src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_pd(fr->ic->k_rf);
 107     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_pd(fr->ic->c_rf);
 109
 110     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 111     rcutoff_scalar   = fr->rcoulomb;
 112     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 139
 140         fix0             = _mm_setzero_pd();
 141         fiy0             = _mm_setzero_pd();
 142         fiz0             = _mm_setzero_pd();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_pd();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               &jx0,&jy0,&jz0);
 163
 164             /* Calculate displacement vector */
 165             dx00             = _mm_sub_pd(ix0,jx0);
 166             dy00             = _mm_sub_pd(iy0,jy0);
 167             dz00             = _mm_sub_pd(iz0,jz0);
 168
 169             /* Calculate squared distance and things based on it */
 170             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 171
 172             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 173
 174             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 175
 176             /* Load parameters for j particles */
 177             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 184             {
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             qq00             = _mm_mul_pd(iq0,jq0);
 188
 189             /* REACTION-FIELD ELECTROSTATICS */
 190             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 191             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 192
 193             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 194
 195             /* Update potential sum for this i atom from the interaction with this j atom. */
 196             velec            = _mm_and_pd(velec,cutoff_mask);
 197             velecsum         = _mm_add_pd(velecsum,velec);
 198
 199             fscal            = felec;
 200
 201             fscal            = _mm_and_pd(fscal,cutoff_mask);
 202
 203             /* Update vectorial force */
 204             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 205             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 206             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 207
 208             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 209                                                    _mm_mul_pd(dx00,fscal),
 210                                                    _mm_mul_pd(dy00,fscal),
 211                                                    _mm_mul_pd(dz00,fscal));
 212
 213             }
 214
 215             /* Inner loop uses 39 flops */
 216         }
 217
 218         if(jidx<j_index_end)
 219         {
 220
 221             jnrA             = jjnr[jidx];
 222             j_coord_offsetA  = DIM*jnrA;
 223
 224             /* load j atom coordinates */
 225             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 226                                               &jx0,&jy0,&jz0);
 227
 228             /* Calculate displacement vector */
 229             dx00             = _mm_sub_pd(ix0,jx0);
 230             dy00             = _mm_sub_pd(iy0,jy0);
 231             dz00             = _mm_sub_pd(iz0,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 235
 236             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 237
 238             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 239
 240             /* Load parameters for j particles */
 241             jq0              = _mm_load_sd(charge+jnrA+0);
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_pd(iq0,jq0);
 252
 253             /* REACTION-FIELD ELECTROSTATICS */
 254             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 255             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 256
 257             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velec            = _mm_and_pd(velec,cutoff_mask);
 261             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 262             velecsum         = _mm_add_pd(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             fscal            = _mm_and_pd(fscal,cutoff_mask);
 267
 268             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 274
 275             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 276                                                    _mm_mul_pd(dx00,fscal),
 277                                                    _mm_mul_pd(dy00,fscal),
 278                                                    _mm_mul_pd(dz00,fscal));
 279
 280             }
 281
 282             /* Inner loop uses 39 flops */
 283         }
 284
 285         /* End of innermost loop */
 286
 287         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 288                                               f+i_coord_offset,fshift+i_shift_offset);
 289
 290         ggid                        = gid[iidx];
 291         /* Update potential energies */
 292         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 293
 294         /* Increment number of inner iterations */
 295         inneriter                  += j_index_end - j_index_start;
 296
 297         /* Outer loop uses 8 flops */
 298     }
 299
 300     /* Increment number of outer iterations */
 301     outeriter        += nri;
 302
 303     /* Update outer/inner flops */
 304
 305     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*39);
 306 }
 307 /*
 308  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_double
 309  * Electrostatics interaction: ReactionField
 310  * VdW interaction:            None
 311  * Geometry:                   Particle-Particle
 312  * Calculate force/pot:        Force
 313  */
 314 void
 315 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_double
 316                     (t_nblist                    * gmx_restrict       nlist,
 317                      rvec                        * gmx_restrict          xx,
 318                      rvec                        * gmx_restrict          ff,
 319                      t_forcerec                  * gmx_restrict          fr,
 320                      t_mdatoms                   * gmx_restrict     mdatoms,
 321                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 322                      t_nrnb                      * gmx_restrict        nrnb)
 323 {
 324     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 325      * just 0 for non-waters.
 326      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 327      * jnr indices corresponding to data put in the four positions in the SIMD register.
