src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
  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 AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  84     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  91     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  92     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  93     real             rswitch_scalar,d_scalar;
  94     __m256d          dummy_mask,cutoff_mask;
  95     __m128           tmpmask0,tmpmask1;
  96     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  97     __m256d          one     = _mm256_set1_pd(1.0);
  98     __m256d          two     = _mm256_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm256_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm256_set1_pd(fr->ic->k_rf);
 113     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 114     crf              = _mm256_set1_pd(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 122     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 123     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 124     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->rvdw_switch;
 132     rswitch          = _mm256_set1_pd(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = _mm256_set1_pd(d_scalar);
 136     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm256_setzero_pd();
 177         fiy0             = _mm256_setzero_pd();
 178         fiz0             = _mm256_setzero_pd();
 179         fix1             = _mm256_setzero_pd();
 180         fiy1             = _mm256_setzero_pd();
 181         fiz1             = _mm256_setzero_pd();
 182         fix2             = _mm256_setzero_pd();
 183         fiy2             = _mm256_setzero_pd();
 184         fiz2             = _mm256_setzero_pd();
 185         fix3             = _mm256_setzero_pd();
 186         fiy3             = _mm256_setzero_pd();
 187         fiz3             = _mm256_setzero_pd();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm256_setzero_pd();
 191         vvdwsum          = _mm256_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 210                                                  &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm256_sub_pd(ix0,jx0);
 214             dy00             = _mm256_sub_pd(iy0,jy0);
 215             dz00             = _mm256_sub_pd(iz0,jz0);
 216             dx10             = _mm256_sub_pd(ix1,jx0);
 217             dy10             = _mm256_sub_pd(iy1,jy0);
 218             dz10             = _mm256_sub_pd(iz1,jz0);
 219             dx20             = _mm256_sub_pd(ix2,jx0);
 220             dy20             = _mm256_sub_pd(iy2,jy0);
 221             dz20             = _mm256_sub_pd(iz2,jz0);
 222             dx30             = _mm256_sub_pd(ix3,jx0);
 223             dy30             = _mm256_sub_pd(iy3,jy0);
 224             dz30             = _mm256_sub_pd(iz3,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 228             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 229             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 230             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 231
 232             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 233             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 234             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 235             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 236
 237             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 238             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 239             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 240             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249
 250             fjx0             = _mm256_setzero_pd();
 251             fjy0             = _mm256_setzero_pd();
 252             fjz0             = _mm256_setzero_pd();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             r00              = _mm256_mul_pd(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 265                                             vdwioffsetptr0+vdwjidx0B,
 266                                             vdwioffsetptr0+vdwjidx0C,
 267                                             vdwioffsetptr0+vdwjidx0D,
 268                                             &c6_00,&c12_00);
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 274             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 275             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 276             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 277
 278             d                = _mm256_sub_pd(r00,rswitch);
 279             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 280             d2               = _mm256_mul_pd(d,d);
 281             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 282
 283             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 284
 285             /* Evaluate switch function */
