src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwNone_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_sse4_1_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              vdwioffset1;
  70     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  71     int              vdwioffset2;
  72     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  73     int              vdwioffset3;
  74     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102
 103     /* Setup water-specific parameters */
 104     inr              = nlist->iinr[0];
 105     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 106     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 107     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 108
 109     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 110     rcutoff_scalar   = fr->rcoulomb;
 111     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 112     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     for(iidx=0;iidx<4*DIM;iidx++)
 125     {
 126         scratch[iidx] = 0.0;
 127     }
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 145                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 146
 147         fix1             = _mm_setzero_ps();
 148         fiy1             = _mm_setzero_ps();
 149         fiz1             = _mm_setzero_ps();
 150         fix2             = _mm_setzero_ps();
 151         fiy2             = _mm_setzero_ps();
 152         fiz2             = _mm_setzero_ps();
 153         fix3             = _mm_setzero_ps();
 154         fiy3             = _mm_setzero_ps();
 155         fiz3             = _mm_setzero_ps();
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_ps();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               x+j_coord_offsetC,x+j_coord_offsetD,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx10             = _mm_sub_ps(ix1,jx0);
 181             dy10             = _mm_sub_ps(iy1,jy0);
 182             dz10             = _mm_sub_ps(iz1,jz0);
 183             dx20             = _mm_sub_ps(ix2,jx0);
 184             dy20             = _mm_sub_ps(iy2,jy0);
 185             dz20             = _mm_sub_ps(iz2,jz0);
 186             dx30             = _mm_sub_ps(ix3,jx0);
 187             dy30             = _mm_sub_ps(iy3,jy0);
 188             dz30             = _mm_sub_ps(iz3,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 192             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 193             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 194
 195             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 196             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 197             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 198
 199             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 200             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 201             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 202
 203             /* Load parameters for j particles */
 204             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 205                                                               charge+jnrC+0,charge+jnrD+0);
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             if (gmx_mm_any_lt(rsq10,rcutoff2))
 212             {
 213
 214             /* Compute parameters for interactions between i and j atoms */
 215             qq10             = _mm_mul_ps(iq1,jq0);
 216
 217             /* REACTION-FIELD ELECTROSTATICS */
 218             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 219             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 220
 221             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 222
 223             /* Update potential sum for this i atom from the interaction with this j atom. */
 224             velec            = _mm_and_ps(velec,cutoff_mask);
 225             velecsum         = _mm_add_ps(velecsum,velec);
 226
 227             fscal            = felec;
 228
 229             fscal            = _mm_and_ps(fscal,cutoff_mask);
 230
 231             /* Calculate temporary vectorial force */
 232             tx               = _mm_mul_ps(fscal,dx10);
 233             ty               = _mm_mul_ps(fscal,dy10);
 234             tz               = _mm_mul_ps(fscal,dz10);
 235
 236             /* Update vectorial force */
 237             fix1             = _mm_add_ps(fix1,tx);
 238             fiy1             = _mm_add_ps(fiy1,ty);
 239             fiz1             = _mm_add_ps(fiz1,tz);
 240
 241             fjptrA             = f+j_coord_offsetA;
 242             fjptrB             = f+j_coord_offsetB;
 243             fjptrC             = f+j_coord_offsetC;
 244             fjptrD             = f+j_coord_offsetD;
 245             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 246
 247             }
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             if (gmx_mm_any_lt(rsq20,rcutoff2))
 254             {
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq20             = _mm_mul_ps(iq2,jq0);
 258
 259             /* REACTION-FIELD ELECTROSTATICS */
 260             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 261             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 262
 263             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velec            = _mm_and_ps(velec,cutoff_mask);
 267             velecsum         = _mm_add_ps(velecsum,velec);
 268
 269             fscal            = felec;
 270
 271             fscal            = _mm_and_ps(fscal,cutoff_mask);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx20);
 275             ty               = _mm_mul_ps(fscal,dy20);
 276             tz               = _mm_mul_ps(fscal,dz20);
 277
