src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwNone_GeomW4P1_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_ElecRFCut_VdwNone_GeomW4P1_VF_sse2_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_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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm_set1_ps(fr->ic->k_rf);
  97     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_ps(fr->ic->c_rf);
  99
 100     /* Setup water-specific parameters */
 101     inr              = nlist->iinr[0];
 102     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 103     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 104     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 105
 106     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 107     rcutoff_scalar   = fr->rcoulomb;
 108     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 109     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126         shX              = shiftvec[i_shift_offset+XX];
 127         shY              = shiftvec[i_shift_offset+YY];
 128         shZ              = shiftvec[i_shift_offset+ZZ];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 140         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 141         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 142         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 143         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 144         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 145         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 146         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 147         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 148
 149         fix1             = _mm_setzero_ps();
 150         fiy1             = _mm_setzero_ps();
 151         fiz1             = _mm_setzero_ps();
 152         fix2             = _mm_setzero_ps();
 153         fiy2             = _mm_setzero_ps();
 154         fiz2             = _mm_setzero_ps();
 155         fix3             = _mm_setzero_ps();
 156         fiy3             = _mm_setzero_ps();
 157         fiz3             = _mm_setzero_ps();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx10             = _mm_sub_ps(ix1,jx0);
 184             dy10             = _mm_sub_ps(iy1,jy0);
 185             dz10             = _mm_sub_ps(iz1,jz0);
 186             dx20             = _mm_sub_ps(ix2,jx0);
 187             dy20             = _mm_sub_ps(iy2,jy0);
 188             dz20             = _mm_sub_ps(iz2,jz0);
 189             dx30             = _mm_sub_ps(ix3,jx0);
 190             dy30             = _mm_sub_ps(iy3,jy0);
 191             dz30             = _mm_sub_ps(iz3,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 195             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 196             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 197
 198             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 199             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 200             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 201
 202             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 203             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 204             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                               charge+jnrC+0,charge+jnrD+0);
 209
 210             /**************************
 211              * CALCULATE INTERACTIONS *
 212              **************************/
 213
 214             if (gmx_mm_any_lt(rsq10,rcutoff2))
 215             {
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             qq10             = _mm_mul_ps(iq1,jq0);
 219
 220             /* REACTION-FIELD ELECTROSTATICS */
 221             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 222             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 223
 224             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velec            = _mm_and_ps(velec,cutoff_mask);
 228             velecsum         = _mm_add_ps(velecsum,velec);
 229
 230             fscal            = felec;
 231
 232             fscal            = _mm_and_ps(fscal,cutoff_mask);
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_ps(fscal,dx10);
 236             ty               = _mm_mul_ps(fscal,dy10);
 237             tz               = _mm_mul_ps(fscal,dz10);
 238
 239             /* Update vectorial force */
 240             fix1             = _mm_add_ps(fix1,tx);
 241             fiy1             = _mm_add_ps(fiy1,ty);
 242             fiz1             = _mm_add_ps(fiz1,tz);
 243
 244             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 245                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 246                                                    tx,ty,tz);
 247
 248             }
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             if (gmx_mm_any_lt(rsq20,rcutoff2))
 255             {
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq20             = _mm_mul_ps(iq2,jq0);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 262             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 263
 264             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velec            = _mm_and_ps(velec,cutoff_mask);
 268             velecsum         = _mm_add_ps(velecsum,velec);
 269
 270             fscal            = felec;
 271
 272             fscal            = _mm_and_ps(fscal,cutoff_mask);
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_ps(fscal,dx20);
 276             ty               = _mm_mul_ps(fscal,dy20);
 277             tz               = _mm_mul_ps(fscal,dz20);
 278
 279             /* Update vectorial force */
 280             fix2             = _mm_add_ps(fix2,tx);
 281             fiy2             = _mm_add_ps(fiy2,ty);
 282             fiz2             = _mm_add_ps(fiz2,tz);
 283
 284             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 285                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 286                                                    tx,ty,tz);
 287
 288             }
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm_any_lt(rsq30,rcutoff2))
 295             {
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq30             = _mm_mul_ps(iq3,jq0);
 299
 300             /* REACTION-FIELD ELECTROSTATICS */
 301             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 302             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 303
 304             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velec            = _mm_and_ps(velec,cutoff_mask);
 308             velecsum         = _mm_add_ps(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             fscal            = _mm_and_ps(fscal,cutoff_mask);
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm_mul_ps(fscal,dx30);
 316             ty               = _mm_mul_ps(fscal,dy30);
 317             tz               = _mm_mul_ps(fscal,dz30);
 318
 319             /* Update vectorial force */
 320             fix3             = _mm_add_ps(fix3,tx);
 321             fiy3             = _mm_add_ps(fiy3,ty);
 322             fiz3             = _mm_add_ps(fiz3,tz);
 323
 324             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 325                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 326                                                    tx,ty,tz);
 327
 328             }
 329
 330             /* Inner loop uses 108 flops */
 331         }
 332
 333         if(jidx<j_index_end)
 334         {
 335
 336             /* Get j neighbor index, and coordinate index */
 337             jnrA             = jjnr[jidx];
 338             jnrB             = jjnr[jidx+1];
 339             jnrC             = jjnr[jidx+2];
 340             jnrD             = jjnr[jidx+3];
 341
 342             /* Sign of each element will be negative for non-real atoms.
