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 NbnxnPairlistCpuWork;
58 struct NbnxnPairlistGpuWork;
59 struct nbnxn_atomdata_t;
60 struct PairlistParams;
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 /* With CPU kernels the i-cluster size is always 4 atoms. */
79 static constexpr int c_nbnxnCpuIClusterSize = 4;
81 /* With GPU kernels the i and j cluster size is 8 atoms for CUDA and can be set at compile time for OpenCL */
82 #if GMX_GPU == GMX_GPU_OPENCL
83 static constexpr int c_nbnxnGpuClusterSize = GMX_OPENCL_NB_CLUSTER_SIZE;
85 static constexpr int c_nbnxnGpuClusterSize = 8;
88 /* The number of clusters in a pair-search cell, used for GPU */
89 static constexpr int c_gpuNumClusterPerCellZ = 2;
90 static constexpr int c_gpuNumClusterPerCellY = 2;
91 static constexpr int c_gpuNumClusterPerCellX = 2;
92 static constexpr int c_gpuNumClusterPerCell = c_gpuNumClusterPerCellZ*c_gpuNumClusterPerCellY*c_gpuNumClusterPerCellX;
95 /* In CUDA the number of threads in a warp is 32 and we have cluster pairs
96 * of 8*8=64 atoms, so it's convenient to store data for cluster pair halves.
98 static constexpr int c_nbnxnGpuClusterpairSplit = 2;
100 /* The fixed size of the exclusion mask array for a half cluster pair */
101 static constexpr int c_nbnxnGpuExclSize = c_nbnxnGpuClusterSize*c_nbnxnGpuClusterSize/c_nbnxnGpuClusterpairSplit;
103 //! The available pair list types
104 enum class PairlistType : int
113 //! Gives the i-cluster size for each pairlist type
114 static constexpr gmx::EnumerationArray<PairlistType, int> IClusterSizePerListType =
116 c_nbnxnCpuIClusterSize,
117 c_nbnxnCpuIClusterSize,
118 c_nbnxnCpuIClusterSize,
119 c_nbnxnGpuClusterSize
121 //! Gives the j-cluster size for each pairlist type
122 static constexpr gmx::EnumerationArray<PairlistType, int> JClusterSizePerListType =
127 c_nbnxnGpuClusterSize
130 /* A buffer data structure of 64 bytes
131 * to be placed at the beginning and end of structs
132 * to avoid cache invalidation of the real contents
133 * of the struct by writes to neighboring memory.
137 } gmx_cache_protect_t;
139 /* This is the actual cluster-pair list j-entry.
140 * cj is the j-cluster.
141 * The interaction bits in excl are indexed i-major, j-minor.
142 * The cj entries are sorted such that ones with exclusions come first.
143 * This means that once a full mask (=NBNXN_INTERACTION_MASK_ALL)
144 * is found, all subsequent j-entries in the i-entry also have full masks.
148 int cj; /* The j-cluster */
149 unsigned int excl; /* The exclusion (interaction) bits */
152 /* In nbnxn_ci_t the integer shift contains the shift in the lower 7 bits.
153 * The upper bits contain information for non-bonded kernel optimization.
154 * Simply calculating LJ and Coulomb for all pairs in a cluster pair is fine.
155 * But three flags can be used to skip interactions, currently only for subc=0
156 * !(shift & NBNXN_CI_DO_LJ(subc)) => we can skip LJ for all pairs
157 * shift & NBNXN_CI_HALF_LJ(subc) => we can skip LJ for the second half of i
158 * !(shift & NBNXN_CI_DO_COUL(subc)) => we can skip Coulomb for all pairs
160 #define NBNXN_CI_SHIFT 127
161 #define NBNXN_CI_DO_LJ(subc) (1<<(7+3*(subc)))
162 #define NBNXN_CI_HALF_LJ(subc) (1<<(8+3*(subc)))
163 #define NBNXN_CI_DO_COUL(subc) (1<<(9+3*(subc)))
165 /* Cluster-pair Interaction masks
166 * Bit i*j-cluster-size + j tells if atom i and j interact.
