src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_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_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  92     real             rswitch_scalar,d_scalar;
  93     __m256           dummy_mask,cutoff_mask;
  94     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  95     __m256           one     = _mm256_set1_ps(1.0);
  96     __m256           two     = _mm256_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm256_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm256_set1_ps(fr->ic->k_rf);
 111     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 112     crf              = _mm256_set1_ps(fr->ic->c_rf);
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 120     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 121     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 122     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 127     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 128
 129     rswitch_scalar   = fr->rvdw_switch;
 130     rswitch          = _mm256_set1_ps(rswitch_scalar);
 131     /* Setup switch parameters */
 132     d_scalar         = rcutoff_scalar-rswitch_scalar;
 133     d                = _mm256_set1_ps(d_scalar);
 134     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 135     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 138     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254             vdwjidx0E        = 2*vdwtype[jnrE+0];
 255             vdwjidx0F        = 2*vdwtype[jnrF+0];
 256             vdwjidx0G        = 2*vdwtype[jnrG+0];
 257             vdwjidx0H        = 2*vdwtype[jnrH+0];
 258
 259             fjx0             = _mm256_setzero_ps();
 260             fjy0             = _mm256_setzero_ps();
 261             fjz0             = _mm256_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 268             {
 269
 270             r00              = _mm256_mul_ps(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm256_mul_ps(iq0,jq0);
 274             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             vdwioffsetptr0+vdwjidx0E,
 279                                             vdwioffsetptr0+vdwjidx0F,
 280                                             vdwioffsetptr0+vdwjidx0G,
 281                                             vdwioffsetptr0+vdwjidx0H,
 282                                             &c6_00,&c12_00);
 283
 284             /* REACTION-FIELD ELECTROSTATICS */
 285             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 286             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 287
 288             /* LENNARD-JONES DISPERSION/REPULSION */
 289
 290             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 291             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 292             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 293             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 294             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 295
 296             d                = _mm256_sub_ps(r00,rswitch);
 297             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 298             d2               = _mm256_mul_ps(d,d);
 299             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 300
 301             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 302
 303             /* Evaluate switch function */
 304             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 305             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 306             vvdw             = _mm256_mul_ps(vvdw,sw);
 307             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm256_and_ps(velec,cutoff_mask);
 311             velecsum         = _mm256_add_ps(velecsum,velec);
 312             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 313             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 314
 315             fscal            = _mm256_add_ps(felec,fvdw);
 316
 317             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_ps(fscal,dx00);
 321             ty               = _mm256_mul_ps(fscal,dy00);
 322             tz               = _mm256_mul_ps(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm256_add_ps(fix0,tx);
 326             fiy0             = _mm256_add_ps(fiy0,ty);
 327             fiz0             = _mm256_add_ps(fiz0,tz);
 328
 329             fjx0             = _mm256_add_ps(fjx0,tx);
 330             fjy0             = _mm256_add_ps(fjy0,ty);
 331             fjz0             = _mm256_add_ps(fjz0,tz);
 332
 333             }
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 340             {
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq10             = _mm256_mul_ps(iq1,jq0);
 344
 345             /* REACTION-FIELD ELECTROSTATICS */
 346             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 347             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 348
 349             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm256_and_ps(velec,cutoff_mask);
 353             velecsum         = _mm256_add_ps(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_ps(fscal,dx10);
 361             ty               = _mm256_mul_ps(fscal,dy10);
 362             tz               = _mm256_mul_ps(fscal,dz10);
 363
 364             /* Update vectorial force */
 365             fix1             = _mm256_add_ps(fix1,tx);
 366             fiy1             = _mm256_add_ps(fiy1,ty);
 367             fiz1             = _mm256_add_ps(fiz1,tz);
 368
 369             fjx0             = _mm256_add_ps(fjx0,tx);
 370             fjy0             = _mm256_add_ps(fjy0,ty);
 371             fjz0             = _mm256_add_ps(fjz0,tz);
 372
 373             }
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 380             {
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq20             = _mm256_mul_ps(iq2,jq0);
 384
 385             /* REACTION-FIELD ELECTROSTATICS */
 386             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 387             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 388
 389             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm256_and_ps(velec,cutoff_mask);
 393             velecsum         = _mm256_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_ps(fscal,dx20);
 401             ty               = _mm256_mul_ps(fscal,dy20);
 402             tz               = _mm256_mul_ps(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm256_add_ps(fix2,tx);
 406             fiy2             = _mm256_add_ps(fiy2,ty);
 407             fiz2             = _mm256_add_ps(fiz2,tz);
 408
