File size: 8,310 Bytes
2a504a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 (c) 2016-2018, NVIDIA CORPORATION.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in the
 *       documentation and/or other materials provided with the distribution.
 *     * Neither the name of the NVIDIA CORPORATION nor the
 *       names of its contributors may be used to endorse or promote products
 *       derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANY
 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 ******************************************************************************/
#pragma once

#include <thrust/detail/config.h>

#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
#  pragma GCC system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
#  pragma clang system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
#  pragma system_header
#endif // no system header
#include <thrust/system/cuda/detail/guarded_cuda_runtime_api.h>
#include <thrust/system/cuda/detail/execution_policy.h>
#include <thrust/system/cuda/detail/util.h>

#include <thrust/detail/allocator_aware_execution_policy.h>

#if THRUST_CPP_DIALECT >= 2011
#  include <thrust/detail/dependencies_aware_execution_policy.h>
#endif


THRUST_NAMESPACE_BEGIN
namespace cuda_cub {

template <class Derived>
struct execute_on_stream_base : execution_policy<Derived>
{
private:
  cudaStream_t stream;

public:
  __thrust_exec_check_disable__
  __host__ __device__
  execute_on_stream_base(cudaStream_t stream_ = default_stream())
      : stream(stream_){}

  THRUST_RUNTIME_FUNCTION
  Derived
  on(cudaStream_t const &s) const
  {
    Derived result = derived_cast(*this);
    result.stream  = s;
    return result;
  }

private:
  friend __host__ __device__
  cudaStream_t
  get_stream(const execute_on_stream_base &exec)
  {
    return exec.stream;
  }
};

template <class Derived>
struct execute_on_stream_nosync_base : execution_policy<Derived>
{
private:
  cudaStream_t stream;

public:
  __host__ __device__
  execute_on_stream_nosync_base(cudaStream_t stream_ = default_stream())
      : stream(stream_){}

  THRUST_RUNTIME_FUNCTION
  Derived
  on(cudaStream_t const &s) const
  {
    Derived result = derived_cast(*this);
    result.stream  = s;
    return result;
  }

private:
  friend __host__ __device__
  cudaStream_t
  get_stream(const execute_on_stream_nosync_base &exec)
  {
    return exec.stream;
  }

  friend __host__ __device__
  bool
  must_perform_optional_stream_synchronization(const execute_on_stream_nosync_base &)
  {
    return false;
  }
};

struct execute_on_stream : execute_on_stream_base<execute_on_stream>
{
  typedef execute_on_stream_base<execute_on_stream> base_t;

  __host__ __device__
  execute_on_stream() : base_t(){};
  __host__ __device__
  execute_on_stream(cudaStream_t stream)
  : base_t(stream){};
};

struct execute_on_stream_nosync : execute_on_stream_nosync_base<execute_on_stream_nosync>
{
  typedef execute_on_stream_nosync_base<execute_on_stream_nosync> base_t;

  __host__ __device__
  execute_on_stream_nosync() : base_t(){};
  __host__ __device__
  execute_on_stream_nosync(cudaStream_t stream)
  : base_t(stream){};
};


struct par_t : execution_policy<par_t>,
  thrust::detail::allocator_aware_execution_policy<
    execute_on_stream_base>
#if THRUST_CPP_DIALECT >= 2011
, thrust::detail::dependencies_aware_execution_policy<
    execute_on_stream_base>
#endif
{
  typedef execution_policy<par_t> base_t;

  __host__ __device__
  constexpr par_t() : base_t() {}

  typedef execute_on_stream stream_attachment_type;

  THRUST_RUNTIME_FUNCTION
  stream_attachment_type
  on(cudaStream_t const &stream) const
  {
    return execute_on_stream(stream);
  }
};

struct par_nosync_t : execution_policy<par_nosync_t>,
  thrust::detail::allocator_aware_execution_policy<
    execute_on_stream_nosync_base>
#if THRUST_CPP_DIALECT >= 2011
, thrust::detail::dependencies_aware_execution_policy<
    execute_on_stream_nosync_base>
#endif
{
  typedef execution_policy<par_nosync_t> base_t;

  __host__ __device__
  constexpr par_nosync_t() : base_t() {}

  typedef execute_on_stream_nosync stream_attachment_type;

  THRUST_RUNTIME_FUNCTION
  stream_attachment_type
  on(cudaStream_t const &stream) const
  {
    return execute_on_stream_nosync(stream);
  }

private:
  //this function is defined to allow non-blocking calls on the default_stream() with thrust::cuda::par_nosync
  //without explicitly using thrust::cuda::par_nosync.on(default_stream())
  friend __host__ __device__
  bool
  must_perform_optional_stream_synchronization(const par_nosync_t &)
  {
    return false;
  }
};

THRUST_INLINE_CONSTANT par_t par;

/*! \p thrust::cuda::par_nosync is a parallel execution policy targeting Thrust's CUDA device backend.
 *  Similar to \p thrust::cuda::par it allows execution of Thrust algorithms in a specific CUDA stream.
 *
 *  \p thrust::cuda::par_nosync indicates that an algorithm is free to avoid any synchronization of the
 *  associated stream that is not strictly required for correctness. Additionally, algorithms may return
 *  before the corresponding kernels are completed, similar to asynchronous kernel launches via <<< >>> syntax.
 *  The user must take care to perform explicit synchronization if necessary.
 *
 *  The following code snippet demonstrates how to use \p thrust::cuda::par_nosync :
 *
 *  \code
 *    #include <thrust/device_vector.h>
 *    #include <thrust/for_each.h>
 *    #include <thrust/execution_policy.h>
 *
 *    struct IncFunctor{
 *        __host__ __device__
 *        void operator()(std::size_t& x){ x = x + 1; };
 *    };
 *
 *    int main(){
 *        std::size_t N = 1000000;
 *        thrust::device_vector<std::size_t> d_vec(N);
 *
 *        cudaStream_t stream;
 *        cudaStreamCreate(&stream);
 *        auto nosync_policy = thrust::cuda::par_nosync.on(stream);
 *
 *        thrust::for_each(nosync_policy, d_vec.begin(), d_vec.end(), IncFunctor{});
 *        thrust::for_each(nosync_policy, d_vec.begin(), d_vec.end(), IncFunctor{});
 *        thrust::for_each(nosync_policy, d_vec.begin(), d_vec.end(), IncFunctor{});
 *
 *        //for_each may return before completion. Could do other cpu work in the meantime
 *        // ...
 *
 *        //Wait for the completion of all for_each kernels
 *        cudaStreamSynchronize(stream);
 *
 *        std::size_t x = thrust::reduce(nosync_policy, d_vec.begin(), d_vec.end());
 *        //Currently, this synchronization is not necessary. reduce will still perform
 *        //implicit synchronization to transfer the reduced value to the host to return it.
 *        cudaStreamSynchronize(stream);
 *        cudaStreamDestroy(stream);
 *    }
 *  \endcode
 *
 */
THRUST_INLINE_CONSTANT par_nosync_t par_nosync;
}    // namespace cuda_

namespace system {
namespace cuda {
  using thrust::cuda_cub::par;
  using thrust::cuda_cub::par_nosync;
  namespace detail {
    using thrust::cuda_cub::par_t;
    using thrust::cuda_cub::par_nosync_t;
  }
} // namesapce cuda
} // namespace system

namespace cuda {
using thrust::cuda_cub::par;
using thrust::cuda_cub::par_nosync;
} // namespace cuda

THRUST_NAMESPACE_END