package bvh import "../collision" import "base:runtime" import "core:container/queue" import "core:log" import "core:math" import lg "core:math/linalg" import "core:mem" import "game:debug" import "libs:tracy" import rl "vendor:raylib" _ :: log _ :: rl _ :: lg Vec3 :: [3]f32 AABB :: struct { min, max: Vec3, } // Helper struct to avoid passing verts/indices separately Mesh :: struct { vertices: []Vec3, indices: []u16, } BVH :: struct { nodes: []Node, // Triangle IDs. first_index = indices[primitive * 3] primitives: []u16, nodes_used: i32, } destroy_bvh :: proc(bvh: ^BVH) { delete(bvh.nodes) delete(bvh.primitives) } // Helper struct to store mesh data together with its bvh for convenience // You don't have to use it Mesh_BVH :: struct { bvh: BVH, mesh: Mesh, } Node :: struct { aabb: AABB, // Index of the left child, right child is left_child + 1 child_or_prim_start: i32, prim_len: i32, } // uvw Bary :: [3]f32 is_leaf_node :: #force_inline proc(node: Node) -> bool { return node.prim_len > 0 } #assert(size_of(Node) == 32) build_bvh_from_mesh :: proc(mesh: Mesh, allocator := context.allocator) -> (mesh_bvh: Mesh_BVH) { tracy.Zone() vertices, indices := mesh.vertices, mesh.indices assert(len(indices) % 3 == 0) bvh := &mesh_bvh.bvh num_triangles := len(indices) / 3 // Caller owned, allocator might be temp_allocator so do this before checkpoint below, otherwise we the result accidentally bvh.nodes, _ = mem.make_aligned([]Node, num_triangles * 2 - 1, size_of(Node), allocator) bvh.primitives = make([]u16, num_triangles, allocator) // Clean up after ourselves temp := runtime.default_temp_allocator_temp_begin() defer runtime.default_temp_allocator_temp_end(temp) // Temp stuff centroids := make([]Vec3, num_triangles, context.temp_allocator) aabbs := make([]AABB, num_triangles, context.temp_allocator) // Calculate centroids and aabbs { tracy.ZoneN("calculate_centroids_and_aabbs") for i in 0 ..< num_triangles { i1, i2, i3 := indices[i * 3], indices[i * 3 + 1], indices[i * 3 + 2] v1, v2, v3 := vertices[i1], vertices[i2], vertices[i3] centroids[i] = (v1 + v2 + v3) * 0.33333333333 aabbs[i].min = Vec3 { min(v1.x, v2.x, v3.x), min(v1.y, v2.y, v3.y), min(v1.z, v2.z, v3.z), } aabbs[i].max = Vec3 { max(v1.x, v2.x, v3.x), max(v1.y, v2.y, v3.y), max(v1.z, v2.z, v3.z), } size := aabbs[i].max - aabbs[i].min assert(size.x >= 0) assert(size.y >= 0) assert(size.z >= 0) bvh.primitives[i] = u16(i) } } bvh.nodes_used = 1 // root root := &bvh.nodes[0] root.child_or_prim_start = 0 root.prim_len = i32(num_triangles) update_node_bounds(bvh, 0, aabbs) subdivide(bvh, 0, centroids, aabbs) return } /// Useful for a top level accel structure build_bvh_from_aabbs :: proc(aabbs: []AABB, allocator := context.allocator) -> (bvh: BVH) { bvh.nodes, _ = mem.make_aligned([]Node, len(aabbs) * 2 - 1, size_of(Node), allocator) bvh.primitives = make([]u16, len(aabbs), allocator) temp := runtime.default_temp_allocator_temp_begin() defer runtime.default_temp_allocator_temp_end(temp) // Temp stuff centroids := make([]Vec3, len(aabbs), context.temp_allocator) // Calculate centroids for i in 0 ..< len(aabbs) { centroids[i] = (aabbs[i].max + aabbs[i].min) * 0.5 bvh.primitives[i] = u16(i) } bvh.nodes_used = 1 root := &bvh.nodes[0] root.prim_len = i32(len(aabbs)) update_node_bounds(&bvh, 0, aabbs) subdivide(&bvh, 0, centroids, aabbs) return } update_node_bounds :: proc(bvh: ^BVH, node_idx: i32, prim_aabbs: []AABB) { tracy.Zone() node := &bvh.nodes[node_idx] node.aabb.min = math.F32_MAX node.aabb.max = -math.F32_MAX for i in node.child_or_prim_start ..