 409             fjx0             = _mm256_add_ps(fjx0,tx);
 410             fjy0             = _mm256_add_ps(fjy0,ty);
 411             fjz0             = _mm256_add_ps(fjz0,tz);
 412
 413             }
 414
 415             fjptrA             = f+j_coord_offsetA;
 416             fjptrB             = f+j_coord_offsetB;
 417             fjptrC             = f+j_coord_offsetC;
 418             fjptrD             = f+j_coord_offsetD;
 419             fjptrE             = f+j_coord_offsetE;
 420             fjptrF             = f+j_coord_offsetF;
 421             fjptrG             = f+j_coord_offsetG;
 422             fjptrH             = f+j_coord_offsetH;
 423
 424             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 425
 426             /* Inner loop uses 145 flops */
 427         }
 428
 429         if(jidx<j_index_end)
 430         {
 431
 432             /* Get j neighbor index, and coordinate index */
 433             jnrlistA         = jjnr[jidx];
 434             jnrlistB         = jjnr[jidx+1];
 435             jnrlistC         = jjnr[jidx+2];
 436             jnrlistD         = jjnr[jidx+3];
 437             jnrlistE         = jjnr[jidx+4];
 438             jnrlistF         = jjnr[jidx+5];
 439             jnrlistG         = jjnr[jidx+6];
 440             jnrlistH         = jjnr[jidx+7];
 441             /* Sign of each element will be negative for non-real atoms.
 442              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 443              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 444              */
 445             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 446                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 447
 448             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 449             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 450             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 451             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 452             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 453             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 454             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 455             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458             j_coord_offsetC  = DIM*jnrC;
 459             j_coord_offsetD  = DIM*jnrD;
 460             j_coord_offsetE  = DIM*jnrE;
 461             j_coord_offsetF  = DIM*jnrF;
 462             j_coord_offsetG  = DIM*jnrG;
 463             j_coord_offsetH  = DIM*jnrH;
 464
 465             /* load j atom coordinates */
 466             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 468                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 469                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 470                                                  &jx0,&jy0,&jz0);
 471
 472             /* Calculate displacement vector */
 473             dx00             = _mm256_sub_ps(ix0,jx0);
 474             dy00             = _mm256_sub_ps(iy0,jy0);
 475             dz00             = _mm256_sub_ps(iz0,jz0);
 476             dx10             = _mm256_sub_ps(ix1,jx0);
 477             dy10             = _mm256_sub_ps(iy1,jy0);
 478             dz10             = _mm256_sub_ps(iz1,jz0);
 479             dx20             = _mm256_sub_ps(ix2,jx0);
 480             dy20             = _mm256_sub_ps(iy2,jy0);
 481             dz20             = _mm256_sub_ps(iz2,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 485             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 486             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 487
 488             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 489             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 490             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 491
 492             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 493             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 494             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 498                                                                  charge+jnrC+0,charge+jnrD+0,
 499                                                                  charge+jnrE+0,charge+jnrF+0,
 500                                                                  charge+jnrG+0,charge+jnrH+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505             vdwjidx0E        = 2*vdwtype[jnrE+0];
 506             vdwjidx0F        = 2*vdwtype[jnrF+0];
 507             vdwjidx0G        = 2*vdwtype[jnrG+0];
 508             vdwjidx0H        = 2*vdwtype[jnrH+0];
 509
 510             fjx0             = _mm256_setzero_ps();
 511             fjy0             = _mm256_setzero_ps();
 512             fjz0             = _mm256_setzero_ps();
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 519             {
 520
 521             r00              = _mm256_mul_ps(rsq00,rinv00);
 522             r00              = _mm256_andnot_ps(dummy_mask,r00);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq00             = _mm256_mul_ps(iq0,jq0);
 526             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 527                                             vdwioffsetptr0+vdwjidx0B,
 528                                             vdwioffsetptr0+vdwjidx0C,
 529                                             vdwioffsetptr0+vdwjidx0D,
 530                                             vdwioffsetptr0+vdwjidx0E,
 531                                             vdwioffsetptr0+vdwjidx0F,
 532                                             vdwioffsetptr0+vdwjidx0G,
 533                                             vdwioffsetptr0+vdwjidx0H,
 534                                             &c6_00,&c12_00);
 535
 536             /* REACTION-FIELD ELECTROSTATICS */
 537             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 538             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 539
 540             /* LENNARD-JONES DISPERSION/REPULSION */
 541
 542             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 543             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 544             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 545             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 546             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 547
 548             d                = _mm256_sub_ps(r00,rswitch);
 549             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 550             d2               = _mm256_mul_ps(d,d);
 551             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 552
