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) 2019,2020, 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/timing/wallcycle.h"
  59 #    include "gromacs/utility/classhelpers.h"
  60
  61 #    include "state_propagator_data_gpu_impl.h"
  62
  63
  64 namespace gmx
  65 {
  66
  67 StatePropagatorDataGpu::Impl::Impl(const void*        pmeStream,
  68                                    const void*        localStream,
  69                                    const void*        nonLocalStream,
  70                                    const void*        deviceContext,
  71                                    GpuApiCallBehavior transferKind,
  72                                    int                paddingSize,
  73                                    gmx_wallcycle*     wcycle) :
  74     transferKind_(transferKind),
  75     paddingSize_(paddingSize),
  76     wcycle_(wcycle)
  77 {
  78     static_assert(GMX_GPU != GMX_GPU_NONE,
  79                   "This object should only be constructed on the GPU code-paths.");
  80
  81     // TODO: Refactor when the StreamManager is introduced.
  82     if (GMX_GPU == GMX_GPU_OPENCL)
  83     {
  84         GMX_ASSERT(deviceContext != nullptr, "GPU context should be set in OpenCL builds.");
  85         GMX_ASSERT(pmeStream != nullptr, "GPU PME stream should be set in OpenCL builds.");
  86
  87         // The update stream is set to the PME stream in OpenCL, since PME stream is the only stream created in the PME context.
  88         pmeStream_     = *static_cast<const CommandStream*>(pmeStream);
  89         updateStream_  = *static_cast<const CommandStream*>(pmeStream);
  90         deviceContext_ = *static_cast<const DeviceContext*>(deviceContext);
  91         GMX_UNUSED_VALUE(localStream);
  92         GMX_UNUSED_VALUE(nonLocalStream);
  93     }
  94
  95     if (GMX_GPU == GMX_GPU_CUDA)
  96     {
  97         if (pmeStream != nullptr)
  98         {
  99             pmeStream_ = *static_cast<const CommandStream*>(pmeStream);
 100         }
 101         if (localStream != nullptr)
 102         {
 103             localStream_ = *static_cast<const CommandStream*>(localStream);
 104         }
 105         if (nonLocalStream != nullptr)
 106         {
 107             nonLocalStream_ = *static_cast<const CommandStream*>(nonLocalStream);
 108         }
 109
 110         // TODO: The update stream should be created only when it is needed.
 111 #    if (GMX_GPU == GMX_GPU_CUDA)
 112         cudaError_t stat;
 113         stat = cudaStreamCreate(&updateStream_);
 114         CU_RET_ERR(stat, "CUDA stream creation failed in StatePropagatorDataGpu");
 115 #    endif
 116         GMX_UNUSED_VALUE(deviceContext);
 117     }
 118
 119     // Map the atom locality to the stream that will be used for coordinates,
 120     // velocities and forces transfers. Same streams are used for H2D and D2H copies.
 121     // Note, that nullptr stream is used here to indicate that the copy is not supported.
