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
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
|
/*
* Copyright 2008-2009 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.
*/
#include <cusp/detail/device/generalized_spmv/csr_scalar.h>
#include <cusp/copy.h>
#include <cusp/array1d.h>
#include <cusp/exception.h>
#include <cusp/coo_matrix.h>
#include <cusp/detail/random.h>
#include <cusp/detail/format_utils.h>
#include <thrust/count.h>
#include <thrust/transform.h>
#include <thrust/transform_scan.h>
#include <thrust/iterator/constant_iterator.h>
#include <thrust/iterator/counting_iterator.h>
#include <thrust/iterator/zip_iterator.h>
namespace cusp
{
namespace graph
{
namespace detail
{
namespace device
{
struct process_mis_nodes
{
template <typename Tuple>
__host__ __device__
void operator()(Tuple t)
{
if (thrust::get<1>(t) == 1) // undecided node
{
if (thrust::get<0>(t) == thrust::get<3>(t)) // i == maximal_index
thrust::get<1>(t) = 2; // mis_node
}
}
};
struct process_non_mis_nodes
{
template <typename Tuple>
__host__ __device__
void operator()(Tuple t)
{
if (thrust::get<0>(t) == 1) // undecided node
{
if (thrust::get<1>(t) == 2) // maximal_state == mis_node
thrust::get<0>(t) = 0; // non_mis_node
}
}
};
struct is_subgraph_edge
{
template <typename Tuple>
__host__ __device__
bool operator()(const Tuple& t) const
{
return thrust::get<0>(t) && thrust::get<1>(t);
}
};
template <typename NodeStateType>
struct is_active_node
{
__host__ __device__
bool operator()(const NodeStateType& s) const
{
return s == 1;
}
};
template <typename Array1,
typename Array2,
typename Array3,
typename Array4>
void compute_mis_states(const size_t k,
const Array1& row_indices,
const Array2& column_indices,
const Array3& random_values,
Array4& states)
{
typedef typename Array1::value_type IndexType;
typedef typename Array3::value_type RandomType;
typedef typename Array4::value_type NodeStateType;
typedef typename Array1::memory_space MemorySpace;
typedef typename thrust::tuple<NodeStateType,RandomType,IndexType> Tuple;
const size_t N = states.size();
const IndexType num_rows = states.size();
//const IndexType num_entries = row_indices.size();
// TODO remove this WAR when generalize COO SpMV problem is resolved
cusp::array1d<IndexType,MemorySpace> row_offsets(num_rows + 1);
cusp::detail::indices_to_offsets(row_indices, row_offsets);
cusp::array1d<NodeStateType,MemorySpace> maximal_states(N);
cusp::array1d<RandomType,MemorySpace> maximal_values(N);
cusp::array1d<IndexType,MemorySpace> maximal_indices(N);
cusp::array1d<NodeStateType,MemorySpace> last_states;
cusp::array1d<RandomType,MemorySpace> last_values;
cusp::array1d<IndexType,MemorySpace> last_indices;;
// TODO choose threshold in a more principled manner
// size_t compaction_threshold = (N < 10000) ? 0 : N / 10;
size_t active_nodes = N;
// size_t num_iters = 0;
do
{
// find the largest (state,value,index) 1-ring neighbor for each node
cusp::detail::device::cuda::spmv_csr_scalar
(num_rows,
row_offsets.begin(), column_indices.begin(), thrust::constant_iterator<Tuple>(Tuple(0,0)), // XXX should we mask explicit zeros? (e.g. DIA, array2d)
thrust::make_zip_iterator(thrust::make_tuple(states.begin(), random_values.begin(), thrust::counting_iterator<IndexType>(0))),
thrust::make_zip_iterator(thrust::make_tuple(states.begin(), random_values.begin(), thrust::counting_iterator<IndexType>(0))),
thrust::make_zip_iterator(thrust::make_tuple(maximal_states.begin(), maximal_values.begin(), maximal_indices.begin())),
thrust::project2nd<Tuple,Tuple>(), thrust::maximum<Tuple>());
//cusp::detail::device::cuda::spmv_coo
// (num_rows, num_entries,
// row_indices.begin(), column_indices.begin(), thrust::constant_iterator<Tuple>(Tuple(0,0)), // XXX should we mask explicit zeros? (e.g. DIA, array2d)