 286             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 287             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 288             vvdw             = _mm256_mul_pd(vvdw,sw);
 289             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 293             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 294
 295             fscal            = fvdw;
 296
 297             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_pd(fscal,dx00);
 301             ty               = _mm256_mul_pd(fscal,dy00);
 302             tz               = _mm256_mul_pd(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm256_add_pd(fix0,tx);
 306             fiy0             = _mm256_add_pd(fiy0,ty);
 307             fiz0             = _mm256_add_pd(fiz0,tz);
 308
 309             fjx0             = _mm256_add_pd(fjx0,tx);
 310             fjy0             = _mm256_add_pd(fjy0,ty);
 311             fjz0             = _mm256_add_pd(fjz0,tz);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 320             {
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm256_mul_pd(iq1,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 327             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 328
 329             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velec            = _mm256_and_pd(velec,cutoff_mask);
 333             velecsum         = _mm256_add_pd(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_pd(fscal,dx10);
 341             ty               = _mm256_mul_pd(fscal,dy10);
 342             tz               = _mm256_mul_pd(fscal,dz10);
 343
 344             /* Update vectorial force */
 345             fix1             = _mm256_add_pd(fix1,tx);
 346             fiy1             = _mm256_add_pd(fiy1,ty);
 347             fiz1             = _mm256_add_pd(fiz1,tz);
 348
 349             fjx0             = _mm256_add_pd(fjx0,tx);
 350             fjy0             = _mm256_add_pd(fjy0,ty);
 351             fjz0             = _mm256_add_pd(fjz0,tz);
 352
 353             }
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 360             {
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq20             = _mm256_mul_pd(iq2,jq0);
 364
 365             /* REACTION-FIELD ELECTROSTATICS */
 366             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 367             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 368
 369             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velec            = _mm256_and_pd(velec,cutoff_mask);
 373             velecsum         = _mm256_add_pd(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm256_mul_pd(fscal,dx20);
 381             ty               = _mm256_mul_pd(fscal,dy20);
 382             tz               = _mm256_mul_pd(fscal,dz20);
 383
 384             /* Update vectorial force */
 385             fix2             = _mm256_add_pd(fix2,tx);
 386             fiy2             = _mm256_add_pd(fiy2,ty);
 387             fiz2             = _mm256_add_pd(fiz2,tz);
 388
 389             fjx0             = _mm256_add_pd(fjx0,tx);
 390             fjy0             = _mm256_add_pd(fjy0,ty);
 391             fjz0             = _mm256_add_pd(fjz0,tz);
 392
 393             }
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 400             {
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq30             = _mm256_mul_pd(iq3,jq0);
 404
 405             /* REACTION-FIELD ELECTROSTATICS */
 406             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 407             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 408
 409             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm256_and_pd(velec,cutoff_mask);
 413             velecsum         = _mm256_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_pd(fscal,dx30);
 421             ty               = _mm256_mul_pd(fscal,dy30);
 422             tz               = _mm256_mul_pd(fscal,dz30);
 423
 424             /* Update vectorial force */
 425             fix3             = _mm256_add_pd(fix3,tx);
 426             fiy3             = _mm256_add_pd(fiy3,ty);
 427             fiz3             = _mm256_add_pd(fiz3,tz);
 428
 429             fjx0             = _mm256_add_pd(fjx0,tx);
 430             fjy0             = _mm256_add_pd(fjy0,ty);
 431             fjz0             = _mm256_add_pd(fjz0,tz);
 432
 433             }
 434
 435             fjptrA             = f+j_coord_offsetA;
 436             fjptrB             = f+j_coord_offsetB;
 437             fjptrC             = f+j_coord_offsetC;
 438             fjptrD             = f+j_coord_offsetD;
 439
 440             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 441
 442             /* Inner loop uses 170 flops */
 443         }
 444
 445         if(jidx<j_index_end)
 446         {
 447
 448             /* Get j neighbor index, and coordinate index */
 449             jnrlistA         = jjnr[jidx];
 450             jnrlistB         = jjnr[jidx+1];
 451             jnrlistC         = jjnr[jidx+2];
 452             jnrlistD         = jjnr[jidx+3];
 453             /* Sign of each element will be negative for non-real atoms.