 122     xCopyStreams_[AtomLocality::Local]    = updateStream_;
 123     xCopyStreams_[AtomLocality::NonLocal] = nonLocalStream_;
 124     xCopyStreams_[AtomLocality::All]      = updateStream_;
 125
 126     vCopyStreams_[AtomLocality::Local]    = updateStream_;
 127     vCopyStreams_[AtomLocality::NonLocal] = nullptr;
 128     vCopyStreams_[AtomLocality::All]      = updateStream_;
 129
 130     fCopyStreams_[AtomLocality::Local]    = localStream_;
 131     fCopyStreams_[AtomLocality::NonLocal] = nonLocalStream_;
 132     fCopyStreams_[AtomLocality::All]      = updateStream_;
 133 }
 134
 135 StatePropagatorDataGpu::Impl::Impl(const void*        pmeStream,
 136                                    const void*        deviceContext,
 137                                    GpuApiCallBehavior transferKind,
 138                                    int                paddingSize,
 139                                    gmx_wallcycle*     wcycle) :
 140     transferKind_(transferKind),
 141     paddingSize_(paddingSize),
 142     wcycle_(wcycle)
 143 {
 144     static_assert(GMX_GPU != GMX_GPU_NONE,
 145                   "This object should only be constructed on the GPU code-paths.");
 146
 147     if (GMX_GPU == GMX_GPU_OPENCL)
 148     {
 149         GMX_ASSERT(deviceContext != nullptr, "GPU context should be set in OpenCL builds.");
 150         deviceContext_ = *static_cast<const DeviceContext*>(deviceContext);
 151     }
 152
 153     GMX_ASSERT(pmeStream != nullptr, "GPU PME stream should be set.");
 154     pmeStream_ = *static_cast<const CommandStream*>(pmeStream);
 155
 156     localStream_    = nullptr;
 157     nonLocalStream_ = nullptr;
 158     updateStream_   = nullptr;
 159
 160
 161     // Only local/all coordinates are allowed to be copied in PME-only rank/ PME tests.
 162     // This it temporary measure to make it safe to use this class in those cases.
 163     xCopyStreams_[AtomLocality::Local]    = pmeStream_;
 164     xCopyStreams_[AtomLocality::NonLocal] = nullptr;
 165     xCopyStreams_[AtomLocality::All]      = pmeStream_;
 166
 167     vCopyStreams_[AtomLocality::Local]    = nullptr;
 168     vCopyStreams_[AtomLocality::NonLocal] = nullptr;
 169     vCopyStreams_[AtomLocality::All]      = nullptr;
 170
 171     fCopyStreams_[AtomLocality::Local]    = nullptr;
 172     fCopyStreams_[AtomLocality::NonLocal] = nullptr;
 173     fCopyStreams_[AtomLocality::All]      = nullptr;
 174 }
 175
 176 StatePropagatorDataGpu::Impl::~Impl() {}
 177
 178 void StatePropagatorDataGpu::Impl::reinit(int numAtomsLocal, int numAtomsAll)
 179 {
 180     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 181     wallcycle_sub_start_nocount(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 182
 183     numAtomsLocal_ = numAtomsLocal;
 184     numAtomsAll_   = numAtomsAll;
 185
 186     int numAtomsPadded;
 187     if (paddingSize_ > 0)
 188     {
 189         numAtomsPadded = ((numAtomsAll_ + paddingSize_ - 1) / paddingSize_) * paddingSize_;
 190     }
 191     else
 192     {
 193         numAtomsPadded = numAtomsAll_;
 194     }
 195
 196     reallocateDeviceBuffer(&d_x_, DIM * numAtomsPadded, &d_xSize_, &d_xCapacity_, deviceContext_);
 197
 198     const size_t paddingAllocationSize = numAtomsPadded - numAtomsAll_;
 199     if (paddingAllocationSize > 0)
 200     {
 201         // The PME stream is used here because the padding region of d_x_ is only in the PME task.
 202         clearDeviceBufferAsync(&d_x_, DIM * numAtomsAll_, DIM * paddingAllocationSize, pmeStream_);
 203     }
 204
 205     reallocateDeviceBuffer(&d_v_, DIM * numAtomsAll_, &d_vSize_, &d_vCapacity_, deviceContext_);
 206     const int d_fOldCapacity = d_fCapacity_;
 207     reallocateDeviceBuffer(&d_f_, DIM * numAtomsAll_, &d_fSize_, &d_fCapacity_, deviceContext_);
 208     // Clearing of the forces can be done in local stream since the nonlocal stream cannot reach
 209     // the force accumulation stage before syncing with the local stream. Only done in CUDA,
 210     // since the force buffer ops are not implemented in OpenCL.
