SIMD support for nonbonded free-energy kernels
[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 #include "gromacs/utility/message_string_collector.h"
52
53 #include "nbnxm_gpu.h"
54 #include "pairlistsets.h"
55 #include "pairsearch.h"
56
57 /*! \cond INTERNAL */
58
59 void nbnxn_put_on_grid(nonbonded_verlet_t*            nb_verlet,
60                        const matrix                   box,
61                        int                            gridIndex,
62                        const rvec                     lowerCorner,
63                        const rvec                     upperCorner,
64                        const gmx::UpdateGroupsCog*    updateGroupsCog,
65                        gmx::Range<int>                atomRange,
66                        real                           atomDensity,
67                        gmx::ArrayRef<const int64_t>   atomInfo,
68                        gmx::ArrayRef<const gmx::RVec> x,
69                        int                            numAtomsMoved,
70                        const int*                     move)
71 {
72     nb_verlet->pairSearch_->putOnGrid(box,
73                                       gridIndex,
74                                       lowerCorner,
75                                       upperCorner,
76                                       updateGroupsCog,
77                                       atomRange,
78                                       atomDensity,
79                                       atomInfo,
80                                       x,
81                                       numAtomsMoved,
82                                       move,
83                                       nb_verlet->nbat.get());
84 }
85
86 /* Calls nbnxn_put_on_grid for all non-local domains */
87 void nbnxn_put_on_grid_nonlocal(nonbonded_verlet_t*              nbv,
88                                 const struct gmx_domdec_zones_t* zones,
89                                 gmx::ArrayRef<const int64_t>     atomInfo,
90                                 gmx::ArrayRef<const gmx::RVec>   x)
91 {
92     for (int zone = 1; zone < zones->n; zone++)
93     {
94         rvec c0, c1;
95         for (int d = 0; d < DIM; d++)
96         {
97             c0[d] = zones->size[zone].bb_x0[d];
98             c1[d] = zones->size[zone].bb_x1[d];
99         }
100
101         nbnxn_put_on_grid(nbv,
102                           nullptr,
103                           zone,
104                           c0,
105                           c1,
106                           nullptr,
107                           { zones->cg_range[zone], zones->cg_range[zone + 1] },
108                           -1,
109                           atomInfo,
110                           x,
111                           0,
112                           nullptr);
113     }
114 }
115
116 bool nonbonded_verlet_t::isDynamicPruningStepCpu(int64_t step) const
117 {
118     return pairlistSets_->isDynamicPruningStepCpu(step);
119 }
120
121 bool nonbonded_verlet_t::isDynamicPruningStepGpu(int64_t step) const
122 {
123     return pairlistSets_->isDynamicPruningStepGpu(step);
124 }
125
126 gmx::ArrayRef<const int> nonbonded_verlet_t::getLocalAtomOrder() const
127 {
128     /* Return the atom order for the home cell (index 0) */
129     const Nbnxm::Grid& grid = pairSearch_->gridSet().grids()[0];
130
131     const int numIndices = grid.atomIndexEnd() - grid.firstAtomInColumn(0);
132
133     return gmx::constArrayRefFromArray(pairSearch_->gridSet().atomIndices().data(), numIndices);
134 }
135
136 void nonbonded_verlet_t::setLocalAtomOrder() const
137 {
138     pairSearch_->setLocalAtomOrder();
139 }
140
141 void nonbonded_verlet_t::setAtomProperties(gmx::ArrayRef<const int>     atomTypes,
142                                            gmx::ArrayRef<const real>    atomCharges,
143                                            gmx::ArrayRef<const int64_t> atomInfo) const
144 {
145     nbnxn_atomdata_set(nbat.get(), pairSearch_->gridSet(), atomTypes, atomCharges, atomInfo);
146 }
147
148 void nonbonded_verlet_t::convertCoordinates(const gmx::AtomLocality        locality,
149                                             gmx::ArrayRef<const gmx::RVec> coordinates)
150 {
151     wallcycle_start(wcycle_, WallCycleCounter::NbXFBufOps);