 278             /* Update vectorial force */
 279             fix2             = _mm_add_ps(fix2,tx);
 280             fiy2             = _mm_add_ps(fiy2,ty);
 281             fiz2             = _mm_add_ps(fiz2,tz);
 282
 283             fjptrA             = f+j_coord_offsetA;
 284             fjptrB             = f+j_coord_offsetB;
 285             fjptrC             = f+j_coord_offsetC;
 286             fjptrD             = f+j_coord_offsetD;
 287             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 288
 289             }
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             if (gmx_mm_any_lt(rsq30,rcutoff2))
 296             {
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq30             = _mm_mul_ps(iq3,jq0);
 300
 301             /* REACTION-FIELD ELECTROSTATICS */
 302             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 303             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 304
 305             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_and_ps(velec,cutoff_mask);
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm_and_ps(fscal,cutoff_mask);
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_ps(fscal,dx30);
 317             ty               = _mm_mul_ps(fscal,dy30);
 318             tz               = _mm_mul_ps(fscal,dz30);
 319
 320             /* Update vectorial force */
 321             fix3             = _mm_add_ps(fix3,tx);
 322             fiy3             = _mm_add_ps(fiy3,ty);
 323             fiz3             = _mm_add_ps(fiz3,tz);
 324
 325             fjptrA             = f+j_coord_offsetA;
 326             fjptrB             = f+j_coord_offsetB;
 327             fjptrC             = f+j_coord_offsetC;
 328             fjptrD             = f+j_coord_offsetD;
 329             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 330
 331             }
 332
 333             /* Inner loop uses 108 flops */
 334         }
 335
 336         if(jidx<j_index_end)
 337         {
 338
 339             /* Get j neighbor index, and coordinate index */
 340             jnrlistA         = jjnr[jidx];
 341             jnrlistB         = jjnr[jidx+1];
 342             jnrlistC         = jjnr[jidx+2];
 343             jnrlistD         = jjnr[jidx+3];
 344             /* Sign of each element will be negative for non-real atoms.
 345              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 346              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 347              */
 348             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 349             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 350             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 351             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 352             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 353             j_coord_offsetA  = DIM*jnrA;
 354             j_coord_offsetB  = DIM*jnrB;
 355             j_coord_offsetC  = DIM*jnrC;
 356             j_coord_offsetD  = DIM*jnrD;
 357
 358             /* load j atom coordinates */
 359             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 360                                               x+j_coord_offsetC,x+j_coord_offsetD,
 361                                               &jx0,&jy0,&jz0);
 362
 363             /* Calculate displacement vector */
 364             dx10             = _mm_sub_ps(ix1,jx0);
 365             dy10             = _mm_sub_ps(iy1,jy0);
 366             dz10             = _mm_sub_ps(iz1,jz0);
 367             dx20             = _mm_sub_ps(ix2,jx0);
 368             dy20             = _mm_sub_ps(iy2,jy0);
 369             dz20             = _mm_sub_ps(iz2,jz0);
 370             dx30             = _mm_sub_ps(ix3,jx0);
 371             dy30             = _mm_sub_ps(iy3,jy0);
 372             dz30             = _mm_sub_ps(iz3,jz0);
 373
 374             /* Calculate squared distance and things based on it */
 375             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 376             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 377             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 378
 379             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 380             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 381             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 382
 383             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 384             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 385             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 386
 387             /* Load parameters for j particles */
 388             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 389                                                               charge+jnrC+0,charge+jnrD+0);
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             if (gmx_mm_any_lt(rsq10,rcutoff2))
 396             {
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq10             = _mm_mul_ps(iq1,jq0);
 400
 401             /* REACTION-FIELD ELECTROSTATICS */
 402             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 403             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 404
 405             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm_and_ps(velec,cutoff_mask);
 409             velec            = _mm_andnot_ps(dummy_mask,velec);
 410             velecsum         = _mm_add_ps(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             fscal            = _mm_and_ps(fscal,cutoff_mask);
 415
 416             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 417
 418             /* Calculate temporary vectorial force */
 419             tx               = _mm_mul_ps(fscal,dx10);
