thread_operators.cuh 8.67 KB
Newer Older
zhuwenwen's avatar
zhuwenwen committed
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
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
/******************************************************************************
 * Copyright (c) 2010-2011, Duane Merrill.  All rights reserved.
 * Copyright (c) 2011-2018, NVIDIA CORPORATION.  All rights reserved.
 * Modifications Copyright (c) 2017-2020, Advanced Micro Devices, Inc.  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.
 *
 ******************************************************************************/

#ifndef HIBCUB_ROCPRIM_THREAD_THREAD_OPERATORS_HPP_
#define HIBCUB_ROCPRIM_THREAD_THREAD_OPERATORS_HPP_

#include "../config.hpp"

#include "../util_type.cuh"

BEGIN_HIPCUB_NAMESPACE

struct Equality
{
    template<class T>
    HIPCUB_HOST_DEVICE inline
    constexpr bool operator()(const T& a, const T& b) const
    {
        return a == b;
    }
};

struct Inequality
{
    template<class T>
    HIPCUB_HOST_DEVICE inline
    constexpr bool operator()(const T& a, const T& b) const
    {
        return a != b;
    }
};

template <class EqualityOp>
struct InequalityWrapper
{
    EqualityOp op;

    HIPCUB_HOST_DEVICE inline
    InequalityWrapper(EqualityOp op) : op(op) {}

    template<class T>
    HIPCUB_HOST_DEVICE inline
    bool operator()(const T &a, const T &b)
    {
        return !op(a, b);
    }
};

struct Sum
{
    template<class T>
    HIPCUB_HOST_DEVICE inline
    constexpr T operator()(const T &a, const T &b) const
    {
        return a + b;
    }
};

struct Difference
{
    template <class T>
    HIPCUB_HOST_DEVICE inline
    constexpr T operator()(const T &a, const T &b) const
    {
        return a - b;
    }
};

struct Division
{
    template <class T>
    HIPCUB_HOST_DEVICE inline
    constexpr T operator()(const T &a, const T &b) const
    {
        return a / b;
    }
};

struct Max
{
    template<class T>
    HIPCUB_HOST_DEVICE inline
    constexpr T operator()(const T &a, const T &b) const
    {
        return a < b ? b : a;
    }
};

struct Min
{
    template<class T>
    HIPCUB_HOST_DEVICE inline
    constexpr T operator()(const T &a, const T &b) const
    {
        return a < b ? a : b;
    }
};

struct ArgMax
{
    template<
        class Key,
        class Value
    >
    HIPCUB_HOST_DEVICE inline
    constexpr KeyValuePair<Key, Value>
    operator()(const KeyValuePair<Key, Value>& a,
               const KeyValuePair<Key, Value>& b) const
    {
        return ((b.value > a.value) || ((a.value == b.value) && (b.key < a.key))) ? b : a;
    }
};

struct ArgMin
{
    template<
        class Key,
        class Value
    >
    HIPCUB_HOST_DEVICE inline
    constexpr KeyValuePair<Key, Value>
    operator()(const KeyValuePair<Key, Value>& a,
               const KeyValuePair<Key, Value>& b) const
    {
        return ((b.value < a.value) || ((a.value == b.value) && (b.key < a.key))) ? b : a;
    }
};

template <typename B>
struct CastOp
{
    template <typename A>
    HIPCUB_HOST_DEVICE inline
    B operator()(const A &a) const
    {
        return (B)a;
    }
};

template <typename ScanOp>
class SwizzleScanOp
{
private:
    ScanOp scan_op;

public:
    HIPCUB_HOST_DEVICE inline
    SwizzleScanOp(ScanOp scan_op) : scan_op(scan_op)
    {
    }

    template <typename T>
    HIPCUB_HOST_DEVICE inline
    T operator()(const T &a, const T &b)
    {
      T _a(a);
      T _b(b);

      return scan_op(_b, _a);
    }
};

template <typename ReductionOpT>
struct ReduceBySegmentOp
{
    ReductionOpT op;

