SceneKit SCNPhysicsBody多核性能优化问询：多球体模拟卡顿

我发现SceneKit的SCNPhysicsBody无法充分利用多核性能。编写了一段代码在立方体内生成大量小球体，受重力下落时，球体数量增加后模拟速度急剧下降。测试不同实例数的运行耗时：

• 1实例：15秒
• 2实例：25秒
• 8实例：35秒
  显然单实例的物理模拟没有优化利用多核资源。请问如何在不损失碰撞精度的前提下提升单实例运行速度？

func createCubeWithSpheres(bottomLeft: SCNVector3, topRight: SCNVector3, radiusOfSpheres: Double) -> SCNNode {
  // Create the box using the bottomLeft and topRight coordinates
  let box = SCNBox(
    width: topRight.x - bottomLeft.x,
    height: topRight.y - bottomLeft.y,
    length: topRight.z - bottomLeft.z,
    chamferRadius: 0.0
  )
  let node = SCNNode(geometry: box)

  // Add a static physics body to the box
  node.physicsBody = SCNPhysicsBody(type: .static, shape: nil)

  // Calculate the dimensions of the box
  let width = box.width
  let height = box.height
  let depth = box.length

  // Calculate the maximum number of spheres that can fit inside the box
  let maxNumSpheresWidth = Int(width / (radiusOfSpheres * 2))
  let maxNumSpheresHeight = Int(height / (radiusOfSpheres * 2))
  let maxNumSpheresDepth = Int(depth / (radiusOfSpheres * 2))
  let maxNumSpheres = maxNumSpheresWidth * maxNumSpheresHeight * maxNumSpheresDepth

  // Create an array to hold the spheres
  var spheres = [Sphere]()

  // Create the spheres and add them to the array
  for zIndex in 0..<maxNumSpheresDepth {
    for yIndex in 0..<maxNumSpheresHeight {
      for xIndex in 0..<maxNumSpheresWidth {
        let xPos = bottomLeft.x + radiusOfSpheres + (radiusOfSpheres * 2) * CGFloat(xIndex)
        let yPos = bottomLeft.y + radiusOfSpheres + (radiusOfSpheres * 2) * CGFloat(yIndex)
        let zPos = bottomLeft.z + radiusOfSpheres + (radiusOfSpheres * 2) * CGFloat(zIndex)
        let sphere = Sphere(radius: radiusOfSpheres, position: SCNVector3(xPos, yPos, zPos))
        spheres.append(sphere)
      }
    }
  }

  // Add the spheres to the box's node
  for sphere in spheres {
    node.addChildNode(sphere.node)
  }
  return node
}

class Sphere {
    let node = SCNNode()
    
    // Initialize the sphere with a radius and position.
    init(radius: CGFloat, position: SCNVector3? = nil) {
        node.geometry = SCNSphere(radius: radius)
        if let position = position {
            node.position = position
        }
        // Add a physics body to the sphere so that it can interact with other objects.
        node.physicsBody = SCNPhysicsBody(type: .dynamic, shape: nil)

        // Add a "stickiness" to the sphere by applying a static electric field to it.
//        let electricField = SCNPhysicsField.electric()
//        electricField.strength = 10 // Adjust this value to increase or decrease the stickiness of the sphere.
//        node.physicsField = electricField
    }
}

let box = createCubeWithSpheres(bottomLeft: SCNVector3(x: 0.0, y: 0.0, z: 0.0), topRight: SCNVector3(x: 5.0, y: 5.0, z: 5.0), radiusOfSpheres: 0.1)
        
scene.rootNode.addChildNode(box)

以下是几个不损失碰撞精度的前提下提升单实例物理模拟速度的方法：

1. 开启物理世界多线程支持 #

SceneKit默认可能未启用物理模拟的多线程加速，手动开启后引擎会自动分配多核资源处理物理计算：

scene.physicsWorld.isMultithreaded = true

同时确保项目Build Settings中Swift Compiler - Code Generation下的Optimization Level设置为Fast, Whole Module Optimization（Release模式默认开启，Debug模式可临时开启测试性能）。

2. 复用物理形状 #

你的代码中每个球体都独立创建物理形状，这会产生大量冗余开销。同半径的球体可以共享同一个SCNPhysicsShape：

// 全局或类级别提前创建共享形状
let sharedSphereShape = SCNPhysicsShape(geometry: SCNSphere(radius: 0.1), options: nil)

// 修改Sphere类初始化方法
class Sphere {
    let node = SCNNode()
    
    init(radius: CGFloat, position: SCNVector3? = nil) {
        node.geometry = SCNSphere(radius: radius)
        if let position = position {
            node.position = position
        }
        node.physicsBody = SCNPhysicsBody(type: .dynamic, shape: sharedSphereShape)
    }
}

这能大幅减少内存占用和形状初始化的计算量。

3. 优化物理迭代参数 #

适当调整物理世界的迭代次数，在精度和性能间取平衡：

• 减少碰撞迭代次数：scene.physicsWorld.contactIterations = 6（默认值10，降低到6-8不会明显损失碰撞精度）
• 减少求解器迭代次数：scene.physicsWorld.solverIterations = 6
• 设置固定时间步长：scene.physicsWorld.timeStep = 1/60，避免动态步长带来的性能波动，同时保证模拟稳定性。

4. 减少不必要的物理计算 #

• 禁用不需要的物理属性：如果球体不需要旋转，添加node.physicsBody?.allowsRotation = false，减少物理引擎的旋转计算量。
• 优化碰撞掩码：如果场景允许，限制球体的碰撞范围。比如只让球体和立方体碰撞，不需要球体间碰撞，可设置：

// 给立方体设置类别掩码
node.physicsBody?.categoryBitMask = 1
// 球体只和类别1的物体碰撞
node.physicsBody?.collisionBitMask = 1

如果必须保留球体间碰撞，此方法不适用。

5. 分批初始化物理体 #

一次性创建所有球体的物理体会导致性能峰值，可分批添加：

// 修改createCubeWithSpheres方法，不直接添加物理体，而是先创建节点，再分批添加
func createCubeWithSpheres(bottomLeft: SCNVector3, topRight: SCNVector3, radiusOfSpheres: Double) -> (container: SCNNode, spheres: [SCNNode]) {
    // ... 原有创建立方体和球体节点的代码，但不给球体添加physicsBody ...
    return (node, spheres.map { $0.node })
}

// 在场景中分批添加物理体
let (box, sphereNodes) = createCubeWithSpheres(...)
scene.rootNode.addChildNode(box)

var currentIndex = 0
Timer.scheduledTimer(withTimeInterval: 0.016, repeats: true) { timer in
    guard currentIndex < sphereNodes.count else {
        timer.invalidate()
        return
    }
    let batchCount = min(100, sphereNodes.count - currentIndex)
    for i in currentIndex..<currentIndex+batchCount {
        sphereNodes[i].physicsBody = SCNPhysicsBody(type: .dynamic, shape: sharedSphereShape)
    }
    currentIndex += batchCount
}

这样能分散初始化压力，避免一次性加载大量物理体导致的卡顿。

内容的提问来源于stack exchange，提问作者bhartsb