// thrust::make_zip_iterator(thrust::make_tuple(states.begin(), random_values.begin(), thrust::counting_iterator<IndexType>(0))),
// thrust::make_zip_iterator(thrust::make_tuple(states.begin(), random_values.begin(), thrust::counting_iterator<IndexType>(0))),
// thrust::make_zip_iterator(thrust::make_tuple(maximal_states.begin(), maximal_values.begin(), maximal_indices.begin())),
// thrust::project2nd<Tuple,Tuple>(), thrust::maximum<Tuple>());
// find the largest (state,value,index) k-ring neighbor for each node (if k > 1)
for(size_t ring = 1; ring < k; ring++)
{
last_states.resize (N); last_states.swap (maximal_states);
last_values.resize (N); last_values.swap (maximal_values);
last_indices.resize(N); last_indices.swap(maximal_indices);
// TODO replace with call to generalized method
cusp::detail::device::cuda::spmv_csr_scalar
(num_rows,
row_offsets.begin(), column_indices.begin(), thrust::constant_iterator<Tuple>(Tuple(0,0)), // XXX should we mask explicit zeros? (e.g. DIA, array2d)
thrust::make_zip_iterator(thrust::make_tuple(last_states.begin(), last_values.begin(), last_indices.begin())),
thrust::make_zip_iterator(thrust::make_tuple(last_states.begin(), last_values.begin(), last_indices.begin())),
thrust::make_zip_iterator(thrust::make_tuple(maximal_states.begin(), maximal_values.begin(), maximal_indices.begin())),
thrust::project2nd<Tuple,Tuple>(), thrust::maximum<Tuple>());
}
// label local maxima as MIS nodes
thrust::for_each(thrust::make_zip_iterator(thrust::make_tuple(thrust::counting_iterator<IndexType>(0), states.begin(), maximal_states.begin(), maximal_indices.begin())),
thrust::make_zip_iterator(thrust::make_tuple(thrust::counting_iterator<IndexType>(0), states.begin(), maximal_states.begin(), maximal_indices.begin())) + N,
process_mis_nodes());
// label k-ring neighbors of MIS nodes as non-MIS nodes
thrust::for_each(thrust::make_zip_iterator(thrust::make_tuple(states.begin(), thrust::make_permutation_iterator(states.begin(), maximal_indices.begin()))),
thrust::make_zip_iterator(thrust::make_tuple(states.begin(), thrust::make_permutation_iterator(states.begin(), maximal_indices.begin()))) + N,
process_non_mis_nodes());
active_nodes = thrust::count(states.begin(), states.end(), 1);
// num_iters++;
// std::cout << "(iter " << num_iters << "," << (double(active_nodes) / double(N)) << ")" << std::endl;
// std::cout << "N= " << N << " iteration=" << num_iters << " active_nodes=" << active_nodes << " compaction_threshold=" << compaction_threshold << "\n";
// std::cout << "states\n";
// cusp::print(states);
//
// if (active_nodes < compaction_threshold)
// {
// cusp::array1d<bool,MemorySpace> retained_nodes(N);
// cusp::array1d<bool,MemorySpace> last_retained_nodes(N);
//
// thrust::transform(maximal_states.begin(), maximal_states.end(), thrust::constant_iterator<NodeStateType>(1), retained_nodes.begin(), thrust::equal_to<NodeStateType>());
//
// // propagate retained region outward
// for(size_t ring = 1; 2*ring <= k; ring++)
// {
// retained_nodes.swap(last_retained_nodes);
//
// // TODO replace with call to generalized method
// cusp::detail::device::cuda::spmv_coo
// (num_rows, num_entries,
// row_indices.begin(), column_indices.begin(), thrust::constant_iterator<bool>(false),
// last_retained_nodes.begin(),
// last_retained_nodes.begin(),
// retained_nodes.begin(),
// thrust::project2nd<bool,bool>(), thrust::logical_or<bool>());
// }
//
// std::cout << "retained nodes\n";
// cusp::print(retained_nodes);
//
// size_t num_subgraph_nodes = thrust::count(retained_nodes.begin(), retained_nodes.end(), true);
// size_t num_subgraph_edges = thrust::count
// (thrust::make_zip_iterator(thrust::make_tuple(thrust::make_permutation_iterator(retained_nodes.begin(), row_indices.begin()),
