src/gromacs/mdtypes/state_propagator_data_gpu_impl_gpu.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2019, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*! \internal \file
  36  *
  37  * \brief Definitions of interfaces for GPU state data propagator object.
  38  *
  39  * \author Artem Zhmurov <zhmurov@gmail.com>
  40  *
  41  * \ingroup module_mdtypes
  42  */
  43 #include "gmxpre.h"
  44
  45 #include "config.h"
  46
  47 #if GMX_GPU != GMX_GPU_NONE
  48
  49 #if GMX_GPU == GMX_GPU_CUDA
  50 #include "gromacs/gpu_utils/cudautils.cuh"
  51 #endif
  52 #include "gromacs/gpu_utils/devicebuffer.h"
  53 #if GMX_GPU == GMX_GPU_OPENCL
  54 #include "gromacs/gpu_utils/oclutils.h"
  55 #endif
  56 #include "gromacs/math/vectypes.h"
  57 #include "gromacs/mdtypes/state_propagator_data_gpu.h"
  58 #include "gromacs/utility/classhelpers.h"
  59
  60 #include "state_propagator_data_gpu_impl.h"
  61
  62 namespace gmx
  63 {
  64
  65 StatePropagatorDataGpu::Impl::Impl(const void            *pmeStream,
  66                                    const void            *localStream,
  67                                    const void            *nonLocalStream,
  68                                    const void            *deviceContext,
  69                                    GpuApiCallBehavior     transferKind,
  70                                    int                    paddingSize) :
  71     transferKind_(transferKind),
  72     paddingSize_(paddingSize)
  73 {
  74     static_assert(GMX_GPU != GMX_GPU_NONE, "This object should only be constructed on the GPU code-paths.");
  75
  76     // TODO: Refactor when the StreamManager is introduced.
  77     if (GMX_GPU == GMX_GPU_OPENCL)
  78     {
  79         GMX_ASSERT(deviceContext != nullptr, "GPU context should be set in OpenCL builds.");
  80         GMX_ASSERT(pmeStream != nullptr, "GPU PME stream should be set in OpenCL builds.");
  81
  82         // The update stream is set to the PME stream in OpenCL, since PME stream is the only stream created in the PME context.
  83         pmeStream_      = *static_cast<const CommandStream*>(pmeStream);
  84         updateStream_   = *static_cast<const CommandStream*>(pmeStream);
  85         deviceContext_  = *static_cast<const DeviceContext*>(deviceContext);
  86         GMX_UNUSED_VALUE(localStream);
  87         GMX_UNUSED_VALUE(nonLocalStream);
  88     }
  89
  90     if (GMX_GPU == GMX_GPU_CUDA)
  91     {
  92         if (pmeStream != nullptr)
  93         {
  94             pmeStream_ = *static_cast<const CommandStream*>(pmeStream);
  95         }
  96         if (localStream != nullptr)
  97         {
  98             localStream_ = *static_cast<const CommandStream*>(localStream);
  99         }
 100         if (nonLocalStream != nullptr)
 101         {
 102             nonLocalStream_ = *static_cast<const CommandStream*>(nonLocalStream);
 103         }
 104
 105         // TODO: The update stream should be created only when it is needed.
 106 #if (GMX_GPU == GMX_GPU_CUDA)
 107         cudaError_t stat;
 108         stat = cudaStreamCreate(&updateStream_);
 109         CU_RET_ERR(stat, "CUDA stream creation failed in StatePropagatorDataGpu");
 110 #endif
 111         GMX_UNUSED_VALUE(deviceContext);
 112     }
 113
 114     // Map the atom locality to the stream that will be used for coordinates,
 115     // velocities and forces transfers. Same streams are used for H2D and D2H copies.
 116     // Note, that nullptr stream is used here to indicate that the copy is not supported.
