• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Contains get_min_count, the core optimization of the spreadsort algorithm.
2 // Also has other helper functions commonly useful across variants.
3 
4 //          Copyright Steven J. Ross 2001 - 2014.
5 // Distributed under the Boost Software License, Version 1.0.
6 //    (See accompanying file LICENSE_1_0.txt or copy at
7 //          http://www.boost.org/LICENSE_1_0.txt)
8 
9 // See http://www.boost.org/libs/sort for library home page.
10 
11 /*
12 Some improvements suggested by:
13 Phil Endecott and Frank Gennari
14 */
15 
16 #ifndef BOOST_SORT_SPREADSORT_DETAIL_SPREAD_SORT_COMMON_HPP
17 #define BOOST_SORT_SPREADSORT_DETAIL_SPREAD_SORT_COMMON_HPP
18 #include <algorithm>
19 #include <vector>
20 #include <cstring>
21 #include <limits>
22 #include <functional>
23 #include <boost/static_assert.hpp>
24 #include <boost/sort/pdqsort/pdqsort.hpp>
25 #include <boost/sort/spreadsort/detail/constants.hpp>
26 #include <boost/cstdint.hpp>
27 
28 namespace boost {
29 namespace sort {
30 namespace spreadsort {
31  namespace detail {
32     //This only works on unsigned data types
33     template <typename T>
34     inline unsigned
rough_log_2_size(const T & input)35     rough_log_2_size(const T& input)
36     {
37       unsigned result = 0;
38       //The && is necessary on some compilers to avoid infinite loops
39       //it doesn't significantly impair performance
40       while ((result < (8*sizeof(T))) && (input >> result)) ++result;
41       return result;
42     }
43 
44     //Gets the minimum size to call spreadsort on to control worst-case runtime.
45     //This is called for a set of bins, instead of bin-by-bin, to minimize
46     //runtime overhead.
47     //This could be replaced by a lookup table of sizeof(Div_type)*8 but this
48     //function is more general.
49     template<unsigned log_mean_bin_size,
50       unsigned log_min_split_count, unsigned log_finishing_count>
51     inline size_t
get_min_count(unsigned log_range)52     get_min_count(unsigned log_range)
53     {
54       const size_t typed_one = 1;
55       const unsigned min_size = log_mean_bin_size + log_min_split_count;
56       //Assuring that constants have valid settings
57       BOOST_STATIC_ASSERT(log_min_split_count <= max_splits &&
58                           log_min_split_count > 0);
59       BOOST_STATIC_ASSERT(max_splits > 1 &&
60                           max_splits < (8 * sizeof(unsigned)));
61       BOOST_STATIC_ASSERT(max_finishing_splits >= max_splits &&
62                           max_finishing_splits < (8 * sizeof(unsigned)));
63       BOOST_STATIC_ASSERT(log_mean_bin_size >= 0);
64       BOOST_STATIC_ASSERT(log_finishing_count >= 0);
65       //if we can complete in one iteration, do so
66       //This first check allows the compiler to optimize never-executed code out
67       if (log_finishing_count < min_size) {
68         if (log_range <= min_size && log_range <= max_splits) {
69           //Return no smaller than a certain minimum limit
70           if (log_range <= log_finishing_count)
71             return typed_one << log_finishing_count;
72           return typed_one << log_range;
73         }
74       }
75       const unsigned base_iterations = max_splits - log_min_split_count;
76       //sum of n to n + x = ((x + 1) * (n + (n + x)))/2 + log_mean_bin_size
77       const unsigned base_range =
78           ((base_iterations + 1) * (max_splits + log_min_split_count))/2
79           + log_mean_bin_size;
80       //Calculating the required number of iterations, and returning
81       //1 << (iteration_count + min_size)
82       if (log_range < base_range) {
83         unsigned result = log_min_split_count;
84         for (unsigned offset = min_size; offset < log_range;
85           offset += ++result);
86         //Preventing overflow; this situation shouldn't occur
87         if ((result + log_mean_bin_size) >= (8 * sizeof(size_t)))
88           return typed_one << ((8 * sizeof(size_t)) - 1);
89         return typed_one << (result + log_mean_bin_size);
90       }
91       //A quick division can calculate the worst-case runtime for larger ranges
92       unsigned remainder = log_range - base_range;
93       //the max_splits - 1 is used to calculate the ceiling of the division
94       unsigned bit_length = ((((max_splits - 1) + remainder)/max_splits)
95         + base_iterations + min_size);
96       //Preventing overflow; this situation shouldn't occur
97       if (bit_length >= (8 * sizeof(size_t)))
98         return typed_one << ((8 * sizeof(size_t)) - 1);
99       //n(log_range)/max_splits + C, optimizing worst-case performance
100       return typed_one << bit_length;
101     }
102 
103     // Resizes the bin cache and bin sizes, and initializes each bin size to 0.
104     // This generates the memory overhead to use in radix sorting.
105     template <class RandomAccessIter>
106     inline RandomAccessIter *
size_bins(size_t * bin_sizes,std::vector<RandomAccessIter> & bin_cache,unsigned cache_offset,unsigned & cache_end,unsigned bin_count)107     size_bins(size_t *bin_sizes, std::vector<RandomAccessIter>
108   &bin_cache, unsigned cache_offset, unsigned &cache_end, unsigned bin_count)
109     {
110       // Clear the bin sizes
111       for (size_t u = 0; u < bin_count; u++)
112         bin_sizes[u] = 0;
113       //Make sure there is space for the bins
114       cache_end = cache_offset + bin_count;
115       if (cache_end > bin_cache.size())
116         bin_cache.resize(cache_end);
117       return &(bin_cache[cache_offset]);
118     }
119   }
120 }
121 }
122 }
123 
124 #endif
125