 343              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 344              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 345              */
 346             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 347             jnrA       = (jnrA>=0) ? jnrA : 0;
 348             jnrB       = (jnrB>=0) ? jnrB : 0;
 349             jnrC       = (jnrC>=0) ? jnrC : 0;
 350             jnrD       = (jnrD>=0) ? jnrD : 0;
 351
 352             j_coord_offsetA  = DIM*jnrA;
 353             j_coord_offsetB  = DIM*jnrB;
 354             j_coord_offsetC  = DIM*jnrC;
 355             j_coord_offsetD  = DIM*jnrD;
 356
 357             /* load j atom coordinates */
 358             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 359                                               x+j_coord_offsetC,x+j_coord_offsetD,
 360                                               &jx0,&jy0,&jz0);
 361
 362             /* Calculate displacement vector */
 363             dx10             = _mm_sub_ps(ix1,jx0);
 364             dy10             = _mm_sub_ps(iy1,jy0);
 365             dz10             = _mm_sub_ps(iz1,jz0);
 366             dx20             = _mm_sub_ps(ix2,jx0);
 367             dy20             = _mm_sub_ps(iy2,jy0);
 368             dz20             = _mm_sub_ps(iz2,jz0);
 369             dx30             = _mm_sub_ps(ix3,jx0);
 370             dy30             = _mm_sub_ps(iy3,jy0);
 371             dz30             = _mm_sub_ps(iz3,jz0);
 372
 373             /* Calculate squared distance and things based on it */
 374             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 375             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 376             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 377
 378             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 379             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 380             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 381
 382             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 383             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 384             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 385
 386             /* Load parameters for j particles */
 387             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 388                                                               charge+jnrC+0,charge+jnrD+0);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             if (gmx_mm_any_lt(rsq10,rcutoff2))
 395             {
 396
 397             /* Compute parameters for interactions between i and j atoms */
 398             qq10             = _mm_mul_ps(iq1,jq0);
 399
 400             /* REACTION-FIELD ELECTROSTATICS */
 401             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 402             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 403
 404             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm_and_ps(velec,cutoff_mask);
 408             velec            = _mm_andnot_ps(dummy_mask,velec);
 409             velecsum         = _mm_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm_and_ps(fscal,cutoff_mask);
 414
 415             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_ps(fscal,dx10);
 419             ty               = _mm_mul_ps(fscal,dy10);
 420             tz               = _mm_mul_ps(fscal,dz10);
 421
 422             /* Update vectorial force */
 423             fix1             = _mm_add_ps(fix1,tx);
 424             fiy1             = _mm_add_ps(fiy1,ty);
 425             fiz1             = _mm_add_ps(fiz1,tz);
 426
 427             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 428                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 429                                                    tx,ty,tz);
 430
 431             }
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             if (gmx_mm_any_lt(rsq20,rcutoff2))
 438             {
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq20             = _mm_mul_ps(iq2,jq0);
 442
 443             /* REACTION-FIELD ELECTROSTATICS */
 444             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 445             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 446
 447             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velec            = _mm_and_ps(velec,cutoff_mask);