 117     xCopyStreams_[AtomLocality::Local]    = updateStream_;
 118     xCopyStreams_[AtomLocality::NonLocal] = nonLocalStream_;
 119     xCopyStreams_[AtomLocality::All]      = updateStream_;
 120
 121     vCopyStreams_[AtomLocality::Local]    = updateStream_;
 122     vCopyStreams_[AtomLocality::NonLocal] = nullptr;
 123     vCopyStreams_[AtomLocality::All]      = updateStream_;
 124
 125     fCopyStreams_[AtomLocality::Local]    = localStream_;
 126     fCopyStreams_[AtomLocality::NonLocal] = nonLocalStream_;
 127     fCopyStreams_[AtomLocality::All]      = updateStream_;
 128 }
 129
 130 StatePropagatorDataGpu::Impl::Impl(const void            *pmeStream,
 131                                    const void            *deviceContext,
 132                                    GpuApiCallBehavior     transferKind,
 133                                    int                    paddingSize) :
 134     transferKind_(transferKind),
 135     paddingSize_(paddingSize)
 136 {
 137     static_assert(GMX_GPU != GMX_GPU_NONE, "This object should only be constructed on the GPU code-paths.");
 138
 139     if (GMX_GPU == GMX_GPU_OPENCL)
 140     {
 141         GMX_ASSERT(deviceContext != nullptr, "GPU context should be set in OpenCL builds.");
 142         deviceContext_  = *static_cast<const DeviceContext*>(deviceContext);
 143     }
 144
 145     GMX_ASSERT(pmeStream != nullptr, "GPU PME stream should be set.");
 146     pmeStream_      = *static_cast<const CommandStream*>(pmeStream);
 147
 148     localStream_    = nullptr;
 149     nonLocalStream_ = nullptr;
 150     updateStream_   = nullptr;
 151
 152
 153     // Only local/all coordinates are allowed to be copied in PME-only rank/ PME tests.
 154     // This it temporary measure to make it safe to use this class in those cases.
 155     xCopyStreams_[AtomLocality::Local]    = pmeStream_;
 156     xCopyStreams_[AtomLocality::NonLocal] = nullptr;
 157     xCopyStreams_[AtomLocality::All]      = pmeStream_;
 158
 159     vCopyStreams_[AtomLocality::Local]    = nullptr;
 160     vCopyStreams_[AtomLocality::NonLocal] = nullptr;
 161     vCopyStreams_[AtomLocality::All]      = nullptr;
 162
 163     fCopyStreams_[AtomLocality::Local]    = nullptr;
 164     fCopyStreams_[AtomLocality::NonLocal] = nullptr;
 165     fCopyStreams_[AtomLocality::All]      = nullptr;
 166 }
 167
 168 StatePropagatorDataGpu::Impl::~Impl()
 169 {
 170 }
 171
 172 void StatePropagatorDataGpu::Impl::reinit(int numAtomsLocal, int numAtomsAll)
 173 {
 174     numAtomsLocal_ = numAtomsLocal;
 175     numAtomsAll_   = numAtomsAll;
 176
 177     int numAtomsPadded;
 178     if (paddingSize_ > 0)
 179     {
 180         numAtomsPadded = ((numAtomsAll_ + paddingSize_ - 1 ) / paddingSize_ )*paddingSize_;
 181     }
 182     else
 183     {
 184         numAtomsPadded = numAtomsAll_;
 185     }
 186
 187     reallocateDeviceBuffer(&d_x_, DIM*numAtomsPadded, &d_xSize_, &d_xCapacity_, deviceContext_);
 188
 189     const size_t paddingAllocationSize = numAtomsPadded - numAtomsAll_;
 190     if (paddingAllocationSize > 0)
 191     {
 192         // The PME stream is used here because the padding region of d_x_ is only in the PME task.
 193         clearDeviceBufferAsync(&d_x_, DIM*numAtomsAll_, DIM*paddingAllocationSize, pmeStream_);
 194     }
 195
 196     reallocateDeviceBuffer(&d_v_, DIM*numAtomsAll_, &d_vSize_, &d_vCapacity_, deviceContext_);
 197     reallocateDeviceBuffer(&d_f_, DIM*numAtomsAll_, &d_fSize_, &d_fCapacity_, deviceContext_);
 198
 199 }
 200
 201 std::tuple<int, int> StatePropagatorDataGpu::Impl::getAtomRangesFromAtomLocality(AtomLocality  atomLocality)
 202 {
 203     int atomsStartAt   = 0;
 204     int numAtomsToCopy = 0;
 205     switch (atomLocality)
 206     {
 207         case AtomLocality::All:
 208             atomsStartAt    = 0;
 209             numAtomsToCopy  = numAtomsAll_;
 210             break;
 211         case AtomLocality::Local:
 212             atomsStartAt    = 0;
 213             numAtomsToCopy  = numAtomsLocal_;
 214             break;
 215         case AtomLocality::NonLocal:
 216             atomsStartAt    = numAtomsLocal_;
 217             numAtomsToCopy  = numAtomsAll_ - numAtomsLocal_;
 218             break;
 219         default:
 220             GMX_RELEASE_ASSERT(false, "Wrong range of atoms requested in GPU state data manager. Should be All, Local or NonLocal.");
 221     }
 222     GMX_ASSERT(atomsStartAt   >= 0, "The first elemtnt to copy has negative index. Probably, the GPU propagator state was not initialized.");
 223     GMX_ASSERT(numAtomsToCopy >= 0, "Number of atoms to copy is negative. Probably, the GPU propagator state was not initialized.");
 224     return std::make_tuple(atomsStartAt, numAtomsToCopy);
 225 }
 226
