1 //---------------------------------------------------------------------------//
2 // Copyright (c) 2013-2014 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 <iostream>
12 #include <iterator>
13 #include <algorithm>
14
15 #include <thrust/device_vector.h>
16 #include <thrust/functional.h>
17 #include <thrust/host_vector.h>
18 #include <thrust/transform.h>
19
20 #include "perf.hpp"
21
22 struct saxpy_functor : public thrust::binary_function<float,float,float>
23 {
24 const float a;
25
saxpy_functorsaxpy_functor26 saxpy_functor(float _a) : a(_a) {}
27
28 __host__ __device__
operator ()saxpy_functor29 float operator()(const float& x, const float& y) const
30 {
31 return a * x + y;
32 }
33 };
34
main(int argc,char * argv[])35 int main(int argc, char *argv[])
36 {
37 perf_parse_args(argc, argv);
38
39 std::cout << "size: " << PERF_N << std::endl;
40 thrust::host_vector<int> host_x(PERF_N);
41 thrust::host_vector<int> host_y(PERF_N);
42 std::generate(host_x.begin(), host_x.end(), rand);
43 std::generate(host_y.begin(), host_y.end(), rand);
44
45 // transfer data to the device
46 thrust::device_vector<int> device_x = host_x;
47 thrust::device_vector<int> device_y = host_y;
48
49 perf_timer t;
50 for(size_t trial = 0; trial < PERF_TRIALS; trial++){
51 t.start();
52 thrust::transform(device_x.begin(), device_x.end(), device_y.begin(), device_y.begin(), saxpy_functor(2.5f));
53 cudaDeviceSynchronize();
54 t.stop();
55 }
56 std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;
57
58 // transfer data back to host
59 thrust::copy(device_x.begin(), device_x.end(), host_x.begin());
60 thrust::copy(device_y.begin(), device_y.end(), host_y.begin());
61
62 return 0;
63 }
64