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
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
|
/*
* -----------------------------------------------------------------
* Programmer(s): Slaven Peles @ LLNL
* -----------------------------------------------------------------
* LLNS Copyright Start
* Copyright (c) 2014, Lawrence Livermore National Security
* This work was performed under the auspices of the U.S. Department
* of Energy by Lawrence Livermore National Laboratory in part under
* Contract W-7405-Eng-48 and in part under Contract DE-AC52-07NA27344.
* Produced at the Lawrence Livermore National Laboratory.
* All rights reserved.
* For details, see the LICENSE file.
* LLNS Copyright End
* -----------------------------------------------------------------
*/
#ifndef _THREAD_PARTITIONING_HPP_
#define _THREAD_PARTITIONING_HPP_
#include <iostream>
#include <cuda_runtime.h>
namespace suncudavec
{
template<class T, class I>
class StreamPartitioning
{
public:
StreamPartitioning(I N, unsigned block)
: block_(block),
grid_((N + block - 1) / block)
{
}
explicit StreamPartitioning(StreamPartitioning<T, I>& p)
: block_(p.block_),
grid_(p.grid_)
{
}
unsigned grid() const
{
return grid_;
}
unsigned block() const
{
return block_;
}
private:
unsigned block_;
unsigned grid_;
};
template<class T, class I=int>
class ReducePartitioning
{
public:
ReducePartitioning(I N, unsigned block)
: block_(block),
grid_((N + (block_ * 2 - 1)) / (block_ * 2)),
shMemSize_(block_*sizeof(T))
{
allocateBuffer();
}
explicit ReducePartitioning(StreamPartitioning<T, I>& p)
: block_(p.block_),
grid_(p.grid_),
shMemSize_(p.shMemSize_)
{
allocateBuffer();
}
~ReducePartitioning()
{
cudaError_t err;
if (bufferSize_ > 0)
free(h_buffer_);
if (bufferSize_ > 0)
{
err = cudaFree(d_buffer_);
if(err != cudaSuccess)
std::cerr << "Failed to free device vector (error code " << err << ")!\n";
}
}
int setPartitioning(I N, unsigned& grid, unsigned& block, unsigned& shMemSize)
{
block = block_;
grid = (N + (block * 2 - 1)) / (block * 2);
shMemSize = block * sizeof(T);
return 0;
}
unsigned grid() const
{
return grid_;
}
unsigned block() const
{
return block_;
}
unsigned shmem() const
{
return shMemSize_;
}
unsigned int buffSize()
{
return bufferSize_;
}
T* devBuffer()
{
return d_buffer_;
}
const T* devBuffer() const
{
return d_buffer_;
}
T* hostBuffer()
{
return h_buffer_;
}
const T* hostBuffer() const
{
return h_buffer_;
}
void copyFromDevBuffer(unsigned int n) const
{
cudaError_t err = cudaMemcpy(h_buffer_, d_buffer_, n*sizeof(T), cudaMemcpyDeviceToHost);
if(err != cudaSuccess)
std::cerr << "Failed to copy vector from device to host (error code " << err << ")!\n";
}
private:
int allocateBuffer()
{
bufferSize_ = grid_ * sizeof(T);
h_buffer_ = static_cast<T*>(malloc(bufferSize_));
if(h_buffer_ == NULL)
std::cerr << "Failed to allocate host vector!\n";
cudaError_t err;
err = cudaMalloc((void**) &d_buffer_, bufferSize_);
if(err != cudaSuccess)
std::cerr << "Failed to allocate device vector (error code " << err << ")!\n";
return 0;
}
private:
unsigned block_;
unsigned grid_;
unsigned shMemSize_;
T* d_buffer_;
T* h_buffer_;
unsigned bufferSize_;
};
} // namespace suncudavec
#endif // _THREAD_PARTITIONING_HPP_
|
