src/gromacs/gmxlib/nonbonded/nonbonded.c

   1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
   2  *
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  *          GROningen MAchine for Chemical Simulations
   9  *
  10  *                        VERSION 3.2.0
  11  * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
  12  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
  13  * Copyright (c) 2001-2004, The GROMACS development team,
  14  * check out http://www.gromacs.org for more information.
  15
  16  * This program is free software; you can redistribute it and/or
  17  * modify it under the terms of the GNU General Public License
  18  * as published by the Free Software Foundation; either version 2
  19  * of the License, or (at your option) any later version.
  20  *
  21  * If you want to redistribute modifications, please consider that
  22  * scientific software is very special. Version control is crucial -
  23  * bugs must be traceable. We will be happy to consider code for
  24  * inclusion in the official distribution, but derived work must not
  25  * be called official GROMACS. Details are found in the README & COPYING
  26  * files - if they are missing, get the official version at www.gromacs.org.
  27  *
  28  * To help us fund GROMACS development, we humbly ask that you cite
  29  * the papers on the package - you can find them in the top README file.
  30  *
  31  * For more info, check our website at http://www.gromacs.org
  32  *
  33  * And Hey:
  34  * GROningen Mixture of Alchemy and Childrens' Stories
  35  */
  36 #ifdef HAVE_CONFIG_H
  37 #include <config.h>
  38 #endif
  39
  40 #ifdef GMX_THREAD_MPI
  41 #include <thread_mpi.h>
  42 #endif
  43
  44 #include <stdio.h>
  45 #include <stdlib.h>
  46 #include "typedefs.h"
  47 #include "txtdump.h"
  48 #include "smalloc.h"
  49 #include "ns.h"
  50 #include "vec.h"
  51 #include "maths.h"
  52 #include "macros.h"
  53 #include "string2.h"
  54 #include "force.h"
  55 #include "names.h"
  56 #include "main.h"
  57 #include "xvgr.h"
  58 #include "gmx_fatal.h"
  59 #include "physics.h"
  60 #include "force.h"
  61 #include "bondf.h"
  62 #include "nrnb.h"
  63 #include "smalloc.h"
  64 #include "nonbonded.h"
  65
  66 #include "nb_kernel.h"
  67 #include "nb_free_energy.h"
  68 #include "nb_generic.h"
  69 #include "nb_generic_cg.h"
  70 #include "nb_generic_adress.h"
  71
  72 /* Different default (c) and accelerated interaction-specific kernels */
  73 #include "nb_kernel_c/nb_kernel_c.h"
  74
  75 #if (defined GMX_CPU_ACCELERATION_X86_SSE2) && !(defined GMX_DOUBLE)
  76 #    include "nb_kernel_sse2_single/nb_kernel_sse2_single.h"
  77 #endif
  78 #if (defined GMX_CPU_ACCELERATION_X86_SSE4_1) && !(defined GMX_DOUBLE)
  79 #    include "nb_kernel_sse4_1_single/nb_kernel_sse4_1_single.h"
  80 #endif
  81 #if (defined GMX_CPU_ACCELERATION_X86_AVX_128_FMA) && !(defined GMX_DOUBLE)
  82 #    include "nb_kernel_avx_128_fma_single/nb_kernel_avx_128_fma_single.h"
  83 #endif
  84 #if (defined GMX_CPU_ACCELERATION_X86_AVX_256) && !(defined GMX_DOUBLE)
  85 #    include "nb_kernel_avx_256_single/nb_kernel_avx_256_single.h"
  86 #endif
  87 #if (defined GMX_CPU_ACCELERATION_X86_SSE2 && defined GMX_DOUBLE)
