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, 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 "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_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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              /* Update vectorial force */
 202             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 203             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 204             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 205
 206             fjptrA             = f+j_coord_offsetA;
 207             fjptrB             = f+j_coord_offsetB;
 208             fjptrC             = f+j_coord_offsetC;
 209             fjptrD             = f+j_coord_offsetD;
 210             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 211                                                    _mm_mul_ps(dx00,fscal),
 212                                                    _mm_mul_ps(dy00,fscal),
 213                                                    _mm_mul_ps(dz00,fscal));
 214
 215             /* Inner loop uses 31 flops */
 216         }
 217
 218         if(jidx<j_index_end)
 219         {
 220
 221             /* Get j neighbor index, and coordinate index */
 222             jnrlistA         = jjnr[jidx];
 223             jnrlistB         = jjnr[jidx+1];
 224             jnrlistC         = jjnr[jidx+2];
 225             jnrlistD         = jjnr[jidx+3];
 226             /* Sign of each element will be negative for non-real atoms.
 227              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 228              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 229              */
 230             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 231             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 232             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 233             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 234             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 235             j_coord_offsetA  = DIM*jnrA;
 236             j_coord_offsetB  = DIM*jnrB;
 237             j_coord_offsetC  = DIM*jnrC;
 238             j_coord_offsetD  = DIM*jnrD;
 239
 240             /* load j atom coordinates */
 241             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 242                                               x+j_coord_offsetC,x+j_coord_offsetD,
 243                                               &jx0,&jy0,&jz0);
 244
 245             /* Calculate displacement vector */
 246             dx00             = _mm_sub_ps(ix0,jx0);
 247             dy00             = _mm_sub_ps(iy0,jy0);
 248             dz00             = _mm_sub_ps(iz0,jz0);
 249
 250             /* Calculate squared distance and things based on it */
 251             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 252
 253             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 254
 255             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 256
 257             /* Load parameters for j particles */
 258             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 259                                                               charge+jnrC+0,charge+jnrD+0);
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq00             = _mm_mul_ps(iq0,jq0);
 267
 268             /* COULOMB ELECTROSTATICS */
 269             velec            = _mm_mul_ps(qq00,rinv00);
 270             felec            = _mm_mul_ps(velec,rinvsq00);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velec            = _mm_andnot_ps(dummy_mask,velec);
 274             velecsum         = _mm_add_ps(velecsum,velec);
 275
 276             fscal            = felec;
 277
 278             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 279
 280              /* Update vectorial force */
 281             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 282             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 283             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 284
 285             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 286             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 287             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 288             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 289             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 290                                                    _mm_mul_ps(dx00,fscal),
 291                                                    _mm_mul_ps(dy00,fscal),
 292                                                    _mm_mul_ps(dz00,fscal));
 293
 294             /* Inner loop uses 31 flops */
 295         }
 296
 297         /* End of innermost loop */
 298
 299         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 300                                               f+i_coord_offset,fshift+i_shift_offset);
 301
 302         ggid                        = gid[iidx];
 303         /* Update potential energies */
 304         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 305
 306         /* Increment number of inner iterations */
 307         inneriter                  += j_index_end - j_index_start;
 308
 309         /* Outer loop uses 8 flops */
 310     }
 311
 312     /* Increment number of outer iterations */
 313     outeriter        += nri;
 314
 315     /* Update outer/inner flops */
 316
 317     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*31);
 318 }
 319 /*
 320  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_single
 321  * Electrostatics interaction: Coulomb
 322  * VdW interaction:            None
 323  * Geometry:                   Particle-Particle
 324  * Calculate force/pot:        Force
 325  */
 326 void
 327 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_single
 328                     (t_nblist                    * gmx_restrict       nlist,
 329                      rvec                        * gmx_restrict          xx,
 330                      rvec                        * gmx_restrict          ff,
 331                      t_forcerec                  * gmx_restrict          fr,
 332                      t_mdatoms                   * gmx_restrict     mdatoms,
 333                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 334                      t_nrnb                      * gmx_restrict        nrnb)
 335 {
 336     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 337      * just 0 for non-waters.
 338      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 339      * jnr indices corresponding to data put in the four positions in the SIMD register.
