1: template<class _RanIt> inline
2: void sort(_RanIt _First, _RanIt _Last)
3: { // order [_First, _Last), using operator<
4: _DEBUG_RANGE(_First, _Last);
5: _Sort(_CHECKED_BASE(_First), _CHECKED_BASE(_Last), _Last - _First);
6: }
1: template<class _RanIt,
2: class _Diff,
3: class _Pr> inline
4: void _Sort(_RanIt _First, _RanIt _Last, _Diff _Ideal, _Pr _Pred)
5: {
6: // order [_First, _Last), using _Pred
7: _Diff _Count;
8: for (; _ISORT_MAX < (_Count = _Last - _First) && 0 < _Ideal; )
9: {
10: // divide and conquer by quicksort
11:
12: // First, we get the middle position, divide the set to two part
13: pair<_RanIt, _RanIt> _Mid =
14: _Unguarded_partition(_First, _Last, _Pred);
15:
16:
17: _Ideal /= 2, _Ideal += _Ideal / 2; // allow 1.5 log2(N) divisions
18:
19: if (_Mid.first - _First < _Last - _Mid.second) // loop on larger half
20: _Sort(_First, _Mid.first, _Ideal, _Pred), _First = _Mid.second;
21: else
22: _Sort(_Mid.second, _Last, _Ideal, _Pred), _Last = _Mid.first;
23: }
24:
25: if (_ISORT_MAX < _Count)
26: {
27: // heap sort if too many divisions
28: std::make_heap(_First, _Last, _Pred);
29: std::sort_heap(_First, _Last, _Pred);
30: }
31: else if (1 < _Count)
32: _Insertion_sort(_First, _Last, _Pred); // small, insertion sort
33: }
其中
1: // COMMON SORT PARAMETERS
2: const int _ISORT_MAX = 32; // maximum size for insertion sort
1. _ISORT_MAX用来控制能够conquer的集合的大小,如果集合已经足够小了,比如这里集合中元素的个数小于32个,则可以使用Insertion_Sort来进行排序操作了;
2. _Ideal用来控制(Divide-and-Conquer)递归的层次,
相当 (N * 1.5_Ideal) >> (_Ideal) 次就达到了0,即2_Ideal == N*1.5_Ideal,
即_Ideal = log4/3N, 但是实际上由于在“除2”操作过程中的低位损耗,_Ideal的值会比公式计算出来的偏小,不过原comment给出的1.5 log2(N)是不准确的,以下做了一些对比
1: int x = 1024;
2: int y = 0;
3: while (x > 0)
4: {
5: x /= 2, x += x / 2;
6: printf("0x%08X ", x);
7: y++;
8: }
N | 1.5 log2(N) | log4/3N | 程序计算出的值y |
1024 | 15 | 24 | 21 |
言归正传,当_Ideal的层次太深的时候,for循环结束
3. 如果不是因为分块已经足够小了而结束的,就调用Heap_Sort进行堆排序
4. 如果分块已经都足够小了,就进行Insertion_Sort排序