for SphereGeometry {
+    fn upload_to_buffers(&mut self) {
+        self.sphere_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(self.spheres.as_ptr()),
+                (self.spheres.len() * size_of::<Sphere>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.sphere_buffer
+            .as_ref()
+            .unwrap()
+            .set_label("sphere buffer");
+        let mut bounding_boxes = Vec::new();
+        for sphere in &self.spheres {
+            bounding_boxes.push(BoundingBox {
+                min: sphere.origin_radius_squared.xyz() - sphere.colour_radius.w,
+                max: sphere.origin_radius_squared.xyz() + sphere.colour_radius.w,
+            });
+        }
+        self.bounding_box_buffer = Some(unsafe {
+            self.device.new_buffer_with_data(
+                transmute(bounding_boxes.as_ptr()),
+                (bounding_boxes.len() * size_of::<BoundingBox>()) as NSUInteger,
+                get_managed_buffer_storage_mode(),
+            )
+        });
+        self.bounding_box_buffer
+            .as_ref()
+            .unwrap()
+            .set_label("bounding box buffer");
+        self.sphere_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.sphere_buffer.as_ref().unwrap().length(),
+            ));
+        self.bounding_box_buffer
+            .as_ref()
+            .unwrap()
+            .did_modify_range(NSRange::new(
+                0,
+                self.bounding_box_buffer.as_ref().unwrap().length(),
+            ));
+    }
+
+    fn clear(&mut self) {
+        self.spheres.clear();
+    }
+
+    fn get_geometry_descriptor(&self) -> AccelerationStructureGeometryDescriptor {
+        let descriptor = AccelerationStructureBoundingBoxGeometryDescriptor::descriptor();
+        descriptor.set_bounding_box_buffer(Some(self.bounding_box_buffer.as_ref().unwrap()));
+        descriptor.set_bounding_box_count(self.spheres.len() as NSUInteger);
+        descriptor.set_primitive_data_buffer(Some(&self.sphere_buffer.as_ref().unwrap()));
+        descriptor.set_primitive_data_stride(size_of::<Sphere>() as NSUInteger);
+        descriptor.set_primitive_data_element_size(size_of::<Sphere>() as NSUInteger);
+        From::from(descriptor)
+    }
+
+    fn get_resources(&self) -> Vec<Resource> {
+        return vec![From::from(self.sphere_buffer.as_ref().unwrap().clone())];
+    }
+
+    fn get_intersection_function_name(&self) -> Option<&str> {
+        Some("sphereIntersectionFunction")
+    }
+}
+
+pub struct GeometryInstance {
+    pub geometry: Arc<dyn Geometry>,
+    pub transform: Mat4,
+    pub mask: u32,
+    pub index_in_scene: NSUInteger,
+}
+
+#[repr(C)]
+pub struct AreaLight {
+    pub position: Vec4,
+    pub forward: Vec4,
+    pub right: Vec4,
+    pub up: Vec4,
+    pub colour: Vec4,
+}

vendor/metal/examples/raytracing/main.rs

@@ -0,0 +1,89 @@
+extern crate objc;
+
+use cocoa::{appkit::NSView, base::id as cocoa_id};
+use core_graphics_types::geometry::CGSize;
+use metal::*;
+use objc::{rc::autoreleasepool, runtime::YES};
+use std::mem;
+use winit::{
+    event::{Event, WindowEvent},
+    event_loop::ControlFlow,
+    raw_window_handle::{HasWindowHandle, RawWindowHandle},
+};
+
+pub mod camera;
+pub mod geometry;
+pub mod renderer;
+pub mod scene;
+
+fn find_raytracing_supporting_device() -> Device {
+    for device in Device::all() {
+        if !device.supports_raytracing() {
+            continue;
+        }
+        if device.is_low_power() {
+            continue;
+        }
+        return device;
+    }
+
+    panic!("No device in this machine supports raytracing!")
+}
+
+fn main() {
+    let event_loop = winit::event_loop::EventLoop::new().unwrap();
+    let size = winit::dpi::LogicalSize::new(800, 600);
+
+    let window = winit::window::WindowBuilder::new()
+        .with_inner_size(size)
+        .with_title("Metal Raytracing Example".to_string())
+        .build(&event_loop)
+        .unwrap();
+
+    let device = find_raytracing_supporting_device();
+
+    let layer = MetalLayer::new();
+    layer.set_device(&device);
+    layer.set_pixel_format(MTLPixelFormat::RGBA16Float);
+    layer.set_presents_with_transaction(false);
+
+    unsafe {
+        if let Ok(RawWindowHandle::AppKit(rw)) = window.window_handle().map(|wh| wh.as_raw()) {
+            let view = rw.ns_view.as_ptr() as cocoa_id;
+            view.setWantsLayer(YES);
+            view.setLayer(mem::transmute(layer.as_ref()));
+        }
+    }
+
+    let draw_size = window.inner_size();
+    let cg_size = CGSize::new(draw_size.width as f64, draw_size.height as f64);
+    layer.set_drawable_size(cg_size);
+
+    let mut renderer = renderer::Renderer::new(device);
+    renderer.window_resized(cg_size);
+
+    event_loop
+        .run(move |event, event_loop| {
+            autoreleasepool(|| {
+                event_loop.set_control_flow(ControlFlow::Poll);
+
+                match event {
+                    Event::AboutToWait => window.request_redraw(),
+                    Event::WindowEvent { event, .. } => match event {
+                        WindowEvent::CloseRequested => event_loop.exit(),
+                        WindowEvent::Resized(size) => {
+                            let size = CGSize::new(size.width as f64, size.height as f64);
+                            layer.set_drawable_size(size);
+                            renderer.window_resized(size);
+                        }
+                        WindowEvent::RedrawRequested => {
+                            renderer.draw(&layer);
+                        }
+                        _ => (),
+                    },
+                    _ => {}
+                }
+            });
+        })
+        .unwrap();
+}

vendor/metal/examples/raytracing/renderer.rs

@@ -0,0 +1,512 @@
+use core_graphics_types::{base::CGFloat, geometry::CGSize};
+use std::{
+    collections::BTreeMap,
+    ffi::c_void,
+    mem::{size_of, transmute},
+    ops::Index,
+    sync::{Arc, Condvar, Mutex},
+};
+
+use glam::{Vec3, Vec4, Vec4Swizzles};
+use rand::{thread_rng, RngCore};
+
+use metal::{foreign_types::ForeignType, *};
+
+use crate::{camera::Camera, geometry::get_managed_buffer_storage_mode, scene::Scene};
+
+#[repr(C)]
+struct Uniforms {
+    pub width: u32,
+    pub height: u32,
+    pub frame_index: u32,
+    pub light_count: u32,
+    pub camera: Camera,
+}
+
+pub const MAX_FRAMES_IN_FLIGHT: NSUInteger = 3;
+pub const ALIGNED_UNIFORMS_SIZE: NSUInteger = (size_of::<Uniforms>() as NSUInteger + 255) & !255;
+pub const UNIFORM_BUFFER_SIZE: NSUInteger = MAX_FRAMES_IN_FLIGHT * ALIGNED_UNIFORMS_SIZE;
+
+#[derive(Clone)]
+struct Semaphore {
+    data: Arc<(Mutex<usize>, Condvar)>,
+}
+
+impl Semaphore {
+    fn new(capacity: usize) -> Self {
+        Self {
+            data: Arc::new((Mutex::new(capacity), Condvar::new())),
+        }
+    }
+
+    fn acquire(&self) {
+        let mut value = self.data.0.lock().unwrap();
+        while *value == 0 {
+            value = self.data.1.wait(value).unwrap();
+        }
+        *value -= 1;
+    }
+
+    fn release(&self) {
+        let mut value = self.data.0.lock().unwrap();
+        *value += 1;
+        self.data.1.notify_one();
+    }
+}
+
+pub struct Renderer {
+    pub device: Device,
