src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwNone_GeomP1P1_sse4_1_single.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 sse4_1_single 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_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse4_1_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107
 108     /* Avoid stupid compiler warnings */
 109     jnrA = jnrB = jnrC = jnrD = 0;
 110     j_coord_offsetA = 0;
 111     j_coord_offsetB = 0;
 112     j_coord_offsetC = 0;
 113     j_coord_offsetD = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 117
 118     for(iidx=0;iidx<4*DIM;iidx++)
 119     {
 120         scratch[iidx] = 0.0;
 121     }
 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_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 139
 140         fix0             = _mm_setzero_ps();
 141         fiy0             = _mm_setzero_ps();
 142         fiz0             = _mm_setzero_ps();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_ps();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             jnrC             = jjnr[jidx+2];
 158             jnrD             = jjnr[jidx+3];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161             j_coord_offsetC  = DIM*jnrC;
 162             j_coord_offsetD  = DIM*jnrD;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               x+j_coord_offsetC,x+j_coord_offsetD,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _mm_sub_ps(ix0,jx0);
 171             dy00             = _mm_sub_ps(iy0,jy0);
 172             dz00             = _mm_sub_ps(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 178
 179             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 183                                                               charge+jnrC+0,charge+jnrD+0);
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             /* Compute parameters for interactions between i and j atoms */
 190             qq00             = _mm_mul_ps(iq0,jq0);
 191
 192             /* COULOMB ELECTROSTATICS */
 193             velec            = _mm_mul_ps(qq00,rinv00);
 194             felec            = _mm_mul_ps(velec,rinvsq00);
 195
 196             /* Update potential sum for this i atom from the interaction with this j atom. */
 197             velecsum         = _mm_add_ps(velecsum,velec);
 198
 199             fscal            = felec;
 200
 201             /* Calculate temporary vectorial force */
 202             tx               = _mm_mul_ps(fscal,dx00);
 203             ty               = _mm_mul_ps(fscal,dy00);
 204             tz               = _mm_mul_ps(fscal,dz00);
 205
 206             /* Update vectorial force */
 207             fix0             = _mm_add_ps(fix0,tx);
 208             fiy0             = _mm_add_ps(fiy0,ty);
 209             fiz0             = _mm_add_ps(fiz0,tz);
 210
 211             fjptrA             = f+j_coord_offsetA;
 212             fjptrB             = f+j_coord_offsetB;
 213             fjptrC             = f+j_coord_offsetC;
 214             fjptrD             = f+j_coord_offsetD;
 215             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 216
 217             /* Inner loop uses 28 flops */
 218         }
 219
 220         if(jidx<j_index_end)
 221         {
 222
 223             /* Get j neighbor index, and coordinate index */
 224             jnrlistA         = jjnr[jidx];
 225             jnrlistB         = jjnr[jidx+1];
 226             jnrlistC         = jjnr[jidx+2];
 227             jnrlistD         = jjnr[jidx+3];
 228             /* Sign of each element will be negative for non-real atoms.
 229              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 230              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 231              */
 232             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 233             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 234             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 235             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 236             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 237             j_coord_offsetA  = DIM*jnrA;
 238             j_coord_offsetB  = DIM*jnrB;
 239             j_coord_offsetC  = DIM*jnrC;
 240             j_coord_offsetD  = DIM*jnrD;
 241
 242             /* load j atom coordinates */
 243             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 244                                               x+j_coord_offsetC,x+j_coord_offsetD,
 245                                               &jx0,&jy0,&jz0);
 246
 247             /* Calculate displacement vector */
 248             dx00             = _mm_sub_ps(ix0,jx0);
 249             dy00             = _mm_sub_ps(iy0,jy0);
 250             dz00             = _mm_sub_ps(iz0,jz0);
 251
 252             /* Calculate squared distance and things based on it */
 253             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 254
 255             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 256
 257             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 258
 259             /* Load parameters for j particles */
 260             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 261                                                               charge+jnrC+0,charge+jnrD+0);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq00             = _mm_mul_ps(iq0,jq0);
 269
 270             /* COULOMB ELECTROSTATICS */
 271             velec            = _mm_mul_ps(qq00,rinv00);
 272             felec            = _mm_mul_ps(velec,rinvsq00);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velec            = _mm_andnot_ps(dummy_mask,velec);
 276             velecsum         = _mm_add_ps(velecsum,velec);
 277
 278             fscal            = felec;
 279
 280             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_ps(fscal,dx00);
