class TOPSComparator:
    """
    跨平台 AI 算力真实对比
    
    将不同平台的标称 TOPS 标准化为"等效 INT8 密集 TOPS"
    """
    
    # 转换系数（基于实测数据估算）
    INT4_TO_INT8_FACTOR = 0.35   # INT4 模型精度损失大，等效折算
    SPARSE_TO_DENSE_FACTOR = 0.70  # 稀疏模型实际场景命中率
    PEAK_TO_SUSTAINED_FACTOR = 0.60  # 持续性能通常是峰值的 60%
    
    @classmethod
    def normalized_tops(cls, platform: str, raw_tops: float, 
                        dtype: str, sparse: bool = False) -> dict:
        """
        计算标准化等效算力
        
        Args:
            platform: 平台名称
            raw_tops: 标称 TOPS
            dtype: 'INT4' | 'INT8' | 'FP16'
            sparse: 是否稀疏加速
        """
        normalized = raw_tops
        
        # INT4 折算
        if dtype == 'INT4':
            normalized *= cls.INT4_TO_INT8_FACTOR
        
        # 稀疏折算
        if sparse:
            normalized *= cls.SPARSE_TO_DENSE_FACTOR
        
        # 持续性能
        sustained = normalized * cls.PEAK_TO_SUSTAINED_FACTOR
        
        return {
            'platform': platform,
            'raw_tops': raw_tops,
            'dtype': dtype,
            'normalized_tops': normalized,
            'sustained_tops': sustained,
            'suitable_for_dms': sustained > 5,  # DMS 最低需求 5 TOPS
        }


# 测试
if __name__ == "__main__":
    platforms = [
        ("Pi AI HAT+ 2", 40, "INT4", False),
        ("Jetson Orin Nano Super", 67, "INT8", True),
        ("QCS8255", 26, "INT8", False),
        ("Jetson Orin NX", 100, "INT8", True),
        ("TI TDA4VM", 8, "INT8", False),
    ]
    
    print("=== 标准化 AI 算力对比 ===\n")
    print(f"{'平台':<25} {'标称TOPS':>8} {'类型':>6} {'等效TOPS':>10} {'持续TOPS':>10} {'DMS可用':>8}")
    print("-" * 75)
    
    for name, tops, dtype, sparse in platforms:
        result = TOPSComparator.normalized_tops(name, tops, dtype, sparse)
        print(f"{result['platform']:<25} {result['raw_tops']:>8.0f} {result['dtype']:>6} "
              f"{result['normalized_tops']:>10.1f} {result['sustained_tops']:>10.1f} "
              f"{'✅' if result['suitable_for_dms'] else '❌':>8}")

class DMSPipelineBenchmark:
    """
    DMS 完整推理流水线延迟分析
    
    流程: 采集 → 预处理 → 推理 → 后处理 → 输出
    """
    
    def __init__(self, platform: str):
        self.platform = platform
        # 各阶段延迟 (ms)，基于实测估算
        self.latencies = {
            'Pi 5 CPU': {'capture': 5, 'preprocess': 3, 'inference': 80, 'postprocess': 2, 'total': 90},
            'Pi 5 + Hailo-8L (13T)': {'capture': 5, 'preprocess': 3, 'inference': 22, 'postprocess': 2, 'total': 32},
            'Pi 5 + HAT+ 2 (40T)': {'capture': 5, 'preprocess': 3, 'inference': 12, 'postprocess': 2, 'total': 22},
            'Jetson Orin Nano Super': {'capture': 3, 'preprocess': 2, 'inference': 8, 'postprocess': 1, 'total': 14},
            'QCS8255': {'capture': 3, 'preprocess': 2, 'inference': 10, 'postprocess': 1, 'total': 16},
        }
    
    def evaluate(self) -> dict:
        lat = self.latencies.get(self.platform, self.latencies['Pi 5 CPU'])
        fps = 1000 / lat['total']
        
        return {
            'platform': self.platform,
            'total_latency_ms': lat['total'],
            'achieved_fps': fps,
            'meets_30fps': fps >= 30,
            'meets_20fps': fps >= 20,
            'latency_breakdown': lat,
        }


# 测试
if __name__ == "__main__":
    print("=== DMS 推理流水线延迟对比 ===\n")
    
    for platform in ['Pi 5 CPU', 'Pi 5 + Hailo-8L (13T)', 'Pi 5 + HAT+ 2 (40T)', 
                     'Jetson Orin Nano Super', 'QCS8255']:
        bench = DMSPipelineBenchmark(platform)
        result = bench.evaluate()
        status = '✅' if result['meets_30fps'] else ('⚠️' if result['meets_20fps'] else '❌')
        print(f"{platform:<30} {result['total_latency_ms']:>5.0f}ms  {result['achieved_fps']:>5.1f}fps  {status}")