343 lines
9.4 KiB
Markdown
343 lines
9.4 KiB
Markdown
# SoA存储优化指南
|
||
|
||
SoA (Structure of Arrays) 存储模式是ECS框架中的高级性能优化特性，适用于大规模实体系统和批量操作场景。
|
||
|
||
## 目录
|
||
|
||
1. [什么是SoA存储](#什么是soa存储)
|
||
2. [适用场景](#适用场景)
|
||
3. [不适用场景](#不适用场景)
|
||
4. [装饰器使用指南](#装饰器使用指南)
|
||
5. [性能对比](#性能对比)
|
||
6. [最佳实践](#最佳实践)
|
||
7. [故障排除](#故障排除)
|
||
|
||
## 什么是SoA存储
|
||
|
||
### AoS vs SoA 对比
|
||
|
||
**传统AoS (Array of Structures):**
|
||
```typescript
|
||
// 数据在内存中的布局
|
||
[{x:1, y:2, z:3}, {x:4, y:5, z:6}, {x:7, y:8, z:9}]
|
||
// 内存布局: x1,y1,z1,x2,y2,z2,x3,y3,z3
|
||
```
|
||
|
||
**SoA (Structure of Arrays):**
|
||
```typescript
|
||
// 数据在内存中的布局
|
||
{
|
||
x: [1, 4, 7], // Float32Array
|
||
y: [2, 5, 8], // Float32Array
|
||
z: [3, 6, 9] // Float32Array
|
||
}
|
||
// 内存布局: x1,x2,x3,y1,y2,y3,z1,z2,z3
|
||
```
|
||
|
||
### SoA的优势
|
||
|
||
- **缓存友好**: 相同类型数据连续存储，提高缓存命中率
|
||
- **向量化优化**: 支持SIMD指令并行处理
|
||
- **内存局部性**: 批量操作时减少缓存miss
|
||
- **类型优化**: 针对不同数据类型使用最优存储格式
|
||
|
||
## 适用场景
|
||
|
||
### ✅ 推荐使用SoA的场景
|
||
|
||
1. **大规模实体系统**
|
||
```typescript
|
||
// 大量相似实体的物理系统
|
||
@EnableSoA
|
||
class PhysicsComponent extends Component {
|
||
@Float64 public x: number = 0;
|
||
@Float64 public y: number = 0;
|
||
@Float32 public velocityX: number = 0;
|
||
@Float32 public velocityY: number = 0;
|
||
}
|
||
```
|
||
|
||
2. **频繁批量更新操作**
|
||
```typescript
|
||
// 每帧更新大量实体位置
|
||
system.performVectorizedOperation((fields, indices) => {
|
||
const x = fields.get('x') as Float32Array;
|
||
const y = fields.get('y') as Float32Array;
|
||
const vx = fields.get('velocityX') as Float32Array;
|
||
const vy = fields.get('velocityY') as Float32Array;
|
||
|
||
// 向量化更新所有实体
|
||
for (let i = 0; i < indices.length; i++) {
|
||
const idx = indices[i];
|
||
x[idx] += vx[idx] * deltaTime;
|
||
y[idx] += vy[idx] * deltaTime;
|
||
}
|
||
});
|
||
```
|
||
|
||
3. **数值密集计算**
|
||
```typescript
|
||
@EnableSoA
|
||
class AIBrainComponent extends Component {
|
||
@Float32 public neuron1: number = 0;
|
||
@Float32 public neuron2: number = 0;
|
||
@Float32 public output: number = 0;
|
||
}
|
||
```
|
||
|
||
## 不适用场景
|
||
|
||
### ❌ 不推荐使用SoA的场景
|
||
|
||
1. **小规模系统**
|
||
- SoA的开销大于收益
|
||
- 原始存储更简单高效
|
||
|
||
2. **随机访问为主**
|
||
```typescript
|
||
// 经常需要随机获取单个组件
|
||
const component = entityManager.getComponent(randomId, SomeComponent);
|
||
```
|
||
|
||
3. **复杂对象为主的组件**
|
||
```typescript
|
||
// 大量复杂对象，序列化开销大
|
||
class UIComponent extends Component {
|
||
public domElement: HTMLElement;
|
||
public eventHandlers: Map<string, Function>;
|
||
public children: UIComponent[];
|
||
}
|
||
```
|
||
|
||
4. **频繁增删实体**
|
||
- SoA在频繁增删时性能不如AoS
|
||
- 适合稳定的实体集合
|
||
|
||
## 装饰器使用指南
|
||
|
||
### 基础装饰器
|
||
|
||
```typescript
|
||