 451             velec            = _mm_andnot_ps(dummy_mask,velec);
 452             velecsum         = _mm_add_ps(velecsum,velec);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm_and_ps(fscal,cutoff_mask);
 457
 458             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm_mul_ps(fscal,dx20);
 462             ty               = _mm_mul_ps(fscal,dy20);
 463             tz               = _mm_mul_ps(fscal,dz20);
 464
 465             /* Update vectorial force */
 466             fix2             = _mm_add_ps(fix2,tx);
 467             fiy2             = _mm_add_ps(fiy2,ty);
 468             fiz2             = _mm_add_ps(fiz2,tz);
 469
 470             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 471                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 472                                                    tx,ty,tz);
 473
 474             }
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             if (gmx_mm_any_lt(rsq30,rcutoff2))
 481             {
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq30             = _mm_mul_ps(iq3,jq0);
 485
 486             /* REACTION-FIELD ELECTROSTATICS */
 487             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 488             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 489
 490             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm_and_ps(velec,cutoff_mask);
 494             velec            = _mm_andnot_ps(dummy_mask,velec);
 495             velecsum         = _mm_add_ps(velecsum,velec);
 496
 497             fscal            = felec;
 498
 499             fscal            = _mm_and_ps(fscal,cutoff_mask);
 500
 501             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 502
 503             /* Calculate temporary vectorial force */
 504             tx               = _mm_mul_ps(fscal,dx30);
 505             ty               = _mm_mul_ps(fscal,dy30);
 506             tz               = _mm_mul_ps(fscal,dz30);
 507
 508             /* Update vectorial force */
 509             fix3             = _mm_add_ps(fix3,tx);
 510             fiy3             = _mm_add_ps(fiy3,ty);
 511             fiz3             = _mm_add_ps(fiz3,tz);
 512
 513             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 514                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 515                                                    tx,ty,tz);
 516
 517             }
 518
 519             /* Inner loop uses 108 flops */
 520         }
 521
 522         /* End of innermost loop */
 523
 524         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 525                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 526
 527         ggid                        = gid[iidx];
 528         /* Update potential energies */
 529         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 530
 531         /* Increment number of inner iterations */
 532         inneriter                  += j_index_end - j_index_start;
 533
 534         /* Outer loop uses 28 flops */
 535     }
 536
 537     /* Increment number of outer iterations */
 538     outeriter        += nri;
 539
 540     /* Update outer/inner flops */
 541
 542     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*28 + inneriter*108);
 543 }
 544 /*
 545  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse2_single
 546  * Electrostatics interaction: ReactionField
 547  * VdW interaction:            None
 548  * Geometry:                   Water4-Particle
 549  * Calculate force/pot:        Force
 550  */
 551 void
 552 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse2_single
 553                     (t_nblist * gmx_restrict                nlist,
 554                      rvec * gmx_restrict                    xx,
 555                      rvec * gmx_restrict                    ff,
 556                      t_forcerec * gmx_restrict              fr,
 557                      t_mdatoms * gmx_restrict               mdatoms,
 558                      nb_kernel_data_t * gmx_restrict        kernel_data,
 559                      t_nrnb * gmx_restrict                  nrnb)
 560 {
 561     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 562      * just 0 for non-waters.
 563      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 564      * jnr indices corresponding to data put in the four positions in the SIMD register.