 454              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 455              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 456              */
 457             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 458
 459             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 460             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 461             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 462
 463             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 464             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 465             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 466             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 467             j_coord_offsetA  = DIM*jnrA;
 468             j_coord_offsetB  = DIM*jnrB;
 469             j_coord_offsetC  = DIM*jnrC;
 470             j_coord_offsetD  = DIM*jnrD;
 471
 472             /* load j atom coordinates */
 473             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 475                                                  &jx0,&jy0,&jz0);
 476
 477             /* Calculate displacement vector */
 478             dx00             = _mm256_sub_pd(ix0,jx0);
 479             dy00             = _mm256_sub_pd(iy0,jy0);
 480             dz00             = _mm256_sub_pd(iz0,jz0);
 481             dx10             = _mm256_sub_pd(ix1,jx0);
 482             dy10             = _mm256_sub_pd(iy1,jy0);
 483             dz10             = _mm256_sub_pd(iz1,jz0);
 484             dx20             = _mm256_sub_pd(ix2,jx0);
 485             dy20             = _mm256_sub_pd(iy2,jy0);
 486             dz20             = _mm256_sub_pd(iz2,jz0);
 487             dx30             = _mm256_sub_pd(ix3,jx0);
 488             dy30             = _mm256_sub_pd(iy3,jy0);
 489             dz30             = _mm256_sub_pd(iz3,jz0);
 490
 491             /* Calculate squared distance and things based on it */
 492             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 493             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 494             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 495             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 496
 497             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 498             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 499             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 500             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 501
 502             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 503             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 504             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 505             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 506
 507             /* Load parameters for j particles */
 508             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 509                                                                  charge+jnrC+0,charge+jnrD+0);
 510             vdwjidx0A        = 2*vdwtype[jnrA+0];
 511             vdwjidx0B        = 2*vdwtype[jnrB+0];
 512             vdwjidx0C        = 2*vdwtype[jnrC+0];
 513             vdwjidx0D        = 2*vdwtype[jnrD+0];
 514
 515             fjx0             = _mm256_setzero_pd();
 516             fjy0             = _mm256_setzero_pd();
 517             fjz0             = _mm256_setzero_pd();
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 524             {
 525
 526             r00              = _mm256_mul_pd(rsq00,rinv00);
 527             r00              = _mm256_andnot_pd(dummy_mask,r00);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 531                                             vdwioffsetptr0+vdwjidx0B,
 532                                             vdwioffsetptr0+vdwjidx0C,
 533                                             vdwioffsetptr0+vdwjidx0D,
 534                                             &c6_00,&c12_00);
 535
 536             /* LENNARD-JONES DISPERSION/REPULSION */
 537
 538             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 539             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 540             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 541             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 542             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 543
 544             d                = _mm256_sub_pd(r00,rswitch);
 545             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 546             d2               = _mm256_mul_pd(d,d);
 547             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 548
 549             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 550
 551             /* Evaluate switch function */
 552             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 553             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 554             vvdw             = _mm256_mul_pd(vvdw,sw);
 555             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 559             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 560             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 561
 562             fscal            = fvdw;
 563
 564             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 565
 566             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_pd(fscal,dx00);
 570             ty               = _mm256_mul_pd(fscal,dy00);
 571             tz               = _mm256_mul_pd(fscal,dz00);
 572
 573             /* Update vectorial force */
 574             fix0             = _mm256_add_pd(fix0,tx);
 575             fiy0             = _mm256_add_pd(fiy0,ty);
 576             fiz0             = _mm256_add_pd(fiz0,tz);
 577
 578             fjx0             = _mm256_add_pd(fjx0,tx);
 579             fjy0             = _mm256_add_pd(fjy0,ty);
 580             fjz0             = _mm256_add_pd(fjz0,tz);
 581
 582             }
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 589             {
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq10             = _mm256_mul_pd(iq1,jq0);
 593
