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
41 #include "gromacs/gpu_utils/hostallocator.h"
42 #include "gromacs/math/vectypes.h"
43 #include "gromacs/mdtypes/nblist.h"
44 #include "gromacs/utility/basedefinitions.h"
45 #include "gromacs/utility/defaultinitializationallocator.h"
46 #include "gromacs/utility/enumerationhelpers.h"
47 #include "gromacs/utility/real.h"
49 // This file with constants is separate from this file to be able
50 // to include it during OpenCL jitting without including config.h
51 #include "gromacs/nbnxm/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 /* A buffer data structure of 64 bytes
69 * to be placed at the beginning and end of structs
70 * to avoid cache invalidation of the real contents
71 * of the struct by writes to neighboring memory.
75 } gmx_cache_protect_t
;
77 /* This is the actual cluster-pair list j-entry.
78 * cj is the j-cluster.
79 * The interaction bits in excl are indexed i-major, j-minor.
80 * The cj entries are sorted such that ones with exclusions come first.
81 * This means that once a full mask (=NBNXN_INTERACTION_MASK_ALL)
82 * is found, all subsequent j-entries in the i-entry also have full masks.
86 int cj
; /* The j-cluster */
87 unsigned int excl
; /* The exclusion (interaction) bits */
90 /* In nbnxn_ci_t the integer shift contains the shift in the lower 7 bits.
91 * The upper bits contain information for non-bonded kernel optimization.
92 * Simply calculating LJ and Coulomb for all pairs in a cluster pair is fine.
93 * But three flags can be used to skip interactions, currently only for subc=0
94 * !(shift & NBNXN_CI_DO_LJ(subc)) => we can skip LJ for all pairs
95 * shift & NBNXN_CI_HALF_LJ(subc) => we can skip LJ for the second half of i
96 * !(shift & NBNXN_CI_DO_COUL(subc)) => we can skip Coulomb for all pairs
98 #define NBNXN_CI_SHIFT 127
99 #define NBNXN_CI_DO_LJ(subc) (1<<(7+3*(subc)))
100 #define NBNXN_CI_HALF_LJ(subc) (1<<(8+3*(subc)))
101 #define NBNXN_CI_DO_COUL(subc) (1<<(9+3*(subc)))
103 /* Cluster-pair Interaction masks
104 * Bit i*j-cluster-size + j tells if atom i and j interact.
106 // TODO: Rename according to convention when moving into Nbnxn namespace
107 /* All interaction mask is the same for all kernels */
108 constexpr unsigned int NBNXN_INTERACTION_MASK_ALL
= 0xffffffffU
;
109 /* 4x4 kernel diagonal mask */
110 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG
= 0x08ceU
;
111 /* 4x2 kernel diagonal masks */
112 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_0
= 0x0002U
;
113 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J2_1
= 0x002fU
;
114 /* 4x8 kernel diagonal masks */
115 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_0
= 0xf0f8fcfeU
;
116 constexpr unsigned int NBNXN_INTERACTION_MASK_DIAG_J8_1
= 0x0080c0e0U
;
118 /* Simple pair-list i-unit */
121 int ci
; /* i-cluster */
122 int shift
; /* Shift vector index plus possible flags, see above */
123 int cj_ind_start
; /* Start index into cj */
124 int cj_ind_end
; /* End index into cj */
127 /* Grouped pair-list i-unit */
129 /* Returns the number of j-cluster groups in this entry */
130 int numJClusterGroups() const
132 return cj4_ind_end
- cj4_ind_start
;
135 int sci
; /* i-super-cluster */
136 int shift
; /* Shift vector index plus possible flags */
137 int cj4_ind_start
; /* Start index into cj4 */
138 int cj4_ind_end
; /* End index into cj4 */
141 /* Interaction data for a j-group for one warp */
144 // The i-cluster interactions mask for 1 warp
145 unsigned int imask
= 0U;
146 // Index into the exclusion array for 1 warp, default index 0 which means no exclusions
151 int cj
[c_nbnxnGpuJgroupSize
]; /* The 4 j-clusters */
152 nbnxn_im_ei_t imei
[c_nbnxnGpuClusterpairSplit
]; /* The i-cluster mask data for 2 warps */
155 /* Struct for storing the atom-pair interaction bits for a cluster pair in a GPU pairlist */
158 /* Constructor, sets no exclusions, so all atom pairs interacting */
161 for (unsigned int &pairEntry
: pair
)
163 pairEntry
= NBNXN_INTERACTION_MASK_ALL
;
167 /* Topology exclusion interaction bits per warp */
168 unsigned int pair
[c_nbnxnGpuExclSize
];
171 /* Cluster pairlist type for use on CPUs */
172 struct NbnxnPairlistCpu
176 gmx_cache_protect_t cp0
;
178 int na_ci
; /* The number of atoms per i-cluster */
179 int na_cj
; /* The number of atoms per j-cluster */
180 real rlist
; /* The radius for constructing the list */
181 FastVector
<nbnxn_ci_t
> ci
; /* The i-cluster list */
182 FastVector
<nbnxn_ci_t
> ciOuter
; /* The outer, unpruned i-cluster list */
184 FastVector
<nbnxn_cj_t
> cj
; /* The j-cluster list, size ncj */
185 FastVector
<nbnxn_cj_t
> cjOuter
; /* The outer, unpruned j-cluster list */
186 int ncjInUse
; /* The number of j-clusters that are used by ci entries in this list, will be <= cj.size() */
188 int nci_tot
; /* The total number of i clusters */
190 /* Working data storage for list construction */
191 std::unique_ptr
<NbnxnPairlistCpuWork
> work
;
193 gmx_cache_protect_t cp1
;
196 /* Cluster pairlist type, with extra hierarchies, for on the GPU
198 * NOTE: for better performance when combining lists over threads,
199 * all vectors should use default initialization. But when
200 * changing this, excl should be intialized when adding entries.
202 struct NbnxnPairlistGpu
206 * \param[in] pinningPolicy Sets the pinning policy for all buffers used on the GPU
208 NbnxnPairlistGpu(gmx::PinningPolicy pinningPolicy
);
210 gmx_cache_protect_t cp0
;
212 int na_ci
; /* The number of atoms per i-cluster */
213 int na_cj
; /* The number of atoms per j-cluster */
214 int na_sc
; /* The number of atoms per super cluster */
215 real rlist
; /* The radius for constructing the list */
216 // The i-super-cluster list, indexes into cj4;
217 gmx::HostVector
<nbnxn_sci_t
> sci
;
218 // The list of 4*j-cluster groups
219 gmx::HostVector
<nbnxn_cj4_t
> cj4
;
220 // Atom interaction bits (non-exclusions)
221 gmx::HostVector
<nbnxn_excl_t
> excl
;
222 // The total number of i-clusters
225 /* Working data storage for list construction */
226 std::unique_ptr
<NbnxnPairlistGpuWork
> work
;
228 gmx_cache_protect_t cp1
;
231 //! Initializes a free-energy pair-list
232 void nbnxn_init_pairlist_fep(t_nblist
*nl
);