import { Component, EnableSoA, HighPrecision, Float64, Int32, SerializeMap, SerializeSet, SerializeArray, DeepCopy } from '@esengine/ecs-framework';
|
||
|
||
@EnableSoA // 启用SoA优化
|
||
class GameComponent extends Component {
|
||
// 数值类型装饰器
|
||
@HighPrecision // 高精度数值，保持完整精度
|
||
public entityId: number = 0;
|
||
|
||
@Float64 // 64位浮点数 (8字节，高精度)
|
||
public precisePosition: number = 0;
|
||
|
||
@Int32 // 32位整数 (4字节，整数优化)
|
||
public health: number = 100;
|
||
|
||
// 普通数值 (默认Float32Array，4字节)
|
||
public x: number = 0;
|
||
public y: number = 0;
|
||
|
||
// 集合类型装饰器
|
||
@SerializeMap
|
||
public playerStats: Map<string, number> = new Map();
|
||
|
||
@SerializeSet
|
||
public achievements: Set<string> = new Set();
|
||
|
||
@SerializeArray
|
||
public inventory: any[] = [];
|
||
|
||
@DeepCopy
|
||
public config: any = { settings: {} };
|
||
|
||
// 未装饰的字段自动选择最优存储
|
||
public name: string = ''; // string[] 数组
|
||
public active: boolean = true; // Float32Array (0/1)
|
||
public metadata: any = null; // 复杂对象存储
|
||
}
|
||
```
|
||
|
||
### 装饰器选择指南
|
||
|
||
| 装饰器 | 用途 | 存储方式 | 开销 | 适用场景 |
|
||
|--------|------|----------|------|----------|
|
||
| `@HighPrecision` | 高精度数值 | 复杂对象 | 高 | ID、时间戳、大整数 |
|
||
| `@Float64` | 双精度浮点 | Float64Array | 中 | 精密计算 |
|
||
| `@Int32` | 32位整数 | Int32Array | 低 | 整数计数、枚举值 |
|
||
| `@SerializeMap` | Map序列化 | JSON字符串 | 高 | 配置映射、属性集合 |
|
||
| `@SerializeSet` | Set序列化 | JSON字符串 | 高 | 标签集合、ID集合 |
|
||
| `@SerializeArray` | Array序列化 | JSON字符串 | 中 | 列表数据、队列 |
|
||
| `@DeepCopy` | 深拷贝对象 | 复杂对象副本 | 高 | 嵌套配置、独立状态 |
|
||
|
||
## 性能对比
|
||
|
||
### 基准测试结果
|
||
|
||
```
|
||
测试场景: 2000个实体，包含位置、速度、生命值组件
|
||
|
||
创建性能:
|
||
- 原始存储: 12.45ms
|
||
- SoA存储: 15.72ms (慢26%)
|
||
|
||
随机访问性能:
|
||
- 原始存储: 8.33ms
|
||
- SoA存储: 14.20ms (慢70%)
|
||
|
||
批量更新性能:
|
||
- 原始存储: 25.67ms
|
||
- SoA存储: 8.91ms (快188%)
|
||
|
||
内存使用:
|
||
- 原始存储: ~45KB (对象开销)
|
||
- SoA存储: ~28KB (TypedArray优化)
|
||
```
|
||
|
||
### 性能权衡总结
|
||
|
||
- **SoA优势**: 批量操作、内存效率、向量化计算
|
||
- **SoA劣势**: 随机访问、创建开销、复杂度增加
|
||
- **建议**: 大规模批量操作场景使用，小规模随机访问避免使用
|
||
|
||
## 最佳实践
|
||
|
||
### 1. 合理的组件设计
|
||
|
||
```typescript
|
||
// ✅ 好的设计：纯数值组件
|
||
@EnableSoA
|
||
class TransformComponent extends Component {
|
||
@Float64 public x: number = 0;
|
||
@Float64 public y: number = 0;
|
||
@Float32 public rotation: number = 0;
|
||
@Float32 public scaleX: number = 1;
|
||
@Float32 public scaleY: number = 1;
|
||
}
|
||
|
||
// ❌ 不好的设计：混合复杂对象
|
||
@EnableSoA
|
||
class MixedComponent extends Component {
|
||
public x: number = 0;
|
||
public domElement: HTMLElement = null; // 复杂对象开销大
|
||
public callback: Function = null; // 无法序列化
|
||
}
|
||
```
|
||
|
||
### 2. 批量操作优化
|
||
|
||
```typescript
|
||
// ✅ 使用向量化操作
|
||
const storage = entityManager.getStorage(TransformComponent) as SoAStorage<TransformComponent>;
