forked from taskflow/taskflow
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtransform.hpp
More file actions
282 lines (222 loc) · 7.13 KB
/
Copy pathtransform.hpp
File metadata and controls
282 lines (222 loc) · 7.13 KB
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
#pragma once
#include "../cudaflow.hpp"
/**
@file taskflow/cuda/algorithm/transform.hpp
@brief cuda parallel-transform algorithms include file
*/
namespace tf {
// ----------------------------------------------------------------------------
// transform
// ----------------------------------------------------------------------------
namespace detail {
/**
@private
*/
template <size_t nt, size_t vt, typename I, typename O, typename C>
__global__ void cuda_transform_kernel(I first, unsigned count, O output, C op) {
auto tid = threadIdx.x;
auto bid = blockIdx.x;
auto tile = cuda_get_tile(bid, nt*vt, count);
cuda_strided_iterate<nt, vt>(
[=]__device__(auto, auto j) {
auto offset = j + tile.begin;
*(output + offset) = op(*(first+offset));
},
tid,
tile.count()
);
}
/**
@private
*/
template <size_t nt, size_t vt, typename I1, typename I2, typename O, typename C>
__global__ void cuda_transform_kernel(
I1 first1, I2 first2, unsigned count, O output, C op
) {
auto tid = threadIdx.x;
auto bid = blockIdx.x;
auto tile = cuda_get_tile(bid, nt*vt, count);
cuda_strided_iterate<nt, vt>(
[=]__device__(auto, auto j) {
auto offset = j + tile.begin;
*(output + offset) = op(*(first1+offset), *(first2+offset));
},
tid,
tile.count()
);
}
} // end of namespace detail -------------------------------------------------
// ----------------------------------------------------------------------------
// CUDA standard algorithms: transform
// ----------------------------------------------------------------------------
/**
@brief performs asynchronous parallel transforms over a range of items
@tparam P execution policy type
@tparam I input iterator type
@tparam O output iterator type
@tparam C unary operator type
@param p execution policy
@param first iterator to the beginning of the range
@param last iterator to the end of the range
@param output iterator to the beginning of the output range
@param op unary operator to apply to transform each item
This method is equivalent to the parallel execution of the following loop on a GPU:
@code{.cpp}
while (first != last) {
*output++ = op(*first++);
}
@endcode
*/
template <typename P, typename I, typename O, typename C>
void cuda_transform(P&& p, I first, I last, O output, C op) {
using E = std::decay_t<P>;
unsigned count = std::distance(first, last);
if(count == 0) {
return;
}
detail::cuda_transform_kernel<E::nt, E::vt, I, O, C>
<<<E::num_blocks(count), E::nt, 0, p.stream()>>> (
first, count, output, op
);
}
/**
@brief performs asynchronous parallel transforms over two ranges of items
@tparam P execution policy type
@tparam I1 first input iterator type
@tparam I2 second input iterator type
@tparam O output iterator type
@tparam C binary operator type
@param p execution policy
@param first1 iterator to the beginning of the first range
@param last1 iterator to the end of the first range
@param first2 iterator to the beginning of the second range
@param output iterator to the beginning of the output range
@param op binary operator to apply to transform each pair of items
This method is equivalent to the parallel execution of the following loop on a GPU:
@code{.cpp}
while (first1 != last1) {
*output++ = op(*first1++, *first2++);
}
@endcode
*/
template <typename P, typename I1, typename I2, typename O, typename C>
void cuda_transform(
P&& p, I1 first1, I1 last1, I2 first2, O output, C op
) {
using E = std::decay_t<P>;
unsigned count = std::distance(first1, last1);
if(count == 0) {
return;
}
detail::cuda_transform_kernel<E::nt, E::vt, I1, I2, O, C>
<<<E::num_blocks(count), E::nt, 0, p.stream()>>> (
first1, first2, count, output, op
);
}
// ----------------------------------------------------------------------------
// cudaFlow
// ----------------------------------------------------------------------------
// Function: transform
template <typename I, typename O, typename C>
cudaTask cudaFlow::transform(I first, I last, O output, C c) {
using E = cudaDefaultExecutionPolicy;
unsigned count = std::distance(first, last);
// TODO:
//if(count == 0) {
// return;
//}
return kernel(
E::num_blocks(count), E::nt, 0,
detail::cuda_transform_kernel<E::nt, E::vt, I, O, C>,
first, count, output, c
);
}
// Function: transform
template <typename I1, typename I2, typename O, typename C>
cudaTask cudaFlow::transform(I1 first1, I1 last1, I2 first2, O output, C c) {
using E = cudaDefaultExecutionPolicy;
unsigned count = std::distance(first1, last1);
// TODO:
//if(count == 0) {
// return;
//}
return kernel(
E::num_blocks(count), E::nt, 0,
detail::cuda_transform_kernel<E::nt, E::vt, I1, I2, O, C>,
first1, first2, count, output, c
);
}
// Function: update transform
template <typename I, typename O, typename C>
void cudaFlow::transform(cudaTask task, I first, I last, O output, C c) {
using E = cudaDefaultExecutionPolicy;
unsigned count = std::distance(first, last);
// TODO:
//if(count == 0) {
// return;
//}
kernel(task,
E::num_blocks(count), E::nt, 0,
detail::cuda_transform_kernel<E::nt, E::vt, I, O, C>,
first, count, output, c
);
}
// Function: update transform
template <typename I1, typename I2, typename O, typename C>
void cudaFlow::transform(
cudaTask task, I1 first1, I1 last1, I2 first2, O output, C c
) {
using E = cudaDefaultExecutionPolicy;
unsigned count = std::distance(first1, last1);
// TODO:
//if(count == 0) {
// return;
//}
kernel(task,
E::num_blocks(count), E::nt, 0,
detail::cuda_transform_kernel<E::nt, E::vt, I1, I2, O, C>,
first1, first2, count, output, c
);
}
// ----------------------------------------------------------------------------
// cudaFlowCapturer
// ----------------------------------------------------------------------------
// Function: transform
template <typename I, typename O, typename C>
cudaTask cudaFlowCapturer::transform(I first, I last, O output, C op) {
return on([=](cudaStream_t stream) mutable {
cudaDefaultExecutionPolicy p(stream);
cuda_transform(p, first, last, output, op);
});
}
// Function: transform
template <typename I1, typename I2, typename O, typename C>
cudaTask cudaFlowCapturer::transform(
I1 first1, I1 last1, I2 first2, O output, C op
) {
return on([=](cudaStream_t stream) mutable {
cudaDefaultExecutionPolicy p(stream);
cuda_transform(p, first1, last1, first2, output, op);
});
}
// Function: transform
template <typename I, typename O, typename C>
void cudaFlowCapturer::transform(
cudaTask task, I first, I last, O output, C op
) {
on(task, [=] (cudaStream_t stream) mutable {
cudaDefaultExecutionPolicy p(stream);
cuda_transform(p, first, last, output, op);
});
}
// Function: transform
template <typename I1, typename I2, typename O, typename C>
void cudaFlowCapturer::transform(
cudaTask task, I1 first1, I1 last1, I2 first2, O output, C op
) {
on(task, [=] (cudaStream_t stream) mutable {
cudaDefaultExecutionPolicy p(stream);
cuda_transform(p, first1, last1, first2, output, op);
});
}
} // end of namespace tf -----------------------------------------------------