+    pub scene: Scene,
+    pub uniform_buffer: Buffer,
+    pub resource_buffer: Buffer,
+    pub instance_acceleration_structure: AccelerationStructure,
+    pub accumulation_targets: [Texture; 2],
+    pub random_texture: Texture,
+    pub frame_index: NSUInteger,
+    pub uniform_buffer_index: NSUInteger,
+    pub uniform_buffer_offset: NSUInteger,
+    pub size: CGSize,
+    semaphore: Semaphore,
+    pub queue: CommandQueue,
+    instance_buffer: Buffer,
+    intersection_function_table: IntersectionFunctionTable,
+    primitive_acceleration_structures: Vec<AccelerationStructure>,
+    raytracing_pipeline: ComputePipelineState,
+    copy_pipeline: RenderPipelineState,
+}
+
+impl Renderer {
+    pub fn new(device: Device) -> Self {
+        let scene = Scene::new(device.clone());
+
+        let library_path = std::path::PathBuf::from(env!("CARGO_MANIFEST_DIR"))
+            .join("examples/raytracing/shaders.metallib");
+        let library = device.new_library_with_file(library_path).unwrap();
+        let queue = device.new_command_queue();
+
+        let buffer_data = [0u8; UNIFORM_BUFFER_SIZE as usize];
+        let uniform_buffer = device.new_buffer_with_data(
+            buffer_data.as_ptr() as *const c_void,
+            UNIFORM_BUFFER_SIZE,
+            get_managed_buffer_storage_mode(),
+        );
+        uniform_buffer.set_label("uniform buffer");
+        let resources_stride = {
+            let mut max = 0;
+            for geometry in &scene.geometries {
+                let s = geometry.get_resources().len();
+                if s > max {
+                    max = s;
+                }
+            }
+            max
+        };
+        let mut resource_buffer_data = vec![0u64; resources_stride * scene.geometries.len()];
+        for geometry_index in 0..scene.geometries.len() {
+            let geometry = scene.geometries[geometry_index].as_ref();
+            let resource_buffer_begin_index = resources_stride * geometry_index;
+            let resources = geometry.get_resources();
+
+            for argument_index in 0..resources.len() {
+                let resource_buffer_index = resource_buffer_begin_index + argument_index;
+                let resource = resources[argument_index].clone();
+                resource_buffer_data[resource_buffer_index] =
+                    if resource.conforms_to_protocol::<MTLBuffer>().unwrap() {
+                        let buffer = unsafe { Buffer::from_ptr(transmute(resource.into_ptr())) };
+                        buffer.gpu_address()
+                    } else if resource.conforms_to_protocol::<MTLTexture>().unwrap() {
+                        let texture = unsafe { Texture::from_ptr(transmute(resource.into_ptr())) };
+                        texture.gpu_resource_id()._impl
+                    } else {
+                        panic!("Unexpected resource!")
+                    }
+            }
+        }
+        let resource_buffer = device.new_buffer_with_data(
+            resource_buffer_data.as_ptr() as *const c_void,
+            (resource_buffer_data.len() * size_of::<u64>()) as NSUInteger,
+            get_managed_buffer_storage_mode(),
+        );
+        resource_buffer.set_label("resource buffer");
+        resource_buffer.did_modify_range(NSRange::new(0, resource_buffer.length()));
+
+        let mut primitive_acceleration_structures = Vec::new();
+        for i in 0..scene.geometries.len() {
+            let mesh = scene.geometries[i].as_ref();
+            let geometry_descriptor = mesh.get_geometry_descriptor();
+            geometry_descriptor.set_intersection_function_table_offset(i as NSUInteger);
+            let geometry_descriptors = Array::from_owned_slice(&[geometry_descriptor]);
+            let accel_descriptor = PrimitiveAccelerationStructureDescriptor::descriptor();
+            accel_descriptor.set_geometry_descriptors(&geometry_descriptors);
+            let accel_descriptor: AccelerationStructureDescriptor = From::from(accel_descriptor);
+            primitive_acceleration_structures.push(
+                Self::new_acceleration_structure_with_descriptor(
+                    &device,
+                    &queue,
+                    &accel_descriptor,
+                ),
+            );
+        }
+
+        let mut instance_descriptors = vec![
+            MTLAccelerationStructureInstanceDescriptor::default();
+            scene.geometry_instances.len()
+        ];
+        for instance_index in 0..scene.geometry_instances.len() {
+            let instance = scene.geometry_instances[instance_index].as_ref();
+            let geometry_index = instance.index_in_scene;
+            instance_descriptors[instance_index].acceleration_structure_index =
+                geometry_index as u32;
+            instance_descriptors[instance_index].options =
+                if instance.geometry.get_intersection_function_name().is_none() {
+                    MTLAccelerationStructureInstanceOptions::Opaque
+                } else {
+                    MTLAccelerationStructureInstanceOptions::None
+                };
+            instance_descriptors[instance_index].intersection_function_table_offset = 0;
+            instance_descriptors[instance_index].mask = instance.mask as u32;
+            for column in 0..4 {
+                for row in 0..3 {
+                    instance_descriptors[instance_index].transformation_matrix[column][row] =
+                        *instance.transform.col(column).index(row);
+                }
+            }
+        }
+        let instance_buffer = device.new_buffer_with_data(
+            instance_descriptors.as_ptr() as *const c_void,
+            (size_of::<MTLAccelerationStructureInstanceDescriptor>()
+                * scene.geometry_instances.len()) as NSUInteger,
+            get_managed_buffer_storage_mode(),
+        );
+        instance_buffer.set_label("instance buffer");
+        instance_buffer.did_modify_range(NSRange::new(0, instance_buffer.length()));
+
+        let accel_descriptor = InstanceAccelerationStructureDescriptor::descriptor();
+        accel_descriptor.set_instanced_acceleration_structures(&Array::from_owned_slice(
+            &primitive_acceleration_structures,
+        ));
+        accel_descriptor.set_instance_count(scene.geometry_instances.len() as NSUInteger);
+        accel_descriptor.set_instance_descriptor_buffer(&instance_buffer);
+        let accel_descriptor: AccelerationStructureDescriptor = From::from(accel_descriptor);
+        let instance_acceleration_structure =
+            Self::new_acceleration_structure_with_descriptor(&device, &queue, &accel_descriptor);
+
+        let mut intersection_functions = BTreeMap::<String, Function>::new();
+        for geometry in &scene.geometries {
+            if let Some(name) = geometry.get_intersection_function_name() {
+                if !intersection_functions.contains_key(name) {
+                    let intersection_function = Self::new_specialised_function_with_name(
+                        &library,
+                        resources_stride as u32,
+                        name,
+                    );
+                    intersection_functions.insert(name.to_string(), intersection_function);