  88 #    include "nb_kernel_sse2_double/nb_kernel_sse2_double.h"
  89 #endif
  90 #if (defined GMX_CPU_ACCELERATION_X86_SSE4_1 && defined GMX_DOUBLE)
  91 #    include "nb_kernel_sse4_1_double/nb_kernel_sse4_1_double.h"
  92 #endif
  93 #if (defined GMX_CPU_ACCELERATION_X86_AVX_128_FMA && defined GMX_DOUBLE)
  94 #    include "nb_kernel_avx_128_fma_double/nb_kernel_avx_128_fma_double.h"
  95 #endif
  96 #if (defined GMX_CPU_ACCELERATION_X86_AVX_256 && defined GMX_DOUBLE)
  97 #    include "nb_kernel_avx_256_double/nb_kernel_avx_256_double.h"
  98 #endif
  99 #if (defined GMX_CPU_ACCELERATION_SPARC64_HPC_ACE && defined GMX_DOUBLE)
 100 #    include "nb_kernel_sparc64_hpc_ace_double/nb_kernel_sparc64_hpc_ace_double.h"
 101 #endif
 102
 103
 104 #ifdef GMX_THREAD_MPI
 105 static tMPI_Thread_mutex_t nonbonded_setup_mutex = TMPI_THREAD_MUTEX_INITIALIZER;
 106 #endif
 107 static gmx_bool            nonbonded_setup_done  = FALSE;
 108
 109
 110 void
 111 gmx_nonbonded_setup(FILE *         fplog,
 112                     t_forcerec *   fr,
 113                     gmx_bool       bGenericKernelOnly)
 114 {
 115 #ifdef GMX_THREAD_MPI
 116     tMPI_Thread_mutex_lock(&nonbonded_setup_mutex);
 117 #endif
 118     /* Here we are guaranteed only one thread made it. */
 119     if (nonbonded_setup_done == FALSE)
 120     {
 121         if (bGenericKernelOnly == FALSE)
 122         {
 123             /* Add the generic kernels to the structure stored statically in nb_kernel.c */
 124             nb_kernel_list_add_kernels(kernellist_c, kernellist_c_size);
 125
 126             if (!(fr != NULL && fr->use_cpu_acceleration == FALSE))
 127             {
 128                 /* Add interaction-specific kernels for different architectures */
 129                 /* Single precision */
 130 #if (defined GMX_CPU_ACCELERATION_X86_SSE2) && !(defined GMX_DOUBLE)
 131                 nb_kernel_list_add_kernels(kernellist_sse2_single, kernellist_sse2_single_size);
 132 #endif
 133 #if (defined GMX_CPU_ACCELERATION_X86_SSE4_1) && !(defined GMX_DOUBLE)
 134                 nb_kernel_list_add_kernels(kernellist_sse4_1_single, kernellist_sse4_1_single_size);
 135 #endif
 136 #if (defined GMX_CPU_ACCELERATION_X86_AVX_128_FMA) && !(defined GMX_DOUBLE)
 137                 nb_kernel_list_add_kernels(kernellist_avx_128_fma_single, kernellist_avx_128_fma_single_size);
 138 #endif
 139 #if (defined GMX_CPU_ACCELERATION_X86_AVX_256) && !(defined GMX_DOUBLE)
 140                 nb_kernel_list_add_kernels(kernellist_avx_256_single, kernellist_avx_256_single_size);
 141 #endif
 142                 /* Double precision */
 143 #if (defined GMX_CPU_ACCELERATION_X86_SSE2 && defined GMX_DOUBLE)
 144                 nb_kernel_list_add_kernels(kernellist_sse2_double, kernellist_sse2_double_size);
 145 #endif
 146 #if (defined GMX_CPU_ACCELERATION_X86_SSE4_1 && defined GMX_DOUBLE)
 147                 nb_kernel_list_add_kernels(kernellist_sse4_1_double, kernellist_sse4_1_double_size);
 148 #endif
 149 #if (defined GMX_CPU_ACCELERATION_X86_AVX_128_FMA && defined GMX_DOUBLE)
 150                 nb_kernel_list_add_kernels(kernellist_avx_128_fma_double, kernellist_avx_128_fma_double_size);