 227 void StatePropagatorDataGpu::Impl::copyToDevice(DeviceBuffer<float>                   d_data,
 228                                                 const gmx::ArrayRef<const gmx::RVec>  h_data,
 229                                                 int                                   dataSize,
 230                                                 AtomLocality                          atomLocality,
 231                                                 CommandStream                         commandStream)
 232 {
 233
 234     GMX_UNUSED_VALUE(dataSize);
 235
 236     GMX_ASSERT(dataSize >= 0, "Trying to copy to device buffer before it was allocated.");
 237
 238     int atomsStartAt, numAtomsToCopy;
 239     std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
 240
 241     int elementsStartAt   = atomsStartAt*DIM;
 242     int numElementsToCopy = numAtomsToCopy*DIM;
 243
 244     if (numAtomsToCopy != 0)
 245     {
 246         GMX_ASSERT(elementsStartAt + numElementsToCopy <= dataSize, "The device allocation is smaller than requested copy range.");
 247         GMX_ASSERT(atomsStartAt + numAtomsToCopy <= h_data.ssize(), "The host buffer is smaller than the requested copy range.");
 248
 249         copyToDeviceBuffer(&d_data, reinterpret_cast<const float *>(&h_data.data()[atomsStartAt]),
 250                            elementsStartAt, numElementsToCopy,
 251                            commandStream, transferKind_, nullptr);
 252     }
 253 }
 254
 255 void StatePropagatorDataGpu::Impl::copyFromDevice(gmx::ArrayRef<gmx::RVec>  h_data,
 256                                                   DeviceBuffer<float>       d_data,
 257                                                   int                       dataSize,
 258                                                   AtomLocality              atomLocality,
 259                                                   CommandStream             commandStream)
 260 {
 261
 262     GMX_UNUSED_VALUE(dataSize);
 263
 264     GMX_ASSERT(dataSize >= 0, "Trying to copy from device buffer before it was allocated.");
 265
 266     int atomsStartAt, numAtomsToCopy;
 267     std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
 268
 269     int elementsStartAt   = atomsStartAt*DIM;
 270     int numElementsToCopy = numAtomsToCopy*DIM;
 271
 272     if (numAtomsToCopy != 0)
 273     {
 274         GMX_ASSERT(elementsStartAt + numElementsToCopy <= dataSize, "The device allocation is smaller than requested copy range.");
 275         GMX_ASSERT(atomsStartAt + numAtomsToCopy <= h_data.ssize(), "The host buffer is smaller than the requested copy range.");
 276
 277         copyFromDeviceBuffer(reinterpret_cast<float*>(&h_data.data()[atomsStartAt]), &d_data,
 278                              elementsStartAt, numElementsToCopy,
 279                              commandStream, transferKind_, nullptr);
 280     }
 281 }
 282
 283 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getCoordinates()
 284 {
 285     return d_x_;
 286 }
 287
 288 void StatePropagatorDataGpu::Impl::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec>  h_x,
 289                                                         AtomLocality                          atomLocality)
 290 {
 291     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 292     CommandStream commandStream = xCopyStreams_[atomLocality];
 293     GMX_ASSERT(commandStream != nullptr, "No stream is valid for copying positions with given atom locality.");
 294
 295     copyToDevice(d_x_, h_x, d_xSize_, atomLocality, commandStream);
 296
 297     // markEvent is skipped in OpenCL as:
 298     //   - it's not needed, copy is done in the same stream as the only consumer task (PME)
 299     //   - we don't consume the events in OpenCL which is not allowed by GpuEventSynchronizer (would leak memory).
 300     // TODO: remove this by adding an event-mark free flavor of this function
 301     if (GMX_GPU == GMX_GPU_CUDA)
 302     {
 303         xReadyOnDevice_[atomLocality].markEvent(commandStream);
 304     }
 305 }
 306
 307 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getCoordinatesReadyOnDeviceEvent(AtomLocality              atomLocality,
 308                                                                                      const SimulationWorkload &simulationWork,
 309                                                                                      const StepWorkload       &stepWork)
 310 {
 311     // The provider of the coordinates may be different for local atoms. If the update is offloaded
 312     // and this is not a neighbor search step, then the consumer needs to wait for the update
 313     // to complete. Otherwise, the coordinates are copied from the host and we need to wait for
 314     // the copy event. Non-local coordinates are always provided by the H2D copy.
