2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018,2019, 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 // FIXME: remove the "__" prefix in front of the group def when we move the
37 // nonbonded code into separate dir.
39 /*! \libinternal \defgroup __module_nbnxm Short-range non-bonded interaction module
40 * \ingroup group_mdrun
42 * \brief Computes forces and energies for short-range pair-interactions
43 * based on the Verlet algorithm. The algorithm uses pair-lists generated
44 * at fixed intervals as well as various flavors of pair interaction kernels
45 * implemented for a wide range of CPU and GPU architectures.
47 * The module includes support for flavors of Coulomb and Lennard-Jones interaction
48 * treatment implemented for a large range of SIMD instruction sets for CPU
49 * architectures as well as in CUDA and OpenCL for GPU architectures.
50 * Additionally there is a reference CPU non-SIMD and a reference CPU
51 * for GPU pair-list setup interaction kernel.
53 * The implementation of the kernels is based on the cluster non-bonded algorithm
54 * which in the code is referred to as the NxM algorithms ("nbnxm_" prefix);
55 * for details of the algorithm see DOI:10.1016/j.cpc.2013.06.003.
57 * Algorithmically, the non-bonded computation has two different modes:
58 * A "classical" mode: generate a list every nstlist steps containing at least
59 * all atom pairs up to a distance of rlistOuter and compute pair interactions
60 * for all pairs that are within the interaction cut-off.
61 * A "dynamic pruning" mode: generate an "outer-list" up to cut-off rlistOuter
62 * every nstlist steps and prune the outer-list using a cut-off of rlistInner
63 * every nstlistPrune steps to obtain a, smaller, "inner-list". This
64 * results in fewer interaction computations and allows for a larger nstlist.
65 * On a GPU, this dynamic pruning is performed in a rolling fashion, pruning
66 * only a sub-part of the list each (second) step. This way it can often
67 * overlap with integration and constraints on the CPU.
68 * Currently a simple heuristic determines which mode will be used.
70 * TODO: add a summary list and brief descriptions of the different submodules:
71 * search, CPU kernels, GPU glue code + kernels.
73 * \author Berk Hess <hess@kth.se>
74 * \author Szilárd Páll <pall.szilard@gmail.com>
75 * \author Mark Abraham <mark.j.abraham@gmail.com>
76 * \author Anca Hamuraru <anca@streamcomputing.eu>
77 * \author Teemu Virolainen <teemu@streamcomputing.eu>
78 * \author Dimitrios Karkoulis <dimitris.karkoulis@gmail.com>
80 * TODO: add more authors!
83 /*! \libinternal \file
85 * \brief This file contains the public interface of the nbnxm module
86 * that implements the NxM atom cluster non-bonded algorithm to efficiently
87 * compute pair forces.
90 * \author Berk Hess <hess@kth.se>
91 * \author Szilárd Páll <pall.szilard@gmail.com>
94 * \ingroup __module_nbnxm
98 #ifndef GMX_NBNXM_NBNXM_H
99 #define GMX_NBNXM_NBNXM_H
103 #include "gromacs/math/vectypes.h"
104 #include "gromacs/nbnxm/pairlist.h"
105 #include "gromacs/nbnxm/pairlistset.h"
106 #include "gromacs/utility/arrayref.h"
107 #include "gromacs/utility/real.h"
109 // TODO: Remove this include and the two nbnxm includes above
110 #include "nbnxm_gpu.h"
112 struct gmx_device_info_t;
113 struct gmx_domdec_zones_t;
114 struct gmx_enerdata_t;
115 struct gmx_hw_info_t;
117 struct interaction_const_t;
126 class UpdateGroupsCog;
129 //! Help pass GPU-emulation parameters with type safety.
130 enum class EmulateGpuNonbonded : bool
132 //! Do not emulate GPUs.
139 /*! \brief Nonbonded NxN kernel types: plain C, CPU SIMD, GPU, GPU emulation */
151 /*! \brief Return a string identifying the kernel type.
153 * \param [in] kernel_type nonbonded kernel types, takes values from the nbnxn_kernel_type enum
154 * \returns a string identifying the kernel corresponding to the type passed as argument
156 const char *lookup_nbnxn_kernel_name(int kernel_type);
158 /*! \brief Ewald exclusion types */
160 ewaldexclTable, ewaldexclAnalytical
163 /*! \brief Atom locality indicator: local, non-local, all.
166 * gridding, pair-search, force calculation, x/f buffer operations
169 eatLocal = 0, eatNonlocal = 1, eatAll
172 /*! \brief Tests for local atom range */
173 #define LOCAL_A(x) ((x) == eatLocal)
174 /*! \brief Tests for non-local atom range */
175 #define NONLOCAL_A(x) ((x) == eatNonlocal)
176 /*! \brief Tests for either local or non-local atom range */
177 #define LOCAL_OR_NONLOCAL_A(x) (LOCAL_A(x) || NONLOCAL_A(x))
179 /*! \brief Interaction locality indicator
181 * Used in pair-list search/calculations in the following manner:
182 * - local interactions require local atom data and affect local output only;
183 * - non-local interactions require both local and non-local atom data and
184 * affect both local- and non-local output.