 211     if (GMX_GPU == GMX_GPU_CUDA && d_fCapacity_ != d_fOldCapacity)
 212     {
 213         clearDeviceBufferAsync(&d_f_, 0, d_fCapacity_, localStream_);
 214     }
 215
 216     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 217     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 218 }
 219
 220 std::tuple<int, int> StatePropagatorDataGpu::Impl::getAtomRangesFromAtomLocality(AtomLocality atomLocality)
 221 {
 222     int atomsStartAt   = 0;
 223     int numAtomsToCopy = 0;
 224     switch (atomLocality)
 225     {
 226         case AtomLocality::All:
 227             atomsStartAt   = 0;
 228             numAtomsToCopy = numAtomsAll_;
 229             break;
 230         case AtomLocality::Local:
 231             atomsStartAt   = 0;
 232             numAtomsToCopy = numAtomsLocal_;
 233             break;
 234         case AtomLocality::NonLocal:
 235             atomsStartAt   = numAtomsLocal_;
 236             numAtomsToCopy = numAtomsAll_ - numAtomsLocal_;
 237             break;
 238         default:
 239             GMX_RELEASE_ASSERT(false,
 240                                "Wrong range of atoms requested in GPU state data manager. Should "
 241                                "be All, Local or NonLocal.");
 242     }
 243     GMX_ASSERT(atomsStartAt >= 0,
 244                "The first elemtnt to copy has negative index. Probably, the GPU propagator state "
 245                "was not initialized.");
 246     GMX_ASSERT(numAtomsToCopy >= 0,
 247                "Number of atoms to copy is negative. Probably, the GPU propagator state was not "
 248                "initialized.");
 249     return std::make_tuple(atomsStartAt, numAtomsToCopy);
 250 }
 251
 252 void StatePropagatorDataGpu::Impl::copyToDevice(DeviceBuffer<float>                  d_data,
 253                                                 const gmx::ArrayRef<const gmx::RVec> h_data,
 254                                                 int                                  dataSize,
 255                                                 AtomLocality                         atomLocality,
 256                                                 CommandStream                        commandStream)
 257 {
 258     GMX_UNUSED_VALUE(dataSize);
 259
 260     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 261
 262     GMX_ASSERT(dataSize >= 0, "Trying to copy to device buffer before it was allocated.");
 263
 264     GMX_ASSERT(commandStream != nullptr,
 265                "No stream is valid for copying with given atom locality.");
 266     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 267     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 268
 269     int atomsStartAt, numAtomsToCopy;
 270     std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
 271
 272     int elementsStartAt   = atomsStartAt * DIM;
 273     int numElementsToCopy = numAtomsToCopy * DIM;
 274
 275     if (numAtomsToCopy != 0)
 276     {
 277         GMX_ASSERT(elementsStartAt + numElementsToCopy <= dataSize,
 278                    "The device allocation is smaller than requested copy range.");
 279         GMX_ASSERT(atomsStartAt + numAtomsToCopy <= h_data.ssize(),
 280                    "The host buffer is smaller than the requested copy range.");
 281
 282         copyToDeviceBuffer(&d_data, reinterpret_cast<const float*>(&h_data.data()[atomsStartAt]),
 283                            elementsStartAt, numElementsToCopy, commandStream, transferKind_, nullptr);
 284     }
 285
 286     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 287     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 288 }
 289
 290 void StatePropagatorDataGpu::Impl::copyFromDevice(gmx::ArrayRef<gmx::RVec> h_data,
 291                                                   DeviceBuffer<float>      d_data,
 292                                                   int                      dataSize,
 293                                                   AtomLocality             atomLocality,
 294                                                   CommandStream            commandStream)
 295 {
 296     GMX_UNUSED_VALUE(dataSize);
 297
 298     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 299
 300     GMX_ASSERT(dataSize >= 0, "Trying to copy from device buffer before it was allocated.");
 301
 302     GMX_ASSERT(commandStream != nullptr,
 303                "No stream is valid for copying with given atom locality.");
 304     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 305     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 306
 307     int atomsStartAt, numAtomsToCopy;
 308     std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
 309
 310     int elementsStartAt   = atomsStartAt * DIM;
