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
|
//@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
#include <Kokkos_Core.hpp>
#include <Kokkos_Timer.hpp>
#include <iostream>
#include <cstdlib>
#include <cstdint>
#include <cinttypes>
namespace TestTeamVectorRange {
struct my_complex {
double re, im;
int dummy;
KOKKOS_INLINE_FUNCTION
my_complex() {
re = 0.0;
im = 0.0;
dummy = 0;
}
KOKKOS_INLINE_FUNCTION
my_complex(const my_complex& src) {
re = src.re;
im = src.im;
dummy = src.dummy;
}
KOKKOS_INLINE_FUNCTION
my_complex& operator=(const my_complex& src) {
re = src.re;
im = src.im;
dummy = src.dummy;
return *this;
}
KOKKOS_INLINE_FUNCTION
my_complex(const double& val) {
re = val;
im = 0.0;
dummy = 0;
}
KOKKOS_INLINE_FUNCTION
my_complex& operator+=(const my_complex& src) {
re += src.re;
im += src.im;
dummy += src.dummy;
return *this;
}
KOKKOS_INLINE_FUNCTION
my_complex operator+(const my_complex& src) {
my_complex tmp = *this;
tmp.re += src.re;
tmp.im += src.im;
tmp.dummy += src.dummy;
return tmp;
}
KOKKOS_INLINE_FUNCTION
my_complex& operator*=(const my_complex& src) {
double re_tmp = re * src.re - im * src.im;
double im_tmp = re * src.im + im * src.re;
re = re_tmp;
im = im_tmp;
dummy *= src.dummy;
return *this;
}
KOKKOS_INLINE_FUNCTION
bool operator==(const my_complex& src) const {
return (re == src.re) && (im == src.im) && (dummy == src.dummy);
}
KOKKOS_INLINE_FUNCTION
bool operator!=(const my_complex& src) const {
return (re != src.re) || (im != src.im) || (dummy != src.dummy);
}
KOKKOS_INLINE_FUNCTION
bool operator!=(const double& val) const {
return (re != val) || (im != 0) || (dummy != 0);
}
KOKKOS_INLINE_FUNCTION
my_complex& operator=(const int& val) {
re = val;
im = 0.0;
dummy = 0;
return *this;
}
KOKKOS_INLINE_FUNCTION
my_complex& operator=(const double& val) {
re = val;
im = 0.0;
dummy = 0;
return *this;
}
KOKKOS_INLINE_FUNCTION
operator double() { return re; }
};
} // namespace TestTeamVectorRange
namespace Kokkos {
template <>
struct reduction_identity<TestTeamVectorRange::my_complex> {
using t_red_ident = reduction_identity<double>;
KOKKOS_FORCEINLINE_FUNCTION static TestTeamVectorRange::my_complex sum() {
return TestTeamVectorRange::my_complex(t_red_ident::sum());
}
KOKKOS_FORCEINLINE_FUNCTION static TestTeamVectorRange::my_complex prod() {
return TestTeamVectorRange::my_complex(t_red_ident::prod());
}
};
} // namespace Kokkos
namespace TestTeamVectorRange {
template <typename Scalar, class ExecutionSpace>
struct functor_teamvector_for {
using policy_type = Kokkos::TeamPolicy<ExecutionSpace>;
using execution_space = ExecutionSpace;
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> flag;
functor_teamvector_for(
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> flag_)
: flag(flag_) {}
using shmem_space = typename ExecutionSpace::scratch_memory_space;
using shared_int =
Kokkos::View<Scalar*, shmem_space, Kokkos::MemoryUnmanaged>;
unsigned team_shmem_size(int /*team_size*/) const {
return shared_int::shmem_size(131);
}
KOKKOS_INLINE_FUNCTION
void operator()(typename policy_type::member_type team) const {
using size_type = typename shmem_space::size_type;
const size_type shmemSize = 131;
shared_int values = shared_int(team.team_shmem(), shmemSize);
if (values.data() == nullptr || values.extent(0) < shmemSize) {
Kokkos::printf("FAILED to allocate shared memory of size %u\n",
static_cast<unsigned int>(shmemSize));
} else {
// Initialize shared memory.
Kokkos::parallel_for(Kokkos::TeamVectorRange(team, 131),
[&](int i) { values(i) = 0; });
// Wait for all memory to be written.
team.team_barrier();
// Accumulate value into per thread shared memory.
// This is non blocking.
Kokkos::parallel_for(Kokkos::TeamVectorRange(team, 131), [&](int i) {
values(i) +=
i - team.league_rank() + team.league_size() + team.team_size();
});
// Wait for all memory to be written.
team.team_barrier();
// One thread per team executes the comparison.