 594             /* REACTION-FIELD ELECTROSTATICS */
 595             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 596             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 597
 598             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm256_and_pd(velec,cutoff_mask);
 602             velec            = _mm256_andnot_pd(dummy_mask,velec);
 603             velecsum         = _mm256_add_pd(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 608
 609             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_pd(fscal,dx10);
 613             ty               = _mm256_mul_pd(fscal,dy10);
 614             tz               = _mm256_mul_pd(fscal,dz10);
 615
 616             /* Update vectorial force */
 617             fix1             = _mm256_add_pd(fix1,tx);
 618             fiy1             = _mm256_add_pd(fiy1,ty);
 619             fiz1             = _mm256_add_pd(fiz1,tz);
 620
 621             fjx0             = _mm256_add_pd(fjx0,tx);
 622             fjy0             = _mm256_add_pd(fjy0,ty);
 623             fjz0             = _mm256_add_pd(fjz0,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 632             {
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq20             = _mm256_mul_pd(iq2,jq0);
 636
 637             /* REACTION-FIELD ELECTROSTATICS */
 638             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 639             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 640
 641             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm256_and_pd(velec,cutoff_mask);
 645             velec            = _mm256_andnot_pd(dummy_mask,velec);
 646             velecsum         = _mm256_add_pd(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 651
 652             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm256_mul_pd(fscal,dx20);
 656             ty               = _mm256_mul_pd(fscal,dy20);
 657             tz               = _mm256_mul_pd(fscal,dz20);
 658
 659             /* Update vectorial force */
 660             fix2             = _mm256_add_pd(fix2,tx);
 661             fiy2             = _mm256_add_pd(fiy2,ty);
 662             fiz2             = _mm256_add_pd(fiz2,tz);
 663
 664             fjx0             = _mm256_add_pd(fjx0,tx);
 665             fjy0             = _mm256_add_pd(fjy0,ty);
 666             fjz0             = _mm256_add_pd(fjz0,tz);
 667
 668             }
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 675             {
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq30             = _mm256_mul_pd(iq3,jq0);
 679
 680             /* REACTION-FIELD ELECTROSTATICS */
 681             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 682             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 683
 684             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velec            = _mm256_and_pd(velec,cutoff_mask);
 688             velec            = _mm256_andnot_pd(dummy_mask,velec);
 689             velecsum         = _mm256_add_pd(velecsum,velec);
 690
 691             fscal            = felec;
 692
 693             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 694
 695             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm256_mul_pd(fscal,dx30);
 699             ty               = _mm256_mul_pd(fscal,dy30);
 700             tz               = _mm256_mul_pd(fscal,dz30);
 701
 702             /* Update vectorial force */
 703             fix3             = _mm256_add_pd(fix3,tx);
 704             fiy3             = _mm256_add_pd(fiy3,ty);
 705             fiz3             = _mm256_add_pd(fiz3,tz);
 706
 707             fjx0             = _mm256_add_pd(fjx0,tx);
 708             fjy0             = _mm256_add_pd(fjy0,ty);
 709             fjz0             = _mm256_add_pd(fjz0,tz);
 710
 711             }
 712
 713             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 714             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 715             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 716             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 717
 718             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 719
 720             /* Inner loop uses 171 flops */
 721         }
 722
 723         /* End of innermost loop */
 724
 725         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 726                                                  f+i_coord_offset,fshift+i_shift_offset);
 727
 728         ggid                        = gid[iidx];
 729         /* Update potential energies */
 730         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 731         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 732
 733         /* Increment number of inner iterations */
 734         inneriter                  += j_index_end - j_index_start;
 735
 736         /* Outer loop uses 26 flops */
 737     }
 738
 739     /* Increment number of outer iterations */
 740     outeriter        += nri;
 741
 742     /* Update outer/inner flops */
 743
 744     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
 745 }
 746 /*
 747  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 748  * Electrostatics interaction: ReactionField
 749  * VdW interaction:            LennardJones
 750  * Geometry:                   Water4-Particle
 751  * Calculate force/pot:        Force
 752  */
 753 void
 754 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 755                     (t_nblist * gmx_restrict                nlist,
 756                      rvec * gmx_restrict                    xx,
 757                      rvec * gmx_restrict                    ff,
 758                      t_forcerec * gmx_restrict              fr,
 759                      t_mdatoms * gmx_restrict               mdatoms,
 760                      nb_kernel_data_t * gmx_restrict        kernel_data,
 761                      t_nrnb * gmx_restrict                  nrnb)
 762 {
 763     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 764      * just 0 for non-waters.
 765      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 766      * jnr indices corresponding to data put in the four positions in the SIMD register.
