src/gromacs/mdlib/nbnxn_pairlist.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015, 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.
   8  *
   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.
  13  *
  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.
  18  *
  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.
  23  *
  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.
  31  *
  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.
  34  */
  35
  36 #ifndef _nbnxn_pairlist_h
  37 #define _nbnxn_pairlist_h
  38
  39 #include <stddef.h>
  40
  41 #include "thread_mpi/atomic.h"
  42
  43 #include "gromacs/math/vectypes.h"
  44 #include "gromacs/mdtypes/nblist.h"
  45 #include "gromacs/utility/basedefinitions.h"
  46 #include "gromacs/utility/bitmask.h"
  47 #include "gromacs/utility/real.h"
  48
  49 #ifdef __cplusplus
  50 extern "C" {
  51 #endif
  52
  53 /* A buffer data structure of 64 bytes
  54  * to be placed at the beginning and end of structs
  55  * to avoid cache invalidation of the real contents
  56  * of the struct by writes to neighboring memory.
  57  */
  58 typedef struct {
  59     int dummy[16];
  60 } gmx_cache_protect_t;
  61
  62 /* Abstract type for pair searching data */
  63 typedef struct nbnxn_search * nbnxn_search_t;
  64
  65 /* Function that should return a pointer *ptr to memory
  66  * of size nbytes.
  67  * Error handling should be done within this function.
  68  */
  69 typedef void nbnxn_alloc_t (void **ptr, size_t nbytes);
  70
  71 /* Function that should free the memory pointed to by *ptr.
  72  * NULL should not be passed to this function.
  73  */
  74 typedef void nbnxn_free_t (void *ptr);
  75
  76 /* This is the actual cluster-pair list j-entry.
  77  * cj is the j-cluster.
  78  * The interaction bits in excl are indexed i-major, j-minor.
  79  * The cj entries are sorted such that ones with exclusions come first.
  80  * This means that once a full mask (=NBNXN_INTERACTION_MASK_ALL)
  81  * is found, all subsequent j-entries in the i-entry also have full masks.
  82  */
  83 typedef struct {
  84     int          cj;    /* The j-cluster                    */
  85     unsigned int excl;  /* The exclusion (interaction) bits */
  86 } nbnxn_cj_t;
  87
  88 /* In nbnxn_ci_t the integer shift contains the shift in the lower 7 bits.
  89  * The upper bits contain information for non-bonded kernel optimization.
  90  * Simply calculating LJ and Coulomb for all pairs in a cluster pair is fine.
  91  * But three flags can be used to skip interactions, currently only for subc=0
  92  * !(shift & NBNXN_CI_DO_LJ(subc))   => we can skip LJ for all pairs
  93  * shift & NBNXN_CI_HALF_LJ(subc)    => we can skip LJ for the second half of i
  94  * !(shift & NBNXN_CI_DO_COUL(subc)) => we can skip Coulomb for all pairs
  95  */
  96 #define NBNXN_CI_SHIFT          127
  97 #define NBNXN_CI_DO_LJ(subc)    (1<<(7+3*(subc)))
  98 #define NBNXN_CI_HALF_LJ(subc)  (1<<(8+3*(subc)))
  99 #define NBNXN_CI_DO_COUL(subc)  (1<<(9+3*(subc)))
 100
 101 /* Simple pair-list i-unit */
 102 typedef struct {
 103     int ci;             /* i-cluster             */
 104     int shift;          /* Shift vector index plus possible flags, see above */
 105     int cj_ind_start;   /* Start index into cj   */
 106     int cj_ind_end;     /* End index into cj     */
 107 } nbnxn_ci_t;
 108
 109 /* Grouped pair-list i-unit */
 110 typedef struct {
 111     int sci;            /* i-super-cluster       */
 112     int shift;          /* Shift vector index plus possible flags */
 113     int cj4_ind_start;  /* Start index into cj4  */
 114     int cj4_ind_end;    /* End index into cj4    */
 115 } nbnxn_sci_t;
 116
 117 typedef struct {
 118     unsigned int imask;    /* The i-cluster interactions mask for 1 warp  */
 119     int          excl_ind; /* Index into the exclusion array for 1 warp   */
 120 } nbnxn_im_ei_t;
 121
 122 typedef struct {
 123     int           cj[4];   /* The 4 j-clusters                            */
 124     nbnxn_im_ei_t imei[2]; /* The i-cluster mask data       for 2 warps   */
 125 } nbnxn_cj4_t;
 126
 127 typedef struct {
 128     unsigned int pair[32]; /* Topology exclusion interaction bits for one warp,
 129                             * each unsigned has bitS for 4*8 i clusters