 565      */
 566     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 567     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 568     int              jnrA,jnrB,jnrC,jnrD;
 569     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 570     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 571     real             shX,shY,shZ,rcutoff_scalar;
 572     real             *shiftvec,*fshift,*x,*f;
 573     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 574     int              vdwioffset1;
 575     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 576     int              vdwioffset2;
 577     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 578     int              vdwioffset3;
 579     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 580     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 581     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 582     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 583     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 584     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 585     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 586     real             *charge;
 587     __m128           dummy_mask,cutoff_mask;
 588     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 589     __m128           one     = _mm_set1_ps(1.0);
 590     __m128           two     = _mm_set1_ps(2.0);
 591     x                = xx[0];
 592     f                = ff[0];
 593
 594     nri              = nlist->nri;
 595     iinr             = nlist->iinr;
 596     jindex           = nlist->jindex;
 597     jjnr             = nlist->jjnr;
 598     shiftidx         = nlist->shift;
 599     gid              = nlist->gid;
 600     shiftvec         = fr->shift_vec[0];
 601     fshift           = fr->fshift[0];
 602     facel            = _mm_set1_ps(fr->epsfac);
 603     charge           = mdatoms->chargeA;
 604     krf              = _mm_set1_ps(fr->ic->k_rf);
 605     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 606     crf              = _mm_set1_ps(fr->ic->c_rf);
 607
 608     /* Setup water-specific parameters */
 609     inr              = nlist->iinr[0];
 610     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 611     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 612     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 613
 614     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 615     rcutoff_scalar   = fr->rcoulomb;
 616     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 617     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 618
 619     /* Avoid stupid compiler warnings */
 620     jnrA = jnrB = jnrC = jnrD = 0;
 621     j_coord_offsetA = 0;
 622     j_coord_offsetB = 0;
 623     j_coord_offsetC = 0;
 624     j_coord_offsetD = 0;
 625
 626     outeriter        = 0;
 627     inneriter        = 0;
 628
 629     /* Start outer loop over neighborlists */
 630     for(iidx=0; iidx<nri; iidx++)
 631     {
 632         /* Load shift vector for this list */
 633         i_shift_offset   = DIM*shiftidx[iidx];
 634         shX              = shiftvec[i_shift_offset+XX];
 635         shY              = shiftvec[i_shift_offset+YY];
 636         shZ              = shiftvec[i_shift_offset+ZZ];
 637
 638         /* Load limits for loop over neighbors */
 639         j_index_start    = jindex[iidx];
 640         j_index_end      = jindex[iidx+1];
 641
 642         /* Get outer coordinate index */
 643         inr              = iinr[iidx];
 644         i_coord_offset   = DIM*inr;
 645
 646         /* Load i particle coords and add shift vector */
 647         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 648         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 649         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 650         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 651         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 652         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 653         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 654         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 655         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 656
 657         fix1             = _mm_setzero_ps();
 658         fiy1             = _mm_setzero_ps();
 659         fiz1             = _mm_setzero_ps();
 660         fix2             = _mm_setzero_ps();
 661         fiy2             = _mm_setzero_ps();
 662         fiz2             = _mm_setzero_ps();
 663         fix3             = _mm_setzero_ps();
 664         fiy3             = _mm_setzero_ps();
 665         fiz3             = _mm_setzero_ps();
 666
 667         /* Start inner kernel loop */
 668         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 669         {
 670
 671             /* Get j neighbor index, and coordinate index */
 672             jnrA             = jjnr[jidx];
 673             jnrB             = jjnr[jidx+1];
 674             jnrC             = jjnr[jidx+2];
 675             jnrD             = jjnr[jidx+3];
 676
 677             j_coord_offsetA  = DIM*jnrA;
 678             j_coord_offsetB  = DIM*jnrB;
 679             j_coord_offsetC  = DIM*jnrC;
 680             j_coord_offsetD  = DIM*jnrD;
 681
 682             /* load j atom coordinates */
 683             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 684                                               x+j_coord_offsetC,x+j_coord_offsetD,
 685                                               &jx0,&jy0,&jz0);
 686
 687             /* Calculate displacement vector */
 688             dx10             = _mm_sub_ps(ix1,jx0);
 689             dy10             = _mm_sub_ps(iy1,jy0);
 690             dz10             = _mm_sub_ps(iz1,jz0);
