2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017,2018,2019, 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 #ifndef GMX_NBNXM_PAIRLIST_H
37 #define GMX_NBNXM_PAIRLIST_H
43 #include "gromacs/gpu_utils/hostallocator.h"
44 #include "gromacs/math/vectypes.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 "gromacs/nbnxm/constants.h"
57 struct NbnxnListParameters;
58 struct NbnxnPairlistCpuWork;
59 struct NbnxnPairlistGpuWork;
60 struct nbnxn_atomdata_t;
67 enum class KernelType;
70 /* Convenience type for vector with aligned memory */
72 using AlignedVector = std::vector < T, gmx::AlignedAllocator < T>>;
74 /* Convenience type for vector that avoids initialization at resize() */
76 using FastVector = std::vector < T, gmx::DefaultInitializationAllocator < T>>;
78 enum class PairlistType : int
87 static constexpr gmx::EnumerationArray<PairlistType, int> IClusterSizePerListType = { { 4, 4, 4, 8 } };
88 static constexpr gmx::EnumerationArray<PairlistType, int> JClusterSizePerListType = { { 2, 4, 8, 8 } };
90 /* With CPU kernels the i-cluster size is always 4 atoms. */
91 static constexpr int c_nbnxnCpuIClusterSize = 4;
93 /* With GPU kernels the i and j cluster size is 8 atoms for CUDA and can be set at compile time for OpenCL */
94 #if GMX_GPU == GMX_GPU_OPENCL
95 static constexpr int c_nbnxnGpuClusterSize = GMX_OPENCL_NB_CLUSTER_SIZE;
97 static constexpr int c_nbnxnGpuClusterSize = 8;
100 /* The number of clusters in a pair-search cell, used for GPU */
101 static constexpr int c_gpuNumClusterPerCellZ = 2;
102 static constexpr int c_gpuNumClusterPerCellY = 2;
103 static constexpr int c_gpuNumClusterPerCellX = 2;
104 static constexpr int c_gpuNumClusterPerCell = c_gpuNumClusterPerCellZ*c_gpuNumClusterPerCellY*c_gpuNumClusterPerCellX;
107 /* In CUDA the number of threads in a warp is 32 and we have cluster pairs
108 * of 8*8=64 atoms, so it's convenient to store data for cluster pair halves.
110 static constexpr int c_nbnxnGpuClusterpairSplit = 2;
112 /* The fixed size of the exclusion mask array for a half cluster pair */
113 static constexpr int c_nbnxnGpuExclSize = c_nbnxnGpuClusterSize*c_nbnxnGpuClusterSize/c_nbnxnGpuClusterpairSplit;
115 /* A buffer data structure of 64 bytes
116 * to be placed at the beginning and end of structs
117 * to avoid cache invalidation of the real contents
118 * of the struct by writes to neighboring memory.
122 } gmx_cache_protect_t;
124 /* This is the actual cluster-pair list j-entry.
125 * cj is the j-cluster.
126 * The interaction bits in excl are indexed i-major, j-minor.
127 * The cj entries are sorted such that ones with exclusions come first.
128 * This means that once a full mask (=NBNXN_INTERACTION_MASK_ALL)
129 * is found, all subsequent j-entries in the i-entry also have full masks.
133 int cj; /* The j-cluster */
134 unsigned int excl; /* The exclusion (interaction) bits */
137 /* In nbnxn_ci_t the integer shift contains the shift in the lower 7 bits.
138 * The upper bits contain information for non-bonded kernel optimization.
139 * Simply calculating LJ and Coulomb for all pairs in a cluster pair is fine.
140 * But three flags can be used to skip interactions, currently only for subc=0
141 * !(shift & NBNXN_CI_DO_LJ(subc)) => we can skip LJ for all pairs
142 * shift & NBNXN_CI_HALF_LJ(subc) => we can skip LJ for the second half of i
143 * !(shift & NBNXN_CI_DO_COUL(subc)) => we can skip Coulomb for all pairs
145 #define NBNXN_CI_SHIFT 127
146 #define NBNXN_CI_DO_LJ(subc) (1<<(7+3*(subc)))
147 #define NBNXN_CI_HALF_LJ(subc) (1<<(8+3*(subc)))
148 #define NBNXN_CI_DO_COUL(subc) (1<<(9+3*(subc)))
150 /* Cluster-pair Interaction masks
151 * Bit i*j-cluster-size + j tells if atom i and j interact.
