src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single 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/math/vec.h"
  46 #include "gromacs/legacyheaders/nrnb.h"
  47
  48 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  49 #include "kernelutil_x86_avx_128_fma_single.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_single
  53  * Electrostatics interaction: Coulomb
  54  * VdW interaction:            None
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_single
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB,jnrC,jnrD;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m128           fscal,rcutoff,rcutoff2,jidxall;
  84     int              vdwioffset0;
  85     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108
 109     /* Avoid stupid compiler warnings */
 110     jnrA = jnrB = jnrC = jnrD = 0;
 111     j_coord_offsetA = 0;
 112     j_coord_offsetB = 0;
 113     j_coord_offsetC = 0;
 114     j_coord_offsetD = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     for(iidx=0;iidx<4*DIM;iidx++)
 120     {
 121         scratch[iidx] = 0.0;
 122     }
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _mm_setzero_ps();
 142         fiy0             = _mm_setzero_ps();
 143         fiz0             = _mm_setzero_ps();
 144
 145         /* Load parameters for i particles */
 146         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm_setzero_ps();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             jnrC             = jjnr[jidx+2];
 159             jnrD             = jjnr[jidx+3];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162             j_coord_offsetC  = DIM*jnrC;
 163             j_coord_offsetD  = DIM*jnrD;
 164
 165             /* load j atom coordinates */
 166             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               x+j_coord_offsetC,x+j_coord_offsetD,
 168                                               &jx0,&jy0,&jz0);
 169
 170             /* Calculate displacement vector */
 171             dx00             = _mm_sub_ps(ix0,jx0);
 172             dy00             = _mm_sub_ps(iy0,jy0);
 173             dz00             = _mm_sub_ps(iz0,jz0);
 174
 175             /* Calculate squared distance and things based on it */
 176             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 177
 178             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 179
 180             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 181
 182             /* Load parameters for j particles */
 183             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 184                                                               charge+jnrC+0,charge+jnrD+0);
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             qq00             = _mm_mul_ps(iq0,jq0);
 192
 193             /* COULOMB ELECTROSTATICS */
 194             velec            = _mm_mul_ps(qq00,rinv00);
 195             felec            = _mm_mul_ps(velec,rinvsq00);
 196
 197             /* Update potential sum for this i atom from the interaction with this j atom. */
 198             velecsum         = _mm_add_ps(velecsum,velec);
 199
 200             fscal            = felec;
 201
 202              /* Update vectorial force */
 203             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 204             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 205             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 206
 207             fjptrA             = f+j_coord_offsetA;
 208             fjptrB             = f+j_coord_offsetB;
 209             fjptrC             = f+j_coord_offsetC;
 210             fjptrD             = f+j_coord_offsetD;
 211             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 212                                                    _mm_mul_ps(dx00,fscal),
 213                                                    _mm_mul_ps(dy00,fscal),
 214                                                    _mm_mul_ps(dz00,fscal));
 215
 216             /* Inner loop uses 31 flops */
 217         }
 218
 219         if(jidx<j_index_end)
 220         {
 221
 222             /* Get j neighbor index, and coordinate index */
 223             jnrlistA         = jjnr[jidx];
 224             jnrlistB         = jjnr[jidx+1];
 225             jnrlistC         = jjnr[jidx+2];
 226             jnrlistD         = jjnr[jidx+3];
 227             /* Sign of each element will be negative for non-real atoms.
