forked from kokkos/kokkos
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Kokkos_OpenACC_ParallelReduce_Team.hpp
449 lines (403 loc) · 20.3 KB
/
Kokkos_OpenACC_ParallelReduce_Team.hpp
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
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
//@HEADER
// ************************************************************************
//
// Kokkos v. 4.0
// Copyright (2022) National Technology & Engineering
// Solutions of Sandia, LLC (NTESS).
//
// Under the terms of Contract DE-NA0003525 with NTESS,
// the U.S. Government retains certain rights in this software.
//
// Part of Kokkos, under the Apache License v2.0 with LLVM Exceptions.
// See https://kokkos.org/LICENSE for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//@HEADER
#ifndef KOKKOS_OPENACC_PARALLEL_REDUCE_TEAM_HPP
#define KOKKOS_OPENACC_PARALLEL_REDUCE_TEAM_HPP
#include <OpenACC/Kokkos_OpenACC_Team.hpp>
#include <OpenACC/Kokkos_OpenACC_FunctorAdapter.hpp>
#include <OpenACC/Kokkos_OpenACC_Macros.hpp>
#ifdef KOKKOS_ENABLE_OPENACC_COLLAPSE_HIERARCHICAL_CONSTRUCTS
#define KOKKOS_IMPL_OPENACC_LOOP_CLAUSE \
Kokkos::Experimental::Impl::RoutineClause::seq
#else
#define KOKKOS_IMPL_OPENACC_LOOP_CLAUSE \
Kokkos::Experimental::Impl::RoutineClause::worker
#endif
//----------------------------------------------------------------------------
//----------------------------------------------------------------------------
// Hierarchical Parallelism -> Team level implementation
namespace Kokkos::Experimental::Impl {
// primary template: catch-all non-implemented custom reducers
template <class Functor, class Reducer, class Policy,
bool = std::is_arithmetic_v<typename Reducer::value_type>>
struct OpenACCParallelReduceTeamHelper {
OpenACCParallelReduceTeamHelper(Functor const&, Reducer const&,
Policy const&) {
static_assert(Kokkos::Impl::always_false<Functor>::value,
"not implemented");
}
};
} // namespace Kokkos::Experimental::Impl
template <class CombinedFunctorReducerType, class... Properties>
class Kokkos::Impl::ParallelReduce<CombinedFunctorReducerType,
Kokkos::TeamPolicy<Properties...>,
Kokkos::Experimental::OpenACC> {
private:
using Policy =
TeamPolicyInternal<Kokkos::Experimental::OpenACC, Properties...>;
using FunctorType = typename CombinedFunctorReducerType::functor_type;
using ReducerType = typename CombinedFunctorReducerType::reducer_type;
using value_type = typename ReducerType::value_type;
using pointer_type = typename ReducerType::pointer_type;
CombinedFunctorReducerType m_functor_reducer;
Policy m_policy;
pointer_type m_result_ptr;
bool m_result_ptr_on_device;
public:
void execute() const {
auto league_size = m_policy.league_size();
auto team_size = m_policy.team_size();
auto vector_length = m_policy.impl_vector_length();
int const async_arg = m_policy.space().acc_async_queue();
value_type val;
const ReducerType& reducer = m_functor_reducer.get_reducer();
reducer.init(&val);
if (league_size <= 0) {
if (m_result_ptr_on_device == false) {
*m_result_ptr = val;
} else {
acc_memcpy_to_device(m_result_ptr, &val, sizeof(value_type));
}
return;
}
Kokkos::Experimental::Impl::OpenACCParallelReduceTeamHelper(
Kokkos::Experimental::Impl::FunctorAdapter<
FunctorType, Policy, KOKKOS_IMPL_OPENACC_LOOP_CLAUSE>(
m_functor_reducer.get_functor()),
std::conditional_t<
std::is_same_v<FunctorType, typename ReducerType::functor_type>,
Sum<value_type>, typename ReducerType::functor_type>(val),
m_policy);
// OpenACC backend supports only built-in Reducer types; thus
// reducer.final() below is a no-op.
reducer.final(&val);
// acc_wait(async_arg) in the below if-else statements is needed because the
// above OpenACC compute kernel can be executed asynchronously and val is a
// local host variable.