+                }
+            }
+        }
+        let raytracing_function = Self::new_specialised_function_with_name(
+            &library,
+            resources_stride as u32,
+            "raytracingKernel",
+        );
+        let intersection_function_array: Vec<&FunctionRef> = intersection_functions
+            .values()
+            .map(|f| -> &FunctionRef { f })
+            .collect();
+        let raytracing_pipeline = Self::new_compute_pipeline_state_with_function(
+            &device,
+            &raytracing_function,
+            &intersection_function_array,
+        );
+        let intersection_function_table_descriptor = IntersectionFunctionTableDescriptor::new();
+        intersection_function_table_descriptor
+            .set_function_count(scene.geometries.len() as NSUInteger);
+        let intersection_function_table = raytracing_pipeline
+            .new_intersection_function_table_with_descriptor(
+                &intersection_function_table_descriptor,
+            );
+        for geometry_index in 0..scene.geometries.len() {
+            let geometry = scene.geometries[geometry_index].as_ref();
+            if let Some(intersection_function_name) = geometry.get_intersection_function_name() {
+                let intersection_function = &intersection_functions[intersection_function_name];
+                let handle = raytracing_pipeline
+                    .function_handle_with_function(intersection_function)
+                    .unwrap();
+                intersection_function_table.set_function(handle, geometry_index as NSUInteger);
+            }
+        }
+        let render_descriptor = RenderPipelineDescriptor::new();
+        render_descriptor
+            .set_vertex_function(Some(&library.get_function("copyVertex", None).unwrap()));
+        render_descriptor
+            .set_fragment_function(Some(&library.get_function("copyFragment", None).unwrap()));
+        render_descriptor
+            .color_attachments()
+            .object_at(0)
+            .unwrap()
+            .set_pixel_format(MTLPixelFormat::RGBA16Float);
+        let copy_pipeline = device
+            .new_render_pipeline_state(&render_descriptor)
+            .unwrap();
+
+        let texture_descriptor = Self::create_target_descriptor(1024, 1024);
+        let accumulation_targets = [
+            device.new_texture(&texture_descriptor),
+            device.new_texture(&texture_descriptor),
+        ];
+        let random_texture = device.new_texture(&texture_descriptor);
+
+        Self {
+            device,
+            scene,
+            uniform_buffer,
+            resource_buffer,
+            instance_acceleration_structure,
+            accumulation_targets,
+            random_texture,
+            frame_index: 0,
+            uniform_buffer_index: 0,
+            uniform_buffer_offset: 0,
+            size: CGSize::new(1024 as CGFloat, 1024 as CGFloat),
+            semaphore: Semaphore::new((MAX_FRAMES_IN_FLIGHT - 2) as usize),
+            instance_buffer,
+            queue,
+            intersection_function_table,
+            primitive_acceleration_structures,
+            raytracing_pipeline,
+            copy_pipeline,
+        }
+    }
+
+    fn create_target_descriptor(width: NSUInteger, height: NSUInteger) -> TextureDescriptor {
+        let texture_descriptor = TextureDescriptor::new();
+        texture_descriptor.set_pixel_format(MTLPixelFormat::RGBA32Float);
+        texture_descriptor.set_texture_type(MTLTextureType::D2);
+        texture_descriptor.set_width(width);
+        texture_descriptor.set_height(height);
+        texture_descriptor.set_storage_mode(MTLStorageMode::Private);
+        texture_descriptor.set_usage(MTLTextureUsage::ShaderRead | MTLTextureUsage::ShaderWrite);
+        texture_descriptor
+    }
+
+    pub fn window_resized(&mut self, size: CGSize) {
+        self.size = size;
+        let texture_descriptor =
+            Self::create_target_descriptor(size.width as NSUInteger, size.height as NSUInteger);
+        self.accumulation_targets[0] = self.device.new_texture(&texture_descriptor);
+        self.accumulation_targets[1] = self.device.new_texture(&texture_descriptor);
+        texture_descriptor.set_pixel_format(MTLPixelFormat::R32Uint);
+        texture_descriptor.set_usage(MTLTextureUsage::ShaderRead);
+        texture_descriptor.set_storage_mode(MTLStorageMode::Managed);
+        self.random_texture = self.device.new_texture(&texture_descriptor);
+        let mut rng = thread_rng();
+        let mut random_values = vec![0u32; (size.width * size.height) as usize];
+        for v in &mut random_values {
+            *v = rng.next_u32();
+        }
+        self.random_texture.replace_region(
+            MTLRegion::new_2d(0, 0, size.width as NSUInteger, size.height as NSUInteger),
+            0,
+            random_values.as_ptr() as *const c_void,
+            size_of::<u32>() as NSUInteger * size.width as NSUInteger,
+        );
+        self.frame_index = 0;
+    }
+
+    fn update_uniforms(&mut self) {
+        self.uniform_buffer_offset = ALIGNED_UNIFORMS_SIZE * self.uniform_buffer_index;
+
+        let uniforms = unsafe {
+            &mut *((self.uniform_buffer.contents() as *mut u8)
+                .add(self.uniform_buffer_offset as usize) as *mut Uniforms)
+        };
+
+        let position = self.scene.camera.position;
+        let target = self.scene.camera.forward;
+        let up = self.scene.camera.up;
+
+        let forward = Vec3::normalize(target.xyz() - position.xyz());
+        let right = Vec3::normalize(Vec3::cross(forward, up.xyz()));
+        let up = Vec3::normalize(Vec3::cross(right, forward));
+
+        uniforms.camera.position = position;
+        uniforms.camera.forward = Vec4::from((forward, 0.0));
+        uniforms.camera.right = Vec4::from((right, 0.0));
+        uniforms.camera.up = Vec4::from((up, 0.0));
+
+        let field_of_view = 45.0 * (std::f32::consts::PI / 180.0);
+        let aspect_ratio = self.size.width as f32 / self.size.height as f32;
+        let image_plane_height = f32::tan(field_of_view / 2.0);
+        let image_plane_width = aspect_ratio * image_plane_height;
+
+        uniforms.camera.right *= image_plane_width;
+        uniforms.camera.up *= image_plane_height;
+
+        uniforms.width = self.size.width as u32;
+        uniforms.height = self.size.height as u32;
+
+        uniforms.frame_index = self.frame_index as u32;
+        self.frame_index += 1;
+
+        uniforms.light_count = self.scene.lights.len() as u32;
+
+        self.uniform_buffer.did_modify_range(NSRange {
+            location: self.uniform_buffer_offset,
+            length: ALIGNED_UNIFORMS_SIZE,
+        });
+
+        self.uniform_buffer_index = (self.uniform_buffer_index + 1) % MAX_FRAMES_IN_FLIGHT;
+    }
+
+    pub fn draw(&mut self, layer: &MetalLayer) {
+        self.semaphore.acquire();
+        self.update_uniforms();
+        let command_buffer = self.queue.new_command_buffer();
+        let sem = self.semaphore.clone();
+        let block = block::ConcreteBlock::new(move |_| {
+            sem.release();
+        })
+        .copy();
+        command_buffer.add_completed_handler(&block);
