1 //---------------------------------------------------------------------------//
2 // Copyright (c) 2014 Kyle Lutz <kyle.r.lutz@gmail.com>
3 //
4 // Distributed under the Boost Software License, Version 1.0
5 // See accompanying file LICENSE_1_0.txt or copy at
6 // http://www.boost.org/LICENSE_1_0.txt
7 //
8 // See http://boostorg.github.com/compute for more information.
9 //---------------------------------------------------------------------------//
10
11 #include <algorithm>
12 #include <iostream>
13 #include <vector>
14
15 #include <tbb/parallel_for.h>
16
17 #include "perf.hpp"
18
19 // example from: http://www.threadingbuildingblocks.org/docs/help/reference/algorithms/parallel_for_func.htm
20 using namespace tbb;
21
22 template<typename Iterator>
23 struct ParallelMergeRange {
24 static size_t grainsize;
25 Iterator begin1, end1; // [begin1,end1) is 1st sequence to be merged
26 Iterator begin2, end2; // [begin2,end2) is 2nd sequence to be merged
27 Iterator out; // where to put merged sequence
emptyParallelMergeRange28 bool empty() const {return (end1-begin1)+(end2-begin2)==0;}
is_divisibleParallelMergeRange29 bool is_divisible() const {
30 return (std::min)( end1-begin1, end2-begin2 ) > grainsize;
31 }
ParallelMergeRangeParallelMergeRange32 ParallelMergeRange( ParallelMergeRange& r, split ) {
33 if( r.end1-r.begin1 < r.end2-r.begin2 ) {
34 std::swap(r.begin1,r.begin2);
35 std::swap(r.end1,r.end2);
36 }
37 Iterator m1 = r.begin1 + (r.end1-r.begin1)/2;
38 Iterator m2 = std::lower_bound( r.begin2, r.end2, *m1 );
39 begin1 = m1;
40 begin2 = m2;
41 end1 = r.end1;
42 end2 = r.end2;
43 out = r.out + (m1-r.begin1) + (m2-r.begin2);
44 r.end1 = m1;
45 r.end2 = m2;
46 }
ParallelMergeRangeParallelMergeRange47 ParallelMergeRange( Iterator begin1_, Iterator end1_,
48 Iterator begin2_, Iterator end2_,
49 Iterator out_ ) :
50 begin1(begin1_), end1(end1_),
51 begin2(begin2_), end2(end2_), out(out_)
52 {}
53 };
54
55 template<typename Iterator>
56 size_t ParallelMergeRange<Iterator>::grainsize = 1000;
57
58 template<typename Iterator>
59 struct ParallelMergeBody {
operator ()ParallelMergeBody60 void operator()( ParallelMergeRange<Iterator>& r ) const {
61 std::merge( r.begin1, r.end1, r.begin2, r.end2, r.out );
62 }
63 };
64
65 template<typename Iterator>
ParallelMerge(Iterator begin1,Iterator end1,Iterator begin2,Iterator end2,Iterator out)66 void ParallelMerge( Iterator begin1, Iterator end1, Iterator begin2, Iterator end2, Iterator out ) {
67 parallel_for(
68 ParallelMergeRange<Iterator>(begin1,end1,begin2,end2,out),
69 ParallelMergeBody<Iterator>(),
70 simple_partitioner()
71 );
72 }
73
main(int argc,char * argv[])74 int main(int argc, char *argv[])
75 {
76 perf_parse_args(argc, argv);
77
78 std::cout << "size: " << PERF_N << std::endl;
79 std::vector<int> v1 = generate_random_vector<int>(PERF_N / 2);
80 std::vector<int> v2 = generate_random_vector<int>(PERF_N / 2);
81 std::vector<int> v3(PERF_N);
82
83 std::sort(v1.begin(), v1.end());
84 std::sort(v2.begin(), v2.end());
85
86 perf_timer t;
87 for(size_t trial = 0; trial < PERF_TRIALS; trial++){
88 t.start();
89 ParallelMerge(v1.begin(), v1.end(), v2.begin(), v2.end(), v3.begin());
90 t.stop();
91 }
92 std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;
93
94 return 0;
95 }
96