 228              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 229              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 230              */
 231             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 232             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 233             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 234             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 235             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 236             j_coord_offsetA  = DIM*jnrA;
 237             j_coord_offsetB  = DIM*jnrB;
 238             j_coord_offsetC  = DIM*jnrC;
 239             j_coord_offsetD  = DIM*jnrD;
 240
 241             /* load j atom coordinates */
 242             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 243                                               x+j_coord_offsetC,x+j_coord_offsetD,
 244                                               &jx0,&jy0,&jz0);
 245
 246             /* Calculate displacement vector */
 247             dx00             = _mm_sub_ps(ix0,jx0);
 248             dy00             = _mm_sub_ps(iy0,jy0);
 249             dz00             = _mm_sub_ps(iz0,jz0);
 250
 251             /* Calculate squared distance and things based on it */
 252             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 253
 254             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 255
 256             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                               charge+jnrC+0,charge+jnrD+0);
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq00             = _mm_mul_ps(iq0,jq0);
 268
 269             /* COULOMB ELECTROSTATICS */
 270             velec            = _mm_mul_ps(qq00,rinv00);
 271             felec            = _mm_mul_ps(velec,rinvsq00);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velec            = _mm_andnot_ps(dummy_mask,velec);
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276
 277             fscal            = felec;
 278
 279             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 280
 281              /* Update vectorial force */
 282             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 283             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 284             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 285
 286             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 287             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 288             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 289             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 290             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 291                                                    _mm_mul_ps(dx00,fscal),
 292                                                    _mm_mul_ps(dy00,fscal),
 293                                                    _mm_mul_ps(dz00,fscal));
 294
 295             /* Inner loop uses 31 flops */
 296         }
 297
 298         /* End of innermost loop */
 299
 300         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 301                                               f+i_coord_offset,fshift+i_shift_offset);
 302
 303         ggid                        = gid[iidx];
 304         /* Update potential energies */
 305         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 306
 307         /* Increment number of inner iterations */
 308         inneriter                  += j_index_end - j_index_start;
 309
 310         /* Outer loop uses 8 flops */
 311     }
 312
 313     /* Increment number of outer iterations */
 314     outeriter        += nri;
 315
 316     /* Update outer/inner flops */
 317
 318     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*31);
 319 }
 320 /*
 321  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_single
 322  * Electrostatics interaction: Coulomb
 323  * VdW interaction:            None
 324  * Geometry:                   Particle-Particle
 325  * Calculate force/pot:        Force
 326  */
 327 void
 328 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_single
 329                     (t_nblist                    * gmx_restrict       nlist,
 330                      rvec                        * gmx_restrict          xx,
 331                      rvec                        * gmx_restrict          ff,
 332                      t_forcerec                  * gmx_restrict          fr,
 333                      t_mdatoms                   * gmx_restrict     mdatoms,
 334                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 335                      t_nrnb                      * gmx_restrict        nrnb)
 336 {
 337     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 338      * just 0 for non-waters.
 339      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 340      * jnr indices corresponding to data put in the four positions in the SIMD register.
 341      */
 342     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 343     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 344     int              jnrA,jnrB,jnrC,jnrD;
 345     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 346     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 347     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 348     real             rcutoff_scalar;
 349     real             *shiftvec,*fshift,*x,*f;
 350     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 351     real             scratch[4*DIM];
 352     __m128           fscal,rcutoff,rcutoff2,jidxall;
 353     int              vdwioffset0;
 354     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 355     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 356     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 357     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 358     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 359     real             *charge;
 360     __m128           dummy_mask,cutoff_mask;
 361     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 362     __m128           one     = _mm_set1_ps(1.0);
 363     __m128           two     = _mm_set1_ps(2.0);
 364     x                = xx[0];
 365     f                = ff[0];
 366
 367     nri              = nlist->nri;
 368     iinr             = nlist->iinr;
 369     jindex           = nlist->jindex;
 370     jjnr             = nlist->jjnr;
 371     shiftidx         = nlist->shift;
 372     gid              = nlist->gid;
 373     shiftvec         = fr->shift_vec[0];
 374     fshift           = fr->fshift[0];
 375     facel            = _mm_set1_ps(fr->epsfac);
 376     charge           = mdatoms->chargeA;
 377
 378     /* Avoid stupid compiler warnings */
 379     jnrA = jnrB = jnrC = jnrD = 0;
 380     j_coord_offsetA = 0;
 381     j_coord_offsetB = 0;
 382     j_coord_offsetC = 0;
 383     j_coord_offsetD = 0;
 384
 385     outeriter        = 0;
 386     inneriter        = 0;
 387
 388     for(iidx=0;iidx<4*DIM;iidx++)
 389     {
 390         scratch[iidx] = 0.0;
 391     }
 392
 393     /* Start outer loop over neighborlists */
 394     for(iidx=0; iidx<nri; iidx++)
