Class Refine¶

Defined in File LegacyRefine.h

Inheritance Relationships¶

Derived Types¶

public DSFM_M_Refine (Class DSFM_M_Refine)
public DSFM_S_Refine (Class DSFM_S_Refine)
public FMRefine (Class FMRefine)
public GreedyRefine (Class GreedyRefine)
public OldRefine (Class OldRefine)
public PMRefine (Class PMRefine)

Class Documentation¶

class Refine¶

FM 类策略的公共基类，封装共享的评估/执行上下文构造。

Subclassed by DSFM_M_Refine, DSFM_S_Refine, FMRefine, GreedyRefine, OldRefine, PMRefine

Public Functions

inline Refine(int max_hop, int large_net_threshold, double penalty, vector<vector<int>> &cut_weights, const fpga &fpgas, const TimingRefineConfig &timing_cfg)¶

Protected Functions

inline GainEvaluatorContext makeGainEvalContext() const¶

inline MoveApplierContext makeMoveApplierContext()¶

Gain pickMove(PerPartBucketSet &buckets, const graph &finest, const vector<VectorXi> &occupied_resources)¶: 从多分区桶集合中选取最佳可行移动。

template<typename PickFn, typename AcceptFn, typename UpdateFn, typename ClearFn> inline double runFMLoop(int max_move, int max_neg_move, const graph &finest, vector<int> &parts, vector<VectorXi> &occupied_resources, vector<bool> &visited_vertices_flag, vector<NetPartition> &partition, PickFn pickFn, AcceptFn acceptFn, UpdateFn updateFn, ClearFn clearFn)¶

增益桶驱动的通用主循环模板。

所有基于收益桶的启发式搜索（PM、FM、DSFM 等）共用此循环。容忍短暂的负收益移动，当无法继续改进时回滚到历史最优状态。

Protected Attributes

int max_hop¶: 最大跳步约束。

int large_net_threshold¶: 大超边过滤阈值。

double penalty¶: TDM 或拥塞惩罚权重。

vector<vector<int>> &cut_weights¶: 割边权重矩阵。

const fpga &fpgas¶: FPGA 平台约束。

TimingRefineConfig timing_cfg¶: 时序细化配置。