+        let width = self.size.width as NSUInteger;
+        let height = self.size.height as NSUInteger;
+        let threads_per_thread_group = MTLSize::new(8, 8, 1);
+        let thread_groups = MTLSize::new(
+            (width + threads_per_thread_group.width - 1) / threads_per_thread_group.width,
+            (height + threads_per_thread_group.height - 1) / threads_per_thread_group.height,
+            1,
+        );
+        let compute_encoder = command_buffer.new_compute_command_encoder();
+        compute_encoder.set_buffer(0, Some(&self.uniform_buffer), self.uniform_buffer_offset);
+        compute_encoder.set_buffer(2, Some(&self.instance_buffer), 0);
+        compute_encoder.set_buffer(3, Some(&self.scene.lights_buffer), 0);
+        compute_encoder.set_acceleration_structure(4, Some(&self.instance_acceleration_structure));
+        compute_encoder.set_intersection_function_table(5, Some(&self.intersection_function_table));
+        compute_encoder.set_texture(0, Some(&self.random_texture));
+        compute_encoder.set_texture(1, Some(&self.accumulation_targets[0]));
+        compute_encoder.set_texture(2, Some(&self.accumulation_targets[1]));
+        for geometry in &self.scene.geometries {
+            for resource in geometry.get_resources() {
+                compute_encoder.use_resource(&resource, MTLResourceUsage::Read);
+            }
+        }
+        for primitive_acceleration_structure in &self.primitive_acceleration_structures {
+            let resource: Resource = From::from(primitive_acceleration_structure.clone());
+            compute_encoder.use_resource(&resource, MTLResourceUsage::Read);
+        }
+        compute_encoder.set_compute_pipeline_state(&self.raytracing_pipeline);
+        compute_encoder.dispatch_thread_groups(thread_groups, threads_per_thread_group);
+        compute_encoder.end_encoding();
+        (self.accumulation_targets[0], self.accumulation_targets[1]) = (
+            self.accumulation_targets[1].clone(),
+            self.accumulation_targets[0].clone(),
+        );
+        if let Some(drawable) = layer.next_drawable() {
+            let render_pass_descriptor = RenderPassDescriptor::new();
+            let colour_attachment = render_pass_descriptor
+                .color_attachments()
+                .object_at(0)
+                .unwrap();
+            colour_attachment.set_texture(Some(drawable.texture()));
+            colour_attachment.set_load_action(MTLLoadAction::Clear);
+            colour_attachment.set_clear_color(MTLClearColor::new(0.0, 0.0, 0.0, 1.0));
+            let render_encoder = command_buffer.new_render_command_encoder(render_pass_descriptor);
+            render_encoder.set_render_pipeline_state(&self.copy_pipeline);
+            render_encoder.set_fragment_texture(0, Some(&self.accumulation_targets[0]));
+            render_encoder.draw_primitives(MTLPrimitiveType::Triangle, 0, 6);
+            render_encoder.end_encoding();
+            command_buffer.present_drawable(&drawable);
+        }
+        command_buffer.commit();
+    }
+
+    fn new_acceleration_structure_with_descriptor(
+        device: &Device,
+        queue: &CommandQueue,
+        descriptor: &AccelerationStructureDescriptorRef,
+    ) -> AccelerationStructure {
+        let accel_sizes = device.acceleration_structure_sizes_with_descriptor(descriptor);
+        let acceleration_structure =
+            device.new_acceleration_structure_with_size(accel_sizes.acceleration_structure_size);
+        let scratch_buffer = device.new_buffer(
+            accel_sizes.build_scratch_buffer_size,
+            MTLResourceOptions::StorageModePrivate,
+        );
+        let command_buffer = queue.new_command_buffer();
+        let command_encoder = command_buffer.new_acceleration_structure_command_encoder();
+        let compacted_size_buffer = device.new_buffer(
+            size_of::<u32>() as NSUInteger,
+            MTLResourceOptions::StorageModeShared,
+        );
+        command_encoder.build_acceleration_structure(
+            &acceleration_structure,
+            &descriptor,
+            &scratch_buffer,
+            0,
+        );
+        command_encoder.write_compacted_acceleration_structure_size(
+            &acceleration_structure,
+            &compacted_size_buffer,
+            0,
+        );
+        command_encoder.end_encoding();
+        command_buffer.commit();
+        command_buffer.wait_until_completed();
+        let compacted_size: *const u32 = unsafe { transmute(compacted_size_buffer.contents()) };
+        let compacted_size = unsafe { *compacted_size } as NSUInteger;
+        let compacted_acceleration_structure =
+            device.new_acceleration_structure_with_size(compacted_size);
+        let command_buffer = queue.new_command_buffer();
+        let command_encoder = command_buffer.new_acceleration_structure_command_encoder();
+        command_encoder.copy_and_compact_acceleration_structure(
+            &acceleration_structure,
+            &compacted_acceleration_structure,
+        );
+        command_encoder.end_encoding();
+        command_buffer.commit();
+        compacted_acceleration_structure
+    }
+
+    fn new_specialised_function_with_name(
+        library: &Library,
+        resources_stride: u32,
+        name: &str,
+    ) -> Function {
+        let constants = FunctionConstantValues::new();
+        let resources_stride = resources_stride * size_of::<u64>() as u32;
+        constants.set_constant_value_at_index(
+            &resources_stride as *const u32 as *const c_void,
+            MTLDataType::UInt,
+            0,
+        );
+        let v = true;
+        constants.set_constant_value_at_index(
+            &v as *const bool as *const c_void,
+            MTLDataType::Bool,
+            1,
+        );
+        constants.set_constant_value_at_index(
+            &v as *const bool as *const c_void,
+            MTLDataType::Bool,
+            2,
+        );
+        library.get_function(name, Some(constants)).unwrap()
+    }
+
+    fn new_compute_pipeline_state_with_function(
+        device: &Device,
+        function: &Function,
+        linked_functions: &[&FunctionRef],
+    ) -> ComputePipelineState {
+        let linked_functions = {
+            let lf = LinkedFunctions::new();
+            lf.set_functions(linked_functions);
+            lf
+        };
+        let descriptor = ComputePipelineDescriptor::new();
+        descriptor.set_compute_function(Some(function));
+        descriptor.set_linked_functions(linked_functions.as_ref());
+        descriptor.set_thread_group_size_is_multiple_of_thread_execution_width(true);
+        device.new_compute_pipeline_state(&descriptor).unwrap()
+    }
+}

vendor/metal/examples/raytracing/scene.rs

@@ -0,0 +1,135 @@
+use std::{ffi::c_void, mem::size_of, sync::Arc};
+
+use glam::{Mat4, Vec3, Vec4};
+use rand::{thread_rng, Rng};
+
+use metal::{Buffer, Device, NSRange, NSUInteger};
+
+use super::{camera::Camera, geometry::*};
+
+pub struct Scene {
+    pub device: Device,
+    pub camera: Camera,
+    pub geometries: Vec<Arc<dyn Geometry>>,
+    pub geometry_instances: Vec<Arc<GeometryInstance>>,
+    pub lights: Vec<AreaLight>,
+    pub lights_buffer: Buffer,
+}
+
+impl Scene {
