1 //---------------------------------------------------------------------------//
2 // Copyright (c) 2015 Jakub Szuppe <j.szuppe@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 <cstdlib>
13 #include <iostream>
14
15 #include <thrust/copy.h>
16 #include <thrust/device_vector.h>
17 #include <thrust/generate.h>
18 #include <thrust/host_vector.h>
19 #include <thrust/reduce.h>
20
21 #include "perf.hpp"
22
rand_int()23 int rand_int()
24 {
25 return static_cast<int>((rand() / double(RAND_MAX)) * 25.0);
26 }
27
28 struct unique_key {
29 int current;
30 int avgValuesNoPerKey;
31
unique_keyunique_key32 unique_key()
33 {
34 current = 0;
35 avgValuesNoPerKey = 512;
36 }
37
operator ()unique_key38 int operator()()
39 {
40 double p = double(1.0) / static_cast<double>(avgValuesNoPerKey);
41 if((rand() / double(RAND_MAX)) <= p)
42 return ++current;
43 return current;
44 }
45 } UniqueKey;
46
main(int argc,char * argv[])47 int main(int argc, char *argv[])
48 {
49 perf_parse_args(argc, argv);
50
51 std::cout << "size: " << PERF_N << std::endl;
52
53 // create vector of keys and random values
54 thrust::host_vector<int> host_keys(PERF_N);
55 thrust::host_vector<int> host_values(PERF_N);
56 std::generate(host_keys.begin(), host_keys.end(), UniqueKey);
57 std::generate(host_values.begin(), host_values.end(), rand_int);
58
59 // transfer data to the device
60 thrust::device_vector<int> device_keys = host_keys;
61 thrust::device_vector<int> device_values = host_values;
62
63 // create device vectors for the results
64 thrust::device_vector<int> device_keys_results(PERF_N);
65 thrust::device_vector<int> device_values_results(PERF_N);
66
67 typedef typename thrust::device_vector<int>::iterator iterType;
68 thrust::pair<iterType, iterType> result;
69
70 perf_timer t;
71 for(size_t trial = 0; trial < PERF_TRIALS; trial++){
72 t.start();
73 result = thrust::reduce_by_key(device_keys.begin(),
74 device_keys.end(),
75 device_values.begin(),
76 device_keys_results.begin(),
77 device_values_results.begin());
78 cudaDeviceSynchronize();
79 t.stop();
80 }
81 std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;
82
83 size_t result_size = thrust::distance(device_keys_results.begin(), result.first);
84 if(result_size != static_cast<size_t>(host_keys[PERF_N-1] + 1)){
85 std::cout << "ERROR: "
86 << "wrong number of keys"
87 << std::endl;
88 return -1;
89 }
90
91 return 0;
92 }
93