 691             dx20             = _mm_sub_ps(ix2,jx0);
 692             dy20             = _mm_sub_ps(iy2,jy0);
 693             dz20             = _mm_sub_ps(iz2,jz0);
 694             dx30             = _mm_sub_ps(ix3,jx0);
 695             dy30             = _mm_sub_ps(iy3,jy0);
 696             dz30             = _mm_sub_ps(iz3,jz0);
 697
 698             /* Calculate squared distance and things based on it */
 699             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 700             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 701             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 702
 703             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 704             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 705             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 706
 707             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 708             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 709             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 710
 711             /* Load parameters for j particles */
 712             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 713                                                               charge+jnrC+0,charge+jnrD+0);
 714
 715             /**************************
 716              * CALCULATE INTERACTIONS *
 717              **************************/
 718
 719             if (gmx_mm_any_lt(rsq10,rcutoff2))
 720             {
 721
 722             /* Compute parameters for interactions between i and j atoms */
 723             qq10             = _mm_mul_ps(iq1,jq0);
 724
 725             /* REACTION-FIELD ELECTROSTATICS */
 726             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 727
 728             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 729
 730             fscal            = felec;
 731
 732             fscal            = _mm_and_ps(fscal,cutoff_mask);
 733
 734             /* Calculate temporary vectorial force */
 735             tx               = _mm_mul_ps(fscal,dx10);
 736             ty               = _mm_mul_ps(fscal,dy10);
 737             tz               = _mm_mul_ps(fscal,dz10);
 738
 739             /* Update vectorial force */
 740             fix1             = _mm_add_ps(fix1,tx);
 741             fiy1             = _mm_add_ps(fiy1,ty);
 742             fiz1             = _mm_add_ps(fiz1,tz);
 743
 744             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 745                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 746                                                    tx,ty,tz);
 747
 748             }
 749
 750             /**************************
 751              * CALCULATE INTERACTIONS *
 752              **************************/
 753
 754             if (gmx_mm_any_lt(rsq20,rcutoff2))
 755             {
 756
 757             /* Compute parameters for interactions between i and j atoms */
 758             qq20             = _mm_mul_ps(iq2,jq0);
 759
 760             /* REACTION-FIELD ELECTROSTATICS */
 761             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 762
 763             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 764
 765             fscal            = felec;
 766
 767             fscal            = _mm_and_ps(fscal,cutoff_mask);
 768
 769             /* Calculate temporary vectorial force */
 770             tx               = _mm_mul_ps(fscal,dx20);
 771             ty               = _mm_mul_ps(fscal,dy20);
 772             tz               = _mm_mul_ps(fscal,dz20);
 773
 774             /* Update vectorial force */
 775             fix2             = _mm_add_ps(fix2,tx);
 776             fiy2             = _mm_add_ps(fiy2,ty);
 777             fiz2             = _mm_add_ps(fiz2,tz);
 778
 779             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 780                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 781                                                    tx,ty,tz);
 782
 783             }
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             if (gmx_mm_any_lt(rsq30,rcutoff2))
 790             {
 791
 792             /* Compute parameters for interactions between i and j atoms */
 793             qq30             = _mm_mul_ps(iq3,jq0);
 794
 795             /* REACTION-FIELD ELECTROSTATICS */
 796             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 797
 798             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 799
 800             fscal            = felec;
 801
 802             fscal            = _mm_and_ps(fscal,cutoff_mask);
 803
 804             /* Calculate temporary vectorial force */
 805             tx               = _mm_mul_ps(fscal,dx30);
 806             ty               = _mm_mul_ps(fscal,dy30);
 807             tz               = _mm_mul_ps(fscal,dz30);
 808
 809             /* Update vectorial force */
 810             fix3             = _mm_add_ps(fix3,tx);
 811             fiy3             = _mm_add_ps(fiy3,ty);
 812             fiz3             = _mm_add_ps(fiz3,tz);
 813
 814             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 815                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 816                                                    tx,ty,tz);
 817
 818             }
 819
 820             /* Inner loop uses 90 flops */
 821         }
 822
 823         if(jidx<j_index_end)
 824         {
 825
 826             /* Get j neighbor index, and coordinate index */
 827             jnrA             = jjnr[jidx];
 828             jnrB             = jjnr[jidx+1];
 829             jnrC             = jjnr[jidx+2];
 830             jnrD             = jjnr[jidx+3];
 831
 832             /* Sign of each element will be negative for non-real atoms.