 553             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 554
 555             /* Evaluate switch function */
 556             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 557             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 558             vvdw             = _mm256_mul_ps(vvdw,sw);
 559             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm256_and_ps(velec,cutoff_mask);
 563             velec            = _mm256_andnot_ps(dummy_mask,velec);
 564             velecsum         = _mm256_add_ps(velecsum,velec);
 565             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 566             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 567             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 568
 569             fscal            = _mm256_add_ps(felec,fvdw);
 570
 571             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 572
 573             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm256_mul_ps(fscal,dx00);
 577             ty               = _mm256_mul_ps(fscal,dy00);
 578             tz               = _mm256_mul_ps(fscal,dz00);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm256_add_ps(fix0,tx);
 582             fiy0             = _mm256_add_ps(fiy0,ty);
 583             fiz0             = _mm256_add_ps(fiz0,tz);
 584
 585             fjx0             = _mm256_add_ps(fjx0,tx);
 586             fjy0             = _mm256_add_ps(fjy0,ty);
 587             fjz0             = _mm256_add_ps(fjz0,tz);
 588
 589             }
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 596             {
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq10             = _mm256_mul_ps(iq1,jq0);
 600
 601             /* REACTION-FIELD ELECTROSTATICS */
 602             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 603             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 604
 605             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm256_and_ps(velec,cutoff_mask);
 609             velec            = _mm256_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm256_add_ps(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 615
 616             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm256_mul_ps(fscal,dx10);
 620             ty               = _mm256_mul_ps(fscal,dy10);
 621             tz               = _mm256_mul_ps(fscal,dz10);
 622
 623             /* Update vectorial force */
 624             fix1             = _mm256_add_ps(fix1,tx);
 625             fiy1             = _mm256_add_ps(fiy1,ty);
 626             fiz1             = _mm256_add_ps(fiz1,tz);
 627
 628             fjx0             = _mm256_add_ps(fjx0,tx);
 629             fjy0             = _mm256_add_ps(fjy0,ty);
 630             fjz0             = _mm256_add_ps(fjz0,tz);
 631
 632             }
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 639             {
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq20             = _mm256_mul_ps(iq2,jq0);
 643
 644             /* REACTION-FIELD ELECTROSTATICS */
 645             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 646             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 647
 648             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm256_and_ps(velec,cutoff_mask);
 652             velec            = _mm256_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm256_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 658
 659             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_ps(fscal,dx20);
 663             ty               = _mm256_mul_ps(fscal,dy20);
 664             tz               = _mm256_mul_ps(fscal,dz20);
 665
 666             /* Update vectorial force */
 667             fix2             = _mm256_add_ps(fix2,tx);
 668             fiy2             = _mm256_add_ps(fiy2,ty);
 669             fiz2             = _mm256_add_ps(fiz2,tz);
 670
 671             fjx0             = _mm256_add_ps(fjx0,tx);
 672             fjy0             = _mm256_add_ps(fjy0,ty);
 673             fjz0             = _mm256_add_ps(fjz0,tz);
 674
 675             }
 676
 677             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 678             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 679             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 680             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 681             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 682             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 683             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 684             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 685
 686             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 687
 688             /* Inner loop uses 146 flops */
 689         }
 690
 691         /* End of innermost loop */
 692
 693         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 694                                                  f+i_coord_offset,fshift+i_shift_offset);
 695
 696         ggid                        = gid[iidx];
 697         /* Update potential energies */
 698         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 699         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 700
 701         /* Increment number of inner iterations */
 702         inneriter                  += j_index_end - j_index_start;
 703
 704         /* Outer loop uses 20 flops */
 705     }
 706
 707     /* Increment number of outer iterations */
 708     outeriter        += nri;
 709
 710     /* Update outer/inner flops */
 711
 712     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
 713 }
 714 /*
 715  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
 716  * Electrostatics interaction: ReactionField
 717  * VdW interaction:            LennardJones
 718  * Geometry:                   Water3-Particle
 719  * Calculate force/pot:        Force
 720  */
 721 void
 722 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
 723                     (t_nblist * gmx_restrict                nlist,
 724                      rvec * gmx_restrict                    xx,
 725                      rvec * gmx_restrict                    ff,
 726                      t_forcerec * gmx_restrict              fr,
 727                      t_mdatoms * gmx_restrict               mdatoms,
 728                      nb_kernel_data_t * gmx_restrict        kernel_data,
 729                      t_nrnb * gmx_restrict                  nrnb)
 730 {
 731     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 732      * just 0 for non-waters.