+    pub fn new(device: Device) -> Self {
+        let mut geometries = Vec::<Arc<dyn Geometry>>::new();
+        let mut light_mesh = TriangleGeometry::new(device.clone(), "light".to_string());
+        let transform = Mat4::from_translation(Vec3::new(0.0, 1.0, 0.0))
+            * Mat4::from_scale(Vec3::new(0.5, 1.98, 0.5));
+        light_mesh.add_cube_with_faces(
+            FACE_MASK_POSITIVE_Y,
+            Vec3::new(1.0, 1.0, 1.0),
+            transform,
+            true,
+        );
+        light_mesh.upload_to_buffers();
+        let light_mesh = Arc::new(light_mesh);
+        geometries.push(light_mesh.clone());
+
+        let mut geometry_mesh = TriangleGeometry::new(device.clone(), "geometry".to_string());
+        let transform = Mat4::from_translation(Vec3::new(0.0, 1.0, 0.0))
+            * Mat4::from_scale(Vec3::new(2.0, 2.0, 2.0));
+        geometry_mesh.add_cube_with_faces(
+            FACE_MASK_NEGATIVE_Y | FACE_MASK_POSITIVE_Y | FACE_MASK_NEGATIVE_Z,
+            Vec3::new(0.725, 0.71, 0.68),
+            transform,
+            true,
+        );
+        geometry_mesh.add_cube_with_faces(
+            FACE_MASK_NEGATIVE_X,
+            Vec3::new(0.63, 0.065, 0.05),
+            transform,
+            true,
+        );
+        geometry_mesh.add_cube_with_faces(
+            FACE_MASK_POSITIVE_X,
+            Vec3::new(0.14, 0.45, 0.091),
+            transform,
+            true,
+        );
+        let transform = Mat4::from_translation(Vec3::new(-0.335, 0.6, -0.29))
+            * Mat4::from_rotation_y(0.3)
+            * Mat4::from_scale(Vec3::new(0.6, 1.2, 0.6));
+        geometry_mesh.add_cube_with_faces(
+            FACE_MASK_ALL,
+            Vec3::new(0.725, 0.71, 0.68),
+            transform,
+            false,
+        );
+        geometry_mesh.upload_to_buffers();
+        let geometry_mesh = Arc::new(geometry_mesh);
+        geometries.push(geometry_mesh.clone());
+
+        let mut sphere_geometry = SphereGeometry::new(device.clone());
+        sphere_geometry.add_sphere_with_origin(
+            Vec3::new(0.3275, 0.3, 0.3725),
+            0.3,
+            Vec3::new(0.725, 0.71, 0.68),
+        );
+        sphere_geometry.upload_to_buffers();
+        let sphere_geometry = Arc::new(sphere_geometry);
+        geometries.push(sphere_geometry.clone());
+
+        let mut rng = thread_rng();
+        let mut geometry_instances = Vec::new();
+        let mut lights = Vec::new();
+        for y in -1..2 {
+            for x in -1..2 {
+                let transform =
+                    Mat4::from_translation(Vec3::new(x as f32 * 2.5, y as f32 * 2.5, 0.0));
+                geometry_instances.push(Arc::new(GeometryInstance {
+                    geometry: light_mesh.clone(),
+                    transform,
+                    mask: GEOMETRY_MASK_LIGHT,
+                    index_in_scene: 0,
+                }));
+                geometry_instances.push(Arc::new(GeometryInstance {
+                    geometry: geometry_mesh.clone(),
+                    transform,
+                    mask: GEOMETRY_MASK_TRIANGLE,
+                    index_in_scene: 1,
+                }));
+                geometry_instances.push(Arc::new(GeometryInstance {
+                    geometry: sphere_geometry.clone(),
+                    transform,
+                    mask: GEOMETRY_MASK_SPHERE,
+                    index_in_scene: 2,
+                }));
+                lights.push(AreaLight {
+                    position: Vec4::new(x as f32 * 2.5, y as f32 * 2.5 + 1.98, 0.0, 0.0),
+                    forward: Vec4::new(0.0, -1.0, 0.0, 0.0),
+                    right: Vec4::new(0.25, 0.0, 0.0, 0.0),
+                    up: Vec4::new(0.0, 0.0, 0.25, 0.0),
+                    colour: Vec4::new(
+                        rng.gen_range(0f32..=1.0),
+                        rng.gen_range(0f32..=1.0),
+                        rng.gen_range(0f32..=1.0),
+                        0.0,
+                    ),
+                });
+            }
+        }
+        let lights_buffer = device.new_buffer_with_data(
+            lights.as_ptr() as *const c_void,
+            (lights.len() * size_of::<AreaLight>()) as NSUInteger,
+            get_managed_buffer_storage_mode(),
+        );
+        lights_buffer.did_modify_range(NSRange::new(0, lights_buffer.length()));
+        lights_buffer.set_label("lights buffer");
+
+        Self {
+            device,
+            camera: Camera::new(),
+            geometries,
+            geometry_instances,
+            lights,
+            lights_buffer,
+        }
+    }
+}

vendor/metal/examples/raytracing/shaders.metal

@@ -0,0 +1,598 @@
+/*
+See LICENSE folder for this sample’s licensing information.
+
+Abstract:
+The Metal shaders used for this sample.
+*/
+
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+using namespace raytracing;
+
+
+#define GEOMETRY_MASK_TRIANGLE 1
+#define GEOMETRY_MASK_SPHERE   2
+#define GEOMETRY_MASK_LIGHT    4
+
+#define GEOMETRY_MASK_GEOMETRY (GEOMETRY_MASK_TRIANGLE | GEOMETRY_MASK_SPHERE)
+
+#define RAY_MASK_PRIMARY   (GEOMETRY_MASK_GEOMETRY | GEOMETRY_MASK_LIGHT)
+#define RAY_MASK_SHADOW    GEOMETRY_MASK_GEOMETRY
+#define RAY_MASK_SECONDARY GEOMETRY_MASK_GEOMETRY
+
+struct Camera {
+    vector_float3 position;
+    vector_float3 right;
+    vector_float3 up;
+    vector_float3 forward;
+};
+
+struct AreaLight {
+    vector_float3 position;
+    vector_float3 forward;
+    vector_float3 right;
+    vector_float3 up;
+    vector_float3 color;
+};
+
+struct Uniforms {
+    unsigned int width;
+    unsigned int height;
+    unsigned int frameIndex;
+    unsigned int lightCount;
+    Camera camera;
+};
+
+struct Sphere {
+    packed_float3 origin;
+    float radiusSquared;
+    packed_float3 color;
+    float radius;
+};
+
+struct Triangle {
+    vector_float3 normals[3];
+    vector_float3 colors[3];
+};
+
+constant unsigned int resourcesStride   [[function_constant(0)]];
+constant bool useIntersectionFunctions  [[function_constant(1)]];
+constant bool usePerPrimitiveData       [[function_constant(2)]];
+constant bool useResourcesBuffer = !usePerPrimitiveData;
+
+constant unsigned int primes[] = {
+    2,   3,  5,  7,
+    11, 13, 17, 19,
+    23, 29, 31, 37,
+    41, 43, 47, 53,
+    59, 61, 67, 71,
+    73, 79, 83, 89
+};
+
+// Returns the i'th element of the Halton sequence using the d'th prime number as a
+// base. The Halton sequence is a low discrepency sequence: the values appear
+// random, but are more evenly distributed than a purely random sequence. Each random
+// value used to render the image uses a different independent dimension, `d`,
+// and each sample (frame) uses a different index `i`. To decorrelate each pixel,
+// you can apply a random offset to `i`.
+float halton(unsigned int i, unsigned int d) {
+    unsigned int b = primes[d];
+
+    float f = 1.0f;
+    float invB = 1.0f / b;
+
+    float r = 0;
+
+    while (i > 0) {
+        f = f * invB;
+        r = r + f * (i % b);
+        i = i / b;
+    }
+
+    return r;
+}
+
+// Interpolates the vertex attribute of an arbitrary type across the surface of a triangle
+// given the barycentric coordinates and triangle index in an intersection structure.