    HIPCUB_HOST_DEVICE inline
    ReduceBySegmentOp()
    {
    }

    HIPCUB_HOST_DEVICE inline
    ReduceBySegmentOp(ReductionOpT op) : op(op)
    {
    }

    template <typename KeyValuePairT>
    HIPCUB_HOST_DEVICE inline
    KeyValuePairT operator()(
        const KeyValuePairT &first,
        const KeyValuePairT &second)
    {
        KeyValuePairT retval;
        retval.key = first.key + second.key;
        retval.value = (second.key) ?
                second.value :
                op(first.value, second.value);
        return retval;
    }
};

template <typename ReductionOpT>
struct ReduceByKeyOp
{
    ReductionOpT op;

    HIPCUB_HOST_DEVICE inline
    ReduceByKeyOp()
    {
    }

    HIPCUB_HOST_DEVICE inline
    ReduceByKeyOp(ReductionOpT op) : op(op)
    {
    }

    template <typename KeyValuePairT>
    HIPCUB_HOST_DEVICE inline
    KeyValuePairT operator()(
        const KeyValuePairT &first,
        const KeyValuePairT &second)
    {
        KeyValuePairT retval = second;

        if (first.key == second.key)
        {
            retval.value = op(first.value, retval.value);
        }
        return retval;
    }
};

template <typename BinaryOpT>
struct BinaryFlip
{
    BinaryOpT binary_op;

    HIPCUB_HOST_DEVICE
    explicit BinaryFlip(BinaryOpT binary_op) : binary_op(binary_op)
    {
    }

    template <typename T, typename U>
    HIPCUB_DEVICE auto
    operator()(T &&t, U &&u) -> decltype(binary_op(std::forward<U>(u),
                                                    std::forward<T>(t)))
    {
        return binary_op(std::forward<U>(u), std::forward<T>(t));
    }
};

template <typename BinaryOpT>
HIPCUB_HOST_DEVICE
BinaryFlip<BinaryOpT> MakeBinaryFlip(BinaryOpT binary_op)
{
    return BinaryFlip<BinaryOpT>(binary_op);
}

namespace detail
{

// CUB uses value_type of OutputIteratorT (if not void) as a type of intermediate results in reduce,
// for example:
//
// /// The output value type
// typedef typename If<(Equals<typename std::iterator_traits<OutputIteratorT>::value_type, void>::VALUE),  // OutputT =  (if output iterator's value type is void) ?
//     typename std::iterator_traits<InputIteratorT>::value_type,                                          // ... then the input iterator's value type,
//     typename std::iterator_traits<OutputIteratorT>::value_type>::Type OutputT;                          // ... else the output iterator's value type
//
// rocPRIM (as well as Thrust) uses result type of BinaryFunction instead (if not void):
//
// using input_type = typename std::iterator_traits<InputIterator>::value_type;
// using result_type = typename ::rocprim::detail::match_result_type<
//     input_type, BinaryFunction
// >::type;
//
// For short -> float using Sum()
// CUB:     float Sum(float, float)
// rocPRIM: short Sum(short, short)
//
// This wrapper allows to have compatibility with CUB in hipCUB.
template<
    class InputIteratorT,
    class OutputIteratorT,
    class BinaryFunction
>
struct convert_result_type_wrapper
{
    using input_type = typename std::iterator_traits<InputIteratorT>::value_type;
    using output_type = typename std::iterator_traits<OutputIteratorT>::value_type;
    using result_type =
        typename std::conditional<
            std::is_void<output_type>::value, input_type, output_type
        >::type;

    convert_result_type_wrapper(BinaryFunction op) : op(op) {}

    template<class T>
    HIPCUB_HOST_DEVICE inline
    constexpr result_type operator()(const T &a, const T &b) const
    {
        return static_cast<result_type>(op(a, b));
    }

    BinaryFunction op;
};

template<
    class InputIteratorT,
    class OutputIteratorT,
    class BinaryFunction
>
inline
convert_result_type_wrapper<InputIteratorT, OutputIteratorT, BinaryFunction>
convert_result_type(BinaryFunction op)
{
    return convert_result_type_wrapper<InputIteratorT, OutputIteratorT, BinaryFunction>(op);
}

} // end detail namespace

END_HIPCUB_NAMESPACE

#endif // HIBCUB_ROCPRIM_THREAD_THREAD_OPERATORS_HPP_