 733      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 734      * jnr indices corresponding to data put in the four positions in the SIMD register.
 735      */
 736     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 737     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 738     int              jnrA,jnrB,jnrC,jnrD;
 739     int              jnrE,jnrF,jnrG,jnrH;
 740     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 741     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 742     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 743     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 744     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 745     real             rcutoff_scalar;
 746     real             *shiftvec,*fshift,*x,*f;
 747     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 748     real             scratch[4*DIM];
 749     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 750     real *           vdwioffsetptr0;
 751     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 752     real *           vdwioffsetptr1;
 753     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 754     real *           vdwioffsetptr2;
 755     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 756     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 757     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 758     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 759     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 760     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 761     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 762     real             *charge;
 763     int              nvdwtype;
 764     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 765     int              *vdwtype;
 766     real             *vdwparam;
 767     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 768     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 769     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 770     real             rswitch_scalar,d_scalar;
 771     __m256           dummy_mask,cutoff_mask;
 772     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 773     __m256           one     = _mm256_set1_ps(1.0);
 774     __m256           two     = _mm256_set1_ps(2.0);
 775     x                = xx[0];
 776     f                = ff[0];
 777
 778     nri              = nlist->nri;
 779     iinr             = nlist->iinr;
 780     jindex           = nlist->jindex;
 781     jjnr             = nlist->jjnr;
 782     shiftidx         = nlist->shift;
 783     gid              = nlist->gid;
 784     shiftvec         = fr->shift_vec[0];
 785     fshift           = fr->fshift[0];
 786     facel            = _mm256_set1_ps(fr->epsfac);
 787     charge           = mdatoms->chargeA;
 788     krf              = _mm256_set1_ps(fr->ic->k_rf);
 789     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 790     crf              = _mm256_set1_ps(fr->ic->c_rf);
 791     nvdwtype         = fr->ntype;
 792     vdwparam         = fr->nbfp;
 793     vdwtype          = mdatoms->typeA;
 794
 795     /* Setup water-specific parameters */
 796     inr              = nlist->iinr[0];
 797     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 798     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 799     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 800     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 801
 802     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 803     rcutoff_scalar   = fr->rcoulomb;
 804     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 805     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 806
 807     rswitch_scalar   = fr->rvdw_switch;
 808     rswitch          = _mm256_set1_ps(rswitch_scalar);
 809     /* Setup switch parameters */
 810     d_scalar         = rcutoff_scalar-rswitch_scalar;
 811     d                = _mm256_set1_ps(d_scalar);
 812     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 813     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 814     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 815     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 816     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 817     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 818
 819     /* Avoid stupid compiler warnings */
 820     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 821     j_coord_offsetA = 0;
 822     j_coord_offsetB = 0;
 823     j_coord_offsetC = 0;
 824     j_coord_offsetD = 0;
 825     j_coord_offsetE = 0;
 826     j_coord_offsetF = 0;
 827     j_coord_offsetG = 0;
 828     j_coord_offsetH = 0;
 829
 830     outeriter        = 0;
 831     inneriter        = 0;
 832
 833     for(iidx=0;iidx<4*DIM;iidx++)
 834     {
 835         scratch[iidx] = 0.0;
 836     }
 837
 838     /* Start outer loop over neighborlists */
 839     for(iidx=0; iidx<nri; iidx++)
 840     {
 841         /* Load shift vector for this list */
 842         i_shift_offset   = DIM*shiftidx[iidx];
 843
 844         /* Load limits for loop over neighbors */
 845         j_index_start    = jindex[iidx];
 846         j_index_end      = jindex[iidx+1];
 847
 848         /* Get outer coordinate index */
 849         inr              = iinr[iidx];
 850         i_coord_offset   = DIM*inr;
