2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2017, 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.
36 * \brief Common functions for the different NBNXN GPU implementations.
38 * \author Szilard Pall <pall.szilard@gmail.com>
40 * \ingroup module_mdlib
43 #ifndef GMX_MDLIB_NBNXN_GPU_COMMON_H
44 #define GMX_MDLIB_NBNXN_GPU_COMMON_H
50 #if GMX_GPU == GMX_GPU_CUDA
51 #include "nbnxn_cuda/nbnxn_cuda_types.h"
54 #if GMX_GPU == GMX_GPU_OPENCL
55 #include "nbnxn_ocl/nbnxn_ocl_types.h"
58 #include "gromacs/gpu_utils/gpu_utils.h"
59 #include "gromacs/math/vec.h"
60 #include "gromacs/mdlib/nbnxn_gpu_types.h"
61 #include "gromacs/pbcutil/ishift.h"
62 #include "gromacs/timing/gpu_timing.h"
63 #include "gromacs/utility/stringutil.h"
65 #include "nbnxn_gpu_common_utils.h"
67 /*! \brief Check that atom locality values are valid for the GPU module.
69 * In the GPU module atom locality "all" is not supported, the local and
70 * non-local ranges are treated separately.
72 * \param[in] atomLocality atom locality specifier
74 static inline void validateGpuAtomLocality(int atomLocality)
76 std::string str = gmx::formatString("Invalid atom locality passed (%d); valid here is only "
77 "local (%d) or nonlocal (%d)", atomLocality, eatLocal, eatNonlocal);
79 GMX_ASSERT(LOCAL_OR_NONLOCAL_A(atomLocality), str.c_str());
82 /*! \brief Convert atom locality to interaction locality.
84 * In the current implementation the this is straightforward conversion:
85 * local to local, non-local to non-local.
87 * \param[in] atomLocality Atom locality specifier
88 * \returns Interaction locality corresponding to the atom locality passed.
90 static inline int gpuAtomToInteractionLocality(int atomLocality)
92 validateGpuAtomLocality(atomLocality);
94 /* determine interaction locality from atom locality */
95 if (LOCAL_A(atomLocality))
99 else if (NONLOCAL_A(atomLocality))
111 /*! \brief Calculate atom range and return start index and length.
113 * \param[in] atomData Atom descriptor data structure
114 * \param[in] atomLocality Atom locality specifier
115 * \param[out] atomRangeBegin Starting index of the atom range in the atom data array.
116 * \param[out] atomRangeLen Atom range length in the atom data array.
118 template <typename AtomDataT>
119 static inline void getGpuAtomRange(const AtomDataT *atomData,
125 validateGpuAtomLocality(atomLocality);
127 /* calculate the atom data index range based on locality */
128 if (LOCAL_A(atomLocality))
131 atomRangeLen = atomData->natoms_local;
135 atomRangeBegin = atomData->natoms_local;
136 atomRangeLen = atomData->natoms - atomData->natoms_local;
141 /*! \brief Count pruning kernel time if either kernel has been triggered
143 * We do the accounting for either of the two pruning kernel flavors:
144 * - 1st pass prune: ran during the current step (prior to the force kernel);
145 * - rolling prune: ran at the end of the previous step (prior to the current step H2D xq);
147 * Note that the resetting of cu_timers_t::didPrune and cu_timers_t::didRollingPrune should happen
148 * after calling this function.
150 * \param[in] timers structs with GPU timer objects
151 * \param[inout] timings GPU task timing data
152 * \param[in] iloc interaction locality
154 template <typename GpuTimers>
155 static void countPruneKernelTime(GpuTimers *timers,
156 gmx_wallclock_gpu_nbnxn_t *timings,
159 // We might have not done any pruning (e.g. if we skipped with empty domains).
160 if (!timers->didPrune[iloc] && !timers->didRollingPrune[iloc])
165 if (timers->didPrune[iloc])
167 timings->pruneTime.c++;
168 timings->pruneTime.t += timers->prune_k[iloc].getLastRangeTime();
171 if (timers->didRollingPrune[iloc])
173 timings->dynamicPruneTime.c++;
174 timings->dynamicPruneTime.t += timers->rollingPrune_k[iloc].getLastRangeTime();
178 /*! \brief Reduce data staged internally in the nbnxn module.
180 * Shift forces and electrostatic/LJ energies copied from the GPU into
181 * a module-internal staging area are immediately reduced (CPU-side buffers passed)
182 * after having waited for the transfers' completion.
184 * Note that this function should always be called after the transfers into the
185 * staging buffers has completed.
187 * \tparam StagingData Type of staging data
188 * \param[in] nbst Nonbonded staging data
189 * \param[in] iLocality Interaction locality specifier
190 * \param[in] reduceEnergies True if energy reduction should be done
191 * \param[in] reduceFshift True if shift force reduction should be done
192 * \param[out] e_lj Variable to accumulate LJ energy into
193 * \param[out] e_el Variable to accumulate electrostatic energy into
194 * \param[out] fshift Pointer to the array of shift forces to accumulate into
196 template <typename StagingData>
197 static inline void nbnxn_gpu_reduce_staged_outputs(const StagingData &nbst,
205 /* add up energies and shift forces (only once at local F wait) */
206 if (LOCAL_I(iLocality))
216 for (int i = 0; i < SHIFTS; i++)
218 rvec_inc(fshift[i], nbst.fshift[i]);
224 /*! \brief Do the per-step timing accounting of the nonbonded tasks.