 311     int numElementsToCopy = numAtomsToCopy * DIM;
 312
 313     if (numAtomsToCopy != 0)
 314     {
 315         GMX_ASSERT(elementsStartAt + numElementsToCopy <= dataSize,
 316                    "The device allocation is smaller than requested copy range.");
 317         GMX_ASSERT(atomsStartAt + numAtomsToCopy <= h_data.ssize(),
 318                    "The host buffer is smaller than the requested copy range.");
 319
 320         copyFromDeviceBuffer(reinterpret_cast<float*>(&h_data.data()[atomsStartAt]), &d_data,
 321                              elementsStartAt, numElementsToCopy, commandStream, transferKind_, nullptr);
 322     }
 323
 324     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 325     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 326 }
 327
 328 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getCoordinates()
 329 {
 330     return d_x_;
 331 }
 332
 333 void StatePropagatorDataGpu::Impl::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec> h_x,
 334                                                         AtomLocality atomLocality)
 335 {
 336     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 337     CommandStream commandStream = xCopyStreams_[atomLocality];
 338     GMX_ASSERT(commandStream != nullptr,
 339                "No stream is valid for copying positions with given atom locality.");
 340
 341     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 342     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 343
 344     copyToDevice(d_x_, h_x, d_xSize_, atomLocality, commandStream);
 345
 346     // markEvent is skipped in OpenCL as:
 347     //   - it's not needed, copy is done in the same stream as the only consumer task (PME)
 348     //   - we don't consume the events in OpenCL which is not allowed by GpuEventSynchronizer (would leak memory).
 349     // TODO: remove this by adding an event-mark free flavor of this function
 350     if (GMX_GPU == GMX_GPU_CUDA)
 351     {
 352         xReadyOnDevice_[atomLocality].markEvent(xCopyStreams_[atomLocality]);
 353     }
 354
 355     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 356     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 357 }
 358
 359 GpuEventSynchronizer*
 360 StatePropagatorDataGpu::Impl::getCoordinatesReadyOnDeviceEvent(AtomLocality atomLocality,
 361                                                                const SimulationWorkload& simulationWork,
 362                                                                const StepWorkload&       stepWork)
 363 {
 364     // The provider of the coordinates may be different for local atoms. If the update is offloaded
 365     // and this is not a neighbor search step, then the consumer needs to wait for the update
 366     // to complete. Otherwise, the coordinates are copied from the host and we need to wait for
 367     // the copy event. Non-local coordinates are always provided by the H2D copy.
 368     //
 369     // TODO: This should be reconsidered to support the halo exchange.
 370     //
 371     // In OpenCL no events are used as coordinate sync is not necessary
 372     if (GMX_GPU == GMX_GPU_OPENCL)
 373     {
 374         return nullptr;
 375     }
 376     if (atomLocality == AtomLocality::Local && simulationWork.useGpuUpdate && !stepWork.doNeighborSearch)
 377     {
 378         return &xUpdatedOnDevice_;
 379     }
 380     else
 381     {
 382         return &xReadyOnDevice_[atomLocality];
 383     }
 384 }
 385
 386 void StatePropagatorDataGpu::Impl::waitCoordinatesCopiedToDevice(AtomLocality atomLocality)
 387 {
 388     wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 389     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 390     xReadyOnDevice_[atomLocality].waitForEvent();
 391     wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 392 }
 393
 394 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::xUpdatedOnDevice()
 395 {
 396     return &xUpdatedOnDevice_;
 397 }
 398
 399 void StatePropagatorDataGpu::Impl::copyCoordinatesFromGpu(gmx::ArrayRef<gmx::RVec> h_x, AtomLocality atomLocality)
 400 {
 401     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 402     CommandStream commandStream = xCopyStreams_[atomLocality];
 403     GMX_ASSERT(commandStream != nullptr,
 404                "No stream is valid for copying positions with given atom locality.");
 405
 406     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 407     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 408
 409     copyFromDevice(h_x, d_x_, d_xSize_, atomLocality, commandStream);
 410     // Note: unlike copyCoordinatesToGpu this is not used in OpenCL, and the conditional is not needed.