 851
 852         /* Load i particle coords and add shift vector */
 853         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 854                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 855
 856         fix0             = _mm256_setzero_ps();
 857         fiy0             = _mm256_setzero_ps();
 858         fiz0             = _mm256_setzero_ps();
 859         fix1             = _mm256_setzero_ps();
 860         fiy1             = _mm256_setzero_ps();
 861         fiz1             = _mm256_setzero_ps();
 862         fix2             = _mm256_setzero_ps();
 863         fiy2             = _mm256_setzero_ps();
 864         fiz2             = _mm256_setzero_ps();
 865
 866         /* Start inner kernel loop */
 867         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 868         {
 869
 870             /* Get j neighbor index, and coordinate index */
 871             jnrA             = jjnr[jidx];
 872             jnrB             = jjnr[jidx+1];
 873             jnrC             = jjnr[jidx+2];
 874             jnrD             = jjnr[jidx+3];
 875             jnrE             = jjnr[jidx+4];
 876             jnrF             = jjnr[jidx+5];
 877             jnrG             = jjnr[jidx+6];
 878             jnrH             = jjnr[jidx+7];
 879             j_coord_offsetA  = DIM*jnrA;
 880             j_coord_offsetB  = DIM*jnrB;
 881             j_coord_offsetC  = DIM*jnrC;
 882             j_coord_offsetD  = DIM*jnrD;
 883             j_coord_offsetE  = DIM*jnrE;
 884             j_coord_offsetF  = DIM*jnrF;
 885             j_coord_offsetG  = DIM*jnrG;
 886             j_coord_offsetH  = DIM*jnrH;
 887
 888             /* load j atom coordinates */
 889             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 890                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 891                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 892                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 893                                                  &jx0,&jy0,&jz0);
 894
 895             /* Calculate displacement vector */
 896             dx00             = _mm256_sub_ps(ix0,jx0);
 897             dy00             = _mm256_sub_ps(iy0,jy0);
 898             dz00             = _mm256_sub_ps(iz0,jz0);
 899             dx10             = _mm256_sub_ps(ix1,jx0);
 900             dy10             = _mm256_sub_ps(iy1,jy0);
 901             dz10             = _mm256_sub_ps(iz1,jz0);
 902             dx20             = _mm256_sub_ps(ix2,jx0);
 903             dy20             = _mm256_sub_ps(iy2,jy0);
 904             dz20             = _mm256_sub_ps(iz2,jz0);
 905
 906             /* Calculate squared distance and things based on it */
 907             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 908             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 909             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 910
 911             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 912             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 913             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 914
 915             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 916             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 917             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 918
 919             /* Load parameters for j particles */
 920             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 921                                                                  charge+jnrC+0,charge+jnrD+0,
 922                                                                  charge+jnrE+0,charge+jnrF+0,
 923                                                                  charge+jnrG+0,charge+jnrH+0);
 924             vdwjidx0A        = 2*vdwtype[jnrA+0];
 925             vdwjidx0B        = 2*vdwtype[jnrB+0];
 926             vdwjidx0C        = 2*vdwtype[jnrC+0];
 927             vdwjidx0D        = 2*vdwtype[jnrD+0];
 928             vdwjidx0E        = 2*vdwtype[jnrE+0];
 929             vdwjidx0F        = 2*vdwtype[jnrF+0];
 930             vdwjidx0G        = 2*vdwtype[jnrG+0];
 931             vdwjidx0H        = 2*vdwtype[jnrH+0];
 932
 933             fjx0             = _mm256_setzero_ps();
 934             fjy0             = _mm256_setzero_ps();
 935             fjz0             = _mm256_setzero_ps();
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 942             {
 943
 944             r00              = _mm256_mul_ps(rsq00,rinv00);
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq00             = _mm256_mul_ps(iq0,jq0);
 948             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 949                                             vdwioffsetptr0+vdwjidx0B,
 950                                             vdwioffsetptr0+vdwjidx0C,
 951                                             vdwioffsetptr0+vdwjidx0D,
 952                                             vdwioffsetptr0+vdwjidx0E,
 953                                             vdwioffsetptr0+vdwjidx0F,
 954                                             vdwioffsetptr0+vdwjidx0G,
 955                                             vdwioffsetptr0+vdwjidx0H,
 956                                             &c6_00,&c12_00);