 151 #endif
 152 #if (defined GMX_CPU_ACCELERATION_X86_AVX_256 && defined GMX_DOUBLE)
 153                 nb_kernel_list_add_kernels(kernellist_avx_256_double, kernellist_avx_256_double_size);
 154 #endif
 155 #if (defined GMX_CPU_ACCELERATION_SPARC64_HPC_ACE && defined GMX_DOUBLE)
 156                 nb_kernel_list_add_kernels(kernellist_sparc64_hpc_ace_double,kernellist_sparc64_hpc_ace_double_size);
 157 #endif
 158                 ; /* empty statement to avoid a completely empty block */
 159             }
 160         }
 161         /* Create a hash for faster lookups */
 162         nb_kernel_list_hash_init();
 163
 164         nonbonded_setup_done = TRUE;
 165     }
 166 #ifdef GMX_THREAD_MPI
 167     tMPI_Thread_mutex_unlock(&nonbonded_setup_mutex);
 168 #endif
 169 }
 170
 171
 172
 173 void
 174 gmx_nonbonded_set_kernel_pointers(FILE *log, t_nblist *nl)
 175 {
 176     const char *     elec;
 177     const char *     elec_mod;
 178     const char *     vdw;
 179     const char *     vdw_mod;
 180     const char *     geom;
 181     const char *     other;
 182     const char *     vf;
 183
 184     struct
 185     {
 186         const char *  arch;
 187         int           simd_padding_width;
 188     }
 189     arch_and_padding[] =
 190     {
 191         /* Single precision */
 192 #if (defined GMX_CPU_ACCELERATION_X86_AVX_256) && !(defined GMX_DOUBLE)
 193         { "avx_256_single", 8 },
 194 #endif
 195 #if (defined GMX_CPU_ACCELERATION_X86_AVX_128_FMA) && !(defined GMX_DOUBLE)
 196         { "avx_128_fma_single", 4 },
 197 #endif
 198 #if (defined GMX_CPU_ACCELERATION_X86_SSE4_1) && !(defined GMX_DOUBLE)
 199         { "sse4_1_single", 4 },
 200 #endif
 201 #if (defined GMX_CPU_ACCELERATION_X86_SSE2) && !(defined GMX_DOUBLE)
 202         { "sse2_single", 4 },
 203 #endif
 204         /* Double precision */
 205 #if (defined GMX_CPU_ACCELERATION_X86_AVX_256 && defined GMX_DOUBLE)
 206         { "avx_256_double", 4 },
 207 #endif
 208 #if (defined GMX_CPU_ACCELERATION_X86_AVX_128_FMA && defined GMX_DOUBLE)
 209         /* Sic. Double precision 2-way SIMD does not require neighbor list padding,
 210          * since the kernels execute a loop unrolled a factor 2, followed by
 211          * a possible single odd-element epilogue.
 212          */
 213         { "avx_128_fma_double", 1 },
 214 #endif
 215 #if (defined GMX_CPU_ACCELERATION_X86_SSE2 && defined GMX_DOUBLE)
 216         /* No padding - see comment above */
 217         { "sse2_double", 1 },
 218 #endif
 219 #if (defined GMX_CPU_ACCELERATION_X86_SSE4_1 && defined GMX_DOUBLE)
 220         /* No padding - see comment above */
 221         { "sse4_1_double", 1 },
 222 #endif
 223 #if (defined GMX_CPU_ACCELERATION_SPARC64_HPC_ACE && defined GMX_DOUBLE)
 224         /* No padding - see comment above */
 225         { "sparc64_hpc_ace_double", 1 },
 226 #endif
 227         { "c", 1 },
 228     };
 229     int              narch = asize(arch_and_padding);
 230     int              i;
 231
 232     if (nonbonded_setup_done == FALSE)
 233     {
 234         /* We typically call this setup routine before starting timers,
 235          * but if that has not been done for whatever reason we do it now.