+template<typename T, typename IndexType>
+inline T interpolateVertexAttribute(device T *attributes,
+                                    IndexType i0,
+                                    IndexType i1,
+                                    IndexType i2,
+                                    float2 uv) {
+    // Look up value for each vertex.
+    const T T0 = attributes[i0];
+    const T T1 = attributes[i1];
+    const T T2 = attributes[i2];
+
+    // Compute the sum of the vertex attributes weighted by the barycentric coordinates.
+    // The barycentric coordinates sum to one.
+    return (1.0f - uv.x - uv.y) * T0 + uv.x * T1 + uv.y * T2;
+}
+
+template<typename T>
+inline T interpolateVertexAttribute(thread T *attributes, float2 uv) {
+    // Look up the value for each vertex.
+    const T T0 = attributes[0];
+    const T T1 = attributes[1];
+    const T T2 = attributes[2];
+
+    // Compute the sum of the vertex attributes weighted by the barycentric coordinates.
+    // The barycentric coordinates sum to one.
+    return (1.0f - uv.x - uv.y) * T0 + uv.x * T1 + uv.y * T2;
+}
+
+// Uses the inversion method to map two uniformly random numbers to a 3D
+// unit hemisphere, where the probability of a given sample is proportional to the cosine
+// of the angle between the sample direction and the "up" direction (0, 1, 0).
+inline float3 sampleCosineWeightedHemisphere(float2 u) {
+    float phi = 2.0f * M_PI_F * u.x;
+
+    float cos_phi;
+    float sin_phi = sincos(phi, cos_phi);
+
+    float cos_theta = sqrt(u.y);
+    float sin_theta = sqrt(1.0f - cos_theta * cos_theta);
+
+    return float3(sin_theta * cos_phi, cos_theta, sin_theta * sin_phi);
+}
+
+// Maps two uniformly random numbers to the surface of a 2D area light
+// source and returns the direction to this point, the amount of light that travels
+// between the intersection point and the sample point on the light source, as well
+// as the distance between these two points.
+
+inline void sampleAreaLight(constant AreaLight & light,
+                            float2 u,
+                            float3 position,
+                            thread float3 & lightDirection,
+                            thread float3 & lightColor,
+                            thread float & lightDistance)
+{
+    // Map to -1..1
+    u = u * 2.0f - 1.0f;
+
+    // Transform into the light's coordinate system.
+    float3 samplePosition = light.position +
+                            light.right * u.x +
+                            light.up * u.y;
+
+    // Compute the vector from sample point on  the light source to intersection point.
+    lightDirection = samplePosition - position;
+
+    lightDistance = length(lightDirection);
+
+    float inverseLightDistance = 1.0f / max(lightDistance, 1e-3f);
+
+    // Normalize the light direction.
+    lightDirection *= inverseLightDistance;
+
+    // Start with the light's color.
+    lightColor = light.color;
+
+    // Light falls off with the inverse square of the distance to the intersection point.
+    lightColor *= (inverseLightDistance * inverseLightDistance);
+
+    // Light also falls off with the cosine of the angle between the intersection point
+    // and the light source.
+    lightColor *= saturate(dot(-lightDirection, light.forward));
+}
+
+// Aligns a direction on the unit hemisphere such that the hemisphere's "up" direction
+// (0, 1, 0) maps to the given surface normal direction.
+inline float3 alignHemisphereWithNormal(float3 sample, float3 normal) {
+    // Set the "up" vector to the normal
+    float3 up = normal;
+
+    // Find an arbitrary direction perpendicular to the normal, which becomes the
+    // "right" vector.
+    float3 right = normalize(cross(normal, float3(0.0072f, 1.0f, 0.0034f)));
+
+    // Find a third vector perpendicular to the previous two, which becomes the
+    // "forward" vector.
+    float3 forward = cross(right, up);
+
+    // Map the direction on the unit hemisphere to the coordinate system aligned
+    // with the normal.
+    return sample.x * right + sample.y * up + sample.z * forward;
+}
+
+// Return the type for a bounding box intersection function.
+struct BoundingBoxIntersection {
+    bool accept    [[accept_intersection]]; // Whether to accept or reject the intersection.
+    float distance [[distance]];            // Distance from the ray origin to the intersection point.
+};
+
+// Resources for a piece of triangle geometry.
+struct TriangleResources {
+    device uint16_t *indices;
+    device float3 *vertexNormals;
+    device float3 *vertexColors;
+};
+
+// Resources for a piece of sphere geometry.
+struct SphereResources {
+    device Sphere *spheres;
+};
+
+/*
+ Custom sphere intersection function. The [[intersection]] keyword marks this as an intersection
+ function. The [[bounding_box]] keyword means that this intersection function handles intersecting rays
+ with bounding box primitives. To create sphere primitives, the sample creates bounding boxes that
+ enclose the sphere primitives.
+
+ The [[triangle_data]] and [[instancing]] keywords indicate that the intersector that calls this
+ intersection function returns barycentric coordinates for triangle intersections and traverses
+ an instance acceleration structure. These keywords must match between the intersection functions,
+ intersection function table, intersector, and intersection result to ensure that Metal propagates
+ data correctly between stages. Using fewer tags when possible may result in better performance,
+ as Metal may need to store less data and pass less data between stages. For example, if you do not
+ need barycentric coordinates, omitting [[triangle_data]] means Metal can avoid computing and storing
+ them.
+
+ The arguments to the intersection function contain information about the ray, primitive to be
+ tested, and so on. The ray intersector provides this datas when it calls the intersection function.
+ Metal provides other built-in arguments, but this sample doesn't use them.
+ */
+[[intersection(bounding_box, triangle_data, instancing)]]
+BoundingBoxIntersection sphereIntersectionFunction(// Ray parameters passed to the ray intersector below
+                                                   float3 origin                        [[origin]],
+                                                   float3 direction                     [[direction]],
+                                                   float minDistance                    [[min_distance]],
+                                                   float maxDistance                    [[max_distance]],
+                                                   // Information about the primitive.
+                                                   unsigned int primitiveIndex          [[primitive_id]],
+                                                   unsigned int geometryIndex           [[geometry_intersection_function_table_offset]],
+                                                   // Custom resources bound to the intersection function table.
+                                                   device void *resources               [[buffer(0), function_constant(useResourcesBuffer)]]
+                                                   ,const device void* perPrimitiveData [[primitive_data]])
+{
+    Sphere sphere;
+    // Look up the resources for this piece of sphere geometry.
+    if (usePerPrimitiveData) {
+        // Per-primitive data points to data from the specified buffer as was configured in the MTLAccelerationStructureBoundingBoxGeometryDescriptor.
+        sphere = *(const device Sphere*)perPrimitiveData;
+    } else
+    {
+        device SphereResources& sphereResources = *(device SphereResources *)((device char *)resources + resourcesStride * geometryIndex);
+        // Get the actual sphere enclosed in this bounding box.
+        sphere = sphereResources.spheres[primitiveIndex];
+    }
+
+    // Check for intersection between the ray and sphere mathematically.
+    float3 oc = origin - sphere.origin;
+
+    float a = dot(direction, direction);
+    float b = 2 * dot(oc, direction);
+    float c = dot(oc, oc) - sphere.radiusSquared;
+
+    float disc = b * b - 4 * a * c;
+
+    BoundingBoxIntersection ret;
+
+    if (disc <= 0.0f) {
+        // If the ray missed the sphere, return false.