 957
 958             /* REACTION-FIELD ELECTROSTATICS */
 959             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 960
 961             /* LENNARD-JONES DISPERSION/REPULSION */
 962
 963             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 964             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 965             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 966             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 967             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 968
 969             d                = _mm256_sub_ps(r00,rswitch);
 970             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 971             d2               = _mm256_mul_ps(d,d);
 972             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 973
 974             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 975
 976             /* Evaluate switch function */
 977             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 978             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 979             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 980
 981             fscal            = _mm256_add_ps(felec,fvdw);
 982
 983             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 984
 985             /* Calculate temporary vectorial force */
 986             tx               = _mm256_mul_ps(fscal,dx00);
 987             ty               = _mm256_mul_ps(fscal,dy00);
 988             tz               = _mm256_mul_ps(fscal,dz00);
 989
 990             /* Update vectorial force */
 991             fix0             = _mm256_add_ps(fix0,tx);
 992             fiy0             = _mm256_add_ps(fiy0,ty);
 993             fiz0             = _mm256_add_ps(fiz0,tz);
 994
 995             fjx0             = _mm256_add_ps(fjx0,tx);
 996             fjy0             = _mm256_add_ps(fjy0,ty);
 997             fjz0             = _mm256_add_ps(fjz0,tz);
 998
 999             }
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1006             {
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq10             = _mm256_mul_ps(iq1,jq0);
1010
1011             /* REACTION-FIELD ELECTROSTATICS */
1012             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1013
1014             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm256_mul_ps(fscal,dx10);
1022             ty               = _mm256_mul_ps(fscal,dy10);
1023             tz               = _mm256_mul_ps(fscal,dz10);
1024
1025             /* Update vectorial force */
1026             fix1             = _mm256_add_ps(fix1,tx);
1027             fiy1             = _mm256_add_ps(fiy1,ty);
1028             fiz1             = _mm256_add_ps(fiz1,tz);
1029
1030             fjx0             = _mm256_add_ps(fjx0,tx);
1031             fjy0             = _mm256_add_ps(fjy0,ty);
1032             fjz0             = _mm256_add_ps(fjz0,tz);
1033
1034             }
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1041             {
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq20             = _mm256_mul_ps(iq2,jq0);
1045
1046             /* REACTION-FIELD ELECTROSTATICS */
1047             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1048
1049             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1050
1051             fscal            = felec;
1052
1053             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1054
1055             /* Calculate temporary vectorial force */
1056             tx               = _mm256_mul_ps(fscal,dx20);
1057             ty               = _mm256_mul_ps(fscal,dy20);
1058             tz               = _mm256_mul_ps(fscal,dz20);
1059
1060             /* Update vectorial force */
1061             fix2             = _mm256_add_ps(fix2,tx);
1062             fiy2             = _mm256_add_ps(fiy2,ty);
1063             fiz2             = _mm256_add_ps(fiz2,tz);
1064
1065             fjx0             = _mm256_add_ps(fjx0,tx);
1066             fjy0             = _mm256_add_ps(fjy0,ty);
1067             fjz0             = _mm256_add_ps(fjz0,tz);
1068
1069             }
1070
1071             fjptrA             = f+j_coord_offsetA;
1072             fjptrB             = f+j_coord_offsetB;
1073             fjptrC             = f+j_coord_offsetC;
1074             fjptrD             = f+j_coord_offsetD;
1075             fjptrE             = f+j_coord_offsetE;
1076             fjptrF             = f+j_coord_offsetF;
1077             fjptrG             = f+j_coord_offsetG;
1078             fjptrH             = f+j_coord_offsetH;
1079
1080             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1081
1082             /* Inner loop uses 124 flops */
1083         }
1084
1085         if(jidx<j_index_end)
1086         {
1087
1088             /* Get j neighbor index, and coordinate index */
1089             jnrlistA         = jjnr[jidx];
1090             jnrlistB         = jjnr[jidx+1];
1091             jnrlistC         = jjnr[jidx+2];
1092             jnrlistD         = jjnr[jidx+3];
1093             jnrlistE         = jjnr[jidx+4];
1094             jnrlistF         = jjnr[jidx+5];
1095             jnrlistG         = jjnr[jidx+6];
1096             jnrlistH         = jjnr[jidx+7];
1097             /* Sign of each element will be negative for non-real atoms.
