2 * This file is part of the GROMACS molecular simulation package.
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.
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.
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.
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.
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.
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.
37 * \brief Definitions of interfaces for GPU state data propagator object.
39 * \author Artem Zhmurov <zhmurov@gmail.com>
41 * \ingroup module_mdtypes
47 #if GMX_GPU != GMX_GPU_NONE
49 # if GMX_GPU == GMX_GPU_CUDA
50 # include "gromacs/gpu_utils/cudautils.cuh"
52 # include "gromacs/gpu_utils/devicebuffer.h"
53 # if GMX_GPU == GMX_GPU_OPENCL
54 # include "gromacs/gpu_utils/oclutils.h"
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"
61 # include "state_propagator_data_gpu_impl.h"
67 StatePropagatorDataGpu::Impl::Impl(const void* pmeStream,
68 const void* localStream,
69 const void* nonLocalStream,
70 const void* deviceContext,
71 GpuApiCallBehavior transferKind,
73 gmx_wallcycle* wcycle) :
74 transferKind_(transferKind),
75 paddingSize_(paddingSize),
78 static_assert(GMX_GPU != GMX_GPU_NONE,
79 "This object should only be constructed on the GPU code-paths.");
81 // TODO: Refactor when the StreamManager is introduced.
82 if (GMX_GPU == GMX_GPU_OPENCL)
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.");
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);
95 if (GMX_GPU == GMX_GPU_CUDA)
97 if (pmeStream != nullptr)
99 pmeStream_ = *static_cast<const CommandStream*>(pmeStream);
101 if (localStream != nullptr)
103 localStream_ = *static_cast<const CommandStream*>(localStream);
105 if (nonLocalStream != nullptr)
107 nonLocalStream_ = *static_cast<const CommandStream*>(nonLocalStream);
110 // TODO: The update stream should be created only when it is needed.
111 # if (GMX_GPU == GMX_GPU_CUDA)
113 stat = cudaStreamCreate(&updateStream_);
114 CU_RET_ERR(stat, "CUDA stream creation failed in StatePropagatorDataGpu");
116 GMX_UNUSED_VALUE(deviceContext);
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_;
126 vCopyStreams_[AtomLocality::Local] = updateStream_;
127 vCopyStreams_[AtomLocality::NonLocal] = nullptr;
128 vCopyStreams_[AtomLocality::All] = updateStream_;
130 fCopyStreams_[AtomLocality::Local] = localStream_;
131 fCopyStreams_[AtomLocality::NonLocal] = nonLocalStream_;
132 fCopyStreams_[AtomLocality::All] = updateStream_;
135 StatePropagatorDataGpu::Impl::Impl(const void* pmeStream,
136 const void* deviceContext,
137 GpuApiCallBehavior transferKind,
139 gmx_wallcycle* wcycle) :
140 transferKind_(transferKind),
141 paddingSize_(paddingSize),
144 static_assert(GMX_GPU != GMX_GPU_NONE,
145 "This object should only be constructed on the GPU code-paths.");
147 if (GMX_GPU == GMX_GPU_OPENCL)
149 GMX_ASSERT(deviceContext != nullptr, "GPU context should be set in OpenCL builds.");
150 deviceContext_ = *static_cast<const DeviceContext*>(deviceContext);
153 GMX_ASSERT(pmeStream != nullptr, "GPU PME stream should be set.");
154 pmeStream_ = *static_cast<const CommandStream*>(pmeStream);
156 localStream_ = nullptr;
157 nonLocalStream_ = nullptr;
158 updateStream_ = nullptr;
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_;
167 vCopyStreams_[AtomLocality::Local] = nullptr;
168 vCopyStreams_[AtomLocality::NonLocal] = nullptr;
169 vCopyStreams_[AtomLocality::All] = nullptr;
171 fCopyStreams_[AtomLocality::Local] = nullptr;
172 fCopyStreams_[AtomLocality::NonLocal] = nullptr;
173 fCopyStreams_[AtomLocality::All] = nullptr;
176 StatePropagatorDataGpu::Impl::~Impl() {}
178 void StatePropagatorDataGpu::Impl::reinit(int numAtomsLocal, int numAtomsAll)
180 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
181 wallcycle_sub_start_nocount(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
183 numAtomsLocal_ = numAtomsLocal;
184 numAtomsAll_ = numAtomsAll;
187 if (paddingSize_ > 0)
189 numAtomsPadded = ((numAtomsAll_ + paddingSize_ - 1) / paddingSize_) * paddingSize_;
193 numAtomsPadded = numAtomsAll_;
196 reallocateDeviceBuffer(&d_x_, DIM * numAtomsPadded, &d_xSize_, &d_xCapacity_, deviceContext_);