 420             ty               = _mm_mul_ps(fscal,dy10);
 421             tz               = _mm_mul_ps(fscal,dz10);
 422
 423             /* Update vectorial force */
 424             fix1             = _mm_add_ps(fix1,tx);
 425             fiy1             = _mm_add_ps(fiy1,ty);
 426             fiz1             = _mm_add_ps(fiz1,tz);
 427
 428             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 429             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 430             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 431             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 432             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 433
 434             }
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             if (gmx_mm_any_lt(rsq20,rcutoff2))
 441             {
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq20             = _mm_mul_ps(iq2,jq0);
 445
 446             /* REACTION-FIELD ELECTROSTATICS */
 447             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 448             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 449
 450             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_and_ps(velec,cutoff_mask);
 454             velec            = _mm_andnot_ps(dummy_mask,velec);
 455             velecsum         = _mm_add_ps(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             fscal            = _mm_and_ps(fscal,cutoff_mask);
 460
 461             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm_mul_ps(fscal,dx20);
 465             ty               = _mm_mul_ps(fscal,dy20);
 466             tz               = _mm_mul_ps(fscal,dz20);
 467
 468             /* Update vectorial force */
 469             fix2             = _mm_add_ps(fix2,tx);
 470             fiy2             = _mm_add_ps(fiy2,ty);
 471             fiz2             = _mm_add_ps(fiz2,tz);
 472
 473             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 474             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 475             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 476             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 477             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 478
 479             }
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             if (gmx_mm_any_lt(rsq30,rcutoff2))
 486             {
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq30             = _mm_mul_ps(iq3,jq0);
 490
 491             /* REACTION-FIELD ELECTROSTATICS */
 492             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 493             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 494
 495             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velec            = _mm_and_ps(velec,cutoff_mask);
 499             velec            = _mm_andnot_ps(dummy_mask,velec);
 500             velecsum         = _mm_add_ps(velecsum,velec);
 501
 502             fscal            = felec;
 503
 504             fscal            = _mm_and_ps(fscal,cutoff_mask);
 505
 506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm_mul_ps(fscal,dx30);
 510             ty               = _mm_mul_ps(fscal,dy30);
 511             tz               = _mm_mul_ps(fscal,dz30);
 512
 513             /* Update vectorial force */
 514             fix3             = _mm_add_ps(fix3,tx);
 515             fiy3             = _mm_add_ps(fiy3,ty);
 516             fiz3             = _mm_add_ps(fiz3,tz);
 517
 518             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 519             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 520             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 521             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 522             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 523
 524             }
 525
 526             /* Inner loop uses 108 flops */
 527         }
 528
 529         /* End of innermost loop */
 530
 531         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 532                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 533
 534         ggid                        = gid[iidx];
 535         /* Update potential energies */
 536         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 537
 538         /* Increment number of inner iterations */
 539         inneriter                  += j_index_end - j_index_start;
 540
 541         /* Outer loop uses 19 flops */
 542     }
 543
 544     /* Increment number of outer iterations */
 545     outeriter        += nri;
 546
 547     /* Update outer/inner flops */
 548
 549     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*108);
 550 }
 551 /*
 552  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse4_1_single
 553  * Electrostatics interaction: ReactionField
 554  * VdW interaction:            None
 555  * Geometry:                   Water4-Particle
 556  * Calculate force/pot:        Force
 557  */
 558 void
 559 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse4_1_single
 560                     (t_nblist * gmx_restrict                nlist,
 561                      rvec * gmx_restrict                    xx,
 562                      rvec * gmx_restrict                    ff,
 563                      t_forcerec * gmx_restrict              fr,
 564                      t_mdatoms * gmx_restrict               mdatoms,
 565                      nb_kernel_data_t * gmx_restrict        kernel_data,
 566                      t_nrnb * gmx_restrict                  nrnb)
 567 {
 568     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 569      * just 0 for non-waters.
 570      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 571      * jnr indices corresponding to data put in the four positions in the SIMD register.