 833              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 834              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 835              */
 836             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 837             jnrA       = (jnrA>=0) ? jnrA : 0;
 838             jnrB       = (jnrB>=0) ? jnrB : 0;
 839             jnrC       = (jnrC>=0) ? jnrC : 0;
 840             jnrD       = (jnrD>=0) ? jnrD : 0;
 841
 842             j_coord_offsetA  = DIM*jnrA;
 843             j_coord_offsetB  = DIM*jnrB;
 844             j_coord_offsetC  = DIM*jnrC;
 845             j_coord_offsetD  = DIM*jnrD;
 846
 847             /* load j atom coordinates */
 848             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                               x+j_coord_offsetC,x+j_coord_offsetD,
 850                                               &jx0,&jy0,&jz0);
 851
 852             /* Calculate displacement vector */
 853             dx10             = _mm_sub_ps(ix1,jx0);
 854             dy10             = _mm_sub_ps(iy1,jy0);
 855             dz10             = _mm_sub_ps(iz1,jz0);
 856             dx20             = _mm_sub_ps(ix2,jx0);
 857             dy20             = _mm_sub_ps(iy2,jy0);
 858             dz20             = _mm_sub_ps(iz2,jz0);
 859             dx30             = _mm_sub_ps(ix3,jx0);
 860             dy30             = _mm_sub_ps(iy3,jy0);
 861             dz30             = _mm_sub_ps(iz3,jz0);
 862
 863             /* Calculate squared distance and things based on it */
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 867
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 871
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 875
 876             /* Load parameters for j particles */
 877             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 878                                                               charge+jnrC+0,charge+jnrD+0);
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             if (gmx_mm_any_lt(rsq10,rcutoff2))
 885             {
 886
 887             /* Compute parameters for interactions between i and j atoms */
 888             qq10             = _mm_mul_ps(iq1,jq0);
 889
 890             /* REACTION-FIELD ELECTROSTATICS */
 891             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 892
 893             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 894
 895             fscal            = felec;
 896
 897             fscal            = _mm_and_ps(fscal,cutoff_mask);
 898
 899             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 900
 901             /* Calculate temporary vectorial force */
 902             tx               = _mm_mul_ps(fscal,dx10);
 903             ty               = _mm_mul_ps(fscal,dy10);
 904             tz               = _mm_mul_ps(fscal,dz10);
 905
 906             /* Update vectorial force */
 907             fix1             = _mm_add_ps(fix1,tx);
 908             fiy1             = _mm_add_ps(fiy1,ty);
 909             fiz1             = _mm_add_ps(fiz1,tz);
 910
 911             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 912                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 913                                                    tx,ty,tz);
 914
 915             }
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             if (gmx_mm_any_lt(rsq20,rcutoff2))
 922             {
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq20             = _mm_mul_ps(iq2,jq0);
 926
 927             /* REACTION-FIELD ELECTROSTATICS */
 928             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 929
 930             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 931
 932             fscal            = felec;
 933
 934             fscal            = _mm_and_ps(fscal,cutoff_mask);
 935
 936             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_ps(fscal,dx20);
 940             ty               = _mm_mul_ps(fscal,dy20);
 941             tz               = _mm_mul_ps(fscal,dz20);
 942
 943             /* Update vectorial force */
 944             fix2             = _mm_add_ps(fix2,tx);
 945             fiy2             = _mm_add_ps(fiy2,ty);
 946             fiz2             = _mm_add_ps(fiz2,tz);
 947
 948             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 949                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 950                                                    tx,ty,tz);
 951
 952             }
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             if (gmx_mm_any_lt(rsq30,rcutoff2))
 959             {
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq30             = _mm_mul_ps(iq3,jq0);
 963
 964             /* REACTION-FIELD ELECTROSTATICS */
 965             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 966
 967             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 968
 969             fscal            = felec;
 970
 971             fscal            = _mm_and_ps(fscal,cutoff_mask);
 972
 973             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 974
 975             /* Calculate temporary vectorial force */
 976             tx               = _mm_mul_ps(fscal,dx30);
 977             ty               = _mm_mul_ps(fscal,dy30);
 978             tz               = _mm_mul_ps(fscal,dz30);
 979
 980             /* Update vectorial force */
 981             fix3             = _mm_add_ps(fix3,tx);
 982             fiy3             = _mm_add_ps(fiy3,ty);
 983             fiz3             = _mm_add_ps(fiz3,tz);
 984
 985             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 986                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 987                                                    tx,ty,tz);
 988
 989             }
 990
 991             /* Inner loop uses 90 flops */
 992         }
 993
 994         /* End of innermost loop */
 995
 996         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 997                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 998
 999         /* Increment number of inner iterations */
1000         inneriter                  += j_index_end - j_index_start;
1001
1002         /* Outer loop uses 27 flops */
1003     }
1004
1005     /* Increment number of outer iterations */
1006     outeriter        += nri;
1007
1008     /* Update outer/inner flops */
1009
1010     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*27 + inneriter*90);
1011 }