198 const size_t paddingAllocationSize = numAtomsPadded - numAtomsAll_;
199 if (paddingAllocationSize > 0)
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_);
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)
213 clearDeviceBufferAsync(&d_f_, 0, d_fCapacity_, localStream_);
216 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
217 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
220 std::tuple<int, int> StatePropagatorDataGpu::Impl::getAtomRangesFromAtomLocality(AtomLocality atomLocality)
222 int atomsStartAt = 0;
223 int numAtomsToCopy = 0;
224 switch (atomLocality)
226 case AtomLocality::All:
228 numAtomsToCopy = numAtomsAll_;
230 case AtomLocality::Local:
232 numAtomsToCopy = numAtomsLocal_;
234 case AtomLocality::NonLocal:
235 atomsStartAt = numAtomsLocal_;
236 numAtomsToCopy = numAtomsAll_ - numAtomsLocal_;
239 GMX_RELEASE_ASSERT(false,
240 "Wrong range of atoms requested in GPU state data manager. Should "
241 "be All, Local or NonLocal.");
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 "
249 return std::make_tuple(atomsStartAt, numAtomsToCopy);
252 void StatePropagatorDataGpu::Impl::copyToDevice(DeviceBuffer<float> d_data,
253 const gmx::ArrayRef<const gmx::RVec> h_data,
255 AtomLocality atomLocality,
256 CommandStream commandStream)
258 GMX_UNUSED_VALUE(dataSize);
260 GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
262 GMX_ASSERT(dataSize >= 0, "Trying to copy to device buffer before it was allocated.");
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);
269 int atomsStartAt, numAtomsToCopy;
270 std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
272 int elementsStartAt = atomsStartAt * DIM;
273 int numElementsToCopy = numAtomsToCopy * DIM;
275 if (numAtomsToCopy != 0)
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.");
282 copyToDeviceBuffer(&d_data, reinterpret_cast<const float*>(&h_data.data()[atomsStartAt]),
283 elementsStartAt, numElementsToCopy, commandStream, transferKind_, nullptr);
286 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
287 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
290 void StatePropagatorDataGpu::Impl::copyFromDevice(gmx::ArrayRef<gmx::RVec> h_data,
291 DeviceBuffer<float> d_data,
293 AtomLocality atomLocality,
294 CommandStream commandStream)
296 GMX_UNUSED_VALUE(dataSize);
298 GMX_ASSERT(atomLocality < AtomLocality::Count, "Wrong atom locality.");
300 GMX_ASSERT(dataSize >= 0, "Trying to copy from device buffer before it was allocated.");
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);
307 int atomsStartAt, numAtomsToCopy;
308 std::tie(atomsStartAt, numAtomsToCopy) = getAtomRangesFromAtomLocality(atomLocality);
310 int elementsStartAt = atomsStartAt * DIM;
311 int numElementsToCopy = numAtomsToCopy * DIM;
313 if (numAtomsToCopy != 0)
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.");
320 copyFromDeviceBuffer(reinterpret_cast<float*>(&h_data.data()[atomsStartAt]), &d_data,
321 elementsStartAt, numElementsToCopy, commandStream, transferKind_, nullptr);
324 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
325 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
328 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getCoordinates()
333 void StatePropagatorDataGpu::Impl::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec> h_x,
334 AtomLocality atomLocality)
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.");
341 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
342 wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
344 copyToDevice(d_x_, h_x, d_xSize_, atomLocality, commandStream);
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)
352 xReadyOnDevice_[atomLocality].markEvent(xCopyStreams_[atomLocality]);
355 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
356 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
359 GpuEventSynchronizer*
360 StatePropagatorDataGpu::Impl::getCoordinatesReadyOnDeviceEvent(AtomLocality atomLocality,
361 const SimulationWorkload& simulationWork,
362 const StepWorkload& stepWork)
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.
369 // TODO: This should be reconsidered to support the halo exchange.