1098              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1099              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1100              */
1101             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1102                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1103
1104             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1105             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1106             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1107             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1108             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1109             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1110             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1111             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1112             j_coord_offsetA  = DIM*jnrA;
1113             j_coord_offsetB  = DIM*jnrB;
1114             j_coord_offsetC  = DIM*jnrC;
1115             j_coord_offsetD  = DIM*jnrD;
1116             j_coord_offsetE  = DIM*jnrE;
1117             j_coord_offsetF  = DIM*jnrF;
1118             j_coord_offsetG  = DIM*jnrG;
1119             j_coord_offsetH  = DIM*jnrH;
1120
1121             /* load j atom coordinates */
1122             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1123                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1124                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1125                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1126                                                  &jx0,&jy0,&jz0);
1127
1128             /* Calculate displacement vector */
1129             dx00             = _mm256_sub_ps(ix0,jx0);
1130             dy00             = _mm256_sub_ps(iy0,jy0);
1131             dz00             = _mm256_sub_ps(iz0,jz0);
1132             dx10             = _mm256_sub_ps(ix1,jx0);
1133             dy10             = _mm256_sub_ps(iy1,jy0);
1134             dz10             = _mm256_sub_ps(iz1,jz0);
1135             dx20             = _mm256_sub_ps(ix2,jx0);
1136             dy20             = _mm256_sub_ps(iy2,jy0);
1137             dz20             = _mm256_sub_ps(iz2,jz0);
1138
1139             /* Calculate squared distance and things based on it */
1140             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1141             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1142             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1143
1144             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1145             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1146             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1147
1148             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1149             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1150             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1151
1152             /* Load parameters for j particles */
1153             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1154                                                                  charge+jnrC+0,charge+jnrD+0,
1155                                                                  charge+jnrE+0,charge+jnrF+0,
1156                                                                  charge+jnrG+0,charge+jnrH+0);
1157             vdwjidx0A        = 2*vdwtype[jnrA+0];
1158             vdwjidx0B        = 2*vdwtype[jnrB+0];
1159             vdwjidx0C        = 2*vdwtype[jnrC+0];
1160             vdwjidx0D        = 2*vdwtype[jnrD+0];
1161             vdwjidx0E        = 2*vdwtype[jnrE+0];
1162             vdwjidx0F        = 2*vdwtype[jnrF+0];
1163             vdwjidx0G        = 2*vdwtype[jnrG+0];
1164             vdwjidx0H        = 2*vdwtype[jnrH+0];
1165
1166             fjx0             = _mm256_setzero_ps();
1167             fjy0             = _mm256_setzero_ps();
1168             fjz0             = _mm256_setzero_ps();
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1175             {
1176
1177             r00              = _mm256_mul_ps(rsq00,rinv00);
1178             r00              = _mm256_andnot_ps(dummy_mask,r00);
1179
1180             /* Compute parameters for interactions between i and j atoms */
1181             qq00             = _mm256_mul_ps(iq0,jq0);
1182             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1183                                             vdwioffsetptr0+vdwjidx0B,
1184                                             vdwioffsetptr0+vdwjidx0C,
1185                                             vdwioffsetptr0+vdwjidx0D,
1186                                             vdwioffsetptr0+vdwjidx0E,
1187                                             vdwioffsetptr0+vdwjidx0F,
1188                                             vdwioffsetptr0+vdwjidx0G,
1189                                             vdwioffsetptr0+vdwjidx0H,
1190                                             &c6_00,&c12_00);
1191
1192             /* REACTION-FIELD ELECTROSTATICS */
1193             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1194
1195             /* LENNARD-JONES DISPERSION/REPULSION */
1196
1197             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1198             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1199             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1200             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1201             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1202