< node.child_or_prim_start + node.prim_len { prim_aabb := prim_aabbs[bvh.primitives[i]] node.aabb.min.x = min(node.aabb.min.x, prim_aabb.min.x) node.aabb.min.y = min(node.aabb.min.y, prim_aabb.min.y) node.aabb.min.z = min(node.aabb.min.z, prim_aabb.min.z) node.aabb.max.x = max(node.aabb.max.x, prim_aabb.max.x) node.aabb.max.y = max(node.aabb.max.y, prim_aabb.max.y) node.aabb.max.z = max(node.aabb.max.z, prim_aabb.max.z) } } subdivide :: proc(bvh: ^BVH, node_idx: i32, centroids: []Vec3, aabbs: []AABB) { tracy.Zone() node := &bvh.nodes[node_idx] if node.prim_len <= 2 { return } size := node.aabb.max - node.aabb.min // Split along longest axis largest_side := size.x split_axis := 0 if size.y > largest_side { split_axis = 1 largest_side = size.y } if size.z > largest_side { split_axis = 2 } split_pos := node.aabb.min[split_axis] + size[split_axis] * 0.5 // Partition i := node.child_or_prim_start j := i + node.prim_len - 1 for i <= j { prim_i := bvh.primitives[i] prim_j := bvh.primitives[j] if centroids[prim_i][split_axis] < split_pos { i += 1 } else { bvh.primitives[i] = prim_j bvh.primitives[j] = prim_i j -= 1 } } left_count := i - node.child_or_prim_start if left_count == 0 || left_count == node.prim_len { return } left_child := bvh.nodes_used right_child := bvh.nodes_used + 1 bvh.nodes_used += 2 prim_start := node.child_or_prim_start node.child_or_prim_start = left_child bvh.nodes[left_child] = {} bvh.nodes[right_child] = {} bvh.nodes[left_child].child_or_prim_start = prim_start bvh.nodes[left_child].prim_len = left_count bvh.nodes[right_child].child_or_prim_start = i bvh.nodes[right_child].prim_len = node.prim_len - left_count node.prim_len = 0 update_node_bounds(bvh, left_child, aabbs) update_node_bounds(bvh, right_child, aabbs) subdivide(bvh, left_child, centroids, aabbs) subdivide(bvh, right_child, centroids, aabbs) } Ray :: struct { origin, dir: Vec3, dir_inv: Vec3, } Collision :: struct { hit: bool, t: f32, // which primitive we hit prim: u16, // Barycentric coords of the hit triangle bary: Bary, // jStats aabb_tests: int, triangle_tests: int, } traverse_bvh_ray_mesh :: proc(bvh: ^BVH, mesh: Mesh, ray: Ray, out_collision: ^Collision) -> bool { tracy.Zone() ray := ray ray.dir_inv.x = 1.0 / ray.dir.x ray.dir_inv.y = 1.0 / ray.dir.y ray.dir_inv.z = 1.0 / ray.dir.z if !out_collision.hit { out_collision.t = math.F32_MAX } prev_t := out_collision.t internal_traverse_bvh_ray_triangles(bvh, mesh, ray, out_collision) return out_collision.hit && out_collision.t < prev_t } internal_traverse_bvh_ray_triangles :: proc( bvh: ^BVH, mesh: Mesh, ray: Ray, out_collision: ^Collision, ) { temp := runtime.default_temp_allocator_temp_begin() defer runtime.default_temp_allocator_temp_end(temp) nodes_to_process: queue.Queue(i32) queue.init(&nodes_to_process, queue.DEFAULT_CAPACITY, context.temp_allocator) queue.push_back(&nodes_to_process, 0) for queue.len(nodes_to_process) > 0 { node_idx := queue.pop_front(&nodes_to_process) assert(node_idx < bvh.nodes_used) node := &bvh.nodes[node_idx] out_collision.aabb_tests += 1 if !internal_ray_aabb_test(ray, node.aabb, out_collision.t) { continue } rl.DrawBoundingBox( {min = node.aabb.min, max = node.aabb.max}, debug.int_to_color(node_idx), ) if is_leaf_node(node^) { for i in node.child_or_prim_start ..< node.child_or_prim_start + node.prim_len { internal_ray_tri_test(ray, mesh, bvh.primitives[i], out_collision) } } else { left_node := node.child_or_prim_start queue.push_back_elems(&nodes_to_process, left_node, left_node + 1) } } } internal_ray_aabb_test :: proc(ray: Ray, box: AABB, min_t: f32) -> bool { ts, ok := collision.intersect_ray_aabb(ray.origin, ray.dir, collision.Aabb{box.min, box.max}) return ok && ts[0] < min_t } // Möller–Trumbore intersection algorithm // https://jacco.ompf2.com/2022/04/13/how-to-build-a-bvh-part-1-basics/ internal_ray_tri_test :: proc(ray: Ray, mesh: Mesh, tri: u16, col: ^Collision) { i1, i2, i3 := mesh.indices[tri * 3], mesh.indices[tri * 3 + 1], mesh.indices[tri * 3 + 2] v1, v2, v3 := mesh.vertices[i1], mesh.vertices[i2], mesh.vertices[i3] col.triangle_tests += 1 // rl.DrawTriangle3D(v1, v2, v3, debug.int_to_color(i32(tri) + 1)) t, _, barycentric, ok := collision.intersect_ray_triangle( {ray.origin, ray.origin + ray.dir}, {v1, v2, v3}, ) if ok && t < col.t { col.hit = true col.t = t col.prim = tri col.bary = barycentric } }