src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwNone_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     __m128           dummy_mask,cutoff_mask;
  77     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  78     __m128           one     = _mm_set1_ps(1.0);
  79     __m128           two     = _mm_set1_ps(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_ps(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93     krf              = _mm_set1_ps(fr->ic->k_rf);
  94     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  95     crf              = _mm_set1_ps(fr->ic->c_rf);
  96
  97     /* Avoid stupid compiler warnings */
  98     jnrA = jnrB = jnrC = jnrD = 0;
  99     j_coord_offsetA = 0;
 100     j_coord_offsetB = 0;
 101     j_coord_offsetC = 0;
 102     j_coord_offsetD = 0;
 103
 104     outeriter        = 0;
 105     inneriter        = 0;
 106
 107     for(iidx=0;iidx<4*DIM;iidx++)
 108     {
 109         scratch[iidx] = 0.0;
 110     }
 111
 112     /* Start outer loop over neighborlists */
 113     for(iidx=0; iidx<nri; iidx++)
 114     {
 115         /* Load shift vector for this list */
 116         i_shift_offset   = DIM*shiftidx[iidx];
 117
 118         /* Load limits for loop over neighbors */
 119         j_index_start    = jindex[iidx];
 120         j_index_end      = jindex[iidx+1];
 121
 122         /* Get outer coordinate index */
 123         inr              = iinr[iidx];
 124         i_coord_offset   = DIM*inr;
 125
 126         /* Load i particle coords and add shift vector */
 127         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 128
 129         fix0             = _mm_setzero_ps();
 130         fiy0             = _mm_setzero_ps();
 131         fiz0             = _mm_setzero_ps();
 132
 133         /* Load parameters for i particles */
 134         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 135
 136         /* Reset potential sums */
 137         velecsum         = _mm_setzero_ps();
 138
 139         /* Start inner kernel loop */
 140         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 141         {
 142
 143             /* Get j neighbor index, and coordinate index */
 144             jnrA             = jjnr[jidx];
 145             jnrB             = jjnr[jidx+1];
 146             jnrC             = jjnr[jidx+2];
 147             jnrD             = jjnr[jidx+3];
 148             j_coord_offsetA  = DIM*jnrA;
 149             j_coord_offsetB  = DIM*jnrB;
 150             j_coord_offsetC  = DIM*jnrC;
 151             j_coord_offsetD  = DIM*jnrD;
 152
 153             /* load j atom coordinates */
 154             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 155                                               x+j_coord_offsetC,x+j_coord_offsetD,
 156                                               &jx0,&jy0,&jz0);
 157
 158             /* Calculate displacement vector */
 159             dx00             = _mm_sub_ps(ix0,jx0);
 160             dy00             = _mm_sub_ps(iy0,jy0);
 161             dz00             = _mm_sub_ps(iz0,jz0);
 162
 163             /* Calculate squared distance and things based on it */
 164             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 165
 166             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 167
 168             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 169
 170             /* Load parameters for j particles */
 171             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 172                                                               charge+jnrC+0,charge+jnrD+0);
 173
 174             /**************************
 175              * CALCULATE INTERACTIONS *
 176              **************************/
 177
 178             /* Compute parameters for interactions between i and j atoms */
 179             qq00             = _mm_mul_ps(iq0,jq0);
 180
 181             /* REACTION-FIELD ELECTROSTATICS */
 182             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 183             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 184
 185             /* Update potential sum for this i atom from the interaction with this j atom. */
 186             velecsum         = _mm_add_ps(velecsum,velec);
 187
 188             fscal            = felec;
 189
 190             /* Calculate temporary vectorial force */
 191             tx               = _mm_mul_ps(fscal,dx00);
 192             ty               = _mm_mul_ps(fscal,dy00);
 193             tz               = _mm_mul_ps(fscal,dz00);
 194
 195             /* Update vectorial force */
 196             fix0             = _mm_add_ps(fix0,tx);
 197             fiy0             = _mm_add_ps(fiy0,ty);
 198             fiz0             = _mm_add_ps(fiz0,tz);
 199
 200             fjptrA             = f+j_coord_offsetA;
 201             fjptrB             = f+j_coord_offsetB;
 202             fjptrC             = f+j_coord_offsetC;
 203             fjptrD             = f+j_coord_offsetD;
 204             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 205
 206             /* Inner loop uses 32 flops */
 207         }
 208
 209         if(jidx<j_index_end)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrlistA         = jjnr[jidx];
 214             jnrlistB         = jjnr[jidx+1];
 215             jnrlistC         = jjnr[jidx+2];
 216             jnrlistD         = jjnr[jidx+3];
 217             /* Sign of each element will be negative for non-real atoms.