 572      */
 573     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 574     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 575     int              jnrA,jnrB,jnrC,jnrD;
 576     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 577     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 578     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 579     real             rcutoff_scalar;
 580     real             *shiftvec,*fshift,*x,*f;
 581     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 582     real             scratch[4*DIM];
 583     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 584     int              vdwioffset1;
 585     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 586     int              vdwioffset2;
 587     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 588     int              vdwioffset3;
 589     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 590     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 591     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 592     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 593     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 594     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 595     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 596     real             *charge;
 597     __m128           dummy_mask,cutoff_mask;
 598     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 599     __m128           one     = _mm_set1_ps(1.0);
 600     __m128           two     = _mm_set1_ps(2.0);
 601     x                = xx[0];
 602     f                = ff[0];
 603
 604     nri              = nlist->nri;
 605     iinr             = nlist->iinr;
 606     jindex           = nlist->jindex;
 607     jjnr             = nlist->jjnr;
 608     shiftidx         = nlist->shift;
 609     gid              = nlist->gid;
 610     shiftvec         = fr->shift_vec[0];
 611     fshift           = fr->fshift[0];
 612     facel            = _mm_set1_ps(fr->epsfac);
 613     charge           = mdatoms->chargeA;
 614     krf              = _mm_set1_ps(fr->ic->k_rf);
 615     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 616     crf              = _mm_set1_ps(fr->ic->c_rf);
 617
 618     /* Setup water-specific parameters */
 619     inr              = nlist->iinr[0];
 620     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 621     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 622     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 623
 624     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 625     rcutoff_scalar   = fr->rcoulomb;
 626     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 627     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 628
 629     /* Avoid stupid compiler warnings */
 630     jnrA = jnrB = jnrC = jnrD = 0;
 631     j_coord_offsetA = 0;
 632     j_coord_offsetB = 0;
 633     j_coord_offsetC = 0;
 634     j_coord_offsetD = 0;
 635
 636     outeriter        = 0;
 637     inneriter        = 0;
 638
 639     for(iidx=0;iidx<4*DIM;iidx++)
 640     {
 641         scratch[iidx] = 0.0;
 642     }
 643
 644     /* Start outer loop over neighborlists */
 645     for(iidx=0; iidx<nri; iidx++)
 646     {
 647         /* Load shift vector for this list */
 648         i_shift_offset   = DIM*shiftidx[iidx];
 649
 650         /* Load limits for loop over neighbors */
 651         j_index_start    = jindex[iidx];
 652         j_index_end      = jindex[iidx+1];
 653
 654         /* Get outer coordinate index */
 655         inr              = iinr[iidx];
 656         i_coord_offset   = DIM*inr;
 657
 658         /* Load i particle coords and add shift vector */
 659         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 660                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 661
 662         fix1             = _mm_setzero_ps();
 663         fiy1             = _mm_setzero_ps();
 664         fiz1             = _mm_setzero_ps();
 665         fix2             = _mm_setzero_ps();
 666         fiy2             = _mm_setzero_ps();
 667         fiz2             = _mm_setzero_ps();
 668         fix3             = _mm_setzero_ps();
 669         fiy3             = _mm_setzero_ps();
 670         fiz3             = _mm_setzero_ps();
 671
 672         /* Start inner kernel loop */
 673         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 674         {
 675
 676             /* Get j neighbor index, and coordinate index */
 677             jnrA             = jjnr[jidx];
 678             jnrB             = jjnr[jidx+1];
 679             jnrC             = jjnr[jidx+2];
 680             jnrD             = jjnr[jidx+3];
 681             j_coord_offsetA  = DIM*jnrA;
 682             j_coord_offsetB  = DIM*jnrB;
 683             j_coord_offsetC  = DIM*jnrC;
 684             j_coord_offsetD  = DIM*jnrD;
 685
 686             /* load j atom coordinates */
 687             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 688                                               x+j_coord_offsetC,x+j_coord_offsetD,
 689                                               &jx0,&jy0,&jz0);
 690
 691             /* Calculate displacement vector */
 692             dx10             = _mm_sub_ps(ix1,jx0);
 693             dy10             = _mm_sub_ps(iy1,jy0);
 694             dz10             = _mm_sub_ps(iz1,jz0);
 695             dx20             = _mm_sub_ps(ix2,jx0);
 696             dy20             = _mm_sub_ps(iy2,jy0);
 697             dz20             = _mm_sub_ps(iz2,jz0);
