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