 315     //
 316     // TODO: This should be reconsidered to support the halo exchange.
 317     //
 318     // In OpenCL no events are used as coordinate sync is not necessary
 319     if (GMX_GPU == GMX_GPU_OPENCL)
 320     {
 321         return nullptr;
 322     }
 323     if (atomLocality == AtomLocality::Local && simulationWork.useGpuUpdate && !stepWork.doNeighborSearch)
 324     {
 325         return &xUpdatedOnDevice_;
 326     }
 327     else
 328     {
 329         return &xReadyOnDevice_[atomLocality];
 330     }
 331 }
 332
 333 void StatePropagatorDataGpu::Impl::waitCoordinatesCopiedToDevice(AtomLocality  atomLocality)
 334 {
 335     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 336     xReadyOnDevice_[atomLocality].waitForEvent();
 337 }
 338
 339 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::xUpdatedOnDevice()
 340 {
 341     return &xUpdatedOnDevice_;
 342 }
 343
 344 void StatePropagatorDataGpu::Impl::copyCoordinatesFromGpu(gmx::ArrayRef<gmx::RVec>  h_x,
 345                                                           AtomLocality              atomLocality)
 346 {
 347     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 348     CommandStream commandStream = xCopyStreams_[atomLocality];
 349     GMX_ASSERT(commandStream != nullptr, "No stream is valid for copying positions with given atom locality.");
 350
 351     copyFromDevice(h_x, d_x_, d_xSize_, atomLocality, commandStream);
 352     // TODO: Remove When event-based synchronization is introduced
 353     gpuStreamSynchronize(commandStream);
 354     // Note: unlike copyCoordinatesToGpu this is not used in OpenCL, and the conditional is not needed.
 355     xReadyOnHost_[atomLocality].markEvent(commandStream);
 356 }
 357
 358 void StatePropagatorDataGpu::Impl::waitCoordinatesReadyOnHost(AtomLocality  atomLocality)
 359 {
 360     xReadyOnHost_[atomLocality].waitForEvent();
 361 }
 362
 363
 364 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getVelocities()
 365 {
 366     return d_v_;
 367 }
 368
 369 void StatePropagatorDataGpu::Impl::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec>  h_v,
 370                                                        AtomLocality                          atomLocality)
 371 {
 372     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 373     CommandStream commandStream = vCopyStreams_[atomLocality];
 374     GMX_ASSERT(commandStream != nullptr, "No stream is valid for copying velocities with given atom locality.");
 375
 376     copyToDevice(d_v_, h_v, d_vSize_, atomLocality, commandStream);
 377     // TODO: Remove When event-based synchronization is introduced
 378     gpuStreamSynchronize(commandStream);
 379     vReadyOnDevice_[atomLocality].markEvent(commandStream);
 380 }
 381
 382 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getVelocitiesReadyOnDeviceEvent(AtomLocality  atomLocality)
 383 {
 384     return &vReadyOnDevice_[atomLocality];
 385 }
 386
 387
 388 void StatePropagatorDataGpu::Impl::copyVelocitiesFromGpu(gmx::ArrayRef<gmx::RVec>  h_v,
 389                                                          AtomLocality              atomLocality)
 390 {
 391     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 392     CommandStream commandStream = vCopyStreams_[atomLocality];
 393     GMX_ASSERT(commandStream != nullptr, "No stream is valid for copying velocities with given atom locality.");
 394
 395     copyFromDevice(h_v, d_v_, d_vSize_, atomLocality, commandStream);
 396     // TODO: Remove When event-based synchronization is introduced
 397     gpuStreamSynchronize(commandStream);
 398     vReadyOnHost_[atomLocality].markEvent(commandStream);
 399 }
 400
 401 void StatePropagatorDataGpu::Impl::waitVelocitiesReadyOnHost(AtomLocality  atomLocality)
 402 {
 403     vReadyOnHost_[atomLocality].waitForEvent();
 404 }
 405
 406
 407 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getForces()
 408 {
 409     return d_f_;
 410 }
 411
 412 void StatePropagatorDataGpu::Impl::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec>  h_f,
 413                                                    AtomLocality                          atomLocality)
 414 {