// thrust::make_permutation_iterator(retained_nodes.begin(), column_indices.begin()))),
// thrust::make_zip_iterator(thrust::make_tuple(thrust::make_permutation_iterator(retained_nodes.begin(), row_indices.end()),
// thrust::make_permutation_iterator(retained_nodes.begin(), column_indices.end()))),
// thrust::make_tuple(true,true));
//
//
// std::cout << "subgraph nodes: " << double(100*num_subgraph_nodes)/N << "% edges " << double(100*num_subgraph_edges)/num_entries << "%" << std::endl;
//
// // map old indices into subgraph indices
// cusp::array1d<IndexType, MemorySpace> index_map(N);
// thrust::transform_exclusive_scan(retained_nodes.begin(), retained_nodes.end(), index_map.begin(), thrust::identity<IndexType>(), IndexType(0), thrust::plus<IndexType>());
//
// std::cout << "index map\n";
// cusp::print(index_map);
//
// // storage for subgraph
// cusp::array1d<IndexType, MemorySpace> subgraph_row_indices(num_subgraph_edges);
// cusp::array1d<IndexType, MemorySpace> subgraph_column_indices(num_subgraph_edges);
// cusp::array1d<NodeStateType, MemorySpace> subgraph_states(num_subgraph_nodes);
// cusp::array1d<RandomType, MemorySpace> subgraph_random_values(num_subgraph_nodes);
//
// thrust::copy_if
// (thrust::make_zip_iterator(thrust::make_tuple(thrust::make_permutation_iterator(index_map.begin(), row_indices.begin()),
// thrust::make_permutation_iterator(index_map.begin(), column_indices.begin()))),
// thrust::make_zip_iterator(thrust::make_tuple(thrust::make_permutation_iterator(index_map.begin(), row_indices.end()),
// thrust::make_permutation_iterator(index_map.begin(), column_indices.end()))),
// thrust::make_zip_iterator(thrust::make_tuple(thrust::make_permutation_iterator(retained_nodes.begin(), row_indices.begin()),
// thrust::make_permutation_iterator(retained_nodes.begin(), column_indices.begin()))),
// thrust::make_zip_iterator(thrust::make_tuple(subgraph_row_indices.begin(),
// subgraph_column_indices.begin())),
// is_subgraph_edge());
//
// thrust::scatter_if
// (thrust::make_zip_iterator(thrust::make_tuple(states.begin(), random_values.begin())),
// thrust::make_zip_iterator(thrust::make_tuple(states.end(), random_values.end())),
// index_map.begin(),
// retained_nodes.begin(),
// thrust::make_zip_iterator(thrust::make_tuple(subgraph_states.begin(), subgraph_random_values.begin())));
//
//
// compute_mis_states(k, subgraph_row_indices, subgraph_column_indices, subgraph_random_values, subgraph_states);
//
// // update active node states from subgraph
// thrust::gather_if(index_map.begin(), index_map.end(),
// retained_nodes.begin(),
// subgraph_states.begin(),
// states.begin());
// return;
// }
} while (active_nodes > 0);
}
//////////////////
// Device Paths //
//////////////////
template <typename Matrix, typename Array>
size_t maximal_independent_set(const Matrix& A, Array& stencil, size_t k)
{
typedef typename Matrix::index_type IndexType;
typedef typename Matrix::value_type ValueType;
typedef typename Matrix::memory_space MemorySpace;
typedef unsigned int RandomType;
typedef unsigned char NodeStateType;
const IndexType N = A.num_rows;
cusp::array1d<RandomType,MemorySpace> random_values(N);
cusp::copy(cusp::detail::random_integers<RandomType>(N), random_values);
cusp::array1d<NodeStateType,MemorySpace> states(N, 1);
compute_mis_states(k, A.row_indices, A.column_indices, random_values, states);
// resize output
stencil.resize(N);
// mark all mis nodes
thrust::transform(states.begin(), states.end(), thrust::constant_iterator<NodeStateType>(2), stencil.begin(), thrust::equal_to<NodeStateType>());
// return the size of the MIS
return thrust::count(stencil.begin(), stencil.end(), typename Array::value_type(true));
}
} // end namespace device
} // end namespace detail
} // end namespace graph
} // end namespace cusp
|