1203             d                = _mm256_sub_ps(r00,rswitch);
1204             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1205             d2               = _mm256_mul_ps(d,d);
1206             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1207
1208             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1209
1210             /* Evaluate switch function */
1211             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1212             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1213             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1214
1215             fscal            = _mm256_add_ps(felec,fvdw);
1216
1217             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1218
1219             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm256_mul_ps(fscal,dx00);
1223             ty               = _mm256_mul_ps(fscal,dy00);
1224             tz               = _mm256_mul_ps(fscal,dz00);
1225
1226             /* Update vectorial force */
1227             fix0             = _mm256_add_ps(fix0,tx);
1228             fiy0             = _mm256_add_ps(fiy0,ty);
1229             fiz0             = _mm256_add_ps(fiz0,tz);
1230
1231             fjx0             = _mm256_add_ps(fjx0,tx);
1232             fjy0             = _mm256_add_ps(fjy0,ty);
1233             fjz0             = _mm256_add_ps(fjz0,tz);
1234
1235             }
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1242             {
1243
1244             /* Compute parameters for interactions between i and j atoms */
1245             qq10             = _mm256_mul_ps(iq1,jq0);
1246
1247             /* REACTION-FIELD ELECTROSTATICS */
1248             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1249
1250             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1251
1252             fscal            = felec;
1253
1254             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1255
1256             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1257
1258             /* Calculate temporary vectorial force */
1259             tx               = _mm256_mul_ps(fscal,dx10);
1260             ty               = _mm256_mul_ps(fscal,dy10);
1261             tz               = _mm256_mul_ps(fscal,dz10);
1262
1263             /* Update vectorial force */
1264             fix1             = _mm256_add_ps(fix1,tx);
1265             fiy1             = _mm256_add_ps(fiy1,ty);
1266             fiz1             = _mm256_add_ps(fiz1,tz);
1267
1268             fjx0             = _mm256_add_ps(fjx0,tx);
1269             fjy0             = _mm256_add_ps(fjy0,ty);
1270             fjz0             = _mm256_add_ps(fjz0,tz);
1271
1272             }
1273
1274             /**************************
1275              * CALCULATE INTERACTIONS *
1276              **************************/
1277
1278             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1279             {
1280
1281             /* Compute parameters for interactions between i and j atoms */
1282             qq20             = _mm256_mul_ps(iq2,jq0);
1283
1284             /* REACTION-FIELD ELECTROSTATICS */
1285             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1286
1287             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1288
1289             fscal            = felec;
1290
1291             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1292
1293             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm256_mul_ps(fscal,dx20);
1297             ty               = _mm256_mul_ps(fscal,dy20);
1298             tz               = _mm256_mul_ps(fscal,dz20);
1299
1300             /* Update vectorial force */
1301             fix2             = _mm256_add_ps(fix2,tx);
1302             fiy2             = _mm256_add_ps(fiy2,ty);
1303             fiz2             = _mm256_add_ps(fiz2,tz);
1304
1305             fjx0             = _mm256_add_ps(fjx0,tx);
1306             fjy0             = _mm256_add_ps(fjy0,ty);
1307             fjz0             = _mm256_add_ps(fjz0,tz);
1308
1309             }
1310
1311             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1312             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1313             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1314             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1315             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1316             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1317             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1318             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1319
1320             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1321
1322             /* Inner loop uses 125 flops */
1323         }
1324
1325         /* End of innermost loop */
1326
1327         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1328                                                  f+i_coord_offset,fshift+i_shift_offset);
1329
1330         /* Increment number of inner iterations */
1331         inneriter                  += j_index_end - j_index_start;
1332
1333         /* Outer loop uses 18 flops */
1334     }
1335
1336     /* Increment number of outer iterations */
1337     outeriter        += nri;
1338
1339     /* Update outer/inner flops */
1340
1341     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);
1342 }