 415     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 416     CommandStream commandStream = fCopyStreams_[atomLocality];
 417     GMX_ASSERT(commandStream != nullptr, "No stream is valid for copying forces with given atom locality.");
 418
 419     copyToDevice(d_f_, h_f, d_fSize_, atomLocality, commandStream);
 420     fReadyOnDevice_[atomLocality].markEvent(commandStream);
 421 }
 422
 423 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getForcesReadyOnDeviceEvent(AtomLocality  atomLocality,
 424                                                                                 bool          useGpuFBufferOps)
 425 {
 426     if ((atomLocality == AtomLocality::Local || atomLocality == AtomLocality::NonLocal) && useGpuFBufferOps)
 427     {
 428         return &fReducedOnDevice_;
 429     }
 430     else
 431     {
 432         return &fReadyOnDevice_[atomLocality];
 433     }
 434 }
 435
 436 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::fReducedOnDevice()
 437 {
 438     return &fReducedOnDevice_;
 439 }
 440
 441 void StatePropagatorDataGpu::Impl::copyForcesFromGpu(gmx::ArrayRef<gmx::RVec>  h_f,
 442                                                      AtomLocality              atomLocality)
 443 {
 444     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 445     CommandStream commandStream = fCopyStreams_[atomLocality];
 446     GMX_ASSERT(commandStream != nullptr, "No stream is valid for copying forces with given atom locality.");
 447
 448     copyFromDevice(h_f, d_f_, d_fSize_, atomLocality, commandStream);
 449     fReadyOnHost_[atomLocality].markEvent(commandStream);
 450 }
 451
 452 void StatePropagatorDataGpu::Impl::waitForcesReadyOnHost(AtomLocality  atomLocality)
 453 {
 454     fReadyOnHost_[atomLocality].waitForEvent();
 455 }
 456
 457 void* StatePropagatorDataGpu::Impl::getUpdateStream()
 458 {
 459     return &updateStream_;
 460 }
 461
 462 int StatePropagatorDataGpu::Impl::numAtomsLocal()
 463 {
 464     return numAtomsLocal_;
 465 }
 466
 467 int StatePropagatorDataGpu::Impl::numAtomsAll()
 468 {
 469     return numAtomsAll_;
 470 }
 471
 472
 473
 474 StatePropagatorDataGpu::StatePropagatorDataGpu(const void        *pmeStream,
 475                                                const void        *localStream,
 476                                                const void        *nonLocalStream,
 477                                                const void        *deviceContext,
 478                                                GpuApiCallBehavior transferKind,
 479                                                int                paddingSize)
 480     : impl_(new Impl(pmeStream,
 481                      localStream,
 482                      nonLocalStream,
 483                      deviceContext,
 484                      transferKind,
 485                      paddingSize))
 486 {
 487 }
 488
 489 StatePropagatorDataGpu::StatePropagatorDataGpu(const void        *pmeStream,
 490                                                const void        *deviceContext,
 491                                                GpuApiCallBehavior transferKind,
 492                                                int                paddingSize)
 493     : impl_(new Impl(pmeStream,
 494                      deviceContext,
 495                      transferKind,
 496                      paddingSize))
 497 {
 498 }
 499
 500 StatePropagatorDataGpu::StatePropagatorDataGpu(StatePropagatorDataGpu && /* other */) noexcept = default;
 501
 502 StatePropagatorDataGpu &StatePropagatorDataGpu::operator=(StatePropagatorDataGpu && /* other */) noexcept = default;
 503
 504 StatePropagatorDataGpu::~StatePropagatorDataGpu() = default;
 505
 506
 507 void StatePropagatorDataGpu::reinit(int numAtomsLocal, int numAtomsAll)
 508 {
 509     return impl_->reinit(numAtomsLocal, numAtomsAll);
 510 }
 511
 512 std::tuple<int, int> StatePropagatorDataGpu::getAtomRangesFromAtomLocality(AtomLocality  atomLocality)
 513 {
 514     return impl_->getAtomRangesFromAtomLocality(atomLocality);
 515 }
 516
 517
 518 DeviceBuffer<float> StatePropagatorDataGpu::getCoordinates()
 519 {
 520     return impl_->getCoordinates();
 521 }
 522