Kokkos::single(Kokkos::PerTeam(team), [&]() {
Scalar test = 0;
Scalar value = 0;
for (int i = 0; i < 131; ++i) {
test +=
i - team.league_rank() + team.league_size() + team.team_size();
}
for (int i = 0; i < 131; ++i) {
value += values(i);
}
if (test != value) {
Kokkos::printf("FAILED teamvector_parallel_for %i %i %lf %lf\n",
team.league_rank(), team.team_rank(),
static_cast<double>(test), static_cast<double>(value));
flag() = 1;
}
});
}
}
};
template <typename Scalar, class ExecutionSpace>
struct functor_teamvector_reduce {
using policy_type = Kokkos::TeamPolicy<ExecutionSpace>;
using execution_space = ExecutionSpace;
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> flag;
functor_teamvector_reduce(
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> flag_)
: flag(flag_) {}
using shmem_space = typename ExecutionSpace::scratch_memory_space;
using shared_scalar_t =
Kokkos::View<Scalar*, shmem_space, Kokkos::MemoryUnmanaged>;
unsigned team_shmem_size(int team_size) const {
return shared_scalar_t::shmem_size(team_size * 13);
}
KOKKOS_INLINE_FUNCTION
void operator()(typename policy_type::member_type team) const {
Scalar value = Scalar();
shared_scalar_t shared_value(team.team_scratch(0), 1);
Kokkos::parallel_reduce(
Kokkos::TeamVectorRange(team, 131),
[&](int i, Scalar& val) {
val += i - team.league_rank() + team.league_size() + team.team_size();
},
shared_value(0));
team.team_barrier();
Kokkos::parallel_reduce(
Kokkos::TeamVectorRange(team, 131),
[&](int i, Scalar& val) {
val += i - team.league_rank() + team.league_size() + team.team_size();
},
value);
// Kokkos::parallel_reduce( Kokkos::TeamVectorRange( team, 131 ), [&] (
// int i, Scalar & val )
// {
// val += i - team.league_rank() + team.league_size() +
// team.team_size();
// }, shared_value(0) );
team.team_barrier();
Kokkos::single(Kokkos::PerTeam(team), [&]() {
Scalar test = 0;
for (int i = 0; i < 131; ++i) {
test += i - team.league_rank() + team.league_size() + team.team_size();
}
if (test != value) {
if (team.league_rank() == 0) {
Kokkos::printf(
"FAILED teamvector_parallel_reduce %i %i %lf %lf %lu\n",
(int)team.league_rank(), (int)team.team_rank(),
static_cast<double>(test), static_cast<double>(value),
static_cast<unsigned long>(sizeof(Scalar)));
}
flag() = 1;
}
if (test != shared_value(0)) {
if (team.league_rank() == 0) {
Kokkos::printf(
"FAILED teamvector_parallel_reduce with shared result %i %i %lf "
"%lf %lu\n",
static_cast<int>(team.league_rank()),
static_cast<int>(team.team_rank()), static_cast<double>(test),
static_cast<double>(shared_value(0)),
static_cast<unsigned long>(sizeof(Scalar)));
}
flag() = 1;
}
});
}
};
template <typename Scalar, class ExecutionSpace>
struct functor_teamvector_reduce_reducer {
using policy_type = Kokkos::TeamPolicy<ExecutionSpace>;
using execution_space = ExecutionSpace;
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> flag;
functor_teamvector_reduce_reducer(
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> flag_)
: flag(flag_) {}
using shmem_space = typename ExecutionSpace::scratch_memory_space;
using shared_scalar_t =
Kokkos::View<Scalar*, shmem_space, Kokkos::MemoryUnmanaged>;
unsigned team_shmem_size(int team_size) const {
return shared_scalar_t::shmem_size(team_size * 13);
}
KOKKOS_INLINE_FUNCTION
void operator()(typename policy_type::member_type team) const {
Scalar value = 0;
shared_scalar_t shared_value(team.team_scratch(0), 1);
Kokkos::parallel_reduce(
Kokkos::TeamVectorRange(team, 131),
[&](int i, Scalar& val) {
val += i - team.league_rank() + team.league_size() + team.team_size();
},
Kokkos::Sum<Scalar>(value));
Kokkos::parallel_reduce(
Kokkos::TeamVectorRange(team, 131),
[&](int i, Scalar& val) {