 411     xReadyOnHost_[atomLocality].markEvent(commandStream);
 412
 413     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 414     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 415 }
 416
 417 void StatePropagatorDataGpu::Impl::waitCoordinatesReadyOnHost(AtomLocality atomLocality)
 418 {
 419     wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 420     xReadyOnHost_[atomLocality].waitForEvent();
 421     wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 422 }
 423
 424
 425 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getVelocities()
 426 {
 427     return d_v_;
 428 }
 429
 430 void StatePropagatorDataGpu::Impl::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec> h_v,
 431                                                        AtomLocality atomLocality)
 432 {
 433     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 434     CommandStream commandStream = vCopyStreams_[atomLocality];
 435     GMX_ASSERT(commandStream != nullptr,
 436                "No stream is valid for copying velocities with given atom locality.");
 437
 438     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 439     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 440
 441     copyToDevice(d_v_, h_v, d_vSize_, atomLocality, commandStream);
 442     vReadyOnDevice_[atomLocality].markEvent(commandStream);
 443
 444     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 445     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 446 }
 447
 448 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getVelocitiesReadyOnDeviceEvent(AtomLocality atomLocality)
 449 {
 450     return &vReadyOnDevice_[atomLocality];
 451 }
 452
 453
 454 void StatePropagatorDataGpu::Impl::copyVelocitiesFromGpu(gmx::ArrayRef<gmx::RVec> h_v, AtomLocality atomLocality)
 455 {
 456     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 457     CommandStream commandStream = vCopyStreams_[atomLocality];
 458     GMX_ASSERT(commandStream != nullptr,
 459                "No stream is valid for copying velocities with given atom locality.");
 460
 461     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 462     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 463
 464     copyFromDevice(h_v, d_v_, d_vSize_, atomLocality, commandStream);
 465     vReadyOnHost_[atomLocality].markEvent(commandStream);
 466
 467     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 468     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 469 }
 470
 471 void StatePropagatorDataGpu::Impl::waitVelocitiesReadyOnHost(AtomLocality atomLocality)
 472 {
 473     wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 474     vReadyOnHost_[atomLocality].waitForEvent();
 475     wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 476 }
 477
 478
 479 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getForces()
 480 {
 481     return d_f_;
 482 }
 483
 484 void StatePropagatorDataGpu::Impl::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec> h_f,
 485                                                    AtomLocality atomLocality)
 486 {
 487     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 488     CommandStream commandStream = fCopyStreams_[atomLocality];
 489     GMX_ASSERT(commandStream != nullptr,
 490                "No stream is valid for copying forces with given atom locality.");
 491
 492     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 493     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 494
 495     copyToDevice(d_f_, h_f, d_fSize_, atomLocality, commandStream);
 496     fReadyOnDevice_[atomLocality].markEvent(commandStream);
 497
 498     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 499     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 500 }
 501
 502 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getForcesReadyOnDeviceEvent(AtomLocality atomLocality,
