1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
|
// Copyright John Maddock 2016.
// Use, modification and distribution are subject to the
// Boost Software License, Version 1.0. (See accompanying file
// LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
#define BOOST_MATH_OVERFLOW_ERROR_POLICY ignore_error
#include <iostream>
#include <iomanip>
#include <vector>
#include <boost/math/special_functions/beta.hpp>
#include <boost/math/special_functions/relative_difference.hpp>
#include <boost/array.hpp>
#include "cuda_managed_ptr.hpp"
#include "stopwatch.hpp"
// For the CUDA runtime routines (prefixed with "cuda_")
#include <cuda_runtime.h>
typedef float float_type;
/**
* CUDA Kernel Device code
*
*/
__global__ void cuda_test(const float_type *in1, const float_type * in2, float_type *out, int numElements)
{
using std::cos;
int i = blockDim.x * blockIdx.x + threadIdx.x;
if (i < numElements)
{
out[i] = boost::math::beta(in1[i], in2[i]);
}
}
template <class T> struct table_type { typedef T type; };
typedef float_type T;
#define SC_(x) static_cast<T>(x)
#include "beta_med_data.ipp"
#include "beta_small_data.ipp"
/**
* Host main routine
*/
int main(void)
{
try{
// Consolidate the test data:
std::vector<float_type> v1, v2;
for(unsigned i = 0; i < beta_med_data.size(); ++i)
{
v1.push_back(beta_med_data[i][0]);
v2.push_back(beta_med_data[i][1]);
}
for(unsigned i = 0; i < beta_small_data.size(); ++i)
{
v1.push_back(beta_small_data[i][0]);
v2.push_back(beta_small_data[i][1]);
}
// Error code to check return values for CUDA calls
cudaError_t err = cudaSuccess;
// Print the vector length to be used, and compute its size
int numElements = 50000;
std::cout << "[Vector operation on " << numElements << " elements]" << std::endl;
// Allocate the managed input vector A
cuda_managed_ptr<float_type> input_vector1(numElements);
cuda_managed_ptr<float_type> input_vector2(numElements);
// Allocate the managed output vector C
cuda_managed_ptr<float_type> output_vector(numElements);
// Initialize the input vectors
for (int i = 0; i < numElements; ++i)
{
int table_id = i % v1.size();
input_vector1[i] = v1[table_id];
input_vector2[i] = v2[table_id];
}
// Launch the Vector Add CUDA Kernel
int threadsPerBlock = 1024;
int blocksPerGrid =(numElements + threadsPerBlock - 1) / threadsPerBlock;
std::cout << "CUDA kernel launch with " << blocksPerGrid << " blocks of " << threadsPerBlock << " threads" << std::endl;
watch w;
cuda_test<<<blocksPerGrid, threadsPerBlock>>>(input_vector1.get(), input_vector2.get(), output_vector.get(), numElements);
cudaDeviceSynchronize();
std::cout << "CUDA kernal done in " << w.elapsed() << "s" << std::endl;
err = cudaGetLastError();
if (err != cudaSuccess)
{
std::cerr << "Failed to launch vectorAdd kernel (error code " << cudaGetErrorString(err) << ")!" << std::endl;
return EXIT_FAILURE;
}
// Verify that the result vector is correct
std::vector<float_type> results;
results.reserve(numElements);
w.reset();
for(int i = 0; i < numElements; ++i)
results.push_back(boost::math::beta(input_vector1[i], input_vector2[i]));
double t = w.elapsed();
// check the results
for(int i = 0; i < numElements; ++i)
{
if (boost::math::epsilon_difference(output_vector[i], results[i]) > 300)
{
std::cerr << "Result verification failed at element " << i << "!" << std::endl;
std::cerr << "Error rate was: " << boost::math::epsilon_difference(output_vector[i], results[i]) << "eps" << std::endl;
return EXIT_FAILURE;
}
}
std::cout << "Test PASSED with calculation time: " << t << "s" << std::endl;
std::cout << "Done\n";
}
catch(const std::exception& e)
{
std::cerr << "Stopped with exception: " << e.what() << std::endl;
}
return 0;
}
|