 767      */
 768     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 769     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 770     int              jnrA,jnrB,jnrC,jnrD;
 771     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 772     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 773     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 774     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 775     real             rcutoff_scalar;
 776     real             *shiftvec,*fshift,*x,*f;
 777     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 778     real             scratch[4*DIM];
 779     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 780     real *           vdwioffsetptr0;
 781     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 782     real *           vdwioffsetptr1;
 783     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 784     real *           vdwioffsetptr2;
 785     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 786     real *           vdwioffsetptr3;
 787     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 788     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 789     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 790     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 791     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 792     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 793     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 794     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 795     real             *charge;
 796     int              nvdwtype;
 797     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 798     int              *vdwtype;
 799     real             *vdwparam;
 800     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 801     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 802     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 803     real             rswitch_scalar,d_scalar;
 804     __m256d          dummy_mask,cutoff_mask;
 805     __m128           tmpmask0,tmpmask1;
 806     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 807     __m256d          one     = _mm256_set1_pd(1.0);
 808     __m256d          two     = _mm256_set1_pd(2.0);
 809     x                = xx[0];
 810     f                = ff[0];
 811
 812     nri              = nlist->nri;
 813     iinr             = nlist->iinr;
 814     jindex           = nlist->jindex;
 815     jjnr             = nlist->jjnr;
 816     shiftidx         = nlist->shift;
 817     gid              = nlist->gid;
 818     shiftvec         = fr->shift_vec[0];
 819     fshift           = fr->fshift[0];
 820     facel            = _mm256_set1_pd(fr->epsfac);
 821     charge           = mdatoms->chargeA;
 822     krf              = _mm256_set1_pd(fr->ic->k_rf);
 823     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 824     crf              = _mm256_set1_pd(fr->ic->c_rf);
 825     nvdwtype         = fr->ntype;
 826     vdwparam         = fr->nbfp;
 827     vdwtype          = mdatoms->typeA;
 828
 829     /* Setup water-specific parameters */
 830     inr              = nlist->iinr[0];
 831     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 832     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 833     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 834     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 835
 836     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 837     rcutoff_scalar   = fr->rcoulomb;
 838     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 839     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 840
 841     rswitch_scalar   = fr->rvdw_switch;
 842     rswitch          = _mm256_set1_pd(rswitch_scalar);
 843     /* Setup switch parameters */
 844     d_scalar         = rcutoff_scalar-rswitch_scalar;
 845     d                = _mm256_set1_pd(d_scalar);
 846     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 847     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 848     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 849     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 850     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 851     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 852
 853     /* Avoid stupid compiler warnings */
 854     jnrA = jnrB = jnrC = jnrD = 0;
 855     j_coord_offsetA = 0;
 856     j_coord_offsetB = 0;
 857     j_coord_offsetC = 0;
 858     j_coord_offsetD = 0;
 859
 860     outeriter        = 0;
 861     inneriter        = 0;
 862
 863     for(iidx=0;iidx<4*DIM;iidx++)
 864     {
 865         scratch[iidx] = 0.0;
 866     }
 867
 868     /* Start outer loop over neighborlists */
 869     for(iidx=0; iidx<nri; iidx++)
 870     {
 871         /* Load shift vector for this list */
 872         i_shift_offset   = DIM*shiftidx[iidx];
 873
 874         /* Load limits for loop over neighbors */
 875         j_index_start    = jindex[iidx];
 876         j_index_end      = jindex[iidx+1];
 877
 878         /* Get outer coordinate index */
 879         inr              = iinr[iidx];
 880         i_coord_offset   = DIM*inr;
 881
 882         /* Load i particle coords and add shift vector */
 883         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 884                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 885
 886         fix0             = _mm256_setzero_pd();
 887         fiy0             = _mm256_setzero_pd();
 888         fiz0             = _mm256_setzero_pd();
 889         fix1             = _mm256_setzero_pd();
 890         fiy1             = _mm256_setzero_pd();
 891         fiz1             = _mm256_setzero_pd();
 892         fix2             = _mm256_setzero_pd();
 893         fiy2             = _mm256_setzero_pd();
 894         fiz2             = _mm256_setzero_pd();
 895         fix3             = _mm256_setzero_pd();
 896         fiy3             = _mm256_setzero_pd();
 897         fiz3             = _mm256_setzero_pd();
 898
 899         /* Start inner kernel loop */
 900         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 901         {
 902
 903             /* Get j neighbor index, and coordinate index */
 904             jnrA             = jjnr[jidx];
 905             jnrB             = jjnr[jidx+1];
 906             jnrC             = jjnr[jidx+2];
 907             jnrD             = jjnr[jidx+3];
 908             j_coord_offsetA  = DIM*jnrA;
 909             j_coord_offsetB  = DIM*jnrB;
 910             j_coord_offsetC  = DIM*jnrC;
 911             j_coord_offsetD  = DIM*jnrD;
 912
 913             /* load j atom coordinates */
