src/gromacs/mdlib/nbnxn_pairlist.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2016,2017, 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 <cstddef>
  40
  41 #include "gromacs/math/vectypes.h"
  42 #include "gromacs/mdtypes/nblist.h"
  43 #include "gromacs/utility/basedefinitions.h"
  44 #include "gromacs/utility/bitmask.h"
  45 #include "gromacs/utility/real.h"
  46
  47 struct tMPI_Atomic;
  48
  49 /*! \cond INTERNAL */
  50
  51 /*! \brief The setup for generating and pruning the nbnxn pair list.
  52  *
  53  * Without dynamic pruning rlistOuter=rlistInner.
  54  */
  55 struct NbnxnListParameters
  56 {
  57     /*! \brief Constructor producing a struct with dynamic pruning disabled
  58      */
  59     NbnxnListParameters(real rlist) :
  60         useDynamicPruning(false),
  61         nstlistPrune(-1),
  62         rlistOuter(rlist),
  63         rlistInner(rlist),
  64         numRollingParts(1)
  65     {
  66     }
  67
  68     bool useDynamicPruning; //!< Are we using dynamic pair-list pruning
  69     int  nstlistPrune;      //!< Pair-list dynamic pruning interval
  70     real rlistOuter;        //!< Cut-off of the larger, outer pair-list
  71     real rlistInner;        //!< Cut-off of the smaller, inner pair-list
  72     int  numRollingParts;   //!< The number parts to divide the pair-list into for rolling pruning, a value of 1 gives no rolling pruning
  73 };
  74
  75 /*! \endcond */
  76
  77 /* With GPU kernels the i and j cluster size is 8 atoms */
  78 static const int c_nbnxnGpuClusterSize = 8;
  79
  80 /* The number of clusters in a super-cluster, used for GPU */
  81 static const int c_nbnxnGpuNumClusterPerSupercluster = 8;
  82
  83 /* With GPU kernels we group cluster pairs in 4 to optimize memory usage
  84  * of integers containing 32 bits.
  85  */
  86 static const int c_nbnxnGpuJgroupSize = 32/c_nbnxnGpuNumClusterPerSupercluster;
  87
  88 /* In CUDA the number of threads in a warp is 32 and we have cluster pairs
  89  * of 8*8=64 atoms, so it's convenient to store data for cluster pair halves.
  90  */
  91 static const int c_nbnxnGpuClusterpairSplit = 2;
  92
  93 /* The fixed size of the exclusion mask array for a half cluster pair */
  94 static const int c_nbnxnGpuExclSize = c_nbnxnGpuClusterSize*c_nbnxnGpuClusterSize/c_nbnxnGpuClusterpairSplit;
  95
  96 /* A buffer data structure of 64 bytes
  97  * to be placed at the beginning and end of structs
  98  * to avoid cache invalidation of the real contents
  99  * of the struct by writes to neighboring memory.
 100  */
 101 typedef struct {
 102     int dummy[16];
 103 } gmx_cache_protect_t;
 104
 105 /* Abstract type for pair searching data */
 106 typedef struct nbnxn_search * nbnxn_search_t;
 107
 108 /* Function that should return a pointer *ptr to memory
 109  * of size nbytes.
 110  * Error handling should be done within this function.
 111  */
 112 typedef void nbnxn_alloc_t (void **ptr, size_t nbytes);
 113
 114 /* Function that should free the memory pointed to by *ptr.
 115  * NULL should not be passed to this function.
 116  */
 117 typedef void nbnxn_free_t (void *ptr);
 118
 119 /* This is the actual cluster-pair list j-entry.
 120  * cj is the j-cluster.
 121  * The interaction bits in excl are indexed i-major, j-minor.
 122  * The cj entries are sorted such that ones with exclusions come first.
 123  * This means that once a full mask (=NBNXN_INTERACTION_MASK_ALL)
 124  * is found, all subsequent j-entries in the i-entry also have full masks.
