Newer
Older
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
342
343
344
345
346
347
348
/***************************************************************************
Atomic.h - description
-------------------
begin : Sep 14, 2018
copyright : (C) 2018 by Tomas Oberhuber et al.
email : tomas.oberhuber@fjfi.cvut.cz
***************************************************************************/
/* See Copyright Notice in tnl/Copyright */
// Implemented by: Jakub Klinkovský
#pragma once
#include <atomic> // std::atomic
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
#include <TNL/param-types.h>
namespace TNL {
template< typename T, typename Device >
class Atomic
{};
template< typename T >
class Atomic< T, Devices::Host >
: public std::atomic< T >
{
public:
Atomic() noexcept = default;
// inherit constructors
using std::atomic< T >::atomic;
// NOTE: std::atomic is not copyable (see https://stackoverflow.com/a/15250851 for
// an explanation), but we need copyability for TNL::Containers::Array. Note that
// this copy-constructor and copy-assignment operator are not atomic as they
// synchronize only with respect to one or the other object.
Atomic( const Atomic& desired ) noexcept
{
this->store(desired.load());
}
Atomic& operator=( const Atomic& desired ) noexcept
{
this->store(desired.load());
return *this;
}
// just for compatibility with TNL::Containers::Array...
static String getType()
{
return "Atomic< " +
TNL::getType< T >() + ", " +
Devices::Host::getDeviceType() + " >";
}
// CAS loops for updating maximum and minimum
// reference: https://stackoverflow.com/a/16190791
T fetch_max( T value ) noexcept
{
const T old = *this;
T prev_value = old;
while(prev_value < value &&
! this->compare_exchange_weak(prev_value, value))
;
return old;
}
T fetch_min( T value ) noexcept
{
const T old = *this;
T prev_value = old;
while(prev_value > value &&
! this->compare_exchange_weak(prev_value, value))
;
return old;
}
};
template< typename T >
class Atomic< T, Devices::Cuda >
{
public:
using value_type = T;
// FIXME
// using difference_type = typename std::atomic< T >::difference_type;
__cuda_callable__
Atomic() noexcept = default;
__cuda_callable__
constexpr Atomic( T desired ) noexcept : value(desired) {}
__cuda_callable__
T operator=( T desired ) noexcept
{
store( desired );
return desired;
}
// NOTE: std::atomic is not copyable (see https://stackoverflow.com/a/15250851 for
// an explanation), but we need copyability for TNL::Containers::Array. Note that
// this copy-constructor and copy-assignment operator are not atomic as they
// synchronize only with respect to one or the other object.
__cuda_callable__
Atomic( const Atomic& desired ) noexcept
{
// FIXME
// *this = desired.load();
*this = desired.value;
}
__cuda_callable__
Atomic& operator=( const Atomic& desired ) noexcept
{
// FIXME
// *this = desired.load();
*this = desired.value;
return *this;
}
// just for compatibility with TNL::Containers::Array...
static String getType()
{
return "Atomic< " +
TNL::getType< T >() + ", " +
Devices::Host::getDeviceType() + " >";
}
bool is_lock_free() const noexcept
{
return true;
}
constexpr bool is_always_lock_free() const noexcept
{
return true;
}
__cuda_callable__
void store( T desired ) noexcept
{
// CUDA does not have a native atomic store, but it can be emulated with atomic exchange
exchange( desired );
}
__cuda_callable__
T load() const noexcept
{
// CUDA does not have a native atomic load:
// https://stackoverflow.com/questions/32341081/how-to-have-atomic-load-in-cuda
return const_cast<Atomic*>(this)->fetch_add( 0 );
}
__cuda_callable__
operator T() const noexcept
{
return load();
}
__cuda_callable__
T exchange( T desired ) noexcept
{
#ifdef __CUDA_ARCH__
return atomicExch( &value, desired );
#else
const T old = value;
value = desired;
return old;
#endif
}
__cuda_callable__
bool compare_exchange_weak( T& expected, T desired ) noexcept
{
return compare_exchange_strong( expected, desired );
}
__cuda_callable__
bool compare_exchange_strong( T& expected, T desired ) noexcept
{
#ifdef __CUDA_ARCH__
const T old = atomicCAS( &value, expected, desired );
const bool result = old == expected;
expected = old;
return result;
#else
if( value == expected ) {
value = desired;
return true;
}
else {
expected = value;
return false;
}
#endif
}
__cuda_callable__
T fetch_add( T arg )
{
#ifdef __CUDA_ARCH__
return atomicAdd( &value, arg );
#else
const T old = value;
value += arg;
return old;
#endif
}
__cuda_callable__
T fetch_sub( T arg )
{
#ifdef __CUDA_ARCH__
return atomicSub( &value, arg );
#else
const T old = value;
value -= arg;
return old;
#endif
}
__cuda_callable__
T fetch_and( T arg )
{
#ifdef __CUDA_ARCH__
return atomicAnd( &value, arg );
#else
const T old = value;
value = value & arg;
return old;
#endif
}
__cuda_callable__
T fetch_or( T arg )
{
#ifdef __CUDA_ARCH__
return atomicOr( &value, arg );
#else
const T old = value;
value = value | arg;
return old;
#endif
}
__cuda_callable__
T fetch_xor( T arg )
{
#ifdef __CUDA_ARCH__
return atomicXor( &value, arg );
#else
const T old = value;
value = value ^ arg;
return old;
#endif
}
__cuda_callable__
T operator+=( T arg ) noexcept
{
return fetch_add( arg ) + arg;
}
__cuda_callable__
T operator-=( T arg ) noexcept
{
return fetch_sub( arg ) - arg;
}
__cuda_callable__
T operator&=( T arg ) noexcept
{
return fetch_and( arg ) & arg;
}
__cuda_callable__
T operator|=( T arg ) noexcept
{
return fetch_or( arg ) | arg;
}
__cuda_callable__
T operator^=( T arg ) noexcept
{
return fetch_xor( arg ) ^ arg;
}
// pre-increment
__cuda_callable__
T operator++() noexcept
{
return fetch_add(1) + 1;
}
// post-increment
__cuda_callable__
T operator++(int) noexcept
{
return fetch_add(1);
}
// pre-decrement
__cuda_callable__
T operator--() noexcept
{
return fetch_sub(1) - 1;
}
// post-decrement
__cuda_callable__
T operator--(int) noexcept
{
return fetch_sub(1);
}
// extensions (methods not present in C++ standards)
__cuda_callable__
T fetch_max( T arg ) noexcept
{
#ifdef __CUDA_ARCH__
return atomicMax( &value, arg );
#else
const T old = value;
value = ( value > arg ) ? value : arg;
return old;
#endif
}
__cuda_callable__
T fetch_min( T arg ) noexcept
{
#ifdef __CUDA_ARCH__
return atomicMin( &value, arg );
#else
const T old = value;
value = ( value < arg ) ? value : arg;
return old;
#endif
}
protected:
T value;
};
} // namespace TNL