src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_ps();
 146         fiy0             = _mm_setzero_ps();
 147         fiz0             = _mm_setzero_ps();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 183
 184             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 188                                                               charge+jnrC+0,charge+jnrD+0);
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm_mul_ps(iq0,jq0);
 196
 197             /* REACTION-FIELD ELECTROSTATICS */
 198             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 199             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 200
 201             /* Update potential sum for this i atom from the interaction with this j atom. */
 202             velecsum         = _mm_add_ps(velecsum,velec);
 203
 204             fscal            = felec;
 205
 206              /* Update vectorial force */
 207             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 208             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 209             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 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,
 216                                                    _mm_mul_ps(dx00,fscal),
 217                                                    _mm_mul_ps(dy00,fscal),
 218                                                    _mm_mul_ps(dz00,fscal));
 219
 220             /* Inner loop uses 35 flops */
 221         }
 222
 223         if(jidx<j_index_end)
 224         {
 225
 226             /* Get j neighbor index, and coordinate index */
 227             jnrlistA         = jjnr[jidx];
 228             jnrlistB         = jjnr[jidx+1];
 229             jnrlistC         = jjnr[jidx+2];
 230             jnrlistD         = jjnr[jidx+3];
 231             /* Sign of each element will be negative for non-real atoms.
 232              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 233              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 234              */
 235             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 236             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 237             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 238             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 239             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 240             j_coord_offsetA  = DIM*jnrA;
 241             j_coord_offsetB  = DIM*jnrB;
 242             j_coord_offsetC  = DIM*jnrC;
 243             j_coord_offsetD  = DIM*jnrD;
 244
 245             /* load j atom coordinates */
 246             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 247                                               x+j_coord_offsetC,x+j_coord_offsetD,
 248                                               &jx0,&jy0,&jz0);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm_sub_ps(ix0,jx0);
 252             dy00             = _mm_sub_ps(iy0,jy0);
 253             dz00             = _mm_sub_ps(iz0,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 257
 258             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 259
 260             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 261
 262             /* Load parameters for j particles */
 263             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 264                                                               charge+jnrC+0,charge+jnrD+0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _mm_mul_ps(iq0,jq0);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 275             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velec            = _mm_andnot_ps(dummy_mask,velec);
 279             velecsum         = _mm_add_ps(velecsum,velec);
 280
 281             fscal            = felec;
 282
 283             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 284
 285              /* Update vectorial force */
 286             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 287             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 288             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 289
 290             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 291             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 292             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 293             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 294             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 295                                                    _mm_mul_ps(dx00,fscal),
 296                                                    _mm_mul_ps(dy00,fscal),
 297                                                    _mm_mul_ps(dz00,fscal));
 298
 299             /* Inner loop uses 35 flops */
 300         }
 301
 302         /* End of innermost loop */
 303
 304         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 305                                               f+i_coord_offset,fshift+i_shift_offset);
 306
 307         ggid                        = gid[iidx];
 308         /* Update potential energies */
 309         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 310
 311         /* Increment number of inner iterations */
 312         inneriter                  += j_index_end - j_index_start;
 313
 314         /* Outer loop uses 8 flops */
 315     }
 316
 317     /* Increment number of outer iterations */
 318     outeriter        += nri;
 319
 320     /* Update outer/inner flops */
 321
 322     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*35);
 323 }
 324 /*
 325  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_single
 326  * Electrostatics interaction: ReactionField
 327  * VdW interaction:            None
 328  * Geometry:                   Particle-Particle
 329  * Calculate force/pot:        Force
 330  */
 331 void
 332 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_single
 333                     (t_nblist                    * gmx_restrict       nlist,
 334                      rvec                        * gmx_restrict          xx,
 335                      rvec                        * gmx_restrict          ff,
 336                      t_forcerec                  * gmx_restrict          fr,
 337                      t_mdatoms                   * gmx_restrict     mdatoms,
 338                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 339                      t_nrnb                      * gmx_restrict        nrnb)
 340 {
 341     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 342      * just 0 for non-waters.
 343      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 344      * jnr indices corresponding to data put in the four positions in the SIMD register.