226 * Does timing accumulation and call-count increments for the nonbonded kernels.
227 * Note that this function should be called after the current step's nonbonded
228 * nonbonded tasks have completed with the exception of the rolling pruning kernels
229 * that are accounted for during the following step.
231 * NOTE: if timing with multiple GPUs (streams) becomes possible, the
232 * counters could end up being inconsistent due to not being incremented
233 * on some of the node when this is skipped on empty local domains!
235 * \tparam GpuTimers GPU timers type
236 * \tparam GpuPairlist Pair list type
237 * \param[out] timings Pointer to the NB GPU timings data
238 * \param[in] timers Pointer to GPU timers data
239 * \param[in] plist Pointer to the pair list data
240 * \param[in] atomLocality Atom locality specifier
241 * \param[in] didEnergyKernels True if energy kernels have been called in the current step
242 * \param[in] doTiming True if timing is enabled.
245 template <typename GpuTimers, typename GpuPairlist>
246 static inline void nbnxn_gpu_accumulate_timings(gmx_wallclock_gpu_nbnxn_t *timings,
248 const GpuPairlist *plist,
250 bool didEnergyKernels,
253 /* timing data accumulation */
259 /* determine interaction locality from atom locality */
260 int iLocality = gpuAtomToInteractionLocality(atomLocality);
262 /* only increase counter once (at local F wait) */
263 if (LOCAL_I(iLocality))
266 timings->ktime[plist->haveFreshList ? 1 : 0][didEnergyKernels ? 1 : 0].c += 1;
270 timings->ktime[plist->haveFreshList ? 1 : 0][didEnergyKernels ? 1 : 0].t +=
271 timers->nb_k[iLocality].getLastRangeTime();
273 /* X/q H2D and F D2H timings */
274 timings->nb_h2d_t += timers->nb_h2d[iLocality].getLastRangeTime();
275 timings->nb_d2h_t += timers->nb_d2h[iLocality].getLastRangeTime();
277 /* Count the pruning kernel times for both cases:1st pass (at search step)
278 and rolling pruning (if called at the previous step).
279 We do the accounting here as this is the only sync point where we
280 know (without checking or additional sync-ing) that prune tasks in
281 in the current stream have completed (having just blocking-waited
282 for the force D2H). */
283 countPruneKernelTime(timers, timings, iLocality);
285 /* only count atdat and pair-list H2D at pair-search step */
286 if (timers->didPairlistH2D[iLocality])
288 /* atdat transfer timing (add only once, at local F wait) */
289 if (LOCAL_A(atomLocality))
292 timings->pl_h2d_t += timers->atdat.getLastRangeTime();
295 timings->pl_h2d_t += timers->pl_h2d[iLocality].getLastRangeTime();
297 /* Clear the timing flag for the next step */
298 timers->didPairlistH2D[iLocality] = false;
302 bool nbnxn_gpu_try_finish_task(gmx_nbnxn_gpu_t *nb,
308 GpuTaskCompletion completionKind)
310 /* determine interaction locality from atom locality */
311 int iLocality = gpuAtomToInteractionLocality(aloc);
313 // We skip when during the non-local phase there was actually no work to do.
314 // This is consistent with nbnxn_gpu_launch_kernel.
315 if (!canSkipWork(nb, iLocality))
317 // Query the state of the GPU stream and return early if we're not done
318 if (completionKind == GpuTaskCompletion::Check)
320 if (!haveStreamTasksCompleted(nb->stream[iLocality]))
322 // Early return to skip the steps below that we have to do only
323 // after the NB task completed
329 gpuStreamSynchronize(nb->stream[iLocality]);
332 bool calcEner = flags & GMX_FORCE_ENERGY;
333 bool calcFshift = flags & GMX_FORCE_VIRIAL;
335 nbnxn_gpu_accumulate_timings(nb->timings, nb->timers, nb->plist[iLocality], aloc, calcEner, nb->bDoTime);
337 nbnxn_gpu_reduce_staged_outputs(nb->nbst, iLocality, calcEner, calcFshift, e_lj, e_el, fshift);
340 /* Always reset both pruning flags (doesn't hurt doing it even when timing is off). */
341 nb->timers->didPrune[iLocality] = nb->timers->didRollingPrune[iLocality] = false;
343 /* Turn off initial list pruning (doesn't hurt if this is not pair-search step). */
344 nb->plist[iLocality]->haveFreshList = false;
350 * Wait for the asynchronously launched nonbonded tasks and data
351 * transfers to finish.
353 * Also does timing accounting and reduction of the internal staging buffers.
354 * As this is called at the end of the step, it also resets the pair list and
357 * \param[in] nb The nonbonded data GPU structure
358 * \param[in] flags Force flags
359 * \param[in] aloc Atom locality identifier
360 * \param[out] e_lj Pointer to the LJ energy output to accumulate into
361 * \param[out] e_el Pointer to the electrostatics energy output to accumulate into
362 * \param[out] fshift Pointer to the shift force buffer to accumulate into
364 void nbnxn_gpu_wait_finish_task(gmx_nbnxn_gpu_t *nb,
371 nbnxn_gpu_try_finish_task(nb, flags, aloc, e_lj, e_el, fshift,
372 GpuTaskCompletion::Wait);