 236          */
 237         gmx_nonbonded_setup(NULL, NULL, FALSE);
 238     }
 239
 240     /* Not used yet */
 241     other = "";
 242
 243     nl->kernelptr_vf = NULL;
 244     nl->kernelptr_v  = NULL;
 245     nl->kernelptr_f  = NULL;
 246
 247     elec     = gmx_nbkernel_elec_names[nl->ielec];
 248     elec_mod = eintmod_names[nl->ielecmod];
 249     vdw      = gmx_nbkernel_vdw_names[nl->ivdw];
 250     vdw_mod  = eintmod_names[nl->ivdwmod];
 251     geom     = gmx_nblist_geometry_names[nl->igeometry];
 252
 253     if (nl->type == GMX_NBLIST_INTERACTION_ADRESS)
 254     {
 255         nl->kernelptr_vf       = (void *) gmx_nb_generic_adress_kernel;
 256         nl->kernelptr_f        = (void *) gmx_nb_generic_adress_kernel;
 257         nl->simd_padding_width = 1;
 258         return;
 259     }
 260
 261     if (nl->type == GMX_NBLIST_INTERACTION_FREE_ENERGY)
 262     {
 263         nl->kernelptr_vf       = (void *) gmx_nb_free_energy_kernel;
 264         nl->kernelptr_f        = (void *) gmx_nb_free_energy_kernel;
 265         nl->simd_padding_width = 1;
 266     }
 267     else if (!gmx_strcasecmp_min(geom, "CG-CG"))
 268     {
 269         nl->kernelptr_vf       = (void *) gmx_nb_generic_cg_kernel;
 270         nl->kernelptr_f        = (void *) gmx_nb_generic_cg_kernel;
 271         nl->simd_padding_width = 1;
 272     }
 273     else
 274     {
 275         /* Try to find a specific kernel first */
 276
 277         for (i = 0; i < narch && nl->kernelptr_vf == NULL; i++)
 278         {
 279             nl->kernelptr_vf       = (void *) nb_kernel_list_findkernel(log, arch_and_padding[i].arch, elec, elec_mod, vdw, vdw_mod, geom, other, "PotentialAndForce");
 280             nl->simd_padding_width = arch_and_padding[i].simd_padding_width;
 281         }
 282         for (i = 0; i < narch && nl->kernelptr_f == NULL; i++)
 283         {
 284             nl->kernelptr_f        = (void *) nb_kernel_list_findkernel(log, arch_and_padding[i].arch, elec, elec_mod, vdw, vdw_mod, geom, other, "Force");
 285             nl->simd_padding_width = arch_and_padding[i].simd_padding_width;
 286
 287             /* If there is not force-only optimized kernel, is there a potential & force one? */
 288             if (nl->kernelptr_f == NULL)
 289             {
 290                 nl->kernelptr_f        = (void *) nb_kernel_list_findkernel(NULL, arch_and_padding[i].arch, elec, elec_mod, vdw, vdw_mod, geom, other, "PotentialAndForce");
 291                 nl->simd_padding_width = arch_and_padding[i].simd_padding_width;
 292             }
 293         }
 294
 295         /* Give up, pick a generic one instead */
 296         if (nl->kernelptr_vf == NULL)
 297         {
 298             nl->kernelptr_vf       = (void *) gmx_nb_generic_kernel;
 299             nl->kernelptr_f        = (void *) gmx_nb_generic_kernel;
 300             nl->simd_padding_width = 1;
 301             if (debug)
 302             {
 303                 fprintf(debug,
 304                         "WARNING - Slow generic NB kernel used for neighborlist with\n"
 305                         "    Elec: '%s', Modifier: '%s'\n"
 306                         "    Vdw:  '%s', Modifier: '%s'\n"
 307                         "    Geom: '%s', Other: '%s'\n\n",
 308                         elec, elec_mod, vdw, vdw_mod, geom, other);
 309             }
 310         }
 311     }
 312
 313     return;
 314 }
 315
 316 void do_nonbonded(t_commrec *cr, t_forcerec *fr,
 317                   rvec x[], rvec f_shortrange[], rvec f_longrange[], t_mdatoms *mdatoms, t_blocka *excl,
 318                   gmx_grppairener_t *grppener, rvec box_size,
 319                   t_nrnb *nrnb, real *lambda, real *dvdl,
 320                   int nls, int eNL, int flags)
 321 {
 322     t_nblist *        nlist;