 284             ty               = _mm_mul_ps(fscal,dy00);
 285             tz               = _mm_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_ps(fix0,tx);
 289             fiy0             = _mm_add_ps(fiy0,ty);
 290             fiz0             = _mm_add_ps(fiz0,tz);
 291
 292             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 293             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 294             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 295             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 297
 298             /* Inner loop uses 28 flops */
 299         }
 300
 301         /* End of innermost loop */
 302
 303         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 304                                               f+i_coord_offset,fshift+i_shift_offset);
 305
 306         ggid                        = gid[iidx];
 307         /* Update potential energies */
 308         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 309
 310         /* Increment number of inner iterations */
 311         inneriter                  += j_index_end - j_index_start;
 312
 313         /* Outer loop uses 8 flops */
 314     }
 315
 316     /* Increment number of outer iterations */
 317     outeriter        += nri;
 318
 319     /* Update outer/inner flops */
 320
 321     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*28);
 322 }
 323 /*
 324  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse4_1_single
 325  * Electrostatics interaction: Coulomb
 326  * VdW interaction:            None
 327  * Geometry:                   Particle-Particle
 328  * Calculate force/pot:        Force
 329  */
 330 void
 331 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse4_1_single
 332                     (t_nblist                    * gmx_restrict       nlist,
 333                      rvec                        * gmx_restrict          xx,
 334                      rvec                        * gmx_restrict          ff,
 335                      t_forcerec                  * gmx_restrict          fr,
 336                      t_mdatoms                   * gmx_restrict     mdatoms,
 337                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 338                      t_nrnb                      * gmx_restrict        nrnb)
 339 {
 340     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 341      * just 0 for non-waters.
 342      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 343      * jnr indices corresponding to data put in the four positions in the SIMD register.
 344      */
 345     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 346     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 347     int              jnrA,jnrB,jnrC,jnrD;
 348     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 349     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 350     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 351     real             rcutoff_scalar;
 352     real             *shiftvec,*fshift,*x,*f;
 353     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 354     real             scratch[4*DIM];
 355     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 356     int              vdwioffset0;
 357     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 358     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 359     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 360     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 361     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 362     real             *charge;
 363     __m128           dummy_mask,cutoff_mask;
 364     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 365     __m128           one     = _mm_set1_ps(1.0);
 366     __m128           two     = _mm_set1_ps(2.0);
 367     x                = xx[0];
 368     f                = ff[0];
 369
 370     nri              = nlist->nri;
 371     iinr             = nlist->iinr;
 372     jindex           = nlist->jindex;
 373     jjnr             = nlist->jjnr;
 374     shiftidx         = nlist->shift;
 375     gid              = nlist->gid;
 376     shiftvec         = fr->shift_vec[0];
 377     fshift           = fr->fshift[0];
 378     facel            = _mm_set1_ps(fr->epsfac);
 379     charge           = mdatoms->chargeA;
 380
 381     /* Avoid stupid compiler warnings */
 382     jnrA = jnrB = jnrC = jnrD = 0;
 383     j_coord_offsetA = 0;
 384     j_coord_offsetB = 0;
 385     j_coord_offsetC = 0;
 386     j_coord_offsetD = 0;
 387
 388     outeriter        = 0;
 389     inneriter        = 0;
 390
 391     for(iidx=0;iidx<4*DIM;iidx++)
 392     {
 393         scratch[iidx] = 0.0;
 394     }
 395
 396     /* Start outer loop over neighborlists */
 397     for(iidx=0; iidx<nri; iidx++)
 398     {
 399         /* Load shift vector for this list */
 400         i_shift_offset   = DIM*shiftidx[iidx];
 401
 402         /* Load limits for loop over neighbors */
 403         j_index_start    = jindex[iidx];
 404         j_index_end      = jindex[iidx+1];
 405
 406         /* Get outer coordinate index */
 407         inr              = iinr[iidx];
 408         i_coord_offset   = DIM*inr;
 409
 410         /* Load i particle coords and add shift vector */
 411         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 412
 413         fix0             = _mm_setzero_ps();
 414         fiy0             = _mm_setzero_ps();
 415         fiz0             = _mm_setzero_ps();
 416
 417         /* Load parameters for i particles */
 418         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 419
 420         /* Start inner kernel loop */
 421         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrA             = jjnr[jidx];
 426             jnrB             = jjnr[jidx+1];
 427             jnrC             = jjnr[jidx+2];
