Use DeviceBuffer in GPU update and NBNXM code
[alexxy/gromacs.git] / src / gromacs / nbnxm / nbnxm.cpp
1 /*
2  * This file is part of the GROMACS molecular simulation package.
3  *
4  * Copyright (c) 2019,2020,2021, 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
36 /*! \internal \file
37  * \brief
38  * Implements the Nbnxm class
39  *
40  * \author Berk Hess <hess@kth.se>
41  * \ingroup module_nbnxm
42  */
43
44 #include "gmxpre.h"
45
46 #include "nbnxm.h"
47
48 #include "gromacs/domdec/domdec_struct.h"
49 #include "gromacs/nbnxm/atomdata.h"
50 #include "gromacs/timing/wallcycle.h"
51
52 #include "nbnxm_gpu.h"
53 #include "pairlistsets.h"
54 #include "pairsearch.h"
55
56 /*! \cond INTERNAL */
57
58 void nbnxn_put_on_grid(nonbonded_verlet_t*            nb_verlet,
59                        const matrix                   box,
60                        int                            gridIndex,
61                        const rvec                     lowerCorner,
62                        const rvec                     upperCorner,
63                        const gmx::UpdateGroupsCog*    updateGroupsCog,
64                        gmx::Range<int>                atomRange,
65                        real                           atomDensity,
66                        gmx::ArrayRef<const int>       atomInfo,
67                        gmx::ArrayRef<const gmx::RVec> x,
68                        int                            numAtomsMoved,
69                        const int*                     move)
70 {
71     nb_verlet->pairSearch_->putOnGrid(box,
72                                       gridIndex,
73                                       lowerCorner,
74                                       upperCorner,
75                                       updateGroupsCog,
76                                       atomRange,
77                                       atomDensity,
78                                       atomInfo,
79                                       x,
80                                       numAtomsMoved,
81                                       move,
82                                       nb_verlet->nbat.get());
83 }
84
85 /* Calls nbnxn_put_on_grid for all non-local domains */
86 void nbnxn_put_on_grid_nonlocal(nonbonded_verlet_t*              nbv,
87                                 const struct gmx_domdec_zones_t* zones,
88                                 gmx::ArrayRef<const int>         atomInfo,
89                                 gmx::ArrayRef<const gmx::RVec>   x)
90 {
91     for (int zone = 1; zone < zones->n; zone++)
92     {
93         rvec c0, c1;
94         for (int d = 0; d < DIM; d++)
95         {
96             c0[d] = zones->size[zone].bb_x0[d];
97             c1[d] = zones->size[zone].bb_x1[d];
98         }
99
100         nbnxn_put_on_grid(nbv,
101                           nullptr,
102                           zone,
103                           c0,
104                           c1,
105                           nullptr,
106                           { zones->cg_range[zone], zones->cg_range[zone + 1] },
107                           -1,
108                           atomInfo,
109                           x,
110                           0,
111                           nullptr);
112     }
113 }
114
115 bool nonbonded_verlet_t::isDynamicPruningStepCpu(int64_t step) const
116 {
117     return pairlistSets_->isDynamicPruningStepCpu(step);
118 }
119
120 bool nonbonded_verlet_t::isDynamicPruningStepGpu(int64_t step) const
121 {
122     return pairlistSets_->isDynamicPruningStepGpu(step);
123 }
124
125 gmx::ArrayRef<const int> nonbonded_verlet_t::getLocalAtomOrder() const
126 {
127     /* Return the atom order for the home cell (index 0) */
128     const Nbnxm::Grid& grid = pairSearch_->gridSet().grids()[0];
129
130     const int numIndices = grid.atomIndexEnd() - grid.firstAtomInColumn(0);
131
132     return gmx::constArrayRefFromArray(pairSearch_->gridSet().atomIndices().data(), numIndices);
133 }
134
135 void nonbonded_verlet_t::setLocalAtomOrder() const
136 {
137     pairSearch_->setLocalAtomOrder();
138 }
139
140 void nonbonded_verlet_t::setAtomProperties(gmx::ArrayRef<const int>  atomTypes,
