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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
|
/*
* Copyright 2008-2013 NVIDIA Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#pragma once
#include <thrust/detail/config.h>
// don't attempt to #include this file without omp support
#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
#include <omp.h>
#endif // omp support
#include <thrust/iterator/iterator_traits.h>
#include <thrust/system/omp/detail/default_decomposition.h>
#include <thrust/system/detail/generic/select_system.h>
#include <thrust/sort.h>
#include <thrust/merge.h>
#include <thrust/detail/seq.h>
#include <thrust/detail/temporary_array.h>
THRUST_NAMESPACE_BEGIN
namespace system
{
namespace omp
{
namespace detail
{
namespace sort_detail
{
template<typename DerivedPolicy,
typename RandomAccessIterator,
typename StrictWeakOrdering>
void inplace_merge(execution_policy<DerivedPolicy> &exec,
RandomAccessIterator first,
RandomAccessIterator middle,
RandomAccessIterator last,
StrictWeakOrdering comp)
{
typedef typename thrust::iterator_value<RandomAccessIterator>::type value_type;
thrust::detail::temporary_array<value_type,DerivedPolicy> a(exec, first, middle);
thrust::detail::temporary_array<value_type,DerivedPolicy> b(exec, middle, last);
thrust::merge(thrust::seq, a.begin(), a.end(), b.begin(), b.end(), first, comp);
}
template<typename DerivedPolicy,
typename RandomAccessIterator1,
typename RandomAccessIterator2,
typename StrictWeakOrdering>
void inplace_merge_by_key(execution_policy<DerivedPolicy> &exec,
RandomAccessIterator1 first1,
RandomAccessIterator1 middle1,
RandomAccessIterator1 last1,
RandomAccessIterator2 first2,
StrictWeakOrdering comp)
{
typedef typename thrust::iterator_value<RandomAccessIterator1>::type value_type1;
typedef typename thrust::iterator_value<RandomAccessIterator2>::type value_type2;
RandomAccessIterator2 middle2 = first2 + (middle1 - first1);
RandomAccessIterator2 last2 = first2 + (last1 - first1);
thrust::detail::temporary_array<value_type1,DerivedPolicy> lhs1(exec, first1, middle1);
thrust::detail::temporary_array<value_type1,DerivedPolicy> rhs1(exec, middle1, last1);
thrust::detail::temporary_array<value_type2,DerivedPolicy> lhs2(exec, first2, middle2);
thrust::detail::temporary_array<value_type2,DerivedPolicy> rhs2(exec, middle2, last2);
thrust::merge_by_key(thrust::seq,
lhs1.begin(), lhs1.end(),
rhs1.begin(), rhs1.end(),
lhs2.begin(), rhs2.begin(),
first1, first2,
comp);
}
} // end sort_detail
template<typename DerivedPolicy,
typename RandomAccessIterator,
typename StrictWeakOrdering>
void stable_sort(execution_policy<DerivedPolicy> &exec,
RandomAccessIterator first,
RandomAccessIterator last,
StrictWeakOrdering comp)
{
// we're attempting to launch an omp kernel, assert we're compiling with omp support
// ========================================================================
// X Note to the user: If you've found this line due to a compiler error, X
// X you need to enable OpenMP support in your compiler. X
// ========================================================================
THRUST_STATIC_ASSERT_MSG(
(thrust::detail::depend_on_instantiation<
RandomAccessIterator, (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
>::value)
, "OpenMP compiler support is not enabled"
);
// Avoid issues on compilers that don't provide `omp_get_num_threads()`.
#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
typedef typename thrust::iterator_difference<RandomAccessIterator>::type IndexType;
if(first == last)
return;
THRUST_PRAGMA_OMP(parallel)
{
thrust::system::detail::internal::uniform_decomposition<IndexType> decomp(last - first, 1, omp_get_num_threads());
// process id
IndexType p_i = omp_get_thread_num();
// every thread sorts its own tile
if(p_i < decomp.size())
{
thrust::stable_sort(thrust::seq,
first + decomp[p_i].begin(),
first + decomp[p_i].end(),
comp);
}
THRUST_PRAGMA_OMP(barrier)
// XXX For some reason, MSVC 2015 yields an error unless we include this meaningless semicolon here
;
IndexType nseg = decomp.size();
IndexType h = 2;
// keep track of which sub-range we're processing
IndexType a=p_i, b=p_i, c=p_i+1;
while(nseg>1)
{
if(c >= decomp.size())
c = decomp.size() - 1;
if((p_i % h) == 0 && c > b)
{
sort_detail::inplace_merge(exec,
first + decomp[a].begin(),
first + decomp[b].end(),
first + decomp[c].end(),
comp);
b = c;
c += h;
}
nseg = (nseg + 1) / 2;
h *= 2;
THRUST_PRAGMA_OMP(barrier)
}
}
#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
}
template<typename DerivedPolicy,
typename RandomAccessIterator1,
typename RandomAccessIterator2,
typename StrictWeakOrdering>
void stable_sort_by_key(execution_policy<DerivedPolicy> &exec,
RandomAccessIterator1 keys_first,
RandomAccessIterator1 keys_last,
RandomAccessIterator2 values_first,
StrictWeakOrdering comp)
{
// we're attempting to launch an omp kernel, assert we're compiling with omp support
// ========================================================================
// X Note to the user: If you've found this line due to a compiler error, X
// X you need to enable OpenMP support in your compiler. X
// ========================================================================
THRUST_STATIC_ASSERT_MSG(
(thrust::detail::depend_on_instantiation<
RandomAccessIterator1, (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
>::value)
, "OpenMP compiler support is not enabled"
);
// Avoid issues on compilers that don't provide `omp_get_num_threads()`.
#if (THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE == THRUST_TRUE)
typedef typename thrust::iterator_difference<RandomAccessIterator1>::type IndexType;
if(keys_first == keys_last)
return;
THRUST_PRAGMA_OMP(parallel)
{
thrust::system::detail::internal::uniform_decomposition<IndexType> decomp(keys_last - keys_first, 1, omp_get_num_threads());
// process id
IndexType p_i = omp_get_thread_num();
// every thread sorts its own tile
if(p_i < decomp.size())
{
thrust::stable_sort_by_key(thrust::seq,
keys_first + decomp[p_i].begin(),
keys_first + decomp[p_i].end(),
values_first + decomp[p_i].begin(),
comp);
}
THRUST_PRAGMA_OMP(barrier)
// XXX For some reason, MSVC 2015 yields an error unless we include this meaningless semicolon here
;
IndexType nseg = decomp.size();
IndexType h = 2;
// keep track of which sub-range we're processing
IndexType a=p_i, b=p_i, c=p_i+1;
while(nseg>1)
{
if(c >= decomp.size())
c = decomp.size() - 1;
if((p_i % h) == 0 && c > b)
{
sort_detail::inplace_merge_by_key(exec,
keys_first + decomp[a].begin(),
keys_first + decomp[b].end(),
keys_first + decomp[c].end(),
values_first + decomp[a].begin(),
comp);
b = c;
c += h;
}
nseg = (nseg + 1) / 2;
h *= 2;
THRUST_PRAGMA_OMP(barrier)
}
}
#endif // THRUST_DEVICE_COMPILER_IS_OMP_CAPABLE
}
} // end namespace detail
} // end namespace omp
} // end namespace system
THRUST_NAMESPACE_END
|
