2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016 by the GROMACS development team.
5 * Copyright (c) 2017,2018,2019,2020, by the GROMACS development team, led by
6 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 * and including many others, as listed in the AUTHORS file in the
8 * top-level source directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
37 #ifndef GMX_NBNXM_PAIRLIST_H
38 #define GMX_NBNXM_PAIRLIST_H
42 #include "gromacs/gpu_utils/hostallocator.h"
43 #include "gromacs/math/vectypes.h"
44 #include "gromacs/mdtypes/locality.h"
45 #include "gromacs/mdtypes/nblist.h"
46 #include "gromacs/utility/basedefinitions.h"
47 #include "gromacs/utility/defaultinitializationallocator.h"
48 #include "gromacs/utility/enumerationhelpers.h"
49 #include "gromacs/utility/real.h"
51 // This file with constants is separate from this file to be able
52 // to include it during OpenCL jitting without including config.h
53 #include "constants.h"
54 #include "pairlistparams.h"
56 struct NbnxnPairlistCpuWork;
57 struct NbnxnPairlistGpuWork;
60 //! Convenience type for vector with aligned memory
62 using AlignedVector = std::vector<T, gmx::AlignedAllocator<T>>;
64 //! Convenience type for vector that avoids initialization at resize()
66 using FastVector = std::vector<T, gmx::DefaultInitializationAllocator<T>>;
68 /*! \brief Cache-line protection buffer
70 * A buffer data structure of 64 bytes
71 * to be placed at the beginning and end of structs
72 * to avoid cache invalidation of the real contents
73 * of the struct by writes to neighboring memory.
77 //! Unused field used to create space to protect cache lines that are in use
79 } gmx_cache_protect_t;
81 /*! \brief This is the actual cluster-pair list j-entry.
83 * cj is the j-cluster.
84 * The interaction bits in excl are indexed i-major, j-minor.
85 * The cj entries are sorted such that ones with exclusions come first.
86 * This means that once a full mask (=NBNXN_INTERACTION_MASK_ALL)
87 * is found, all subsequent j-entries in the i-entry also have full masks.
93 //! The exclusion (interaction) bits
97 /*! \brief Constants for interpreting interaction flags
99 * In nbnxn_ci_t the integer shift contains the shift in the lower 7 bits.
100 * The upper bits contain information for non-bonded kernel optimization.
101 * Simply calculating LJ and Coulomb for all pairs in a cluster pair is fine.
102 * But three flags can be used to skip interactions, currently only for subc=0
103 * !(shift & NBNXN_CI_DO_LJ(subc)) => we can skip LJ for all pairs
104 * shift & NBNXN_CI_HALF_LJ(subc) => we can skip LJ for the second half of i
105 * !(shift & NBNXN_CI_DO_COUL(subc)) => we can skip Coulomb for all pairs
108 #define NBNXN_CI_SHIFT 127
109 #define NBNXN_CI_DO_LJ(subc) (1 << (7 + 3 * (subc)))
110 #define NBNXN_CI_HALF_LJ(subc) (1 << (8 + 3 * (subc)))
111 #define NBNXN_CI_DO_COUL(subc) (1 << (9 + 3 * (subc)))
114 /*! \brief Cluster-pair Interaction masks
116 * Bit i*j-cluster-size + j tells if atom i and j interact.
119 // TODO: Rename according to convention when moving into Nbnxn namespace
120 //! All interaction mask is the same for all kernels
121 constexpr unsigned int NBNXN_INTERACTION_MASK_ALL = 0xffffffffU;
122 //! 4x4 kernel diagonal mask
123 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG = 0x08ceU;
124 //! 4x2 kernel diagonal masks
126 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_0 = 0x0002U;
127 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_1 = 0x002fU;
129 //! 4x8 kernel diagonal masks
131 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_0 = 0xf0f8fcfeU;
132 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_1 = 0x0080c0e0U;
137 * \brief Simple pair-list i-unit
143 //! Shift vector index plus possible flags, see above
145 //! Start index into cj
147 //! End index into cj
151 //! Grouped pair-list i-unit
154 //! Returns the number of j-cluster groups in this entry
155 int numJClusterGroups() const { return cj4_ind_end - cj4_ind_start; }
159 //! Shift vector index plus possible flags
161 //! Start index into cj4
163 //! End index into cj4
167 //! Interaction data for a j-group for one warp
170 //! The i-cluster interactions mask for 1 warp
171 unsigned int imask = 0U;
172 //! Index into the exclusion array for 1 warp, default index 0 which means no exclusions
176 //! Four-way j-cluster lists
180 int cj[c_nbnxnGpuJgroupSize];
181 //! The i-cluster mask data for 2 warps
182 nbnxn_im_ei_t imei[c_nbnxnGpuClusterpairSplit];
185 //! Struct for storing the atom-pair interaction bits for a cluster pair in a GPU pairlist
188 //! Constructor, sets no exclusions, so all atom pairs interacting
191 for (unsigned int& pairEntry : pair)
193 pairEntry = NBNXN_INTERACTION_MASK_ALL;
197 //! Topology exclusion interaction bits per warp
198 unsigned int pair[c_nbnxnGpuExclSize];
201 //! Cluster pairlist type for use on CPUs
202 struct NbnxnPairlistCpu
207 gmx_cache_protect_t cp0;
209 //! The number of atoms per i-cluster
211 //! The number of atoms per j-cluster
213 //! The radius for constructing the list
215 //! The i-cluster list
216 FastVector<nbnxn_ci_t> ci;
217 //! The outer, unpruned i-cluster list
218 FastVector<nbnxn_ci_t> ciOuter;
220 //! The j-cluster list, size ncj
221 FastVector<nbnxn_cj_t> cj;
222 //! The outer, unpruned j-cluster list
223 FastVector<nbnxn_cj_t> cjOuter;
224 //! The number of j-clusters that are used by ci entries in this list, will be <= cj.size()
227 //! The total number of i clusters
230 //! Working data storage for list construction
231 std::unique_ptr<NbnxnPairlistCpuWork> work;
234 gmx_cache_protect_t cp1;
237 /* Cluster pairlist type, with extra hierarchies, for on the GPU
239 * NOTE: for better performance when combining lists over threads,
240 * all vectors should use default initialization. But when
241 * changing this, excl should be intialized when adding entries.
243 struct NbnxnPairlistGpu
245 /*! \brief Constructor
247 * \param[in] pinningPolicy Sets the pinning policy for all buffers used on the GPU
249 NbnxnPairlistGpu(gmx::PinningPolicy pinningPolicy);
252 gmx_cache_protect_t cp0;
254 //! The number of atoms per i-cluster
256 //! The number of atoms per j-cluster
258 //! The number of atoms per super cluster
260 //! The radius for constructing the list
262 // The i-super-cluster list, indexes into cj4;
263 gmx::HostVector<nbnxn_sci_t> sci;
264 // The list of 4*j-cluster groups
265 gmx::HostVector<nbnxn_cj4_t> cj4;
266 // Atom interaction bits (non-exclusions)
267 gmx::HostVector<nbnxn_excl_t> excl;
268 // The total number of i-clusters
271 //! Working data storage for list construction
272 std::unique_ptr<NbnxnPairlistGpuWork> work;
275 gmx_cache_protect_t cp1;
278 //! Initializes a free-energy pair-list
279 void nbnxn_init_pairlist_fep(t_nblist* nl);