 345      */
 346     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 347     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 348     int              jnrA,jnrB,jnrC,jnrD;
 349     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 350     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 351     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 352     real             rcutoff_scalar;
 353     real             *shiftvec,*fshift,*x,*f;
 354     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 355     real             scratch[4*DIM];
 356     __m128           fscal,rcutoff,rcutoff2,jidxall;
 357     int              vdwioffset0;
 358     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 359     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 360     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 361     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 362     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 363     real             *charge;
 364     __m128           dummy_mask,cutoff_mask;
 365     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 366     __m128           one     = _mm_set1_ps(1.0);
 367     __m128           two     = _mm_set1_ps(2.0);
 368     x                = xx[0];
 369     f                = ff[0];
 370
 371     nri              = nlist->nri;
 372     iinr             = nlist->iinr;
 373     jindex           = nlist->jindex;
 374     jjnr             = nlist->jjnr;
 375     shiftidx         = nlist->shift;
 376     gid              = nlist->gid;
 377     shiftvec         = fr->shift_vec[0];
 378     fshift           = fr->fshift[0];
 379     facel            = _mm_set1_ps(fr->epsfac);
 380     charge           = mdatoms->chargeA;
 381     krf              = _mm_set1_ps(fr->ic->k_rf);
 382     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 383     crf              = _mm_set1_ps(fr->ic->c_rf);
 384
 385     /* Avoid stupid compiler warnings */
 386     jnrA = jnrB = jnrC = jnrD = 0;
 387     j_coord_offsetA = 0;
 388     j_coord_offsetB = 0;
 389     j_coord_offsetC = 0;
 390     j_coord_offsetD = 0;
 391
 392     outeriter        = 0;
 393     inneriter        = 0;
 394
 395     for(iidx=0;iidx<4*DIM;iidx++)
 396     {
 397         scratch[iidx] = 0.0;
 398     }
 399
 400     /* Start outer loop over neighborlists */
 401     for(iidx=0; iidx<nri; iidx++)
 402     {
 403         /* Load shift vector for this list */
 404         i_shift_offset   = DIM*shiftidx[iidx];
 405
 406         /* Load limits for loop over neighbors */
 407         j_index_start    = jindex[iidx];
 408         j_index_end      = jindex[iidx+1];
 409
 410         /* Get outer coordinate index */
 411         inr              = iinr[iidx];
 412         i_coord_offset   = DIM*inr;
 413
 414         /* Load i particle coords and add shift vector */
 415         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 416
 417         fix0             = _mm_setzero_ps();
 418         fiy0             = _mm_setzero_ps();
 419         fiz0             = _mm_setzero_ps();
 420
 421         /* Load parameters for i particles */
 422         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 423
 424         /* Start inner kernel loop */
 425         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 426         {
 427
 428             /* Get j neighbor index, and coordinate index */
 429             jnrA             = jjnr[jidx];
 430             jnrB             = jjnr[jidx+1];
 431             jnrC             = jjnr[jidx+2];
 432             jnrD             = jjnr[jidx+3];
 433             j_coord_offsetA  = DIM*jnrA;
 434             j_coord_offsetB  = DIM*jnrB;
 435             j_coord_offsetC  = DIM*jnrC;
 436             j_coord_offsetD  = DIM*jnrD;
 437
 438             /* load j atom coordinates */
 439             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 440                                               x+j_coord_offsetC,x+j_coord_offsetD,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_ps(ix0,jx0);
 445             dy00             = _mm_sub_ps(iy0,jy0);
 446             dz00             = _mm_sub_ps(iz0,jz0);
 447
 448             /* Calculate squared distance and things based on it */
 449             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 450
 451             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 452
 453             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 454
 455             /* Load parameters for j particles */
 456             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 457                                                               charge+jnrC+0,charge+jnrD+0);
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             qq00             = _mm_mul_ps(iq0,jq0);
 465
 466             /* REACTION-FIELD ELECTROSTATICS */
 467             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 468
 469             fscal            = felec;
 470
 471              /* Update vectorial force */
 472             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 473             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 474             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 475
 476             fjptrA             = f+j_coord_offsetA;
 477             fjptrB             = f+j_coord_offsetB;
 478             fjptrC             = f+j_coord_offsetC;
 479             fjptrD             = f+j_coord_offsetD;
 480             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 481                                                    _mm_mul_ps(dx00,fscal),
 482                                                    _mm_mul_ps(dy00,fscal),
 483                                                    _mm_mul_ps(dz00,fscal));
 484
 485             /* Inner loop uses 30 flops */
 486         }
 487
 488         if(jidx<j_index_end)
 489         {
 490
 491             /* Get j neighbor index, and coordinate index */
 492             jnrlistA         = jjnr[jidx];
 493             jnrlistB         = jjnr[jidx+1];
 494             jnrlistC         = jjnr[jidx+2];
 495             jnrlistD         = jjnr[jidx+3];
 496             /* Sign of each element will be negative for non-real atoms.
 497              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 498              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 499              */
 500             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 501             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 502             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 503             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 504             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 505             j_coord_offsetA  = DIM*jnrA;
 506             j_coord_offsetB  = DIM*jnrB;
 507             j_coord_offsetC  = DIM*jnrC;
 508             j_coord_offsetD  = DIM*jnrD;
 509
 510             /* load j atom coordinates */
 511             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                               x+j_coord_offsetC,x+j_coord_offsetD,
 513                                               &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm_sub_ps(ix0,jx0);
 517             dy00             = _mm_sub_ps(iy0,jy0);
 518             dz00             = _mm_sub_ps(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 522
 523             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 524
 525             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 529                                                               charge+jnrC+0,charge+jnrD+0);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq00             = _mm_mul_ps(iq0,jq0);
 537
 538             /* REACTION-FIELD ELECTROSTATICS */
 539             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545              /* Update vectorial force */
 546             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 547             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 548             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 549
 550             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 551             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 552             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 553             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 554             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 555                                                    _mm_mul_ps(dx00,fscal),
 556                                                    _mm_mul_ps(dy00,fscal),
 557                                                    _mm_mul_ps(dz00,fscal));
 558
 559             /* Inner loop uses 30 flops */
 560         }
 561
 562         /* End of innermost loop */
 563
 564         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 565                                               f+i_coord_offset,fshift+i_shift_offset);
 566
 567         /* Increment number of inner iterations */
 568         inneriter                  += j_index_end - j_index_start;
 569
 570         /* Outer loop uses 7 flops */
 571     }
 572
 573     /* Increment number of outer iterations */
 574     outeriter        += nri;
 575
 576     /* Update outer/inner flops */
 577
 578     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 579 }