 503                                                                                 bool useGpuFBufferOps)
 504 {
 505     if ((atomLocality == AtomLocality::Local || atomLocality == AtomLocality::NonLocal) && useGpuFBufferOps)
 506     {
 507         return &fReducedOnDevice_;
 508     }
 509     else
 510     {
 511         return &fReadyOnDevice_[atomLocality];
 512     }
 513 }
 514
 515 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::fReducedOnDevice()
 516 {
 517     return &fReducedOnDevice_;
 518 }
 519
 520 void StatePropagatorDataGpu::Impl::copyForcesFromGpu(gmx::ArrayRef<gmx::RVec> h_f, AtomLocality atomLocality)
 521 {
 522     GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
 523     CommandStream commandStream = fCopyStreams_[atomLocality];
 524     GMX_ASSERT(commandStream != nullptr,
 525                "No stream is valid for copying forces with given atom locality.");
 526
 527     wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
 528     wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 529
 530     copyFromDevice(h_f, d_f_, d_fSize_, atomLocality, commandStream);
 531     fReadyOnHost_[atomLocality].markEvent(commandStream);
 532
 533     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
 534     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
 535 }
 536
 537 void StatePropagatorDataGpu::Impl::waitForcesReadyOnHost(AtomLocality atomLocality)
 538 {
 539     wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 540     fReadyOnHost_[atomLocality].waitForEvent();
 541     wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
 542 }
 543
 544 void* StatePropagatorDataGpu::Impl::getUpdateStream()
 545 {
 546     return &updateStream_;
 547 }
 548
 549 int StatePropagatorDataGpu::Impl::numAtomsLocal()
 550 {
 551     return numAtomsLocal_;
 552 }
 553
 554 int StatePropagatorDataGpu::Impl::numAtomsAll()
 555 {
 556     return numAtomsAll_;
 557 }
 558
 559
 560 StatePropagatorDataGpu::StatePropagatorDataGpu(const void*        pmeStream,
 561                                                const void*        localStream,
 562                                                const void*        nonLocalStream,
 563                                                const void*        deviceContext,
 564                                                GpuApiCallBehavior transferKind,
 565                                                int                paddingSize,
 566                                                gmx_wallcycle*     wcycle) :
 567     impl_(new Impl(pmeStream, localStream, nonLocalStream, deviceContext, transferKind, paddingSize, wcycle))
 568 {
 569 }
 570
 571 StatePropagatorDataGpu::StatePropagatorDataGpu(const void*        pmeStream,
 572                                                const void*        deviceContext,
 573                                                GpuApiCallBehavior transferKind,
 574                                                int                paddingSize,
 575                                                gmx_wallcycle*     wcycle) :
 576     impl_(new Impl(pmeStream, deviceContext, transferKind, paddingSize, wcycle))
 577 {
 578 }
 579
 580 StatePropagatorDataGpu::StatePropagatorDataGpu(StatePropagatorDataGpu&& /* other */) noexcept = default;
 581
 582 StatePropagatorDataGpu& StatePropagatorDataGpu::operator=(StatePropagatorDataGpu&& /* other */) noexcept = default;
 583
 584 StatePropagatorDataGpu::~StatePropagatorDataGpu() = default;
 585
 586
 587 void StatePropagatorDataGpu::reinit(int numAtomsLocal, int numAtomsAll)
 588 {
 589     return impl_->reinit(numAtomsLocal, numAtomsAll);
 590 }
 591
 592 std::tuple<int, int> StatePropagatorDataGpu::getAtomRangesFromAtomLocality(AtomLocality atomLocality)
 593 {
 594     return impl_->getAtomRangesFromAtomLocality(atomLocality);
 595 }
 596
 597
 598 DeviceBuffer<float> StatePropagatorDataGpu::getCoordinates()
 599 {
 600     return impl_->getCoordinates();