371 // In OpenCL no events are used as coordinate sync is not necessary
372 if (GMX_GPU == GMX_GPU_OPENCL)
376 if (atomLocality == AtomLocality::Local && simulationWork.useGpuUpdate && !stepWork.doNeighborSearch)
378 return &xUpdatedOnDevice_;
382 return &xReadyOnDevice_[atomLocality];
386 void StatePropagatorDataGpu::Impl::waitCoordinatesCopiedToDevice(AtomLocality atomLocality)
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);
394 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::xUpdatedOnDevice()
396 return &xUpdatedOnDevice_;
399 void StatePropagatorDataGpu::Impl::copyCoordinatesFromGpu(gmx::ArrayRef<gmx::RVec> h_x, AtomLocality atomLocality)
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.");
406 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
407 wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
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);
413 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
414 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
417 void StatePropagatorDataGpu::Impl::waitCoordinatesReadyOnHost(AtomLocality atomLocality)
419 wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
420 xReadyOnHost_[atomLocality].waitForEvent();
421 wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
425 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getVelocities()
430 void StatePropagatorDataGpu::Impl::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec> h_v,
431 AtomLocality atomLocality)
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.");
438 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
439 wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
441 copyToDevice(d_v_, h_v, d_vSize_, atomLocality, commandStream);
442 vReadyOnDevice_[atomLocality].markEvent(commandStream);
444 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
445 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
448 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getVelocitiesReadyOnDeviceEvent(AtomLocality atomLocality)
450 return &vReadyOnDevice_[atomLocality];
454 void StatePropagatorDataGpu::Impl::copyVelocitiesFromGpu(gmx::ArrayRef<gmx::RVec> h_v, AtomLocality atomLocality)
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.");
461 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
462 wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
464 copyFromDevice(h_v, d_v_, d_vSize_, atomLocality, commandStream);
465 vReadyOnHost_[atomLocality].markEvent(commandStream);
467 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
468 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
471 void StatePropagatorDataGpu::Impl::waitVelocitiesReadyOnHost(AtomLocality atomLocality)
473 wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
474 vReadyOnHost_[atomLocality].waitForEvent();
475 wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
479 DeviceBuffer<float> StatePropagatorDataGpu::Impl::getForces()
484 void StatePropagatorDataGpu::Impl::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec> h_f,
485 AtomLocality atomLocality)
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.");
492 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
493 wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
495 copyToDevice(d_f_, h_f, d_fSize_, atomLocality, commandStream);
496 fReadyOnDevice_[atomLocality].markEvent(commandStream);
498 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
499 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
502 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::getForcesReadyOnDeviceEvent(AtomLocality atomLocality,
503 bool useGpuFBufferOps)
505 if ((atomLocality == AtomLocality::Local || atomLocality == AtomLocality::NonLocal) && useGpuFBufferOps)
507 return &fReducedOnDevice_;
511 return &fReadyOnDevice_[atomLocality];
515 GpuEventSynchronizer* StatePropagatorDataGpu::Impl::fReducedOnDevice()
517 return &fReducedOnDevice_;
520 void StatePropagatorDataGpu::Impl::copyForcesFromGpu(gmx::ArrayRef<gmx::RVec> h_f, AtomLocality atomLocality)
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.");
527 wallcycle_start_nocount(wcycle_, ewcLAUNCH_GPU);
528 wallcycle_sub_start(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
530 copyFromDevice(h_f, d_f_, d_fSize_, atomLocality, commandStream);
531 fReadyOnHost_[atomLocality].markEvent(commandStream);
533 wallcycle_sub_stop(wcycle_, ewcsLAUNCH_STATE_PROPAGATOR_DATA);
534 wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
537 void StatePropagatorDataGpu::Impl::waitForcesReadyOnHost(AtomLocality atomLocality)
539 wallcycle_start(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