 125  */
 126 typedef struct {
 127     int          cj;    /* The j-cluster                    */
 128     unsigned int excl;  /* The exclusion (interaction) bits */
 129 } nbnxn_cj_t;
 130
 131 /* In nbnxn_ci_t the integer shift contains the shift in the lower 7 bits.
 132  * The upper bits contain information for non-bonded kernel optimization.
 133  * Simply calculating LJ and Coulomb for all pairs in a cluster pair is fine.
 134  * But three flags can be used to skip interactions, currently only for subc=0
 135  * !(shift & NBNXN_CI_DO_LJ(subc))   => we can skip LJ for all pairs
 136  * shift & NBNXN_CI_HALF_LJ(subc)    => we can skip LJ for the second half of i
 137  * !(shift & NBNXN_CI_DO_COUL(subc)) => we can skip Coulomb for all pairs
 138  */
 139 #define NBNXN_CI_SHIFT          127
 140 #define NBNXN_CI_DO_LJ(subc)    (1<<(7+3*(subc)))
 141 #define NBNXN_CI_HALF_LJ(subc)  (1<<(8+3*(subc)))
 142 #define NBNXN_CI_DO_COUL(subc)  (1<<(9+3*(subc)))
 143
 144 /* Simple pair-list i-unit */
 145 typedef struct {
 146     int ci;             /* i-cluster             */
 147     int shift;          /* Shift vector index plus possible flags, see above */
 148     int cj_ind_start;   /* Start index into cj   */
 149     int cj_ind_end;     /* End index into cj     */
 150 } nbnxn_ci_t;
 151
 152 /* Grouped pair-list i-unit */
 153 typedef struct {
 154     int sci;            /* i-super-cluster       */
 155     int shift;          /* Shift vector index plus possible flags */
 156     int cj4_ind_start;  /* Start index into cj4  */
 157     int cj4_ind_end;    /* End index into cj4    */
 158 } nbnxn_sci_t;
 159
 160 typedef struct {
 161     unsigned int imask;    /* The i-cluster interactions mask for 1 warp  */
 162     int          excl_ind; /* Index into the exclusion array for 1 warp   */
 163 } nbnxn_im_ei_t;
 164
 165 typedef struct {
 166     int           cj[c_nbnxnGpuJgroupSize];         /* The 4 j-clusters */
 167     nbnxn_im_ei_t imei[c_nbnxnGpuClusterpairSplit]; /* The i-cluster mask data       for 2 warps   */
 168 } nbnxn_cj4_t;
 169
 170 typedef struct {
 171     unsigned int pair[c_nbnxnGpuExclSize]; /* Topology exclusion interaction bits for one warp,
 172                                             * each unsigned has bitS for 4*8 i clusters
 173                                             */
 174 } nbnxn_excl_t;
 175
 176 typedef struct nbnxn_pairlist_t {
 177     gmx_cache_protect_t cp0;
 178
 179     nbnxn_alloc_t      *alloc;
 180     nbnxn_free_t       *free;
 181
 182     gmx_bool            bSimple;         /* Simple list has na_sc=na_s and uses cj   *
 183                                           * Complex list uses cj4                    */
 184
 185     int                     na_ci;       /* The number of atoms per i-cluster        */
 186     int                     na_cj;       /* The number of atoms per j-cluster        */
 187     int                     na_sc;       /* The number of atoms per super cluster    */
 188     real                    rlist;       /* The radius for constructing the list     */
 189     int                     nci;         /* The number of i-clusters in the list     */
 190     int                     nciOuter;    /* The number of i-clusters in the outer, unpruned list, -1 when invalid */
 191     nbnxn_ci_t             *ci;          /* The i-cluster list, size nci             */
 192     nbnxn_ci_t             *ciOuter;     /* The outer, unpruned i-cluster list, size nciOuter(=-1 when invalid) */
 193     int                     ci_nalloc;   /* The allocation size of ci/ciOuter        */
 194     int                     nsci;        /* The number of i-super-clusters in the list */
 195     nbnxn_sci_t            *sci;         /* The i-super-cluster list                 */