 328      */
 329     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 330     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 331     int              jnrA,jnrB;
 332     int              j_coord_offsetA,j_coord_offsetB;
 333     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 334     real             rcutoff_scalar;
 335     real             *shiftvec,*fshift,*x,*f;
 336     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 337     int              vdwioffset0;
 338     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 339     int              vdwjidx0A,vdwjidx0B;
 340     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 341     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 342     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 343     real             *charge;
 344     __m128d          dummy_mask,cutoff_mask;
 345     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 346     __m128d          one     = _mm_set1_pd(1.0);
 347     __m128d          two     = _mm_set1_pd(2.0);
 348     x                = xx[0];
 349     f                = ff[0];
 350
 351     nri              = nlist->nri;
 352     iinr             = nlist->iinr;
 353     jindex           = nlist->jindex;
 354     jjnr             = nlist->jjnr;
 355     shiftidx         = nlist->shift;
 356     gid              = nlist->gid;
 357     shiftvec         = fr->shift_vec[0];
 358     fshift           = fr->fshift[0];
 359     facel            = _mm_set1_pd(fr->epsfac);
 360     charge           = mdatoms->chargeA;
 361     krf              = _mm_set1_pd(fr->ic->k_rf);
 362     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 363     crf              = _mm_set1_pd(fr->ic->c_rf);
 364
 365     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 366     rcutoff_scalar   = fr->rcoulomb;
 367     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 368     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 369
 370     /* Avoid stupid compiler warnings */
 371     jnrA = jnrB = 0;
 372     j_coord_offsetA = 0;
 373     j_coord_offsetB = 0;
 374
 375     outeriter        = 0;
 376     inneriter        = 0;
 377
 378     /* Start outer loop over neighborlists */
 379     for(iidx=0; iidx<nri; iidx++)
 380     {
 381         /* Load shift vector for this list */
 382         i_shift_offset   = DIM*shiftidx[iidx];
 383
 384         /* Load limits for loop over neighbors */
 385         j_index_start    = jindex[iidx];
 386         j_index_end      = jindex[iidx+1];
 387
 388         /* Get outer coordinate index */
 389         inr              = iinr[iidx];
 390         i_coord_offset   = DIM*inr;
 391
 392         /* Load i particle coords and add shift vector */
 393         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 394
 395         fix0             = _mm_setzero_pd();
 396         fiy0             = _mm_setzero_pd();
 397         fiz0             = _mm_setzero_pd();
 398
 399         /* Load parameters for i particles */
 400         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 401
 402         /* Start inner kernel loop */
 403         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrA             = jjnr[jidx];
 408             jnrB             = jjnr[jidx+1];
 409             j_coord_offsetA  = DIM*jnrA;
 410             j_coord_offsetB  = DIM*jnrB;
 411
 412             /* load j atom coordinates */
 413             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm_sub_pd(ix0,jx0);
 418             dy00             = _mm_sub_pd(iy0,jy0);
 419             dz00             = _mm_sub_pd(iz0,jz0);
 420
 421             /* Calculate squared distance and things based on it */
 422             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 423
 424             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 425
 426             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 427
 428             /* Load parameters for j particles */
 429             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             if (gmx_mm_any_lt(rsq00,rcutoff2))
 436             {
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq00             = _mm_mul_pd(iq0,jq0);
 440
 441             /* REACTION-FIELD ELECTROSTATICS */
 442             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 443
 444             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 445
 446             fscal            = felec;
 447
 448             fscal            = _mm_and_pd(fscal,cutoff_mask);
 449
 450             /* Update vectorial force */
 451             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 452             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 453             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 454
 455             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 456                                                    _mm_mul_pd(dx00,fscal),
 457                                                    _mm_mul_pd(dy00,fscal),
 458                                                    _mm_mul_pd(dz00,fscal));
 459
 460             }
 461
 462             /* Inner loop uses 33 flops */
 463         }
 464
 465         if(jidx<j_index_end)
 466         {
 467
 468             jnrA             = jjnr[jidx];
 469             j_coord_offsetA  = DIM*jnrA;
 470
 471             /* load j atom coordinates */
 472             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 473                                               &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm_sub_pd(ix0,jx0);
 477             dy00             = _mm_sub_pd(iy0,jy0);
 478             dz00             = _mm_sub_pd(iz0,jz0);
 479
 480             /* Calculate squared distance and things based on it */
 481             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 482
 483             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 484
 485             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 486
 487             /* Load parameters for j particles */
 488             jq0              = _mm_load_sd(charge+jnrA+0);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_mm_any_lt(rsq00,rcutoff2))
 495             {
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq00             = _mm_mul_pd(iq0,jq0);
 499
 500             /* REACTION-FIELD ELECTROSTATICS */
 501             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 502
 503             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm_and_pd(fscal,cutoff_mask);
 508
 509             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 510
 511             /* Update vectorial force */
 512             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 513             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 514             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 515
 516             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 517                                                    _mm_mul_pd(dx00,fscal),
 518                                                    _mm_mul_pd(dy00,fscal),
 519                                                    _mm_mul_pd(dz00,fscal));
 520
 521             }
 522
 523             /* Inner loop uses 33 flops */
 524         }
 525
 526         /* End of innermost loop */
 527
 528         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 529                                               f+i_coord_offset,fshift+i_shift_offset);
 530
 531         /* Increment number of inner iterations */
 532         inneriter                  += j_index_end - j_index_start;
 533
 534         /* Outer loop uses 7 flops */
 535     }
 536
 537     /* Increment number of outer iterations */
 538     outeriter        += nri;
 539
 540     /* Update outer/inner flops */
 541
 542     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*33);
 543 }