 698             dx30             = _mm_sub_ps(ix3,jx0);
 699             dy30             = _mm_sub_ps(iy3,jy0);
 700             dz30             = _mm_sub_ps(iz3,jz0);
 701
 702             /* Calculate squared distance and things based on it */
 703             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 704             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 705             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 706
 707             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 708             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 709             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 710
 711             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 712             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 713             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 714
 715             /* Load parameters for j particles */
 716             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 717                                                               charge+jnrC+0,charge+jnrD+0);
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             if (gmx_mm_any_lt(rsq10,rcutoff2))
 724             {
 725
 726             /* Compute parameters for interactions between i and j atoms */
 727             qq10             = _mm_mul_ps(iq1,jq0);
 728
 729             /* REACTION-FIELD ELECTROSTATICS */
 730             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 731
 732             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 733
 734             fscal            = felec;
 735
 736             fscal            = _mm_and_ps(fscal,cutoff_mask);
 737
 738             /* Calculate temporary vectorial force */
 739             tx               = _mm_mul_ps(fscal,dx10);
 740             ty               = _mm_mul_ps(fscal,dy10);
 741             tz               = _mm_mul_ps(fscal,dz10);
 742
 743             /* Update vectorial force */
 744             fix1             = _mm_add_ps(fix1,tx);
 745             fiy1             = _mm_add_ps(fiy1,ty);
 746             fiz1             = _mm_add_ps(fiz1,tz);
 747
 748             fjptrA             = f+j_coord_offsetA;
 749             fjptrB             = f+j_coord_offsetB;
 750             fjptrC             = f+j_coord_offsetC;
 751             fjptrD             = f+j_coord_offsetD;
 752             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 753
 754             }
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             if (gmx_mm_any_lt(rsq20,rcutoff2))
 761             {
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq20             = _mm_mul_ps(iq2,jq0);
 765
 766             /* REACTION-FIELD ELECTROSTATICS */
 767             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 768
 769             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 770
 771             fscal            = felec;
 772
 773             fscal            = _mm_and_ps(fscal,cutoff_mask);
 774
 775             /* Calculate temporary vectorial force */
 776             tx               = _mm_mul_ps(fscal,dx20);
 777             ty               = _mm_mul_ps(fscal,dy20);
 778             tz               = _mm_mul_ps(fscal,dz20);
 779
 780             /* Update vectorial force */
 781             fix2             = _mm_add_ps(fix2,tx);
 782             fiy2             = _mm_add_ps(fiy2,ty);
 783             fiz2             = _mm_add_ps(fiz2,tz);
 784
 785             fjptrA             = f+j_coord_offsetA;
 786             fjptrB             = f+j_coord_offsetB;
 787             fjptrC             = f+j_coord_offsetC;
 788             fjptrD             = f+j_coord_offsetD;
 789             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 790
 791             }
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             if (gmx_mm_any_lt(rsq30,rcutoff2))
 798             {
 799
 800             /* Compute parameters for interactions between i and j atoms */
 801             qq30             = _mm_mul_ps(iq3,jq0);
 802
 803             /* REACTION-FIELD ELECTROSTATICS */
 804             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 805
 806             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 807
 808             fscal            = felec;
 809
 810             fscal            = _mm_and_ps(fscal,cutoff_mask);
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm_mul_ps(fscal,dx30);
 814             ty               = _mm_mul_ps(fscal,dy30);
 815             tz               = _mm_mul_ps(fscal,dz30);
 816
 817             /* Update vectorial force */
 818             fix3             = _mm_add_ps(fix3,tx);
 819             fiy3             = _mm_add_ps(fiy3,ty);
 820             fiz3             = _mm_add_ps(fiz3,tz);
 821
 822             fjptrA             = f+j_coord_offsetA;
 823             fjptrB             = f+j_coord_offsetB;
 824             fjptrC             = f+j_coord_offsetC;
 825             fjptrD             = f+j_coord_offsetD;
 826             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 827
 828             }
 829
 830             /* Inner loop uses 90 flops */
 831         }
 832
 833         if(jidx<j_index_end)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrlistA         = jjnr[jidx];
 838             jnrlistB         = jjnr[jidx+1];
 839             jnrlistC         = jjnr[jidx+2];
 840             jnrlistD         = jjnr[jidx+3];
 841             /* Sign of each element will be negative for non-real atoms.