 130                             */
 131 } nbnxn_excl_t;
 132
 133 typedef struct nbnxn_pairlist_t {
 134     gmx_cache_protect_t cp0;
 135
 136     nbnxn_alloc_t      *alloc;
 137     nbnxn_free_t       *free;
 138
 139     gmx_bool            bSimple;         /* Simple list has na_sc=na_s and uses cj   *
 140                                           * Complex list uses cj4                    */
 141
 142     int                     na_ci;       /* The number of atoms per i-cluster        */
 143     int                     na_cj;       /* The number of atoms per j-cluster        */
 144     int                     na_sc;       /* The number of atoms per super cluster    */
 145     real                    rlist;       /* The radius for constructing the list     */
 146     int                     nci;         /* The number of i-clusters in the list     */
 147     nbnxn_ci_t             *ci;          /* The i-cluster list, size nci             */
 148     int                     ci_nalloc;   /* The allocation size of ci                */
 149     int                     nsci;        /* The number of i-super-clusters in the list */
 150     nbnxn_sci_t            *sci;         /* The i-super-cluster list                 */
 151     int                     sci_nalloc;  /* The allocation size of sci               */
 152
 153     int                     ncj;         /* The number of j-clusters in the list     */
 154     nbnxn_cj_t             *cj;          /* The j-cluster list, size ncj             */
 155     int                     cj_nalloc;   /* The allocation size of cj                */
 156
 157     int                     ncj4;        /* The total number of 4*j clusters         */
 158     nbnxn_cj4_t            *cj4;         /* The 4*j cluster list, size ncj4          */
 159     int                     cj4_nalloc;  /* The allocation size of cj4               */
 160     int                     nexcl;       /* The count for excl                       */
 161     nbnxn_excl_t           *excl;        /* Atom interaction bits (non-exclusions)   */
 162     int                     excl_nalloc; /* The allocation size for excl             */
 163     int                     nci_tot;     /* The total number of i clusters           */
 164
 165     struct nbnxn_list_work *work;
 166
 167     gmx_cache_protect_t     cp1;
 168 } nbnxn_pairlist_t;
 169
 170 typedef struct {
 171     int                nnbl;        /* number of lists */
 172     nbnxn_pairlist_t **nbl;         /* lists */
 173     gmx_bool           bCombined;   /* TRUE if lists get combined into one (the 1st) */
 174     gmx_bool           bSimple;     /* TRUE if the list of of type "simple"
 175                                        (na_sc=na_s, no super-clusters used) */
 176     int                natpair_ljq; /* Total number of atom pairs for LJ+Q kernel */
 177     int                natpair_lj;  /* Total number of atom pairs for LJ kernel   */
 178     int                natpair_q;   /* Total number of atom pairs for Q kernel    */
 179     t_nblist         **nbl_fep;
 180 } nbnxn_pairlist_set_t;
 181
 182 enum {
 183     nbatXYZ, nbatXYZQ, nbatX4, nbatX8
 184 };
 185
 186 typedef struct {
 187     real *f;      /* f, size natoms*fstride                             */
 188     real *fshift; /* Shift force array, size SHIFTS*DIM                 */
 189     int   nV;     /* The size of *Vvdw and *Vc                          */
 190     real *Vvdw;   /* Temporary Van der Waals group energy storage       */
 191     real *Vc;     /* Temporary Coulomb group energy storage             */
 192     int   nVS;    /* The size of *VSvdw and *VSc                        */
 193     real *VSvdw;  /* Temporary SIMD Van der Waals group energy storage  */
 194     real *VSc;    /* Temporary SIMD Coulomb group energy storage        */
 195 } nbnxn_atomdata_output_t;
 196
 197 /* Block size in atoms for the non-bonded thread force-buffer reduction,
 198  * should be a multiple of all cell and x86 SIMD sizes (i.e. 2, 4 and 8).
 199  * Should be small to reduce the reduction and zeroing cost,
 200  * but too small will result in overhead.
 201  * Currently the block size is NBNXN_BUFFERFLAG_SIZE*3*sizeof(real)=192 bytes.