 323     int               n, n0, n1, i, i0, i1, sz, range;
 324     t_nblists *       nblists;
 325     nb_kernel_data_t  kernel_data;
 326     nb_kernel_t *     kernelptr = NULL;
 327     rvec *            f;
 328
 329     kernel_data.flags                   = flags;
 330     kernel_data.exclusions              = excl;
 331     kernel_data.lambda                  = lambda;
 332     kernel_data.dvdl                    = dvdl;
 333
 334     if (fr->bAllvsAll)
 335     {
 336         gmx_incons("All-vs-all kernels have not been implemented in version 4.6");
 337         return;
 338     }
 339
 340     if (eNL >= 0)
 341     {
 342         i0 = eNL;
 343         i1 = i0+1;
 344     }
 345     else
 346     {
 347         i0 = 0;
 348         i1 = eNL_NR;
 349     }
 350
 351     if (nls >= 0)
 352     {
 353         n0 = nls;
 354         n1 = nls+1;
 355     }
 356     else
 357     {
 358         n0 = 0;
 359         n1 = fr->nnblists;
 360     }
 361
 362     for (n = n0; (n < n1); n++)
 363     {
 364         nblists = &fr->nblists[n];
 365
 366         kernel_data.table_elec              = &nblists->table_elec;
 367         kernel_data.table_vdw               = &nblists->table_vdw;
 368         kernel_data.table_elec_vdw          = &nblists->table_elec_vdw;
 369
 370         for (range = 0; range < 2; range++)
 371         {
 372             /* Are we doing short/long-range? */
 373             if (range == 0)
 374             {
 375                 /* Short-range */
 376                 if (!(flags & GMX_NONBONDED_DO_SR))
 377                 {
 378                     continue;
 379                 }
 380                 kernel_data.energygrp_elec          = grppener->ener[egCOULSR];
 381                 kernel_data.energygrp_vdw           = grppener->ener[fr->bBHAM ? egBHAMSR : egLJSR];
 382                 kernel_data.energygrp_polarization  = grppener->ener[egGB];
 383                 nlist = nblists->nlist_sr;
 384                 f                                   = f_shortrange;
 385             }
 386             else if (range == 1)
 387             {
 388                 /* Long-range */
 389                 if (!(flags & GMX_NONBONDED_DO_LR))
 390                 {
 391                     continue;
 392                 }
 393                 kernel_data.energygrp_elec          = grppener->ener[egCOULLR];
 394                 kernel_data.energygrp_vdw           = grppener->ener[fr->bBHAM ? egBHAMLR : egLJLR];
 395                 kernel_data.energygrp_polarization  = grppener->ener[egGB];
 396                 nlist = nblists->nlist_lr;
 397                 f                                   = f_longrange;
 398             }
 399
 400             for (i = i0; (i < i1); i++)
 401             {
 402                 if (nlist[i].nri > 0)
 403                 {
 404                     if (flags & GMX_NONBONDED_DO_POTENTIAL)
 405                     {
 406                         /* Potential and force */
 407                         kernelptr = (nb_kernel_t *)nlist[i].kernelptr_vf;
 408                     }
 409                     else
 410                     {
 411                         /* Force only, no potential */
 412                         kernelptr = (nb_kernel_t *)nlist[i].kernelptr_f;
 413                     }
 414
 415                     if (nlist[i].type != GMX_NBLIST_INTERACTION_FREE_ENERGY && (flags & GMX_NONBONDED_DO_FOREIGNLAMBDA))
 416                     {
 417                         /* We don't need the non-perturbed interactions */
 418                         continue;
 419                     }
 420                     (*kernelptr)(&(nlist[i]), x, f, fr, mdatoms, &kernel_data, nrnb);
 421                 }
 422             }
 423         }
 424     }
 425 }
 426
 427 static void
