Algorithm 1: Phase II—Per-Switch Local Refinement

Input: Baseline GCL from Phase I, device-specific caps C_cap(S) Output: Refined GCLs satisfying hardware entry limits per switch     1: function PHASE_II_REFINE(GCL, {C_cap(S)})  2:   // Distribute to switches and refine in parallel   3:   parallel for each switch S do   4:    windows ← PARTITION_BY_PORT(GCL, S)   5:    for iter = 1 to I do              // I = 8   6:     for each port P in S.egress_ports do   7:      if COUNT_ENTRIES(windows[P]) ≤ C_cap(S) then continue   8:   9:         // Step 1: Gap filling-absorb sub-slot intervals 10:      windows[P] ← GAP_FILL(windows[P], t_slot) 11: 12:      // Step 2: Greedy merge-coalesce compatible windows 13:      windows[P] ← GREEDY_MERGE(windows[P]) 14: 15:      // Step 3: Offset alignment-adjust timing for merging 16:      if COUNT_ENTRIES(windows[P]) > C_cap(S) then 17:       windows[P] ← ALIGN_OFFSETS(windows[P], C_cap(S)) 18: 19:      // Step 4: Cap enforcement-force merge if needed 20:      if COUNT_ENTRIES(windows[P]) > C_cap(S) then 21:         windows[P] ← FORCE_MERGE_TO_CAP(windows[P], C_cap(S)) 22: 23:    GCL_refined[S] ← RECONSTRUCT(windows) 24: 25:   // Synchronization and delay recomputation 26:   return MERGE_ALL(GCL_refined), RECOMPUTE_DELAYS() 27: function GREEDY_MERGE(windows) 28:     // Sweep-line merge of overlapping/contiguous TT intervals after GAP_FILL/ALIGN_OFFSETS 29:   windows ← SORT_BY_START(windows) 30:   merged ← [] 31:   for each (st, en) in windows do 32:    if merged is not empty and st ≤ merged[-1].end then 33:     merged[-1].end ← max(merged[-1].end, en) 34:    else 35:     merged.append((st, en)) 36:   return merged