153 // TODO: Rename according to convention when moving into Nbnxn namespace
154 /* All interaction mask is the same for all kernels */
155 constexpr unsigned int NBNXN_INTERACTION_MASK_ALL = 0xffffffffU;
156 /* 4x4 kernel diagonal mask */
157 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG = 0x08ceU;
158 /* 4x2 kernel diagonal masks */
159 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_0 = 0x0002U;
160 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_1 = 0x002fU;
161 /* 4x8 kernel diagonal masks */
162 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_0 = 0xf0f8fcfeU;
163 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_1 = 0x0080c0e0U;
165 /* Simple pair-list i-unit */
168 int ci; /* i-cluster */
169 int shift; /* Shift vector index plus possible flags, see above */
170 int cj_ind_start; /* Start index into cj */
171 int cj_ind_end; /* End index into cj */
174 /* Grouped pair-list i-unit */
176 /* Returns the number of j-cluster groups in this entry */
177 int numJClusterGroups() const
179 return cj4_ind_end - cj4_ind_start;
182 int sci; /* i-super-cluster */
183 int shift; /* Shift vector index plus possible flags */
184 int cj4_ind_start; /* Start index into cj4 */
185 int cj4_ind_end; /* End index into cj4 */
188 /* Interaction data for a j-group for one warp */
191 // The i-cluster interactions mask for 1 warp
192 unsigned int imask = 0U;
193 // Index into the exclusion array for 1 warp, default index 0 which means no exclusions
198 int cj[c_nbnxnGpuJgroupSize]; /* The 4 j-clusters */
199 nbnxn_im_ei_t imei[c_nbnxnGpuClusterpairSplit]; /* The i-cluster mask data for 2 warps */
202 /* Struct for storing the atom-pair interaction bits for a cluster pair in a GPU pairlist */
205 /* Constructor, sets no exclusions, so all atom pairs interacting */
208 for (unsigned int &pairEntry : pair)
210 pairEntry = NBNXN_INTERACTION_MASK_ALL;
214 /* Topology exclusion interaction bits per warp */
215 unsigned int pair[c_nbnxnGpuExclSize];
218 /* Cluster pairlist type for use on CPUs */
219 struct NbnxnPairlistCpu
223 gmx_cache_protect_t cp0;
225 int na_ci; /* The number of atoms per i-cluster */
226 int na_cj; /* The number of atoms per j-cluster */
227 real rlist; /* The radius for constructing the list */
228 FastVector<nbnxn_ci_t> ci; /* The i-cluster list */
229 FastVector<nbnxn_ci_t> ciOuter; /* The outer, unpruned i-cluster list */
231 FastVector<nbnxn_cj_t> cj; /* The j-cluster list, size ncj */
232 FastVector<nbnxn_cj_t> cjOuter; /* The outer, unpruned j-cluster list */
233 int ncjInUse; /* The number of j-clusters that are used by ci entries in this list, will be <= cj.size() */
235 int nci_tot; /* The total number of i clusters */
237 /* Working data storage for list construction */
238 std::unique_ptr<NbnxnPairlistCpuWork> work;
240 gmx_cache_protect_t cp1;
243 /* Cluster pairlist type, with extra hierarchies, for on the GPU
245 * NOTE: for better performance when combining lists over threads,
246 * all vectors should use default initialization. But when
247 * changing this, excl should be intialized when adding entries.
249 struct NbnxnPairlistGpu
253 * \param[in] pinningPolicy Sets the pinning policy for all buffers used on the GPU
255 NbnxnPairlistGpu(gmx::PinningPolicy pinningPolicy);
257 gmx_cache_protect_t cp0;
259 int na_ci; /* The number of atoms per i-cluster */
260 int na_cj; /* The number of atoms per j-cluster */
261 int na_sc; /* The number of atoms per super cluster */
262 real rlist; /* The radius for constructing the list */
263 // The i-super-cluster list, indexes into cj4;
264 gmx::HostVector<nbnxn_sci_t> sci;
265 // The list of 4*j-cluster groups
266 gmx::HostVector<nbnxn_cj4_t> cj4;
267 // Atom interaction bits (non-exclusions)
268 gmx::HostVector<nbnxn_excl_t> excl;
269 // The total number of i-clusters
272 /* Working data storage for list construction */
273 std::unique_ptr<NbnxnPairlistGpuWork> work;
275 gmx_cache_protect_t cp1;
279 * \brief An object that holds the local or non-local pairlists
284 //! Constructor: initializes the pairlist set as empty
285 PairlistSet(Nbnxm::InteractionLocality locality,
286 const NbnxnListParameters &listParams);
290 //! Constructs the pairlists in the set using the coordinates in \p nbat
291 void constructPairlists(PairSearch *pairSearch,
292 nbnxn_atomdata_t *nbat,
293 const t_blocka *excl,
294 Nbnxm::KernelType kernelType,
295 int minimumIlistCountForGpuBalancing,
298 //! Dispatch the kernel for dynamic pairlist pruning
299 void dispatchPruneKernel(const nbnxn_atomdata_t *nbat,
300 const rvec *shift_vec,
301 Nbnxm::KernelType kernelType);
303 //! Returns the locality
304 Nbnxm::InteractionLocality locality() const
309 //! Returns the lists of CPU pairlists
310 gmx::ArrayRef<const NbnxnPairlistCpu> cpuLists() const
315 //! Returns a pointer to the GPU pairlist, nullptr when not present
316 const NbnxnPairlistGpu *gpuList() const
318 if (!gpuLists_.empty())
320 return &gpuLists_[0];
328 //! Returns the lists of free-energy pairlists, empty when nonbonded interactions are not perturbed
329 gmx::ArrayRef<t_nblist const * const> fepLists() const
335 //! The locality of the pairlist set
336 Nbnxm::InteractionLocality locality_;
337 //! List of pairlists in CPU layout
338 std::vector<NbnxnPairlistCpu> cpuLists_;
339 //! List of working list for rebalancing CPU lists
340 std::vector<NbnxnPairlistCpu> cpuListsWork_;
341 //! List of pairlists in GPU layout
342 std::vector<NbnxnPairlistGpu> gpuLists_;
343 //! Pairlist parameters describing setup and ranges
344 const NbnxnListParameters ¶ms_;
345 //! Tells whether multiple lists get merged into one (the first) after creation
347 //! Tells whether the lists is of CPU type, otherwise GPU type
349 //! Lists for perturbed interactions in simple atom-atom layout
350 std::vector<t_nblist *> fepLists_;
353 /* Pair counts for flop counting */
354 //! Total number of atom pairs for LJ+Q kernel
356 //! Total number of atom pairs for LJ kernel
358 //! Total number of atom pairs for Q kernel
362 //! Initializes a free-energy pair-list
363 void nbnxn_init_pairlist_fep(t_nblist *nl);