187 eintLocal = 0, eintNonlocal = 1
190 /*! \brief Tests for local interaction indicator */
191 #define LOCAL_I(x) ((x) == eintLocal)
192 /*! \brief Tests for non-local interaction indicator */
193 #define NONLOCAL_I(x) ((x) == eintNonlocal)
195 /*! \brief Flag to tell the nonbonded kernels whether to clear the force output buffers */
197 enbvClearFNo, enbvClearFYes
201 * \brief Non-bonded interaction group data structure. */
202 typedef struct nonbonded_verlet_group_t {
203 nbnxn_pairlist_set_t nbl_lists; /**< pair list(s) */
204 int kernel_type; /**< non-bonded kernel - see enum above */
205 int ewald_excl; /**< Ewald exclusion - see enum above */
206 } nonbonded_verlet_group_t;
209 * \brief Top-level non-bonded data structure for the Verlet-type cut-off scheme. */
210 typedef struct nonbonded_verlet_t {
211 std::unique_ptr<NbnxnListParameters> listParams; /**< Parameters for the search and list pruning setup */
212 std::unique_ptr<nbnxn_search> nbs; /**< n vs n atom pair searching data */
213 int ngrp; /**< number of interaction groups */
214 nonbonded_verlet_group_t grp[2]; /**< local and non-local interaction group */
215 nbnxn_atomdata_t *nbat; /**< atom data */
217 gmx_bool bUseGPU; /**< TRUE when non-bonded interactions are computed on a physical GPU */
218 EmulateGpuNonbonded emulateGpu; /**< true when non-bonded interactions are computed on the CPU using GPU-style pair lists */
219 gmx_nbnxn_gpu_t *gpu_nbv; /**< pointer to GPU nb verlet data */
220 int min_ci_balanced; /**< pair list balancing parameter
221 used for the 8x8x8 GPU kernels */
222 } nonbonded_verlet_t;
224 /*! \brief Initializes the nbnxn module */
225 void init_nb_verlet(const gmx::MDLogger &mdlog,
226 nonbonded_verlet_t **nb_verlet,
227 gmx_bool bFEP_NonBonded,
228 const t_inputrec *ir,
229 const t_forcerec *fr,
231 const gmx_hw_info_t &hardwareInfo,
232 const gmx_device_info_t *deviceInfo,
233 const gmx_mtop_t *mtop,
236 /*! \brief Put the atoms on the pair search grid.
238 * Only atoms atomStart to atomEnd in x are put on the grid.
239 * The atom_density is used to determine the grid size.
240 * When atomDensity<=0, the density is determined from atomEnd-atomStart and the corners.
241 * With domain decomposition part of the n particles might have migrated,
242 * but have not been removed yet. This count is given by nmoved.
243 * When move[i] < 0 particle i has migrated and will not be put on the grid.
244 * Without domain decomposition move will be NULL.
246 void nbnxn_put_on_grid(nbnxn_search_t nbs,
250 const rvec lowerCorner,
251 const rvec upperCorner,
252 const gmx::UpdateGroupsCog *updateGroupsCog,
257 gmx::ArrayRef<const gmx::RVec> x,
261 nbnxn_atomdata_t *nbat);
263 /*! \brief As nbnxn_put_on_grid, but for the non-local atoms
265 * with domain decomposition. Should be called after calling
266 * nbnxn_search_put_on_grid for the local atoms / home zone.
268 void nbnxn_put_on_grid_nonlocal(nbnxn_search_t nbs,
269 const struct gmx_domdec_zones_t *zones,
271 gmx::ArrayRef<const gmx::RVec> x,
273 nbnxn_atomdata_t *nbat);
275 /*! \brief Returns the number of x and y cells in the local grid */
276 void nbnxn_get_ncells(nbnxn_search_t nbs, int *ncx, int *ncy);
278 /*! \brief Returns the order indices of the atoms on the pairlist search grid */
279 gmx::ArrayRef<const int> nbnxn_get_atomorder(const nbnxn_search* nbs);
281 /*! \brief Renumbers the atom indices on the grid to consecutive order */
282 void nbnxn_set_atomorder(nbnxn_search_t nbs);
284 /*! \brief Returns the index position of the atoms on the pairlist search grid */
285 gmx::ArrayRef<const int> nbnxn_get_gridindices(const nbnxn_search* nbs);
287 /*! \brief Prune all pair-lists with given locality (currently CPU only)
289 * For all pair-lists with given locality, takes the outer list and prunes out
290 * pairs beyond the pairlist inner radius and writes the result to a list that is
291 * to be consumed by the non-bonded kernel.
293 void NbnxnDispatchPruneKernel(nonbonded_verlet_t *nbv,
295 const rvec *shift_vec);
297 /*! \brief Executes the non-bonded kernel of the GPU or launches it on the GPU */
298 void NbnxnDispatchKernel(nonbonded_verlet_t *nbv,
300 const interaction_const_t &ic,
304 gmx_enerdata_t *enerd,
307 #endif // GMX_NBNXN_NBNXN_H