152     wallcycle_sub_start(wcycle_, WallCycleSubCounter::NBXBufOps);
153
154     nbnxn_atomdata_copy_x_to_nbat_x(
155             pairSearch_->gridSet(), locality, as_rvec_array(coordinates.data()), nbat.get());
156
157     wallcycle_sub_stop(wcycle_, WallCycleSubCounter::NBXBufOps);
158     wallcycle_stop(wcycle_, WallCycleCounter::NbXFBufOps);
159 }
160
161 void nonbonded_verlet_t::convertCoordinatesGpu(const gmx::AtomLocality locality,
162                                                DeviceBuffer<gmx::RVec> d_x,
163                                                GpuEventSynchronizer*   xReadyOnDevice)
164 {
165     wallcycle_start(wcycle_, WallCycleCounter::LaunchGpu);
166     wallcycle_sub_start(wcycle_, WallCycleSubCounter::LaunchGpuNBXBufOps);
167
168     nbnxn_atomdata_x_to_nbat_x_gpu(pairSearch_->gridSet(), locality, gpu_nbv, d_x, xReadyOnDevice);
169
170     wallcycle_sub_stop(wcycle_, WallCycleSubCounter::LaunchGpuNBXBufOps);
171     wallcycle_stop(wcycle_, WallCycleCounter::LaunchGpu);
172 }
173
174 gmx::ArrayRef<const int> nonbonded_verlet_t::getGridIndices() const
175 {
176     return pairSearch_->gridSet().cells();
177 }
178
179 void nonbonded_verlet_t::atomdata_add_nbat_f_to_f(const gmx::AtomLocality  locality,
180                                                   gmx::ArrayRef<gmx::RVec> force)
181 {
182
183     /* Skip the reduction if there was no short-range GPU work to do
184      * (either NB or both NB and bonded work). */
185     if (!pairlistIsSimple() && !Nbnxm::haveGpuShortRangeWork(gpu_nbv, atomToInteractionLocality(locality)))
186     {
187         return;
188     }
189
190     wallcycle_start(wcycle_, WallCycleCounter::NbXFBufOps);
191     wallcycle_sub_start(wcycle_, WallCycleSubCounter::NBFBufOps);
192
193     reduceForces(nbat.get(), locality, pairSearch_->gridSet(), as_rvec_array(force.data()));
194
195     wallcycle_sub_stop(wcycle_, WallCycleSubCounter::NBFBufOps);
196     wallcycle_stop(wcycle_, WallCycleCounter::NbXFBufOps);
197 }
198
199 int nonbonded_verlet_t::getNumAtoms(const gmx::AtomLocality locality) const
200 {
201     int numAtoms = 0;
202     switch (locality)
203     {
204         case gmx::AtomLocality::All: numAtoms = pairSearch_->gridSet().numRealAtomsTotal(); break;
205         case gmx::AtomLocality::Local: numAtoms = pairSearch_->gridSet().numRealAtomsLocal(); break;
206         case gmx::AtomLocality::NonLocal:
207             numAtoms = pairSearch_->gridSet().numRealAtomsTotal()
208                        - pairSearch_->gridSet().numRealAtomsLocal();
209             break;
210         case gmx::AtomLocality::Count:
211             GMX_ASSERT(false, "Count is invalid locality specifier");
212             break;
213     }
214     return numAtoms;
215 }
216
217 real nonbonded_verlet_t::pairlistInnerRadius() const
218 {
219     return pairlistSets_->params().rlistInner;
220 }
221
222 real nonbonded_verlet_t::pairlistOuterRadius() const
223 {
224     return pairlistSets_->params().rlistOuter;
225 }
226
227 void nonbonded_verlet_t::changePairlistRadii(real rlistOuter, real rlistInner) const
228 {
229     pairlistSets_->changePairlistRadii(rlistOuter, rlistInner);
230 }
231
232 void nonbonded_verlet_t::setupGpuShortRangeWork(const gmx::ListedForcesGpu*    listedForcesGpu,
233                                                 const gmx::InteractionLocality iLocality) const
234 {
235     if (useGpu() && !emulateGpu())
236     {
237         Nbnxm::setupGpuShortRangeWork(gpu_nbv, listedForcesGpu, iLocality);
238     }
239 }
240
241 void nonbonded_verlet_t::atomdata_init_copy_x_to_nbat_x_gpu() const
242 {
243     Nbnxm::nbnxn_gpu_init_x_to_nbat_x(pairSearch_->gridSet(), gpu_nbv);
244 }
245
246 bool buildSupportsNonbondedOnGpu(std::string* error)
247 {
248     gmx::MessageStringCollector errorReasons;
249     // Before changing the prefix string, make sure that it is not searched for in regression tests.
250     errorReasons.startContext("Nonbonded interactions on GPUs are not supported in:");
251     errorReasons.appendIf(GMX_DOUBLE, "Double precision build of GROMACS");
252     errorReasons.appendIf(!GMX_GPU, "Non-GPU build of GROMACS.");
253     errorReasons.finishContext();
254     if (error != nullptr)
255     {
256         *error = errorReasons.toString();
257     }
258     return errorReasons.isEmpty();
259 }
260
261 /*! \endcond */