val += i - team.league_rank() + team.league_size() + team.team_size();
},
Kokkos::Sum<Scalar>(shared_value(0)));
team.team_barrier();
Kokkos::single(Kokkos::PerTeam(team), [&]() {
Scalar test = 0;
for (int i = 0; i < 131; ++i) {
test += i - team.league_rank() + team.league_size() + team.team_size();
}
if (test != value) {
Kokkos::printf(
"FAILED teamvector_parallel_reduce_reducer %i %i %lf %lf\n",
team.league_rank(), team.team_rank(), static_cast<double>(test),
static_cast<double>(value));
flag() = 1;
}
if (test != shared_value(0)) {
Kokkos::printf(
"FAILED teamvector_parallel_reduce_reducer shared value %i %i %lf "
"%lf\n",
team.league_rank(), team.team_rank(), static_cast<double>(test),
static_cast<double>(shared_value(0)));
flag() = 1;
}
});
}
};
template <typename Scalar, class ExecutionSpace>
bool test_scalar(int nteams, int team_size, int test) {
Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace> d_flag("flag");
typename Kokkos::View<int, Kokkos::LayoutLeft, ExecutionSpace>::HostMirror
h_flag("h_flag");
h_flag() = 0;
Kokkos::deep_copy(d_flag, h_flag);
Kokkos::TeamPolicy<ExecutionSpace> policy(nteams, team_size, 8);
// FIXME_OPENMPTARGET - Need to allocate scratch space via set_scratch_space
// for the OPENMPTARGET backend.
#ifdef KOKKOS_ENABLE_OPENMPTARGET
using scratch_t = Kokkos::View<Scalar*, ExecutionSpace,
Kokkos::MemoryTraits<Kokkos::Unmanaged> >;
int scratch_size = 0;
if (test == 0) {
scratch_size = scratch_t::shmem_size(131);
} else {
// FIXME_OPENMPTARGET - Currently allocating more than one team for nested
// reduction leads to runtime errors of illegal memory access, caused mostly
// due to the OpenMP memory allocation constraints.
policy = Kokkos::TeamPolicy<ExecutionSpace>(1, team_size, 8);
scratch_size = scratch_t::shmem_size(1);
}
policy.set_scratch_size(0, Kokkos::PerTeam(scratch_size));
#endif
if (test == 0) {
Kokkos::parallel_for(
"Test::TeamVectorFor", policy,
functor_teamvector_for<Scalar, ExecutionSpace>(d_flag));
} else if (test == 1) {
Kokkos::parallel_for(
"Test::TeamVectorReduce", policy,
functor_teamvector_reduce<Scalar, ExecutionSpace>(d_flag));
} else if (test == 2) {
Kokkos::parallel_for(
"Test::TeamVectorReduceReducer",
Kokkos::TeamPolicy<ExecutionSpace>(nteams, team_size, 8),
functor_teamvector_reduce_reducer<Scalar, ExecutionSpace>(d_flag));
}
Kokkos::deep_copy(h_flag, d_flag);
return (h_flag() == 0);
}
template <class ExecutionSpace>
bool Test(int test) {
bool passed = true;
// With SYCL 33*8 exceeds the maximum work group size
#ifdef KOKKOS_ENABLE_SYCL
int team_size = 31;
#else
int team_size = 33;
#endif
int const concurrency = ExecutionSpace().concurrency();
if (team_size > concurrency) team_size = concurrency;
passed = passed && test_scalar<int, ExecutionSpace>(317, team_size, test);
passed = passed &&
test_scalar<long long int, ExecutionSpace>(317, team_size, test);
passed = passed && test_scalar<float, ExecutionSpace>(317, team_size, test);
passed = passed && test_scalar<double, ExecutionSpace>(317, team_size, test);
// FIXME_OPENMPTARGET - Use of custom reducers currently results in runtime
// memory errors.
#if !defined(KOKKOS_ENABLE_OPENMPTARGET)
passed =
passed && test_scalar<my_complex, ExecutionSpace>(317, team_size, test);
#endif
return passed;
}
} // namespace TestTeamVectorRange
namespace Test {
TEST(TEST_CATEGORY, team_teamvector_range) {
ASSERT_TRUE((TestTeamVectorRange::Test<TEST_EXECSPACE>(0)));
ASSERT_TRUE((TestTeamVectorRange::Test<TEST_EXECSPACE>(1)));
// FIXME_OPENMPTARGET - Use of kokkos reducers currently results in runtime
// memory errors.
#if !defined(KOKKOS_ENABLE_OPENMPTARGET)
ASSERT_TRUE((TestTeamVectorRange::Test<TEST_EXECSPACE>(2)));
#endif
}
} // namespace Test
|