 523 void StatePropagatorDataGpu::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec>  h_x,
 524                                                   AtomLocality                          atomLocality)
 525 {
 526     return impl_->copyCoordinatesToGpu(h_x, atomLocality);
 527 }
 528
 529 GpuEventSynchronizer* StatePropagatorDataGpu::getCoordinatesReadyOnDeviceEvent(AtomLocality              atomLocality,
 530                                                                                const SimulationWorkload &simulationWork,
 531                                                                                const StepWorkload       &stepWork)
 532 {
 533     return impl_->getCoordinatesReadyOnDeviceEvent(atomLocality, simulationWork, stepWork);
 534 }
 535
 536 void StatePropagatorDataGpu::waitCoordinatesCopiedToDevice(AtomLocality  atomLocality)
 537 {
 538     return impl_->waitCoordinatesCopiedToDevice(atomLocality);
 539 }
 540
 541 GpuEventSynchronizer* StatePropagatorDataGpu::xUpdatedOnDevice()
 542 {
 543     return impl_->xUpdatedOnDevice();
 544 }
 545
 546 void StatePropagatorDataGpu::copyCoordinatesFromGpu(gmx::ArrayRef<RVec>  h_x,
 547                                                     AtomLocality         atomLocality)
 548 {
 549     return impl_->copyCoordinatesFromGpu(h_x, atomLocality);
 550 }
 551
 552 void StatePropagatorDataGpu::waitCoordinatesReadyOnHost(AtomLocality  atomLocality)
 553 {
 554     return impl_->waitCoordinatesReadyOnHost(atomLocality);
 555 }
 556
 557
 558 DeviceBuffer<float> StatePropagatorDataGpu::getVelocities()
 559 {
 560     return impl_->getVelocities();
 561 }
 562
 563 void StatePropagatorDataGpu::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec>  h_v,
 564                                                  AtomLocality                          atomLocality)
 565 {
 566     return impl_->copyVelocitiesToGpu(h_v, atomLocality);
 567 }
 568
 569 GpuEventSynchronizer* StatePropagatorDataGpu::getVelocitiesReadyOnDeviceEvent(AtomLocality  atomLocality)
 570 {
 571     return impl_->getVelocitiesReadyOnDeviceEvent(atomLocality);
 572 }
 573
 574 void StatePropagatorDataGpu::copyVelocitiesFromGpu(gmx::ArrayRef<RVec>  h_v,
 575                                                    AtomLocality         atomLocality)
 576 {
 577     return impl_->copyVelocitiesFromGpu(h_v, atomLocality);
 578 }
 579
 580 void StatePropagatorDataGpu::waitVelocitiesReadyOnHost(AtomLocality  atomLocality)
 581 {
 582     return impl_->waitVelocitiesReadyOnHost(atomLocality);
 583 }
 584
 585
 586 DeviceBuffer<float> StatePropagatorDataGpu::getForces()
 587 {
 588     return impl_->getForces();
 589 }
 590
 591 void StatePropagatorDataGpu::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec>  h_f,
 592                                              AtomLocality                          atomLocality)
 593 {
 594     return impl_->copyForcesToGpu(h_f, atomLocality);
 595 }
 596
 597 GpuEventSynchronizer* StatePropagatorDataGpu::getForcesReadyOnDeviceEvent(AtomLocality  atomLocality,
 598                                                                           bool          useGpuFBufferOps)
 599 {
 600     return impl_->getForcesReadyOnDeviceEvent(atomLocality, useGpuFBufferOps);
 601 }
 602
 603 GpuEventSynchronizer* StatePropagatorDataGpu::fReducedOnDevice()
 604 {
 605     return impl_->fReducedOnDevice();
 606 }
 607
 608 void StatePropagatorDataGpu::copyForcesFromGpu(gmx::ArrayRef<RVec>  h_f,
 609                                                AtomLocality         atomLocality)
 610 {
 611     return impl_->copyForcesFromGpu(h_f, atomLocality);
 612 }
 613
 614 void StatePropagatorDataGpu::waitForcesReadyOnHost(AtomLocality  atomLocality)
 615 {
 616     return impl_->waitForcesReadyOnHost(atomLocality);
 617 }
 618
 619
 620 void* StatePropagatorDataGpu::getUpdateStream()
 621 {
 622     return impl_->getUpdateStream();
 623 }
 624
 625 int StatePropagatorDataGpu::numAtomsLocal()
 626 {
 627     return impl_->numAtomsLocal();
 628 }
 629
 630 int StatePropagatorDataGpu::numAtomsAll()
 631 {
 632     return impl_->numAtomsAll();
 633 }
 634
 635 }      // namespace gmx
 636
 637 #endif // GMX_GPU == GMX_GPU_NONE