if (m_result_ptr_on_device == false) {
acc_wait(async_arg);
*m_result_ptr = val;
} else {
acc_memcpy_to_device_async(m_result_ptr, &val, sizeof(value_type),
async_arg);
acc_wait(async_arg);
}
}
template <class ViewType>
ParallelReduce(const CombinedFunctorReducerType& arg_functor_reducer,
const Policy& arg_policy, const ViewType& arg_result_view)
: m_functor_reducer(arg_functor_reducer),
m_policy(arg_policy),
m_result_ptr(arg_result_view.data()),
m_result_ptr_on_device(
MemorySpaceAccess<Kokkos::Experimental::OpenACCSpace,
typename ViewType::memory_space>::accessible) {}
};
namespace Kokkos {
// Hierarchical Parallelism -> Team thread level implementation
// FIXME_OPENACC: custom reduction is not implemented.
template <typename iType, class Lambda, typename ValueType, class JoinType>
KOKKOS_INLINE_FUNCTION void parallel_reduce(
const Impl::TeamThreadRangeBoundariesStruct<iType, Impl::OpenACCTeamMember>&
loop_boundaries,
const Lambda& lambda, const JoinType& join, ValueType& init_result) {
static_assert(Kokkos::Impl::always_false<Lambda>::value,
"custom reduction is not implemented");
}
// Hierarchical Parallelism -> Thread vector level implementation
// FIXME_OPENACC: custom reduction is not implemented.
template <typename iType, class Lambda, typename ValueType, class JoinType>
KOKKOS_INLINE_FUNCTION void parallel_reduce(
const Impl::ThreadVectorRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, const JoinType& join, ValueType& init_result) {
static_assert(Kokkos::Impl::always_false<Lambda>::value,
"custom reduction is not implemented");
}
} // namespace Kokkos
#ifdef KOKKOS_ENABLE_OPENACC_COLLAPSE_HIERARCHICAL_CONSTRUCTS
#define KOKKOS_IMPL_ACC_REDUCE_TEAM_PRAGMA \
vector vector_length(team_size* vector_length)
#define KOKKOS_IMPL_ACC_REDUCE_TEAM_ITRS league_size* team_size* vector_length
#define KOKKOS_IMPL_ACC_REDUCE_TEAM_LEAGUE_ID_INIT \
i / (team_size * vector_length)
namespace Kokkos {
// Hierarchical Parallelism -> Team thread level implementation
#pragma acc routine seq
template <typename iType, class Lambda, typename ValueType>
KOKKOS_INLINE_FUNCTION std::enable_if_t<!Kokkos::is_reducer_v<ValueType>>
parallel_reduce(const Impl::TeamThreadRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, ValueType& result) {
ValueType tmp = ValueType();
iType j_start =
loop_boundaries.team.team_rank() / loop_boundaries.team.vector_length();
if (j_start == 0) {
#pragma acc loop seq
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++)
lambda(i, tmp);
result = tmp;
}
}
#pragma acc routine seq
template <typename iType, class Lambda, typename ReducerType>
KOKKOS_INLINE_FUNCTION std::enable_if_t<Kokkos::is_reducer_v<ReducerType>>
parallel_reduce(const Impl::TeamThreadRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, const ReducerType& reducer) {
using ValueType = typename ReducerType::value_type;
ValueType tmp;
reducer.init(tmp);
iType j_start =
loop_boundaries.team.team_rank() / loop_boundaries.team.vector_length();
if (j_start == 0) {
#pragma acc loop seq
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++)
lambda(i, tmp);
reducer.reference() = tmp;
}
}
// Hierarchical Parallelism -> Thread vector level implementation
#pragma acc routine seq
template <typename iType, class Lambda, typename ValueType>
KOKKOS_INLINE_FUNCTION std::enable_if_t<!Kokkos::is_reducer_v<ValueType>>
parallel_reduce(const Impl::ThreadVectorRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, ValueType& result) {
ValueType tmp = ValueType();
iType j_start =
loop_boundaries.team.team_rank() % loop_boundaries.team.vector_length();
if (j_start == 0) {
#pragma acc loop seq
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) {
lambda(i, tmp);
}
result = tmp;
}
}
#pragma acc routine seq
template <typename iType, class Lambda, typename ReducerType>
KOKKOS_INLINE_FUNCTION std::enable_if_t<Kokkos::is_reducer_v<ReducerType>>
parallel_reduce(const Impl::ThreadVectorRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, const ReducerType& reducer) {
using ValueType = typename ReducerType::value_type;
ValueType tmp;
reducer.init(tmp);
iType j_start =
loop_boundaries.team.team_rank() % loop_boundaries.team.vector_length();
if (j_start == 0) {
#pragma acc loop seq