141                                            gmx::ArrayRef<const real> atomCharges,
142                                            gmx::ArrayRef<const int>  atomInfo) const
143 {
144     nbnxn_atomdata_set(nbat.get(), pairSearch_->gridSet(), atomTypes, atomCharges, atomInfo);
145 }
146
147 void nonbonded_verlet_t::convertCoordinates(const gmx::AtomLocality        locality,
148                                             gmx::ArrayRef<const gmx::RVec> coordinates)
149 {
150     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
151     wallcycle_sub_start(wcycle_, ewcsNB_X_BUF_OPS);
152
153     nbnxn_atomdata_copy_x_to_nbat_x(
154             pairSearch_->gridSet(), locality, as_rvec_array(coordinates.data()), nbat.get());
155
156     wallcycle_sub_stop(wcycle_, ewcsNB_X_BUF_OPS);
157     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
158 }
159
160 void nonbonded_verlet_t::convertCoordinatesGpu(const gmx::AtomLocality locality,
161                                                DeviceBuffer<gmx::RVec> d_x,
162                                                GpuEventSynchronizer*   xReadyOnDevice)
163 {
164     wallcycle_start(wcycle_, ewcLAUNCH_GPU);
165     wallcycle_sub_start(wcycle_, ewcsLAUNCH_GPU_NB_X_BUF_OPS);
166
167     nbnxn_atomdata_x_to_nbat_x_gpu(pairSearch_->gridSet(), locality, gpu_nbv, d_x, xReadyOnDevice);
168
169     wallcycle_sub_stop(wcycle_, ewcsLAUNCH_GPU_NB_X_BUF_OPS);
170     wallcycle_stop(wcycle_, ewcLAUNCH_GPU);
171 }
172
173 gmx::ArrayRef<const int> nonbonded_verlet_t::getGridIndices() const
174 {
175     return pairSearch_->gridSet().cells();
176 }
177
178 void nonbonded_verlet_t::atomdata_add_nbat_f_to_f(const gmx::AtomLocality  locality,
179                                                   gmx::ArrayRef<gmx::RVec> force)
180 {
181
182     /* Skip the reduction if there was no short-range GPU work to do
183      * (either NB or both NB and bonded work). */
184     if (!pairlistIsSimple() && !Nbnxm::haveGpuShortRangeWork(gpu_nbv, locality))
185     {
186         return;
187     }
188
189     wallcycle_start(wcycle_, ewcNB_XF_BUF_OPS);
190     wallcycle_sub_start(wcycle_, ewcsNB_F_BUF_OPS);
191
192     reduceForces(nbat.get(), locality, pairSearch_->gridSet(), as_rvec_array(force.data()));
193
194     wallcycle_sub_stop(wcycle_, ewcsNB_F_BUF_OPS);
195     wallcycle_stop(wcycle_, ewcNB_XF_BUF_OPS);
196 }
197
198 int nonbonded_verlet_t::getNumAtoms(const gmx::AtomLocality locality) const
199 {
200     int numAtoms = 0;
201     switch (locality)
202     {
203         case gmx::AtomLocality::All: numAtoms = pairSearch_->gridSet().numRealAtomsTotal(); break;
204         case gmx::AtomLocality::Local: numAtoms = pairSearch_->gridSet().numRealAtomsLocal(); break;
205         case gmx::AtomLocality::NonLocal:
206             numAtoms = pairSearch_->gridSet().numRealAtomsTotal()
207                        - pairSearch_->gridSet().numRealAtomsLocal();
208             break;
209         case gmx::AtomLocality::Count:
210             GMX_ASSERT(false, "Count is invalid locality specifier");
211             break;
212     }
213     return numAtoms;
214 }
215
216 DeviceBuffer<gmx::RVec> nonbonded_verlet_t::getGpuForces() const
217 {
218     return Nbnxm::getGpuForces(gpu_nbv);
219 }
220
221 real nonbonded_verlet_t::pairlistInnerRadius() const
222 {
223     return pairlistSets_->params().rlistInner;
224 }
225
226 real nonbonded_verlet_t::pairlistOuterRadius() const
227 {
228     return pairlistSets_->params().rlistOuter;
229 }
230
231 void nonbonded_verlet_t::changePairlistRadii(real rlistOuter, real rlistInner) const
232 {
233     pairlistSets_->changePairlistRadii(rlistOuter, rlistInner);
234 }
235
236 void nonbonded_verlet_t::setupGpuShortRangeWork(const gmx::GpuBonded*          gpuBonded,
237                                                 const gmx::InteractionLocality iLocality) const
238 {
239     if (useGpu() && !emulateGpu())
240     {
241         Nbnxm::setupGpuShortRangeWork(gpu_nbv, gpuBonded, iLocality);
242     }
243 }
244
245 void nonbonded_verlet_t::atomdata_init_copy_x_to_nbat_x_gpu() const
246 {
247     Nbnxm::nbnxn_gpu_init_x_to_nbat_x(pairSearch_->gridSet(), gpu_nbv);
248 }
249
250 /*! \endcond */