+        ret.accept = false;
+    }
+    else {
+        // Otherwise, compute the intersection distance.
+        ret.distance = (-b - sqrt(disc)) / (2 * a);
+
+        // The intersection function must also check whether the intersection distance is
+        // within the acceptable range. Intersection functions do not run in any particular order,
+        // so the maximum distance may be different from the one passed into the ray intersector.
+        ret.accept = ret.distance >= minDistance && ret.distance <= maxDistance;
+    }
+
+    return ret;
+}
+
+__attribute__((always_inline))
+float3 transformPoint(float3 p, float4x4 transform) {
+    return (transform * float4(p.x, p.y, p.z, 1.0f)).xyz;
+}
+
+__attribute__((always_inline))
+float3 transformDirection(float3 p, float4x4 transform) {
+    return (transform * float4(p.x, p.y, p.z, 0.0f)).xyz;
+}
+
+// Main ray tracing kernel.
+kernel void raytracingKernel(
+     uint2                                                  tid                       [[thread_position_in_grid]],
+     constant Uniforms &                                    uniforms                  [[buffer(0)]],
+     texture2d<unsigned int>                                randomTex                 [[texture(0)]],
+     texture2d<float>                                       prevTex                   [[texture(1)]],
+     texture2d<float, access::write>                        dstTex                    [[texture(2)]],
+     device void                                           *resources                 [[buffer(1), function_constant(useResourcesBuffer)]],
+     constant MTLAccelerationStructureInstanceDescriptor   *instances                 [[buffer(2)]],
+     constant AreaLight                                    *areaLights                [[buffer(3)]],
+     instance_acceleration_structure                        accelerationStructure     [[buffer(4)]],
+     intersection_function_table<triangle_data, instancing> intersectionFunctionTable [[buffer(5)]]
+)
+{
+    // The sample aligns the thread count to the threadgroup size, which means the thread count
+    // may be different than the bounds of the texture. Test to make sure this thread
+    // is referencing a pixel within the bounds of the texture.
+    if (tid.x >= uniforms.width || tid.y >= uniforms.height) return;
+
+    // The ray to cast.
+    ray ray;
+
+    // Pixel coordinates for this thread.
+    float2 pixel = (float2)tid;
+
+    // Apply a random offset to the random number index to decorrelate pixels.
+    unsigned int offset = randomTex.read(tid).x;
+
+    // Add a random offset to the pixel coordinates for antialiasing.
+    float2 r = float2(halton(offset + uniforms.frameIndex, 0),
+                      halton(offset + uniforms.frameIndex, 1));
+
+    pixel += r;
+
+    // Map pixel coordinates to -1..1.
+    float2 uv = (float2)pixel / float2(uniforms.width, uniforms.height);
+    uv = uv * 2.0f - 1.0f;
+
+    constant Camera & camera = uniforms.camera;
+
+    // Rays start at the camera position.
+    ray.origin = camera.position;
+
+    // Map normalized pixel coordinates into camera's coordinate system.
+    ray.direction = normalize(uv.x * camera.right +
+                              uv.y * camera.up +
+                              camera.forward);
+
+    // Don't limit intersection distance.
+    ray.max_distance = INFINITY;
+
+    // Start with a fully white color. The kernel scales the light each time the
+    // ray bounces off of a surface, based on how much of each light component
+    // the surface absorbs.
+    float3 color = float3(1.0f, 1.0f, 1.0f);
+
+    float3 accumulatedColor = float3(0.0f, 0.0f, 0.0f);
+
+    // Create an intersector to test for intersection between the ray and the geometry in the scene.
+    intersector<triangle_data, instancing> i;
+
+    // If the sample isn't using intersection functions, provide some hints to Metal for
+    // better performance.
+    if (!useIntersectionFunctions) {
+        i.assume_geometry_type(geometry_type::triangle);
+        i.force_opacity(forced_opacity::opaque);
+    }
+
+    typename intersector<triangle_data, instancing>::result_type intersection;
+
+    // Simulate up to three ray bounces. Each bounce propagates light backward along the
+    // ray's path toward the camera.
+    for (int bounce = 0; bounce < 3; bounce++) {
+        // Get the closest intersection, not the first intersection. This is the default, but
+        // the sample adjusts this property below when it casts shadow rays.
+        i.accept_any_intersection(false);
+
+        // Check for intersection between the ray and the acceleration structure. If the sample
+        // isn't using intersection functions, it doesn't need to include one.
+        if (useIntersectionFunctions)
+            intersection = i.intersect(ray, accelerationStructure, bounce == 0 ? RAY_MASK_PRIMARY : RAY_MASK_SECONDARY, intersectionFunctionTable);
+        else
+            intersection = i.intersect(ray, accelerationStructure, bounce == 0 ? RAY_MASK_PRIMARY : RAY_MASK_SECONDARY);
+
+        // Stop if the ray didn't hit anything and has bounced out of the scene.
+        if (intersection.type == intersection_type::none)
+            break;
+
+        unsigned int instanceIndex = intersection.instance_id;
+
+        // Look up the mask for this instance, which indicates what type of geometry the ray hit.
+        unsigned int mask = instances[instanceIndex].mask;
+
+        // If the ray hit a light source, set the color to white, and stop immediately.
+        if (mask == GEOMETRY_MASK_LIGHT) {
+            accumulatedColor = float3(1.0f, 1.0f, 1.0f);
+            break;
+        }
+
+        // The ray hit something. Look up the transformation matrix for this instance.
+        float4x4 objectToWorldSpaceTransform(1.0f);
+
+        for (int column = 0; column < 4; column++)
+            for (int row = 0; row < 3; row++)
+                objectToWorldSpaceTransform[column][row] = instances[instanceIndex].transformationMatrix[column][row];
+
+        // Compute the intersection point in world space.
+        float3 worldSpaceIntersectionPoint = ray.origin + ray.direction * intersection.distance;
+
+        unsigned primitiveIndex = intersection.primitive_id;
+        unsigned int geometryIndex = instances[instanceIndex].accelerationStructureIndex;
+        float2 barycentric_coords = intersection.triangle_barycentric_coord;
+
+        float3 worldSpaceSurfaceNormal = 0.0f;
+        float3 surfaceColor = 0.0f;
+
+        if (mask & GEOMETRY_MASK_TRIANGLE) {
+            Triangle triangle;
+
+            float3 objectSpaceSurfaceNormal;
+            if (usePerPrimitiveData) {
+                // Per-primitive data points to data from the specified buffer as was configured in the MTLAccelerationStructureTriangleGeometryDescriptor.
+                triangle = *(const device Triangle*)intersection.primitive_data;
+            } else
+            {
+                // The ray hit a triangle. Look up the corresponding geometry's normal and UV buffers.
+                device TriangleResources & triangleResources = *(device TriangleResources *)((device char *)resources + resourcesStride * geometryIndex);
+
+                triangle.normals[0] =  triangleResources.vertexNormals[triangleResources.indices[primitiveIndex * 3 + 0]];
+                triangle.normals[1] =  triangleResources.vertexNormals[triangleResources.indices[primitiveIndex * 3 + 1]];
+                triangle.normals[2] =  triangleResources.vertexNormals[triangleResources.indices[primitiveIndex * 3 + 2]];
+
+                triangle.colors[0] =  triangleResources.vertexColors[triangleResources.indices[primitiveIndex * 3 + 0]];
+                triangle.colors[1] =  triangleResources.vertexColors[triangleResources.indices[primitiveIndex * 3 + 1]];
+                triangle.colors[2] =  triangleResources.vertexColors[triangleResources.indices[primitiveIndex * 3 + 2]];
+            }
+
+            // Interpolate the vertex normal at the intersection point.