 196     int                     sci_nalloc;  /* The allocation size of sci               */
 197
 198     int                     ncj;         /* The number of j-clusters in the list     */
 199     nbnxn_cj_t             *cj;          /* The j-cluster list, size ncj             */
 200     nbnxn_cj_t             *cjOuter;     /* The outer, unpruned j-cluster list, size ncj    */
 201     int                     cj_nalloc;   /* The allocation size of cj/cj0            */
 202     int                     ncjInUse;    /* The number of j-clusters that are used by ci entries in this list, will be <= ncj */
 203
 204     int                     ncj4;        /* The total number of 4*j clusters         */
 205     nbnxn_cj4_t            *cj4;         /* The 4*j cluster list, size ncj4          */
 206     int                     cj4_nalloc;  /* The allocation size of cj4               */
 207     int                     nexcl;       /* The count for excl                       */
 208     nbnxn_excl_t           *excl;        /* Atom interaction bits (non-exclusions)   */
 209     int                     excl_nalloc; /* The allocation size for excl             */
 210     int                     nci_tot;     /* The total number of i clusters           */
 211
 212     struct nbnxn_list_work *work;
 213
 214     gmx_cache_protect_t     cp1;
 215 } nbnxn_pairlist_t;
 216
 217 typedef struct {
 218     int                nnbl;                  /* number of lists */
 219     nbnxn_pairlist_t **nbl;                   /* lists */
 220     nbnxn_pairlist_t **nbl_work;              /* work space for rebalancing lists */
 221     gmx_bool           bCombined;             /* TRUE if lists get combined into one (the 1st) */
 222     gmx_bool           bSimple;               /* TRUE if the list of of type "simple"
 223                                                  (na_sc=na_s, no super-clusters used) */
 224     int                natpair_ljq;           /* Total number of atom pairs for LJ+Q kernel */
 225     int                natpair_lj;            /* Total number of atom pairs for LJ kernel   */
 226     int                natpair_q;             /* Total number of atom pairs for Q kernel    */
 227     t_nblist         **nbl_fep;               /* List of free-energy atom pair interactions */
 228     gmx_int64_t        outerListCreationStep; /* Step at which the outer list was created */
 229 } nbnxn_pairlist_set_t;
 230
 231 enum {
 232     nbatXYZ, nbatXYZQ, nbatX4, nbatX8
 233 };
 234
 235 typedef struct {
 236     real *f;      /* f, size natoms*fstride                             */
 237     real *fshift; /* Shift force array, size SHIFTS*DIM                 */
 238     int   nV;     /* The size of *Vvdw and *Vc                          */
 239     real *Vvdw;   /* Temporary Van der Waals group energy storage       */
 240     real *Vc;     /* Temporary Coulomb group energy storage             */
 241     int   nVS;    /* The size of *VSvdw and *VSc                        */
 242     real *VSvdw;  /* Temporary SIMD Van der Waals group energy storage  */
 243     real *VSc;    /* Temporary SIMD Coulomb group energy storage        */
 244 } nbnxn_atomdata_output_t;
 245
 246 /* Block size in atoms for the non-bonded thread force-buffer reduction,
 247  * should be a multiple of all cell and x86 SIMD sizes (i.e. 2, 4 and 8).
 248  * Should be small to reduce the reduction and zeroing cost,
 249  * but too small will result in overhead.
 250  * Currently the block size is NBNXN_BUFFERFLAG_SIZE*3*sizeof(real)=192 bytes.
 251  */
 252 #if GMX_DOUBLE
 253 #define NBNXN_BUFFERFLAG_SIZE   8
 254 #else
 255 #define NBNXN_BUFFERFLAG_SIZE  16
 256 #endif
 257
 258 /* We store the reduction flags as gmx_bitmask_t.
 259  * This limits the number of flags to BITMASK_SIZE.