 601 }
 602
 603 void StatePropagatorDataGpu::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec> h_x,
 604                                                   AtomLocality                         atomLocality)
 605 {
 606     return impl_->copyCoordinatesToGpu(h_x, atomLocality);
 607 }
 608
 609 GpuEventSynchronizer*
 610 StatePropagatorDataGpu::getCoordinatesReadyOnDeviceEvent(AtomLocality              atomLocality,
 611                                                          const SimulationWorkload& simulationWork,
 612                                                          const StepWorkload&       stepWork)
 613 {
 614     return impl_->getCoordinatesReadyOnDeviceEvent(atomLocality, simulationWork, stepWork);
 615 }
 616
 617 void StatePropagatorDataGpu::waitCoordinatesCopiedToDevice(AtomLocality atomLocality)
 618 {
 619     return impl_->waitCoordinatesCopiedToDevice(atomLocality);
 620 }
 621
 622 GpuEventSynchronizer* StatePropagatorDataGpu::xUpdatedOnDevice()
 623 {
 624     return impl_->xUpdatedOnDevice();
 625 }
 626
 627 void StatePropagatorDataGpu::copyCoordinatesFromGpu(gmx::ArrayRef<RVec> h_x, AtomLocality atomLocality)
 628 {
 629     return impl_->copyCoordinatesFromGpu(h_x, atomLocality);
 630 }
 631
 632 void StatePropagatorDataGpu::waitCoordinatesReadyOnHost(AtomLocality atomLocality)
 633 {
 634     return impl_->waitCoordinatesReadyOnHost(atomLocality);
 635 }
 636
 637
 638 DeviceBuffer<float> StatePropagatorDataGpu::getVelocities()
 639 {
 640     return impl_->getVelocities();
 641 }
 642
 643 void StatePropagatorDataGpu::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec> h_v,
 644                                                  AtomLocality                         atomLocality)
 645 {
 646     return impl_->copyVelocitiesToGpu(h_v, atomLocality);
 647 }
 648
 649 GpuEventSynchronizer* StatePropagatorDataGpu::getVelocitiesReadyOnDeviceEvent(AtomLocality atomLocality)
 650 {
 651     return impl_->getVelocitiesReadyOnDeviceEvent(atomLocality);
 652 }
 653
 654 void StatePropagatorDataGpu::copyVelocitiesFromGpu(gmx::ArrayRef<RVec> h_v, AtomLocality atomLocality)
 655 {
 656     return impl_->copyVelocitiesFromGpu(h_v, atomLocality);
 657 }
 658
 659 void StatePropagatorDataGpu::waitVelocitiesReadyOnHost(AtomLocality atomLocality)
 660 {
 661     return impl_->waitVelocitiesReadyOnHost(atomLocality);
 662 }
 663
 664
 665 DeviceBuffer<float> StatePropagatorDataGpu::getForces()
 666 {
 667     return impl_->getForces();
 668 }
 669
 670 void StatePropagatorDataGpu::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec> h_f, AtomLocality atomLocality)
 671 {
 672     return impl_->copyForcesToGpu(h_f, atomLocality);
 673 }
 674
 675 GpuEventSynchronizer* StatePropagatorDataGpu::getForcesReadyOnDeviceEvent(AtomLocality atomLocality,
 676                                                                           bool useGpuFBufferOps)
 677 {
 678     return impl_->getForcesReadyOnDeviceEvent(atomLocality, useGpuFBufferOps);
 679 }
 680
 681 GpuEventSynchronizer* StatePropagatorDataGpu::fReducedOnDevice()
 682 {
 683     return impl_->fReducedOnDevice();
 684 }
 685
 686 void StatePropagatorDataGpu::copyForcesFromGpu(gmx::ArrayRef<RVec> h_f, AtomLocality atomLocality)
 687 {
 688     return impl_->copyForcesFromGpu(h_f, atomLocality);
 689 }
 690
 691 void StatePropagatorDataGpu::waitForcesReadyOnHost(AtomLocality atomLocality)
 692 {
 693     return impl_->waitForcesReadyOnHost(atomLocality);
 694 }
 695
 696
 697 void* StatePropagatorDataGpu::getUpdateStream()
 698 {
 699     return impl_->getUpdateStream();
 700 }
 701
 702 int StatePropagatorDataGpu::numAtomsLocal()
 703 {
 704     return impl_->numAtomsLocal();
 705 }
 706
 707 int StatePropagatorDataGpu::numAtomsAll()
 708 {
 709     return impl_->numAtomsAll();
 710 }
 711
 712 } // namespace gmx
 713
 714 #endif // GMX_GPU == GMX_GPU_NONE