 428 nb_listed_warning_rlimit(const rvec *x, int ai, int aj, int * global_atom_index, real r, real rlimit)
 429 {
 430     gmx_warning("Listed nonbonded interaction between particles %d and %d\n"
 431                 "at distance %.3f which is larger than the table limit %.3f nm.\n\n"
 432                 "This is likely either a 1,4 interaction, or a listed interaction inside\n"
 433                 "a smaller molecule you are decoupling during a free energy calculation.\n"
 434                 "Since interactions at distances beyond the table cannot be computed,\n"
 435                 "they are skipped until they are inside the table limit again. You will\n"
 436                 "only see this message once, even if it occurs for several interactions.\n\n"
 437                 "IMPORTANT: This should not happen in a stable simulation, so there is\n"
 438                 "probably something wrong with your system. Only change the table-extension\n"
 439                 "distance in the mdp file if you are really sure that is the reason.\n",
 440                 glatnr(global_atom_index, ai), glatnr(global_atom_index, aj), r, rlimit);
 441
 442     if (debug)
 443     {
 444         fprintf(debug,
 445                 "%8f %8f %8f\n%8f %8f %8f\n1-4 (%d,%d) interaction not within cut-off! r=%g. Ignored\n",
 446                 x[ai][XX], x[ai][YY], x[ai][ZZ], x[aj][XX], x[aj][YY], x[aj][ZZ],
 447                 glatnr(global_atom_index, ai), glatnr(global_atom_index, aj), r);
 448     }
 449 }
 450
 451
 452
 453 /* This might logically belong better in the nb_generic.c module, but it is only
 454  * used in do_nonbonded_listed(), and we want it to be inlined there to avoid an
 455  * extra functional call for every single pair listed in the topology.
 456  */
 457 static real
 458 nb_evaluate_single(real r2, real tabscale, real *vftab,
 459                    real qq, real c6, real c12, real *velec, real *vvdw)
 460 {
 461     real       rinv, r, rtab, eps, eps2, Y, F, Geps, Heps2, Fp, VVe, FFe, VVd, FFd, VVr, FFr, fscal;
 462     int        ntab;
 463
 464     /* Do the tabulated interactions - first table lookup */
 465     rinv             = gmx_invsqrt(r2);
 466     r                = r2*rinv;
 467     rtab             = r*tabscale;
 468     ntab             = rtab;
 469     eps              = rtab-ntab;
 470     eps2             = eps*eps;
 471     ntab             = 12*ntab;
 472     /* Electrostatics */
 473     Y                = vftab[ntab];
 474     F                = vftab[ntab+1];
 475     Geps             = eps*vftab[ntab+2];
 476     Heps2            = eps2*vftab[ntab+3];
 477     Fp               = F+Geps+Heps2;
 478     VVe              = Y+eps*Fp;
 479     FFe              = Fp+Geps+2.0*Heps2;
 480     /* Dispersion */
 481     Y                = vftab[ntab+4];
 482     F                = vftab[ntab+5];
 483     Geps             = eps*vftab[ntab+6];
 484     Heps2            = eps2*vftab[ntab+7];
 485     Fp               = F+Geps+Heps2;
 486     VVd              = Y+eps*Fp;
 487     FFd              = Fp+Geps+2.0*Heps2;
 488     /* Repulsion */
 489     Y                = vftab[ntab+8];
 490     F                = vftab[ntab+9];
 491     Geps             = eps*vftab[ntab+10];
 492     Heps2            = eps2*vftab[ntab+11];
 493     Fp               = F+Geps+Heps2;
 494     VVr              = Y+eps*Fp;
 495     FFr              = Fp+Geps+2.0*Heps2;
 496
 497     *velec           = qq*VVe;
 498     *vvdw            = c6*VVd+c12*VVr;
 499
 500     fscal            = -(qq*FFe+c6*FFd+c12*FFr)*tabscale*rinv;
 501
 502     return fscal;
 503 }
 504
 505
 506 real
 507 do_nonbonded_listed(int ftype, int nbonds,
 508                     const t_iatom iatoms[], const t_iparams iparams[],
 509                     const rvec x[], rvec f[], rvec fshift[],