 428             jnrD             = jjnr[jidx+3];
 429             j_coord_offsetA  = DIM*jnrA;
 430             j_coord_offsetB  = DIM*jnrB;
 431             j_coord_offsetC  = DIM*jnrC;
 432             j_coord_offsetD  = DIM*jnrD;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                               x+j_coord_offsetC,x+j_coord_offsetD,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_ps(ix0,jx0);
 441             dy00             = _mm_sub_ps(iy0,jy0);
 442             dz00             = _mm_sub_ps(iz0,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 446
 447             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 448
 449             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 450
 451             /* Load parameters for j particles */
 452             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 453                                                               charge+jnrC+0,charge+jnrD+0);
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             qq00             = _mm_mul_ps(iq0,jq0);
 461
 462             /* COULOMB ELECTROSTATICS */
 463             velec            = _mm_mul_ps(qq00,rinv00);
 464             felec            = _mm_mul_ps(velec,rinvsq00);
 465
 466             fscal            = felec;
 467
 468             /* Calculate temporary vectorial force */
 469             tx               = _mm_mul_ps(fscal,dx00);
 470             ty               = _mm_mul_ps(fscal,dy00);
 471             tz               = _mm_mul_ps(fscal,dz00);
 472
 473             /* Update vectorial force */
 474             fix0             = _mm_add_ps(fix0,tx);
 475             fiy0             = _mm_add_ps(fiy0,ty);
 476             fiz0             = _mm_add_ps(fiz0,tz);
 477
 478             fjptrA             = f+j_coord_offsetA;
 479             fjptrB             = f+j_coord_offsetB;
 480             fjptrC             = f+j_coord_offsetC;
 481             fjptrD             = f+j_coord_offsetD;
 482             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 483
 484             /* Inner loop uses 27 flops */
 485         }
 486
 487         if(jidx<j_index_end)
 488         {
 489
 490             /* Get j neighbor index, and coordinate index */
 491             jnrlistA         = jjnr[jidx];
 492             jnrlistB         = jjnr[jidx+1];
 493             jnrlistC         = jjnr[jidx+2];
 494             jnrlistD         = jjnr[jidx+3];
 495             /* Sign of each element will be negative for non-real atoms.
 496              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 497              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 498              */
 499             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 500             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 501             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 502             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 503             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 504             j_coord_offsetA  = DIM*jnrA;
 505             j_coord_offsetB  = DIM*jnrB;
 506             j_coord_offsetC  = DIM*jnrC;
 507             j_coord_offsetD  = DIM*jnrD;
 508
 509             /* load j atom coordinates */
 510             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                               x+j_coord_offsetC,x+j_coord_offsetD,
 512                                               &jx0,&jy0,&jz0);
 513
 514             /* Calculate displacement vector */
 515             dx00             = _mm_sub_ps(ix0,jx0);
 516             dy00             = _mm_sub_ps(iy0,jy0);
 517             dz00             = _mm_sub_ps(iz0,jz0);
 518
 519             /* Calculate squared distance and things based on it */
 520             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 521
 522             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 523
 524             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 525
 526             /* Load parameters for j particles */
 527             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 528                                                               charge+jnrC+0,charge+jnrD+0);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq00             = _mm_mul_ps(iq0,jq0);
 536
 537             /* COULOMB ELECTROSTATICS */
 538             velec            = _mm_mul_ps(qq00,rinv00);
 539             felec            = _mm_mul_ps(velec,rinvsq00);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx00);
 547             ty               = _mm_mul_ps(fscal,dy00);
 548             tz               = _mm_mul_ps(fscal,dz00);
 549
 550             /* Update vectorial force */
 551             fix0             = _mm_add_ps(fix0,tx);
 552             fiy0             = _mm_add_ps(fiy0,ty);
 553             fiz0             = _mm_add_ps(fiz0,tz);
 554
 555             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 556             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 557             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 558             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 560
 561             /* Inner loop uses 27 flops */
 562         }
 563
 564         /* End of innermost loop */
 565
 566         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 567                                               f+i_coord_offset,fshift+i_shift_offset);
 568
 569         /* Increment number of inner iterations */
 570         inneriter                  += j_index_end - j_index_start;
 571
 572         /* Outer loop uses 7 flops */
 573     }
 574
 575     /* Increment number of outer iterations */
 576     outeriter        += nri;
 577
 578     /* Update outer/inner flops */
 579
 580     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 581 }