 218              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 219              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 220              */
 221             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 222             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 223             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 224             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 225             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228             j_coord_offsetC  = DIM*jnrC;
 229             j_coord_offsetD  = DIM*jnrD;
 230
 231             /* load j atom coordinates */
 232             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 233                                               x+j_coord_offsetC,x+j_coord_offsetD,
 234                                               &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm_sub_ps(ix0,jx0);
 238             dy00             = _mm_sub_ps(iy0,jy0);
 239             dz00             = _mm_sub_ps(iz0,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 243
 244             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 245
 246             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 247
 248             /* Load parameters for j particles */
 249             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 250                                                               charge+jnrC+0,charge+jnrD+0);
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_ps(iq0,jq0);
 258
 259             /* REACTION-FIELD ELECTROSTATICS */
 260             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 261             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velec            = _mm_andnot_ps(dummy_mask,velec);
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266
 267             fscal            = felec;
 268
 269             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 282             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 283             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 284             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 285             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 286
 287             /* Inner loop uses 32 flops */
 288         }
 289
 290         /* End of innermost loop */
 291
 292         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 293                                               f+i_coord_offset,fshift+i_shift_offset);
 294
 295         ggid                        = gid[iidx];
 296         /* Update potential energies */
 297         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 298
 299         /* Increment number of inner iterations */
 300         inneriter                  += j_index_end - j_index_start;
 301
 302         /* Outer loop uses 8 flops */
 303     }
 304
 305     /* Increment number of outer iterations */
 306     outeriter        += nri;
 307
 308     /* Update outer/inner flops */
 309
 310     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 311 }
 312 /*
 313  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_single
 314  * Electrostatics interaction: ReactionField
 315  * VdW interaction:            None
 316  * Geometry:                   Particle-Particle
 317  * Calculate force/pot:        Force
 318  */
 319 void
 320 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_single
 321                     (t_nblist * gmx_restrict                nlist,
 322                      rvec * gmx_restrict                    xx,
 323                      rvec * gmx_restrict                    ff,
 324                      t_forcerec * gmx_restrict              fr,
 325                      t_mdatoms * gmx_restrict               mdatoms,
 326                      nb_kernel_data_t * gmx_restrict        kernel_data,
 327                      t_nrnb * gmx_restrict                  nrnb)
 328 {
 329     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 330      * just 0 for non-waters.
 331      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 332      * jnr indices corresponding to data put in the four positions in the SIMD register.
 333      */
 334     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 335     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 336     int              jnrA,jnrB,jnrC,jnrD;
 337     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 338     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 339     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 340     real             rcutoff_scalar;
 341     real             *shiftvec,*fshift,*x,*f;
 342     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 343     real             scratch[4*DIM];
 344     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 345     int              vdwioffset0;
 346     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 347     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 348     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 349     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 350     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 351     real             *charge;
 352     __m128           dummy_mask,cutoff_mask;
 353     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 354     __m128           one     = _mm_set1_ps(1.0);
 355     __m128           two     = _mm_set1_ps(2.0);
 356     x                = xx[0];
 357     f                = ff[0];
 358
 359     nri              = nlist->nri;
 360     iinr             = nlist->iinr;
 361     jindex           = nlist->jindex;
 362     jjnr             = nlist->jjnr;
 363     shiftidx         = nlist->shift;
 364     gid              = nlist->gid;
 365     shiftvec         = fr->shift_vec[0];
 366     fshift           = fr->fshift[0];
 367     facel            = _mm_set1_ps(fr->epsfac);
 368     charge           = mdatoms->chargeA;
 369     krf              = _mm_set1_ps(fr->ic->k_rf);
 370     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 371     crf              = _mm_set1_ps(fr->ic->c_rf);
 372
 373     /* Avoid stupid compiler warnings */
 374     jnrA = jnrB = jnrC = jnrD = 0;
 375     j_coord_offsetA = 0;
 376     j_coord_offsetB = 0;
 377     j_coord_offsetC = 0;
 378     j_coord_offsetD = 0;
 379
 380     outeriter        = 0;
 381     inneriter        = 0;
 382
 383     for(iidx=0;iidx<4*DIM;iidx++)
 384     {
 385         scratch[iidx] = 0.0;
 386     }
 387
 388     /* Start outer loop over neighborlists */