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) {
lambda(i, tmp);
}
reducer.reference() = tmp;
}
}
// Hierarchical Parallelism -> Team vector level implementation
#pragma acc routine seq
template <typename iType, class Lambda, typename ValueType>
KOKKOS_INLINE_FUNCTION void parallel_reduce(
const Impl::TeamVectorRangeBoundariesStruct<iType, Impl::OpenACCTeamMember>&
loop_boundaries,
const Lambda& lambda, ValueType& result) {
ValueType tmp = ValueType();
iType j_start =
loop_boundaries.team.team_rank() % loop_boundaries.team.vector_length();
if (j_start == 0) {
#pragma acc loop seq
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) {
lambda(i, tmp);
}
result = tmp;
}
}
} // namespace Kokkos
#else /* #ifdef KOKKOS_ENABLE_OPENACC_COLLAPSE_HIERARCHICAL_CONSTRUCTS */
#define KOKKOS_IMPL_ACC_REDUCE_TEAM_PRAGMA \
num_workers(team_size) vector_length(vector_length)
#define KOKKOS_IMPL_ACC_REDUCE_TEAM_ITRS league_size
#define KOKKOS_IMPL_ACC_REDUCE_TEAM_LEAGUE_ID_INIT i
// FIXME_OPENACC: below implementation conforms to the OpenACC standard, but
// the NVHPC compiler (V22.11) fails due to the lack of support for lambda
// expressions containing parallel loops.
namespace Kokkos {
// Hierarchical Parallelism -> Team thread level implementation
#pragma acc routine worker
template <typename iType, class Lambda, typename ValueType>
KOKKOS_INLINE_FUNCTION std::enable_if_t<!Kokkos::is_reducer_v<ValueType>>
parallel_reduce(const Impl::TeamThreadRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, ValueType& result) {
ValueType tmp = ValueType();
#pragma acc loop worker reduction(+ : tmp)
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++)
lambda(i, tmp);
result = tmp;
}
#define KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(REDUCER, OPERATOR) \
KOKKOS_IMPL_ACC_PRAGMA(routine worker) \
template <typename iType, class Lambda, class Scalar, class Space> \
KOKKOS_INLINE_FUNCTION \
std::enable_if_t<Kokkos::is_reducer_v<Kokkos::REDUCER<Scalar, Space>>> \
parallel_reduce(const Impl::TeamThreadRangeBoundariesStruct< \
iType, Impl::OpenACCTeamMember>& loop_boundaries, \
const Lambda& lambda, \
const Kokkos::REDUCER<Scalar, Space>& reducer) { \
using ValueType = typename Kokkos::REDUCER<Scalar, Space>::value_type; \
ValueType tmp = ValueType(); \
reducer.init(tmp); \
KOKKOS_IMPL_ACC_PRAGMA(loop worker reduction(OPERATOR : tmp)) \
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) \
lambda(i, tmp); \
reducer.reference() = tmp; \
}
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(Sum, +);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(Prod, *);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(Min, min);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(Max, max);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(LAnd, &&);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(LOr, ||);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(BAnd, &);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD(BOr, |);
// Hierarchical Parallelism -> Thread vector level implementation
#pragma acc routine vector
template <typename iType, class Lambda, typename ValueType>
KOKKOS_INLINE_FUNCTION std::enable_if_t<!Kokkos::is_reducer_v<ValueType>>
parallel_reduce(const Impl::ThreadVectorRangeBoundariesStruct<
iType, Impl::OpenACCTeamMember>& loop_boundaries,
const Lambda& lambda, ValueType& result) {
ValueType tmp = ValueType();
#pragma acc loop vector reduction(+ : tmp)
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) {
lambda(i, tmp);
}
result = tmp;
}
#define KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(REDUCER, OPERATOR) \
KOKKOS_IMPL_ACC_PRAGMA(routine vector) \
template <typename iType, class Lambda, class Scalar, class Space> \
KOKKOS_INLINE_FUNCTION \
std::enable_if_t<Kokkos::is_reducer_v<Kokkos::REDUCER<Scalar, Space>>> \
parallel_reduce(const Impl::ThreadVectorRangeBoundariesStruct< \
iType, Impl::OpenACCTeamMember>& loop_boundaries, \
const Lambda& lambda, \
const Kokkos::REDUCER<Scalar, Space>& reducer) { \
using ValueType = typename Kokkos::REDUCER<Scalar, Space>::value_type; \
ValueType tmp; \
reducer.init(tmp); \
KOKKOS_IMPL_ACC_PRAGMA(loop vector reduction(OPERATOR : tmp)) \
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) { \
lambda(i, tmp); \
