我试图用来tbb::parallel_sort同时排序2个数组.英特尔的文档在这里说https://software.intel.com/en-us/node/506167 The requirements on the iterator and sequence are the same as for std::sort..似乎并非如此.我的自定义迭代器工作得很好,std::sort但产生编译错误tbb::parallel_sort.请看下面的代码:
int main()//needs boost and tbb to compile
{
int values_size = 6;
int nums1[] = {5, 8, 7, 89, 56, 4};
int nums2[] = {2, 1, 1, 4, 9, 2};
//WORKS!
std::sort(do_dual_sort.make_iter(nums1, nums2),
do_dual_sort.make_iter(nums1+values_size, nums2+values_size),
do_dual_sort.make_comp_desc(nums1, nums2));
//DOESN'T COMPILE
tbb::parallel_sort(do_dual_sort.make_iter(nums1, nums2),
do_dual_sort.make_iter(nums1+values_size, nums2+values_size),
do_dual_sort.make_comp_desc(nums1, nums2));
for(unsigned int i = 0; i < values_size; i++) cout << "nums1[" << i << "] " << nums1[i] << " | nums2[" << i << "] " << nums2[i] << "\n";
return 0;
}
class dual_sort
{
public:
template <class T, class T2>
struct helper_type {
public:
typedef boost::tuple<typename iterator_traits<T>::value_type, typename iterator_traits<T2>::value_type> value_type;
typedef boost::tuple<typename iterator_traits<T>::value_type&, typename iterator_traits<T2>::value_type&> ref_type;
};
template <typename T1, typename T2>
class dual_iterator : public boost::iterator_facade<dual_iterator<T1, T2>,
typename helper_type<T1, T2>::value_type,
boost::random_access_traversal_tag,
typename helper_type<T1, T2>::ref_type> {
public:
explicit dual_iterator(T1 iter1, T2 iter2) : mIter1(iter1), mIter2(iter2) {}
typedef typename iterator_traits<T1>::difference_type difference_type;
private:
void increment() { ++mIter1; ++mIter2; }
void decrement() { --mIter1; --mIter2; }
bool equal(dual_iterator const& other) const { return mIter1 == other.mIter1; }
typename helper_type<T1, T2>::ref_type dereference() const { return (typename helper_type<T1, T2>::ref_type(*mIter1, *mIter2)); }
difference_type distance_to(dual_iterator const& other) const { return other.mIter1 - mIter1; }
void advance(difference_type n) { mIter1 += n; mIter2 += n; }
T1 mIter1;
T2 mIter2;
friend class boost::iterator_core_access;
};
template <typename T1, typename T2>
dual_iterator<T1, T2> make_iter(T1 t1, T2 t2) { return dual_iterator<T1, T2>(t1, t2); }
template <class T1, class T2> struct iter_comp_desc {
typedef typename helper_type<T1, T2>::value_type T;
bool operator()(const T& t1, const T& t2) const { return get<0>(t1) > get<0>(t2); }
bool operator()(const char*& t1, const char*& t2) const { return strcmp(get<0>(t1), get<0>(t2)) == 1; }
};
template <class T1, class T2> iter_comp_desc<T1, T2> make_comp_desc(T1 t1, T2 t2) { return iter_comp_desc<T1, T2>(); }
} do_dual_sort;
我得到的编译错误是:
error C2512: 'dual_sort::dual_iterator<T1,T2>' : no appropriate default constructor available
with
[
T1=int *,
T2=int *
]
tbb44_20150728oss\include\tbb/parallel_sort.h(201) : see reference to function template instantiation 'void tbb::internal::parallel_quick_sort<RandomAccessIterator,Compare>(RandomAccessIterator,RandomAccessIterator,const Compare &)' being compiled
with
[
RandomAccessIterator=dual_sort::dual_iterator<int *,int *>,
Compare=dual_sort::iter_comp_desc<int *,int *>
]
main.cpp(1125) : see reference to function template instantiation 'void tbb::parallel_sort<dual_sort::dual_iterator<T1,T2>,dual_sort::iter_comp_desc<T1,T2>>(RandomAccessIterator,RandomAccessIterator,const Compare &)' being compiled
with
[
T1=int *,
T2=int *,
RandomAccessIterator=dual_sort::dual_iterator<int *,int *>,
Compare=dual_sort::iter_comp_desc<int *,int *>
]
编辑:我使用的编译器是Visual Studio 2012.您可以尝试将一些boost函数替换为一些std以使其正常工作g++.
该class quick_sort_range中tbb/parallel_sort.h含有RandomAccessIterator begin;部件,其副本初始化一个构造函数和默认初始化,然后在其他的构造函数分配.因此,它需要默认的&可复制构造和可分配的迭代器.
因此,TBB文档声称具有相同的要求是正确的,std::sort因为后者只需要随机访问迭代器,这些迭代器不需要可分配,而TBB实现需要版本<= 4.4.
可以修复默认可构造和可分配的需求,但仍可能保留移动或可复制构造(使文档中的声明正确).您可以在TBB论坛上报告此问题.
您可以安全地将默认和复制构造函数和赋值运算符添加到您的代码中,以便tbb::parallel_sort我尽可能地编译它.
以下是使用您的代码段的在线编译器:http://coliru.stacked-crooked.com/a/47dafd091d36a9c4
对于RandomAccessIterator,reference必须是对它的引用value_type.它不能是引用的元组.
因此,您的双迭代器不是有效的RandomAccessIterator.
许多算法仍然有效,但这并不能使您的代码有效.
要求相同并不意味着任何适用于给定实现的std::sort内容也将起作用tbb::parallel_sort:给定的实现std::sort不必强制执行标准中的所有要求.
无论文档如何,如果实现不适用于您的代码,它将无法与您的代码一起使用.
最简单的方法可能是在原始数组中创建一对伪索引(或迭代器)数组,然后对其进行排序.你只需要<正确覆盖.