|
||
storage.performVectorizedOperation((fields, indices) => {
|
||
const x = fields.get('x') as Float64Array;
|
||
const y = fields.get('y') as Float64Array;
|
||
|
||
// 批量处理，利用缓存局部性
|
||
for (let i = 0; i < indices.length; i++) {
|
||
const idx = indices[i];
|
||
x[idx] += deltaX;
|
||
y[idx] += deltaY;
|
||
}
|
||
});
|
||
|
||
// ❌ 避免逐个访问
|
||
for (const entity of entities) {
|
||
const transform = entity.getComponent(TransformComponent);
|
||
transform.x += deltaX;
|
||
transform.y += deltaY;
|
||
}
|
||
```
|
||
|
||
### 3. 组件分离策略
|
||
|
||
```typescript
|
||
// ✅ 将频繁批量操作的数据分离
|
||
@EnableSoA
|
||
class PositionComponent extends Component {
|
||
@Float32 public x: number = 0;
|
||
@Float32 public y: number = 0;
|
||
}
|
||
|
||
// 复杂数据使用普通组件
|
||
class MetadataComponent extends Component {
|
||
public name: string = '';
|
||
public config: any = {};
|
||
public references: any[] = [];
|
||
}
|
||
```
|
||
|
||
### 4. 性能监控
|
||
|
||
```typescript
|
||
// 监控SoA存储使用情况
|
||
const storage = entityManager.getStorage(MyComponent) as SoAStorage<MyComponent>;
|
||
const stats = storage.getStats();
|
||
|
||
console.log('SoA存储统计:', {
|
||
size: stats.size,
|
||
capacity: stats.capacity,
|
||
memoryUsage: stats.memoryUsage,
|
||
fragmentation: stats.fragmentation,
|
||
fieldStats: stats.fieldStats
|
||
});
|
||
```
|
||
|
||
## 故障排除
|
||
|
||
### 常见问题
|
||
|
||
1. **精度丢失**
|
||
```typescript
|
||
// 问题：大整数精度丢失
|
||
public bigId: number = Number.MAX_SAFE_INTEGER;
|
||
|
||
// 解决：使用高精度装饰器
|
||
@HighPrecision
|
||
public bigId: number = Number.MAX_SAFE_INTEGER;
|
||
```
|
||
|
||
2. **序列化失败**
|
||
```typescript
|
||
// 问题：循环引用导致序列化失败
|
||
@SerializeMap
|
||
public cyclicMap: Map<string, any> = new Map();
|
||
|
||
// 解决：避免循环引用或使用DeepCopy
|
||
@DeepCopy
|
||
public cyclicData: any = {};
|
||
```
|
||
|
||
3. **性能反向优化**
|
||
```typescript
|
||
// 问题：小规模数据使用SoA
|
||
@EnableSoA // 只有10个实体，不需要SoA
|
||
class SmallComponent extends Component {}
|
||
|
||
// 解决：移除@EnableSoA装饰器
|
||
class SmallComponent extends Component {}
|
||
```
|
||
|
||
### 调试技巧
|
||
|
||
```typescript
|
||
// 检查存储类型
|
||
const storage = entityManager.getStorage(MyComponent);
|
||
console.log('存储类型:', storage.constructor.name);
|
||
// 输出: 'SoAStorage' 或 'ComponentStorage'
|
||
|
||
// 检查字段存储方式
|
||
if (storage instanceof SoAStorage) {
|
||
const fieldArray = storage.getFieldArray('myField');
|
||
console.log('字段类型:', fieldArray?.constructor.name);
|
||
// 输出: 'Float32Array', 'Float64Array', 'Int32Array', 或 null
|
||
}
|
||
```
|
||
|
||
## 总结
|
||
|
||
SoA存储是一个强大的性能优化工具，但需要在合适的场景下使用：
|
||
|
||
- **适合**: 大规模、批量操作、数值密集的场景
|
||
- **不适合**: 小规模、随机访问、复杂对象为主的场景
|
||
- **关键**: 通过性能测试验证优化效果，避免过度优化
|
||
|
||
正确使用SoA存储可以显著提升ECS系统性能，但滥用会带来相反的效果。建议在实际项目中先进行基准测试，确认优化效果后再应用到生产环境。 |