 395     {
 396         /* Load shift vector for this list */
 397         i_shift_offset   = DIM*shiftidx[iidx];
 398
 399         /* Load limits for loop over neighbors */
 400         j_index_start    = jindex[iidx];
 401         j_index_end      = jindex[iidx+1];
 402
 403         /* Get outer coordinate index */
 404         inr              = iinr[iidx];
 405         i_coord_offset   = DIM*inr;
 406
 407         /* Load i particle coords and add shift vector */
 408         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 409
 410         fix0             = _mm_setzero_ps();
 411         fiy0             = _mm_setzero_ps();
 412         fiz0             = _mm_setzero_ps();
 413
 414         /* Load parameters for i particles */
 415         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 416
 417         /* Start inner kernel loop */
 418         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 419         {
 420
 421             /* Get j neighbor index, and coordinate index */
 422             jnrA             = jjnr[jidx];
 423             jnrB             = jjnr[jidx+1];
 424             jnrC             = jjnr[jidx+2];
 425             jnrD             = jjnr[jidx+3];
 426             j_coord_offsetA  = DIM*jnrA;
 427             j_coord_offsetB  = DIM*jnrB;
 428             j_coord_offsetC  = DIM*jnrC;
 429             j_coord_offsetD  = DIM*jnrD;
 430
 431             /* load j atom coordinates */
 432             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 433                                               x+j_coord_offsetC,x+j_coord_offsetD,
 434                                               &jx0,&jy0,&jz0);
 435
 436             /* Calculate displacement vector */
 437             dx00             = _mm_sub_ps(ix0,jx0);
 438             dy00             = _mm_sub_ps(iy0,jy0);
 439             dz00             = _mm_sub_ps(iz0,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 443
 444             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 445
 446             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 447
 448             /* Load parameters for j particles */
 449             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 450                                                               charge+jnrC+0,charge+jnrD+0);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq00             = _mm_mul_ps(iq0,jq0);
 458
 459             /* COULOMB ELECTROSTATICS */
 460             velec            = _mm_mul_ps(qq00,rinv00);
 461             felec            = _mm_mul_ps(velec,rinvsq00);
 462
 463             fscal            = felec;
 464
 465              /* Update vectorial force */
 466             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 467             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 468             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 469
 470             fjptrA             = f+j_coord_offsetA;
 471             fjptrB             = f+j_coord_offsetB;
 472             fjptrC             = f+j_coord_offsetC;
 473             fjptrD             = f+j_coord_offsetD;
 474             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 475                                                    _mm_mul_ps(dx00,fscal),
 476                                                    _mm_mul_ps(dy00,fscal),
 477                                                    _mm_mul_ps(dz00,fscal));
 478
 479             /* Inner loop uses 30 flops */
 480         }
 481
 482         if(jidx<j_index_end)
 483         {
 484
 485             /* Get j neighbor index, and coordinate index */
 486             jnrlistA         = jjnr[jidx];
 487             jnrlistB         = jjnr[jidx+1];
 488             jnrlistC         = jjnr[jidx+2];
 489             jnrlistD         = jjnr[jidx+3];
 490             /* Sign of each element will be negative for non-real atoms.
 491              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 492              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 493              */
 494             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 495             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 496             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 497             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 498             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 499             j_coord_offsetA  = DIM*jnrA;
 500             j_coord_offsetB  = DIM*jnrB;
 501             j_coord_offsetC  = DIM*jnrC;
 502             j_coord_offsetD  = DIM*jnrD;
 503
 504             /* load j atom coordinates */
 505             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 506                                               x+j_coord_offsetC,x+j_coord_offsetD,
 507                                               &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm_sub_ps(ix0,jx0);
 511             dy00             = _mm_sub_ps(iy0,jy0);
 512             dz00             = _mm_sub_ps(iz0,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 516
 517             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 518
 519             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 523                                                               charge+jnrC+0,charge+jnrD+0);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq00             = _mm_mul_ps(iq0,jq0);
 531
 532             /* COULOMB ELECTROSTATICS */
 533             velec            = _mm_mul_ps(qq00,rinv00);
 534             felec            = _mm_mul_ps(velec,rinvsq00);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 539
 540              /* Update vectorial force */
 541             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 542             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 543             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 544
 545             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 546             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 547             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 548             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 549             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 550                                                    _mm_mul_ps(dx00,fscal),
 551                                                    _mm_mul_ps(dy00,fscal),
 552                                                    _mm_mul_ps(dz00,fscal));
 553
 554             /* Inner loop uses 30 flops */
 555         }
 556
 557         /* End of innermost loop */
 558
 559         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 560                                               f+i_coord_offset,fshift+i_shift_offset);
 561
 562         /* Increment number of inner iterations */
 563         inneriter                  += j_index_end - j_index_start;
 564
 565         /* Outer loop uses 7 flops */
 566     }
 567
 568     /* Increment number of outer iterations */
 569     outeriter        += nri;
 570
 571     /* Update outer/inner flops */
 572
 573     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 574 }