} \
reducer.reference() = tmp; \
}
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(Sum, +);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(Prod, *);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(Min, min);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(Max, max);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(LAnd, &&);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(LOr, ||);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(BAnd, &);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR(BOr, |);
// Hierarchical Parallelism -> Team vector level implementation
#pragma acc routine vector
template <typename iType, class Lambda, typename ValueType>
KOKKOS_INLINE_FUNCTION void parallel_reduce(
const Impl::TeamVectorRangeBoundariesStruct<iType, Impl::OpenACCTeamMember>&
loop_boundaries,
const Lambda& lambda, ValueType& result) {
ValueType tmp = ValueType();
#pragma acc loop vector reduction(+ : tmp)
for (iType i = loop_boundaries.start; i < loop_boundaries.end; i++) {
lambda(i, tmp);
}
result = tmp;
}
} // namespace Kokkos
#undef KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_THREAD
#undef KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_THREAD_VECTOR
#endif /* #ifdef KOKKOS_ENABLE_OPENACC_COLLAPSE_HIERARCHICAL_CONSTRUCTS */
#define KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_DISPATCH_SCHEDULE(REDUCER, \
OPERATOR) \
namespace Kokkos::Experimental::Impl { \
template <class Policy, class ValueType, class Functor> \
void OpenACCParallelReduceTeam##REDUCER(Policy const policy, \
ValueType& aval, \
Functor const& afunctor, \
int async_arg) { \
auto const functor = afunctor; \
auto val = aval; \
auto const league_size = policy.league_size(); \
auto const team_size = policy.team_size(); \
auto const vector_length = policy.impl_vector_length(); \
/* clang-format off */ \
KOKKOS_IMPL_ACC_PRAGMA(parallel loop gang num_gangs(league_size) KOKKOS_IMPL_ACC_REDUCE_TEAM_PRAGMA reduction(OPERATOR : val) copyin(functor) async(async_arg)) \
/* clang-format on */ \
for (int i = 0; i < KOKKOS_IMPL_ACC_REDUCE_TEAM_ITRS; i++) { \
int league_id = KOKKOS_IMPL_ACC_REDUCE_TEAM_LEAGUE_ID_INIT; \
typename Policy::member_type team(league_id, league_size, team_size, \
vector_length); \
functor(team, val); \
} \
aval = val; \
} \
} // namespace Kokkos::Experimental::Impl
#define KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(REDUCER, OPERATOR) \
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_DISPATCH_SCHEDULE(REDUCER, OPERATOR) \
\
template <class Functor, class Scalar, class Space, class... Traits> \
struct Kokkos::Experimental::Impl::OpenACCParallelReduceTeamHelper< \
Functor, Kokkos::REDUCER<Scalar, Space>, \
Kokkos::Impl::TeamPolicyInternal<Traits...>, true> { \
using Policy = Kokkos::Impl::TeamPolicyInternal<Traits...>; \
using Reducer = REDUCER<Scalar, Space>; \
using ValueType = typename Reducer::value_type; \
\
OpenACCParallelReduceTeamHelper(Functor const& functor, \
Reducer const& reducer, \
Policy const& policy) { \
auto league_size = policy.league_size(); \
auto team_size = policy.team_size(); \
auto vector_length = policy.impl_vector_length(); \
\
if (league_size <= 0) { \
return; \
} \
\
ValueType val; \
reducer.init(val); \
\
int const async_arg = policy.space().acc_async_queue(); \
\
OpenACCParallelReduceTeam##REDUCER(policy, val, functor, async_arg); \
\
reducer.reference() = val; \
} \
}
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(Sum, +);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(Prod, *);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(Min, min);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(Max, max);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(LAnd, &&);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(LOr, ||);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(BAnd, &);
KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER(BOr, |);
#undef KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_TEAM_HELPER
#undef KOKKOS_IMPL_OPENACC_PARALLEL_REDUCE_DISPATCH_SCHEDULE
#undef KOKKOS_IMPL_ACC_REDUCE_TEAM_PRAGMA
#undef KOKKOS_IMPL_ACC_REDUCE_TEAM_ITRS
#undef KOKKOS_IMPL_ACC_REDUCE_TEAM_LEAGUE_ID_INIT
#endif /* #ifndef KOKKOS_OPENACC_PARALLEL_REDUCE_TEAM_HPP */