 842              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 843              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 844              */
 845             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 846             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 847             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 848             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 849             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852             j_coord_offsetC  = DIM*jnrC;
 853             j_coord_offsetD  = DIM*jnrD;
 854
 855             /* load j atom coordinates */
 856             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                               x+j_coord_offsetC,x+j_coord_offsetD,
 858                                               &jx0,&jy0,&jz0);
 859
 860             /* Calculate displacement vector */
 861             dx10             = _mm_sub_ps(ix1,jx0);
 862             dy10             = _mm_sub_ps(iy1,jy0);
 863             dz10             = _mm_sub_ps(iz1,jz0);
 864             dx20             = _mm_sub_ps(ix2,jx0);
 865             dy20             = _mm_sub_ps(iy2,jy0);
 866             dz20             = _mm_sub_ps(iz2,jz0);
 867             dx30             = _mm_sub_ps(ix3,jx0);
 868             dy30             = _mm_sub_ps(iy3,jy0);
 869             dz30             = _mm_sub_ps(iz3,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 873             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 874             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 875
 876             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 877             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 878             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 879
 880             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 881             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 882             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 883
 884             /* Load parameters for j particles */
 885             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 886                                                               charge+jnrC+0,charge+jnrD+0);
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             if (gmx_mm_any_lt(rsq10,rcutoff2))
 893             {
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             qq10             = _mm_mul_ps(iq1,jq0);
 897
 898             /* REACTION-FIELD ELECTROSTATICS */
 899             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 900
 901             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm_and_ps(fscal,cutoff_mask);
 906
 907             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 908
 909             /* Calculate temporary vectorial force */
 910             tx               = _mm_mul_ps(fscal,dx10);
 911             ty               = _mm_mul_ps(fscal,dy10);
 912             tz               = _mm_mul_ps(fscal,dz10);
 913
 914             /* Update vectorial force */
 915             fix1             = _mm_add_ps(fix1,tx);
 916             fiy1             = _mm_add_ps(fiy1,ty);
 917             fiz1             = _mm_add_ps(fiz1,tz);
 918
 919             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 920             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 921             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 922             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 923             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 924
 925             }
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             if (gmx_mm_any_lt(rsq20,rcutoff2))
 932             {
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq20             = _mm_mul_ps(iq2,jq0);
 936
 937             /* REACTION-FIELD ELECTROSTATICS */
 938             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 939
 940             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm_and_ps(fscal,cutoff_mask);
 945
 946             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 947
 948             /* Calculate temporary vectorial force */
 949             tx               = _mm_mul_ps(fscal,dx20);
 950             ty               = _mm_mul_ps(fscal,dy20);
 951             tz               = _mm_mul_ps(fscal,dz20);
 952
 953             /* Update vectorial force */
 954             fix2             = _mm_add_ps(fix2,tx);
 955             fiy2             = _mm_add_ps(fiy2,ty);
 956             fiz2             = _mm_add_ps(fiz2,tz);
 957
 958             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 959             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 960             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 961             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 962             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 963
 964             }
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm_any_lt(rsq30,rcutoff2))
 971             {
 972
 973             /* Compute parameters for interactions between i and j atoms */
 974             qq30             = _mm_mul_ps(iq3,jq0);
 975
 976             /* REACTION-FIELD ELECTROSTATICS */
 977             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 978
 979             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 980
 981             fscal            = felec;
 982
 983             fscal            = _mm_and_ps(fscal,cutoff_mask);
 984
 985             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm_mul_ps(fscal,dx30);
 989             ty               = _mm_mul_ps(fscal,dy30);
 990             tz               = _mm_mul_ps(fscal,dz30);
 991
 992             /* Update vectorial force */
 993             fix3             = _mm_add_ps(fix3,tx);
 994             fiy3             = _mm_add_ps(fiy3,ty);
 995             fiz3             = _mm_add_ps(fiz3,tz);
 996
 997             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 998             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 999             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1000             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1001             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1002
1003             }
1004
1005             /* Inner loop uses 90 flops */
1006         }
1007
1008         /* End of innermost loop */
1009
1010         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1011                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1012
1013         /* Increment number of inner iterations */
1014         inneriter                  += j_index_end - j_index_start;
1015
1016         /* Outer loop uses 18 flops */
1017     }
1018
1019     /* Increment number of outer iterations */
1020     outeriter        += nri;
1021
1022     /* Update outer/inner flops */
1023
1024     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*90);
1025 }