168 // TODO: Rename according to convention when moving into Nbnxn namespace
169 /* All interaction mask is the same for all kernels */
170 constexpr unsigned int NBNXN_INTERACTION_MASK_ALL = 0xffffffffU;
171 /* 4x4 kernel diagonal mask */
172 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG = 0x08ceU;
173 /* 4x2 kernel diagonal masks */
174 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_0 = 0x0002U;
175 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_1 = 0x002fU;
176 /* 4x8 kernel diagonal masks */
177 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_0 = 0xf0f8fcfeU;
178 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_1 = 0x0080c0e0U;
180 /* Simple pair-list i-unit */
183 int ci; /* i-cluster */
184 int shift; /* Shift vector index plus possible flags, see above */
185 int cj_ind_start; /* Start index into cj */
186 int cj_ind_end; /* End index into cj */
189 /* Grouped pair-list i-unit */
191 /* Returns the number of j-cluster groups in this entry */
192 int numJClusterGroups() const
194 return cj4_ind_end - cj4_ind_start;
197 int sci; /* i-super-cluster */
198 int shift; /* Shift vector index plus possible flags */
199 int cj4_ind_start; /* Start index into cj4 */
200 int cj4_ind_end; /* End index into cj4 */
203 /* Interaction data for a j-group for one warp */
206 // The i-cluster interactions mask for 1 warp
207 unsigned int imask = 0U;
208 // Index into the exclusion array for 1 warp, default index 0 which means no exclusions
213 int cj[c_nbnxnGpuJgroupSize]; /* The 4 j-clusters */
214 nbnxn_im_ei_t imei[c_nbnxnGpuClusterpairSplit]; /* The i-cluster mask data for 2 warps */
217 /* Struct for storing the atom-pair interaction bits for a cluster pair in a GPU pairlist */
220 /* Constructor, sets no exclusions, so all atom pairs interacting */
223 for (unsigned int &pairEntry : pair)
225 pairEntry = NBNXN_INTERACTION_MASK_ALL;
229 /* Topology exclusion interaction bits per warp */
230 unsigned int pair[c_nbnxnGpuExclSize];
233 /* Cluster pairlist type for use on CPUs */
234 struct NbnxnPairlistCpu
238 gmx_cache_protect_t cp0;
240 int na_ci; /* The number of atoms per i-cluster */
241 int na_cj; /* The number of atoms per j-cluster */
242 real rlist; /* The radius for constructing the list */
243 FastVector<nbnxn_ci_t> ci; /* The i-cluster list */
244 FastVector<nbnxn_ci_t> ciOuter; /* The outer, unpruned i-cluster list */
246 FastVector<nbnxn_cj_t> cj; /* The j-cluster list, size ncj */
247 FastVector<nbnxn_cj_t> cjOuter; /* The outer, unpruned j-cluster list */
248 int ncjInUse; /* The number of j-clusters that are used by ci entries in this list, will be <= cj.size() */
250 int nci_tot; /* The total number of i clusters */
252 /* Working data storage for list construction */
253 std::unique_ptr<NbnxnPairlistCpuWork> work;
255 gmx_cache_protect_t cp1;
258 /* Cluster pairlist type, with extra hierarchies, for on the GPU
260 * NOTE: for better performance when combining lists over threads,
261 * all vectors should use default initialization. But when
262 * changing this, excl should be intialized when adding entries.
264 struct NbnxnPairlistGpu
268 * \param[in] pinningPolicy Sets the pinning policy for all buffers used on the GPU
270 NbnxnPairlistGpu(gmx::PinningPolicy pinningPolicy);
272 gmx_cache_protect_t cp0;
274 int na_ci; /* The number of atoms per i-cluster */
275 int na_cj; /* The number of atoms per j-cluster */
276 int na_sc; /* The number of atoms per super cluster */
277 real rlist; /* The radius for constructing the list */
278 // The i-super-cluster list, indexes into cj4;
279 gmx::HostVector<nbnxn_sci_t> sci;
280 // The list of 4*j-cluster groups
281 gmx::HostVector<nbnxn_cj4_t> cj4;
282 // Atom interaction bits (non-exclusions)
283 gmx::HostVector<nbnxn_excl_t> excl;
284 // The total number of i-clusters
287 /* Working data storage for list construction */
288 std::unique_ptr<NbnxnPairlistGpuWork> work;
290 gmx_cache_protect_t cp1;
294 * \brief An object that holds the local or non-local pairlists
299 //! Constructor: initializes the pairlist set as empty
300 PairlistSet(Nbnxm::InteractionLocality locality,
301 const PairlistParams &listParams);
305 //! Constructs the pairlists in the set using the coordinates in \p nbat
306 void constructPairlists(PairSearch *pairSearch,
307 nbnxn_atomdata_t *nbat,
308 const t_blocka *excl,
309 Nbnxm::KernelType kernelType,
310 int minimumIlistCountForGpuBalancing,
313 //! Dispatch the kernel for dynamic pairlist pruning
314 void dispatchPruneKernel(const nbnxn_atomdata_t *nbat,
315 const rvec *shift_vec,
316 Nbnxm::KernelType kernelType);
318 //! Returns the locality
319 Nbnxm::InteractionLocality locality() const
324 //! Returns the lists of CPU pairlists
325 gmx::ArrayRef<const NbnxnPairlistCpu> cpuLists() const
330 //! Returns a pointer to the GPU pairlist, nullptr when not present
331 const NbnxnPairlistGpu *gpuList() const
333 if (!gpuLists_.empty())
335 return &gpuLists_[0];
343 //! Returns the lists of free-energy pairlists, empty when nonbonded interactions are not perturbed
344 gmx::ArrayRef<t_nblist const * const> fepLists() const
350 //! The locality of the pairlist set
351 Nbnxm::InteractionLocality locality_;
352 //! List of pairlists in CPU layout
353 std::vector<NbnxnPairlistCpu> cpuLists_;
354 //! List of working list for rebalancing CPU lists
355 std::vector<NbnxnPairlistCpu> cpuListsWork_;
356 //! List of pairlists in GPU layout
357 std::vector<NbnxnPairlistGpu> gpuLists_;
358 //! Pairlist parameters describing setup and ranges
359 const PairlistParams ¶ms_;
360 //! Tells whether multiple lists get merged into one (the first) after creation
362 //! Tells whether the lists is of CPU type, otherwise GPU type
364 //! Lists for perturbed interactions in simple atom-atom layout
365 std::vector<t_nblist *> fepLists_;
368 /* Pair counts for flop counting */
369 //! Total number of atom pairs for LJ+Q kernel
371 //! Total number of atom pairs for LJ kernel
373 //! Total number of atom pairs for Q kernel
377 //! Initializes a free-energy pair-list
378 void nbnxn_init_pairlist_fep(t_nblist *nl);