 510                     const t_pbc *pbc, const t_graph *g,
 511                     real *lambda, real *dvdl,
 512                     const t_mdatoms *md,
 513                     const t_forcerec *fr, gmx_grppairener_t *grppener,
 514                     int *global_atom_index)
 515 {
 516     int              ielec, ivdw;
 517     real             qq, c6, c12;
 518     rvec             dx;
 519     ivec             dt;
 520     int              i, j, itype, ai, aj, gid;
 521     int              fshift_index;
 522     real             r2, rinv;
 523     real             fscal, velec, vvdw;
 524     real *           energygrp_elec;
 525     real *           energygrp_vdw;
 526     static gmx_bool  warned_rlimit = FALSE;
 527     /* Free energy stuff */
 528     gmx_bool         bFreeEnergy;
 529     real             LFC[2], LFV[2], DLF[2], lfac_coul[2], lfac_vdw[2], dlfac_coul[2], dlfac_vdw[2];
 530     real             qqB, c6B, c12B, sigma2_def, sigma2_min;
 531
 532
 533     switch (ftype)
 534     {
 535         case F_LJ14:
 536         case F_LJC14_Q:
 537             energygrp_elec = grppener->ener[egCOUL14];
 538             energygrp_vdw  = grppener->ener[egLJ14];
 539             break;
 540         case F_LJC_PAIRS_NB:
 541             energygrp_elec = grppener->ener[egCOULSR];
 542             energygrp_vdw  = grppener->ener[egLJSR];
 543             break;
 544         default:
 545             energygrp_elec = NULL; /* Keep compiler happy */
 546             energygrp_vdw  = NULL; /* Keep compiler happy */
 547             gmx_fatal(FARGS, "Unknown function type %d in do_nonbonded14", ftype);
 548             break;
 549     }
 550
 551     if (fr->efep != efepNO)
 552     {
 553         /* Lambda factor for state A=1-lambda and B=lambda */
 554         LFC[0] = 1.0 - lambda[efptCOUL];
 555         LFV[0] = 1.0 - lambda[efptVDW];
 556         LFC[1] = lambda[efptCOUL];
 557         LFV[1] = lambda[efptVDW];
 558
 559         /*derivative of the lambda factor for state A and B */
 560         DLF[0] = -1;
 561         DLF[1] = 1;
 562
 563         /* precalculate */
 564         sigma2_def = pow(fr->sc_sigma6_def, 1.0/3.0);
 565         sigma2_min = pow(fr->sc_sigma6_min, 1.0/3.0);
 566
 567         for (i = 0; i < 2; i++)
 568         {
 569             lfac_coul[i]  = (fr->sc_power == 2 ? (1-LFC[i])*(1-LFC[i]) : (1-LFC[i]));
 570             dlfac_coul[i] = DLF[i]*fr->sc_power/fr->sc_r_power*(fr->sc_power == 2 ? (1-LFC[i]) : 1);
 571             lfac_vdw[i]   = (fr->sc_power == 2 ? (1-LFV[i])*(1-LFV[i]) : (1-LFV[i]));
 572             dlfac_vdw[i]  = DLF[i]*fr->sc_power/fr->sc_r_power*(fr->sc_power == 2 ? (1-LFV[i]) : 1);
 573         }
 574     }
 575     else
 576     {
 577         sigma2_min = sigma2_def = 0;
 578     }
 579
 580     bFreeEnergy = FALSE;
 581     for (i = 0; (i < nbonds); )
 582     {
 583         itype = iatoms[i++];
 584         ai    = iatoms[i++];
 585         aj    = iatoms[i++];
 586         gid   = GID(md->cENER[ai], md->cENER[aj], md->nenergrp);
 587
 588         /* Get parameters */
 589         switch (ftype)
 590         {
 591             case F_LJ14:
 592                 bFreeEnergy =
 593                     (fr->efep != efepNO &&
 594                      ((md->nPerturbed && (md->bPerturbed[ai] || md->bPerturbed[aj])) ||
 595                       iparams[itype].lj14.c6A != iparams[itype].lj14.c6B ||
 596                       iparams[itype].lj14.c12A != iparams[itype].lj14.c12B));
 597                 qq               = md->chargeA[ai]*md->chargeA[aj]*fr->epsfac*fr->fudgeQQ;
 598                 c6               = iparams[itype].lj14.c6A;
 599                 c12              = iparams[itype].lj14.c12A;
 600                 break;
 601             case F_LJC14_Q:
 602                 qq               = iparams[itype].ljc14.qi*iparams[itype].ljc14.qj*fr->epsfac*iparams[itype].ljc14.fqq;
 603                 c6               = iparams[itype].ljc14.c6;
 604                 c12              = iparams[itype].ljc14.c12;
 605                 break;
 606             case F_LJC_PAIRS_NB:
 607                 qq               = iparams[itype].ljcnb.qi*iparams[itype].ljcnb.qj*fr->epsfac;
 608                 c6               = iparams[itype].ljcnb.c6;
 609                 c12              = iparams[itype].ljcnb.c12;
 610                 break;
 611             default:
 612                 /* Cannot happen since we called gmx_fatal() above in this case */
 613                 qq = c6 = c12 = 0; /* Keep compiler happy */
 614                 break;
 615         }
 616
 617         /* To save flops in the optimized kernels, c6/c12 have 6.0/12.0 derivative prefactors
 618          * included in the general nfbp array now. This means the tables are scaled down by the
 619          * same factor, so when we use the original c6/c12 parameters from iparams[] they must
 620          * be scaled up.
 621          */
 622         c6  *= 6.0;
 623         c12 *= 12.0;
 624
 625         /* Do we need to apply full periodic boundary conditions? */
 626         if (fr->bMolPBC == TRUE)
 627         {
 628             fshift_index = pbc_dx_aiuc(pbc, x[ai], x[aj], dx);
 629         }
 630         else
 631         {
 632             fshift_index = CENTRAL;
 633             rvec_sub(x[ai], x[aj], dx);
 634         }
 635         r2           = norm2(dx);
 636
 637         if (r2 >= fr->tab14.r*fr->tab14.r)
 638         {
 639             if (warned_rlimit == FALSE)
 640             {
 641                 nb_listed_warning_rlimit(x, ai, aj, global_atom_index, sqrt(r2), fr->tab14.r);
 642                 warned_rlimit = TRUE;
 643             }
 644             continue;
 645         }
 646
 647         if (bFreeEnergy)
 648         {
 649             /* Currently free energy is only supported for F_LJ14, so no need to check for that if we got here */
 650             qqB              = md->chargeB[ai]*md->chargeB[aj]*fr->epsfac*fr->fudgeQQ;
 651             c6B              = iparams[itype].lj14.c6B*6.0;
 652             c12B             = iparams[itype].lj14.c12B*12.0;
 653
 654             fscal            = nb_free_energy_evaluate_single(r2, fr->sc_r_power, fr->sc_alphacoul, fr->sc_alphavdw,
 655                                                               fr->tab14.scale, fr->tab14.data, qq, c6, c12, qqB, c6B, c12B,
 656                                                               LFC, LFV, DLF, lfac_coul, lfac_vdw, dlfac_coul, dlfac_vdw,
 657                                                               fr->sc_sigma6_def, fr->sc_sigma6_min, sigma2_def, sigma2_min, &velec, &vvdw, dvdl);
 658         }
 659         else
 660         {
 661             /* Evaluate tabulated interaction without free energy */
 662             fscal            = nb_evaluate_single(r2, fr->tab14.scale, fr->tab14.data, qq, c6, c12, &velec, &vvdw);
 663         }
 664
 665         energygrp_elec[gid]  += velec;
 666         energygrp_vdw[gid]   += vvdw;
 667         svmul(fscal, dx, dx);
 668
 669         /* Add the forces */
 670         rvec_inc(f[ai], dx);
 671         rvec_dec(f[aj], dx);
 672
 673         if (g)
 674         {
 675             /* Correct the shift forces using the graph */
 676             ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
 677             fshift_index = IVEC2IS(dt);
 678         }
 679         if (fshift_index != CENTRAL)
 680         {
 681             rvec_inc(fshift[fshift_index], dx);
 682             rvec_dec(fshift[CENTRAL], dx);
 683         }
 684     }
 685     return 0.0;
 686 }