 340      */
 341     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 342     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 343     int              jnrA,jnrB,jnrC,jnrD;
 344     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 345     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 346     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 347     real             rcutoff_scalar;
 348     real             *shiftvec,*fshift,*x,*f;
 349     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 350     real             scratch[4*DIM];
 351     __m128           fscal,rcutoff,rcutoff2,jidxall;
 352     int              vdwioffset0;
 353     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 354     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 355     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 356     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 357     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 358     real             *charge;
 359     __m128           dummy_mask,cutoff_mask;
 360     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 361     __m128           one     = _mm_set1_ps(1.0);
 362     __m128           two     = _mm_set1_ps(2.0);
 363     x                = xx[0];
 364     f                = ff[0];
 365
 366     nri              = nlist->nri;
 367     iinr             = nlist->iinr;
 368     jindex           = nlist->jindex;
 369     jjnr             = nlist->jjnr;
 370     shiftidx         = nlist->shift;
 371     gid              = nlist->gid;
 372     shiftvec         = fr->shift_vec[0];
 373     fshift           = fr->fshift[0];
 374     facel            = _mm_set1_ps(fr->epsfac);
 375     charge           = mdatoms->chargeA;
 376
 377     /* Avoid stupid compiler warnings */
 378     jnrA = jnrB = jnrC = jnrD = 0;
 379     j_coord_offsetA = 0;
 380     j_coord_offsetB = 0;
 381     j_coord_offsetC = 0;
 382     j_coord_offsetD = 0;
 383
 384     outeriter        = 0;
 385     inneriter        = 0;
 386
 387     for(iidx=0;iidx<4*DIM;iidx++)
 388     {
 389         scratch[iidx] = 0.0;
 390     }
 391
 392     /* Start outer loop over neighborlists */
 393     for(iidx=0; iidx<nri; iidx++)
 394     {
 395         /* Load shift vector for this list */
 396         i_shift_offset   = DIM*shiftidx[iidx];
 397
 398         /* Load limits for loop over neighbors */
 399         j_index_start    = jindex[iidx];
 400         j_index_end      = jindex[iidx+1];
 401
 402         /* Get outer coordinate index */
 403         inr              = iinr[iidx];
 404         i_coord_offset   = DIM*inr;
 405
 406         /* Load i particle coords and add shift vector */
 407         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 408
 409         fix0             = _mm_setzero_ps();
 410         fiy0             = _mm_setzero_ps();
 411         fiz0             = _mm_setzero_ps();
 412
 413         /* Load parameters for i particles */
 414         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 415
 416         /* Start inner kernel loop */
 417         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrA             = jjnr[jidx];
 422             jnrB             = jjnr[jidx+1];
 423             jnrC             = jjnr[jidx+2];
 424             jnrD             = jjnr[jidx+3];
 425             j_coord_offsetA  = DIM*jnrA;
 426             j_coord_offsetB  = DIM*jnrB;
 427             j_coord_offsetC  = DIM*jnrC;
 428             j_coord_offsetD  = DIM*jnrD;
 429
 430             /* load j atom coordinates */
 431             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 432                                               x+j_coord_offsetC,x+j_coord_offsetD,
 433                                               &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm_sub_ps(ix0,jx0);
 437             dy00             = _mm_sub_ps(iy0,jy0);
 438             dz00             = _mm_sub_ps(iz0,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 442
 443             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 444
 445             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 449                                                               charge+jnrC+0,charge+jnrD+0);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq00             = _mm_mul_ps(iq0,jq0);
 457
 458             /* COULOMB ELECTROSTATICS */
 459             velec            = _mm_mul_ps(qq00,rinv00);
 460             felec            = _mm_mul_ps(velec,rinvsq00);
 461
 462             fscal            = felec;
 463
 464              /* Update vectorial force */
 465             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 466             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 467             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 468
 469             fjptrA             = f+j_coord_offsetA;
 470             fjptrB             = f+j_coord_offsetB;
 471             fjptrC             = f+j_coord_offsetC;
 472             fjptrD             = f+j_coord_offsetD;
 473             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 474                                                    _mm_mul_ps(dx00,fscal),
 475                                                    _mm_mul_ps(dy00,fscal),
 476                                                    _mm_mul_ps(dz00,fscal));
 477
 478             /* Inner loop uses 30 flops */
 479         }
 480
 481         if(jidx<j_index_end)
 482         {
 483
 484             /* Get j neighbor index, and coordinate index */
 485             jnrlistA         = jjnr[jidx];
 486             jnrlistB         = jjnr[jidx+1];
 487             jnrlistC         = jjnr[jidx+2];
 488             jnrlistD         = jjnr[jidx+3];
 489             /* Sign of each element will be negative for non-real atoms.
 490              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 491              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 492              */
 493             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 494             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 495             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 496             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 497             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 498             j_coord_offsetA  = DIM*jnrA;
 499             j_coord_offsetB  = DIM*jnrB;
 500             j_coord_offsetC  = DIM*jnrC;
 501             j_coord_offsetD  = DIM*jnrD;
 502
 503             /* load j atom coordinates */
 504             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                               x+j_coord_offsetC,x+j_coord_offsetD,
 506                                               &jx0,&jy0,&jz0);
 507
 508             /* Calculate displacement vector */
 509             dx00             = _mm_sub_ps(ix0,jx0);
 510             dy00             = _mm_sub_ps(iy0,jy0);
 511             dz00             = _mm_sub_ps(iz0,jz0);
 512
 513             /* Calculate squared distance and things based on it */
 514             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 515
 516             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 517
 518             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 519
 520             /* Load parameters for j particles */
 521             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 522                                                               charge+jnrC+0,charge+jnrD+0);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             /* Compute parameters for interactions between i and j atoms */
 529             qq00             = _mm_mul_ps(iq0,jq0);
 530
 531             /* COULOMB ELECTROSTATICS */
 532             velec            = _mm_mul_ps(qq00,rinv00);
 533             felec            = _mm_mul_ps(velec,rinvsq00);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 538
 539              /* Update vectorial force */
 540             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 543
 544             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 545             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 546             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 547             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 548             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 549                                                    _mm_mul_ps(dx00,fscal),
 550                                                    _mm_mul_ps(dy00,fscal),
 551                                                    _mm_mul_ps(dz00,fscal));
 552
 553             /* Inner loop uses 30 flops */
 554         }
 555
 556         /* End of innermost loop */
 557
 558         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 559                                               f+i_coord_offset,fshift+i_shift_offset);
 560
 561         /* Increment number of inner iterations */
 562         inneriter                  += j_index_end - j_index_start;
 563
 564         /* Outer loop uses 7 flops */
 565     }
 566
 567     /* Increment number of outer iterations */
 568     outeriter        += nri;
 569
 570     /* Update outer/inner flops */
 571
 572     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 573 }