 260  */
 261 #define NBNXN_BUFFERFLAG_MAX_THREADS  (BITMASK_SIZE)
 262
 263 /* Flags for telling if threads write to force output buffers */
 264 typedef struct {
 265     int               nflag;       /* The number of flag blocks                         */
 266     gmx_bitmask_t    *flag;        /* Bit i is set when thread i writes to a cell-block */
 267     int               flag_nalloc; /* Allocation size of cxy_flag                       */
 268 } nbnxn_buffer_flags_t;
 269
 270 /* LJ combination rules: geometric, Lorentz-Berthelot, none */
 271 enum {
 272     ljcrGEOM, ljcrLB, ljcrNONE, ljcrNR
 273 };
 274
 275 typedef struct nbnxn_atomdata_t {
 276     nbnxn_alloc_t           *alloc;
 277     nbnxn_free_t            *free;
 278     int                      ntype;           /* The number of different atom types                 */
 279     real                    *nbfp;            /* Lennard-Jones 6*C6 and 12*C12 params, size ntype^2*2 */
 280     int                      comb_rule;       /* Combination rule, see enum above                   */
 281     real                    *nbfp_comb;       /* LJ parameter per atom type, size ntype*2           */
 282     real                    *nbfp_aligned;    /* As nbfp, but with an alignment (stride) suitable
 283                                                * for the present SIMD architectures
 284                                                */
 285     int                      natoms;          /* Number of atoms                                    */
 286     int                      natoms_local;    /* Number of local atoms                           */
 287     int                     *type;            /* Atom types                                         */
 288     real                    *lj_comb;         /* LJ parameters per atom for combining for pairs     */
 289     int                      XFormat;         /* The format of x (and q), enum                      */
 290     int                      FFormat;         /* The format of f, enum                              */
 291     real                    *q;               /* Charges, can be NULL if incorporated in x          */
 292     int                      na_c;            /* The number of atoms per cluster                    */
 293     int                      nenergrp;        /* The number of energy groups                        */
 294     int                      neg_2log;        /* Log2 of nenergrp                                   */
 295     int                     *energrp;         /* The energy groups per cluster, can be NULL         */
 296     gmx_bool                 bDynamicBox;     /* Do we need to update shift_vec every step?    */
 297     rvec                    *shift_vec;       /* Shift vectors, copied from t_forcerec              */
 298     int                      xstride;         /* stride for a coordinate in x (usually 3 or 4)      */
 299     int                      fstride;         /* stride for a coordinate in f (usually 3 or 4)      */
 300     real                    *x;               /* x and possibly q, size natoms*xstride              */
 301
 302     /* j-atom minus i-atom index for generating self and Newton exclusions
 303      * cluster-cluster pairs of the diagonal, for 4xn and 2xnn kernels.
 304      */
 305     real                    *simd_4xn_diagonal_j_minus_i;
 306     real                    *simd_2xnn_diagonal_j_minus_i;
 307     /* Filters for topology exclusion masks for the SIMD kernels. */
 308     gmx_uint32_t            *simd_exclusion_filter;
 309     gmx_uint64_t            *simd_exclusion_filter64; //!< Used for double w/o SIMD int32 logical support
 310     real                    *simd_interaction_array;  /* Array of masks needed for exclusions */
 311     int                      nout;                    /* The number of force arrays                         */
 312     nbnxn_atomdata_output_t *out;                     /* Output data structures               */
 313     int                      nalloc;                  /* Allocation size of all arrays (for x/f *x/fstride) */
 314     gmx_bool                 bUseBufferFlags;         /* Use the flags or operate on all atoms     */
 315     nbnxn_buffer_flags_t     buffer_flags;            /* Flags for buffer zeroing+reduc.  */
 316     gmx_bool                 bUseTreeReduce;          /* Use tree for force reduction */
 317     tMPI_Atomic             *syncStep;                /* Synchronization step for tree reduce */
 318 } nbnxn_atomdata_t;
 319
 320 #endif