 389     for(iidx=0; iidx<nri; iidx++)
 390     {
 391         /* Load shift vector for this list */
 392         i_shift_offset   = DIM*shiftidx[iidx];
 393
 394         /* Load limits for loop over neighbors */
 395         j_index_start    = jindex[iidx];
 396         j_index_end      = jindex[iidx+1];
 397
 398         /* Get outer coordinate index */
 399         inr              = iinr[iidx];
 400         i_coord_offset   = DIM*inr;
 401
 402         /* Load i particle coords and add shift vector */
 403         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 404
 405         fix0             = _mm_setzero_ps();
 406         fiy0             = _mm_setzero_ps();
 407         fiz0             = _mm_setzero_ps();
 408
 409         /* Load parameters for i particles */
 410         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 411
 412         /* Start inner kernel loop */
 413         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 414         {
 415
 416             /* Get j neighbor index, and coordinate index */
 417             jnrA             = jjnr[jidx];
 418             jnrB             = jjnr[jidx+1];
 419             jnrC             = jjnr[jidx+2];
 420             jnrD             = jjnr[jidx+3];
 421             j_coord_offsetA  = DIM*jnrA;
 422             j_coord_offsetB  = DIM*jnrB;
 423             j_coord_offsetC  = DIM*jnrC;
 424             j_coord_offsetD  = DIM*jnrD;
 425
 426             /* load j atom coordinates */
 427             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 428                                               x+j_coord_offsetC,x+j_coord_offsetD,
 429                                               &jx0,&jy0,&jz0);
 430
 431             /* Calculate displacement vector */
 432             dx00             = _mm_sub_ps(ix0,jx0);
 433             dy00             = _mm_sub_ps(iy0,jy0);
 434             dz00             = _mm_sub_ps(iz0,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 438
 439             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 440
 441             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 442
 443             /* Load parameters for j particles */
 444             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 445                                                               charge+jnrC+0,charge+jnrD+0);
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             qq00             = _mm_mul_ps(iq0,jq0);
 453
 454             /* REACTION-FIELD ELECTROSTATICS */
 455             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 456
 457             fscal            = felec;
 458
 459             /* Calculate temporary vectorial force */
 460             tx               = _mm_mul_ps(fscal,dx00);
 461             ty               = _mm_mul_ps(fscal,dy00);
 462             tz               = _mm_mul_ps(fscal,dz00);
 463
 464             /* Update vectorial force */
 465             fix0             = _mm_add_ps(fix0,tx);
 466             fiy0             = _mm_add_ps(fiy0,ty);
 467             fiz0             = _mm_add_ps(fiz0,tz);
 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,tx,ty,tz);
 474
 475             /* Inner loop uses 27 flops */
 476         }
 477
 478         if(jidx<j_index_end)
 479         {
 480
 481             /* Get j neighbor index, and coordinate index */
 482             jnrlistA         = jjnr[jidx];
 483             jnrlistB         = jjnr[jidx+1];
 484             jnrlistC         = jjnr[jidx+2];
 485             jnrlistD         = jjnr[jidx+3];
 486             /* Sign of each element will be negative for non-real atoms.
 487              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 488              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 489              */
 490             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 491             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 492             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 493             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 494             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 495             j_coord_offsetA  = DIM*jnrA;
 496             j_coord_offsetB  = DIM*jnrB;
 497             j_coord_offsetC  = DIM*jnrC;
 498             j_coord_offsetD  = DIM*jnrD;
 499
 500             /* load j atom coordinates */
 501             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                               x+j_coord_offsetC,x+j_coord_offsetD,
 503                                               &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _mm_sub_ps(ix0,jx0);
 507             dy00             = _mm_sub_ps(iy0,jy0);
 508             dz00             = _mm_sub_ps(iz0,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 512
 513             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 514
 515             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 519                                                               charge+jnrC+0,charge+jnrD+0);
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             qq00             = _mm_mul_ps(iq0,jq0);
 527
 528             /* REACTION-FIELD ELECTROSTATICS */
 529             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 530
 531             fscal            = felec;
 532
 533             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm_mul_ps(fscal,dx00);
 537             ty               = _mm_mul_ps(fscal,dy00);
 538             tz               = _mm_mul_ps(fscal,dz00);
 539
 540             /* Update vectorial force */
 541             fix0             = _mm_add_ps(fix0,tx);
 542             fiy0             = _mm_add_ps(fiy0,ty);
 543             fiz0             = _mm_add_ps(fiz0,tz);
 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,tx,ty,tz);
 550
 551             /* Inner loop uses 27 flops */
 552         }
 553
 554         /* End of innermost loop */
 555
 556         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 557                                               f+i_coord_offset,fshift+i_shift_offset);
 558
 559         /* Increment number of inner iterations */
 560         inneriter                  += j_index_end - j_index_start;
 561
 562         /* Outer loop uses 7 flops */
 563     }
 564
 565     /* Increment number of outer iterations */
 566     outeriter        += nri;
 567
 568     /* Update outer/inner flops */
 569
 570     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 571 }