 202  */
 203 #ifdef GMX_DOUBLE
 204 #define NBNXN_BUFFERFLAG_SIZE   8
 205 #else
 206 #define NBNXN_BUFFERFLAG_SIZE  16
 207 #endif
 208
 209 /* We store the reduction flags as gmx_bitmask_t.
 210  * This limits the number of flags to BITMASK_SIZE.
 211  */
 212 #define NBNXN_BUFFERFLAG_MAX_THREADS  (BITMASK_SIZE)
 213
 214 /* Flags for telling if threads write to force output buffers */
 215 typedef struct {
 216     int               nflag;       /* The number of flag blocks                         */
 217     gmx_bitmask_t    *flag;        /* Bit i is set when thread i writes to a cell-block */
 218     int               flag_nalloc; /* Allocation size of cxy_flag                       */
 219 } nbnxn_buffer_flags_t;
 220
 221 /* LJ combination rules: geometric, Lorentz-Berthelot, none */
 222 enum {
 223     ljcrGEOM, ljcrLB, ljcrNONE, ljcrNR
 224 };
 225
 226 typedef struct nbnxn_atomdata_t {
 227     nbnxn_alloc_t           *alloc;
 228     nbnxn_free_t            *free;
 229     int                      ntype;           /* The number of different atom types                 */
 230     real                    *nbfp;            /* Lennard-Jones 6*C6 and 12*C12 params, size ntype^2*2 */
 231     int                      comb_rule;       /* Combination rule, see enum above                   */
 232     real                    *nbfp_comb;       /* LJ parameter per atom type, size ntype*2           */
 233     real                    *nbfp_s4;         /* As nbfp, but with stride 4, size ntype^2*4. This
 234                                                * might suit 4-wide SIMD loads of two values (e.g.
 235                                                * two floats in single precision on x86).            */
 236     int                      natoms;          /* Number of atoms                                    */
 237     int                      natoms_local;    /* Number of local atoms                           */
 238     int                     *type;            /* Atom types                                         */
 239     real                    *lj_comb;         /* LJ parameters per atom for combining for pairs     */
 240     int                      XFormat;         /* The format of x (and q), enum                      */
 241     int                      FFormat;         /* The format of f, enum                              */
 242     real                    *q;               /* Charges, can be NULL if incorporated in x          */
 243     int                      na_c;            /* The number of atoms per cluster                    */
 244     int                      nenergrp;        /* The number of energy groups                        */
 245     int                      neg_2log;        /* Log2 of nenergrp                                   */
 246     int                     *energrp;         /* The energy groups per cluster, can be NULL         */
 247     gmx_bool                 bDynamicBox;     /* Do we need to update shift_vec every step?    */
 248     rvec                    *shift_vec;       /* Shift vectors, copied from t_forcerec              */
 249     int                      xstride;         /* stride for a coordinate in x (usually 3 or 4)      */
 250     int                      fstride;         /* stride for a coordinate in f (usually 3 or 4)      */
 251     real                    *x;               /* x and possibly q, size natoms*xstride              */
 252
 253     /* j-atom minus i-atom index for generating self and Newton exclusions
 254      * cluster-cluster pairs of the diagonal, for 4xn and 2xnn kernels.
 255      */
 256     real                    *simd_4xn_diagonal_j_minus_i;
 257     real                    *simd_2xnn_diagonal_j_minus_i;
 258     /* Filters for topology exclusion masks for the SIMD kernels.
 259      * filter2 is the same as filter1, but with each element duplicated.
 260      */
 261     unsigned int            *simd_exclusion_filter1;
 262     unsigned int            *simd_exclusion_filter2;
 263     real                    *simd_interaction_array; /* Array of masks needed for exclusions on QPX */
 264     int                      nout;                   /* The number of force arrays                         */
 265     nbnxn_atomdata_output_t *out;                    /* Output data structures               */
 266     int                      nalloc;                 /* Allocation size of all arrays (for x/f *x/fstride) */
 267     gmx_bool                 bUseBufferFlags;        /* Use the flags or operate on all atoms     */
 268     nbnxn_buffer_flags_t     buffer_flags;           /* Flags for buffer zeroing+reduc.  */
 269     gmx_bool                 bUseTreeReduce;         /* Use tree for force reduction */
 270     tMPI_Atomic_t           *syncStep;               /* Synchronization step for tree reduce */
 271 } nbnxn_atomdata_t;
 272
 273 #ifdef __cplusplus
 274 }
 275 #endif
 276
 277 #endif