+            objectSpaceSurfaceNormal = interpolateVertexAttribute(triangle.normals, barycentric_coords);
+
+            // Interpolate the vertex color at the intersection point.
+            surfaceColor = interpolateVertexAttribute(triangle.colors, barycentric_coords);
+
+            // Transform the normal from object to world space.
+            worldSpaceSurfaceNormal = normalize(transformDirection(objectSpaceSurfaceNormal, objectToWorldSpaceTransform));
+        }
+        else if (mask & GEOMETRY_MASK_SPHERE) {
+            Sphere sphere;
+            if (usePerPrimitiveData) {
+                // Per-primitive data points to data from the specified buffer as was configured in the MTLAccelerationStructureBoundingBoxGeometryDescriptor.
+                sphere = *(const device Sphere*)intersection.primitive_data;
+            } else
+            {
+                // The ray hit a sphere. Look up the corresponding sphere buffer.
+                device SphereResources & sphereResources = *(device SphereResources *)((device char *)resources + resourcesStride * geometryIndex);
+                sphere = sphereResources.spheres[primitiveIndex];
+            }
+
+            // Transform the sphere's origin from object space to world space.
+            float3 worldSpaceOrigin = transformPoint(sphere.origin, objectToWorldSpaceTransform);
+
+            // Compute the surface normal directly in world space.
+            worldSpaceSurfaceNormal = normalize(worldSpaceIntersectionPoint - worldSpaceOrigin);
+
+            // The sphere is a uniform color, so you don't need to interpolate the color across the surface.
+            surfaceColor = sphere.color;
+        }
+
+        dstTex.write(float4(accumulatedColor, 1.0f), tid);
+
+        // Choose a random light source to sample.
+        float lightSample = halton(offset + uniforms.frameIndex, 2 + bounce * 5 + 0);
+        unsigned int lightIndex = min((unsigned int)(lightSample * uniforms.lightCount), uniforms.lightCount - 1);
+
+        // Choose a random point to sample on the light source.
+        float2 r = float2(halton(offset + uniforms.frameIndex, 2 + bounce * 5 + 1),
+                          halton(offset + uniforms.frameIndex, 2 + bounce * 5 + 2));
+
+        float3 worldSpaceLightDirection;
+        float3 lightColor;
+        float lightDistance;
+
+        // Sample the lighting between the intersection point and the point on the area light.
+        sampleAreaLight(areaLights[lightIndex], r, worldSpaceIntersectionPoint, worldSpaceLightDirection,
+                        lightColor, lightDistance);
+
+        // Scale the light color by the cosine of the angle between the light direction and
+        // surface normal.
+        lightColor *= saturate(dot(worldSpaceSurfaceNormal, worldSpaceLightDirection));
+
+        // Scale the light color by the number of lights to compensate for the fact that
+        // the sample samples only one light source at random.
+        lightColor *= uniforms.lightCount;
+
+        // Scale the ray color by the color of the surface to simulate the surface absorbing light.
+        color *= surfaceColor;
+
+        // Compute the shadow ray. The shadow ray checks whether the sample position on the
+        // light source is visible from the current intersection point.
+        // If it is, the kernel adds lighting to the output image.
+        struct ray shadowRay;
+
+        // Add a small offset to the intersection point to avoid intersecting the same
+        // triangle again.
+        shadowRay.origin = worldSpaceIntersectionPoint + worldSpaceSurfaceNormal * 1e-3f;
+
+        // Travel toward the light source.
+        shadowRay.direction = worldSpaceLightDirection;
+
+        // Don't overshoot the light source.
+        shadowRay.max_distance = lightDistance - 1e-3f;
+
+        // Shadow rays check only whether there is an object between the intersection point
+        // and the light source. Tell Metal to return after finding any intersection.
+        i.accept_any_intersection(true);
+
+        if (useIntersectionFunctions)
+            intersection = i.intersect(shadowRay, accelerationStructure, RAY_MASK_SHADOW, intersectionFunctionTable);
+        else
+            intersection = i.intersect(shadowRay, accelerationStructure, RAY_MASK_SHADOW);
+
+        // If there was no intersection, then the light source is visible from the original
+        // intersection  point. Add the light's contribution to the image.
+        if (intersection.type == intersection_type::none)
+            accumulatedColor += lightColor * color;
+
+        // Choose a random direction to continue the path of the ray. This causes light to
+        // bounce between surfaces. An app might evaluate a more complicated equation to
+        // calculate the amount of light that reflects between intersection points.  However,
+        // all the math in this kernel cancels out because this app assumes a simple diffuse
+        // BRDF and samples the rays with a cosine distribution over the hemisphere (importance
+        // sampling). This requires that the kernel only multiply the colors together. This
+        // sampling strategy also reduces the amount of noise in the output image.
+        r = float2(halton(offset + uniforms.frameIndex, 2 + bounce * 5 + 3),
+                   halton(offset + uniforms.frameIndex, 2 + bounce * 5 + 4));
+
+        float3 worldSpaceSampleDirection = sampleCosineWeightedHemisphere(r);
+        worldSpaceSampleDirection = alignHemisphereWithNormal(worldSpaceSampleDirection, worldSpaceSurfaceNormal);
+
+        ray.origin = worldSpaceIntersectionPoint + worldSpaceSurfaceNormal * 1e-3f;
+        ray.direction = worldSpaceSampleDirection;
+    }
+
+    // Average this frame's sample with all of the previous frames.
+    if (uniforms.frameIndex > 0) {
+        float3 prevColor = prevTex.read(tid).xyz;
+        prevColor *= uniforms.frameIndex;
+
+        accumulatedColor += prevColor;
+        accumulatedColor /= (uniforms.frameIndex + 1);
+    }
+
+    dstTex.write(float4(accumulatedColor, 1.0f), tid);
+}
+
+// Screen filling quad in normalized device coordinates.
+constant float2 quadVertices[] = {
+    float2(-1, -1),
+    float2(-1,  1),
+    float2( 1,  1),
+    float2(-1, -1),
+    float2( 1,  1),
+    float2( 1, -1)
+};
+
+struct CopyVertexOut {
+    float4 position [[position]];
+    float2 uv;
+};
+
+// Simple vertex shader that passes through NDC quad positions.
+vertex CopyVertexOut copyVertex(unsigned short vid [[vertex_id]]) {
+    float2 position = quadVertices[vid];
+
+    CopyVertexOut out;
+
+    out.position = float4(position, 0, 1);
+    out.uv = position * 0.5f + 0.5f;
+
+    return out;
+}
+
+// Simple fragment shader that copies a texture and applies a simple tonemapping function.
+fragment float4 copyFragment(CopyVertexOut in [[stage_in]],
+                             texture2d<float> tex)
+{
+    constexpr sampler sam(min_filter::nearest, mag_filter::nearest, mip_filter::none);
+
+    float3 color = tex.sample(sam, in.uv).xyz;
+
+    // Apply a simple tonemapping function to reduce the dynamic range of the
+    // input image into a range which the screen can display.
+    color = color / (1.0f + color);
+
+    return float4(color, 1.0f);
+}