540 fReadyOnHost_[atomLocality].waitForEvent();
541 wallcycle_stop(wcycle_, ewcWAIT_GPU_STATE_PROPAGATOR_DATA);
544 void* StatePropagatorDataGpu::Impl::getUpdateStream()
546 return &updateStream_;
549 int StatePropagatorDataGpu::Impl::numAtomsLocal()
551 return numAtomsLocal_;
554 int StatePropagatorDataGpu::Impl::numAtomsAll()
560 StatePropagatorDataGpu::StatePropagatorDataGpu(const void* pmeStream,
561 const void* localStream,
562 const void* nonLocalStream,
563 const void* deviceContext,
564 GpuApiCallBehavior transferKind,
566 gmx_wallcycle* wcycle) :
567 impl_(new Impl(pmeStream, localStream, nonLocalStream, deviceContext, transferKind, paddingSize, wcycle))
571 StatePropagatorDataGpu::StatePropagatorDataGpu(const void* pmeStream,
572 const void* deviceContext,
573 GpuApiCallBehavior transferKind,
575 gmx_wallcycle* wcycle) :
576 impl_(new Impl(pmeStream, deviceContext, transferKind, paddingSize, wcycle))
580 StatePropagatorDataGpu::StatePropagatorDataGpu(StatePropagatorDataGpu&& /* other */) noexcept = default;
582 StatePropagatorDataGpu& StatePropagatorDataGpu::operator=(StatePropagatorDataGpu&& /* other */) noexcept = default;
584 StatePropagatorDataGpu::~StatePropagatorDataGpu() = default;
587 void StatePropagatorDataGpu::reinit(int numAtomsLocal, int numAtomsAll)
589 return impl_->reinit(numAtomsLocal, numAtomsAll);
592 std::tuple<int, int> StatePropagatorDataGpu::getAtomRangesFromAtomLocality(AtomLocality atomLocality)
594 return impl_->getAtomRangesFromAtomLocality(atomLocality);
598 DeviceBuffer<float> StatePropagatorDataGpu::getCoordinates()
600 return impl_->getCoordinates();
603 void StatePropagatorDataGpu::copyCoordinatesToGpu(const gmx::ArrayRef<const gmx::RVec> h_x,
604 AtomLocality atomLocality)
606 return impl_->copyCoordinatesToGpu(h_x, atomLocality);
609 GpuEventSynchronizer*
610 StatePropagatorDataGpu::getCoordinatesReadyOnDeviceEvent(AtomLocality atomLocality,
611 const SimulationWorkload& simulationWork,
612 const StepWorkload& stepWork)
614 return impl_->getCoordinatesReadyOnDeviceEvent(atomLocality, simulationWork, stepWork);
617 void StatePropagatorDataGpu::waitCoordinatesCopiedToDevice(AtomLocality atomLocality)
619 return impl_->waitCoordinatesCopiedToDevice(atomLocality);
622 GpuEventSynchronizer* StatePropagatorDataGpu::xUpdatedOnDevice()
624 return impl_->xUpdatedOnDevice();
627 void StatePropagatorDataGpu::copyCoordinatesFromGpu(gmx::ArrayRef<RVec> h_x, AtomLocality atomLocality)
629 return impl_->copyCoordinatesFromGpu(h_x, atomLocality);
632 void StatePropagatorDataGpu::waitCoordinatesReadyOnHost(AtomLocality atomLocality)
634 return impl_->waitCoordinatesReadyOnHost(atomLocality);
638 DeviceBuffer<float> StatePropagatorDataGpu::getVelocities()
640 return impl_->getVelocities();
643 void StatePropagatorDataGpu::copyVelocitiesToGpu(const gmx::ArrayRef<const gmx::RVec> h_v,
644 AtomLocality atomLocality)
646 return impl_->copyVelocitiesToGpu(h_v, atomLocality);
649 GpuEventSynchronizer* StatePropagatorDataGpu::getVelocitiesReadyOnDeviceEvent(AtomLocality atomLocality)
651 return impl_->getVelocitiesReadyOnDeviceEvent(atomLocality);
654 void StatePropagatorDataGpu::copyVelocitiesFromGpu(gmx::ArrayRef<RVec> h_v, AtomLocality atomLocality)
656 return impl_->copyVelocitiesFromGpu(h_v, atomLocality);
659 void StatePropagatorDataGpu::waitVelocitiesReadyOnHost(AtomLocality atomLocality)
661 return impl_->waitVelocitiesReadyOnHost(atomLocality);
665 DeviceBuffer<float> StatePropagatorDataGpu::getForces()
667 return impl_->getForces();
670 void StatePropagatorDataGpu::copyForcesToGpu(const gmx::ArrayRef<const gmx::RVec> h_f, AtomLocality atomLocality)
672 return impl_->copyForcesToGpu(h_f, atomLocality);
675 GpuEventSynchronizer* StatePropagatorDataGpu::getForcesReadyOnDeviceEvent(AtomLocality atomLocality,
676 bool useGpuFBufferOps)
678 return impl_->getForcesReadyOnDeviceEvent(atomLocality, useGpuFBufferOps);
681 GpuEventSynchronizer* StatePropagatorDataGpu::fReducedOnDevice()
683 return impl_->fReducedOnDevice();
686 void StatePropagatorDataGpu::copyForcesFromGpu(gmx::ArrayRef<RVec> h_f, AtomLocality atomLocality)
688 return impl_->copyForcesFromGpu(h_f, atomLocality);
691 void StatePropagatorDataGpu::waitForcesReadyOnHost(AtomLocality atomLocality)
693 return impl_->waitForcesReadyOnHost(atomLocality);
697 void* StatePropagatorDataGpu::getUpdateStream()
699 return impl_->getUpdateStream();
702 int StatePropagatorDataGpu::numAtomsLocal()
704 return impl_->numAtomsLocal();
707 int StatePropagatorDataGpu::numAtomsAll()
709 return impl_->numAtomsAll();
714 #endif // GMX_GPU == GMX_GPU_NONE