 914             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 915                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 916                                                  &jx0,&jy0,&jz0);
 917
 918             /* Calculate displacement vector */
 919             dx00             = _mm256_sub_pd(ix0,jx0);
 920             dy00             = _mm256_sub_pd(iy0,jy0);
 921             dz00             = _mm256_sub_pd(iz0,jz0);
 922             dx10             = _mm256_sub_pd(ix1,jx0);
 923             dy10             = _mm256_sub_pd(iy1,jy0);
 924             dz10             = _mm256_sub_pd(iz1,jz0);
 925             dx20             = _mm256_sub_pd(ix2,jx0);
 926             dy20             = _mm256_sub_pd(iy2,jy0);
 927             dz20             = _mm256_sub_pd(iz2,jz0);
 928             dx30             = _mm256_sub_pd(ix3,jx0);
 929             dy30             = _mm256_sub_pd(iy3,jy0);
 930             dz30             = _mm256_sub_pd(iz3,jz0);
 931
 932             /* Calculate squared distance and things based on it */
 933             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 934             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 935             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 936             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 937
 938             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 939             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 940             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 941             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 942
 943             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 944             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 945             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 946             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 947
 948             /* Load parameters for j particles */
 949             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 950                                                                  charge+jnrC+0,charge+jnrD+0);
 951             vdwjidx0A        = 2*vdwtype[jnrA+0];
 952             vdwjidx0B        = 2*vdwtype[jnrB+0];
 953             vdwjidx0C        = 2*vdwtype[jnrC+0];
 954             vdwjidx0D        = 2*vdwtype[jnrD+0];
 955
 956             fjx0             = _mm256_setzero_pd();
 957             fjy0             = _mm256_setzero_pd();
 958             fjz0             = _mm256_setzero_pd();
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 965             {
 966
 967             r00              = _mm256_mul_pd(rsq00,rinv00);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 971                                             vdwioffsetptr0+vdwjidx0B,
 972                                             vdwioffsetptr0+vdwjidx0C,
 973                                             vdwioffsetptr0+vdwjidx0D,
 974                                             &c6_00,&c12_00);
 975
 976             /* LENNARD-JONES DISPERSION/REPULSION */
 977
 978             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 979             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 980             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 981             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 982             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 983
 984             d                = _mm256_sub_pd(r00,rswitch);
 985             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 986             d2               = _mm256_mul_pd(d,d);
 987             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 988
 989             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 990
 991             /* Evaluate switch function */
 992             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 993             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 994             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 995
 996             fscal            = fvdw;
 997
 998             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_pd(fscal,dx00);
1002             ty               = _mm256_mul_pd(fscal,dy00);
1003             tz               = _mm256_mul_pd(fscal,dz00);
1004
1005             /* Update vectorial force */
1006             fix0             = _mm256_add_pd(fix0,tx);
1007             fiy0             = _mm256_add_pd(fiy0,ty);
1008             fiz0             = _mm256_add_pd(fiz0,tz);
1009
1010             fjx0             = _mm256_add_pd(fjx0,tx);
1011             fjy0             = _mm256_add_pd(fjy0,ty);
1012             fjz0             = _mm256_add_pd(fjz0,tz);
1013
1014             }
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1021             {
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq10             = _mm256_mul_pd(iq1,jq0);
1025
1026             /* REACTION-FIELD ELECTROSTATICS */
1027             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1028
1029             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm256_mul_pd(fscal,dx10);
1037             ty               = _mm256_mul_pd(fscal,dy10);
1038             tz               = _mm256_mul_pd(fscal,dz10);
1039
1040             /* Update vectorial force */
1041             fix1             = _mm256_add_pd(fix1,tx);
1042             fiy1             = _mm256_add_pd(fiy1,ty);
1043             fiz1             = _mm256_add_pd(fiz1,tz);
1044
1045             fjx0             = _mm256_add_pd(fjx0,tx);
1046             fjy0             = _mm256_add_pd(fjy0,ty);
1047             fjz0             = _mm256_add_pd(fjz0,tz);
1048
1049             }
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1056             {
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq20             = _mm256_mul_pd(iq2,jq0);
1060
1061             /* REACTION-FIELD ELECTROSTATICS */
1062             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1063
1064             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1065
1066             fscal            = felec;
1067
1068             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_pd(fscal,dx20);
1072             ty               = _mm256_mul_pd(fscal,dy20);
1073             tz               = _mm256_mul_pd(fscal,dz20);
1074
1075             /* Update vectorial force */
1076             fix2             = _mm256_add_pd(fix2,tx);
1077             fiy2             = _mm256_add_pd(fiy2,ty);
1078             fiz2             = _mm256_add_pd(fiz2,tz);
1079
1080             fjx0             = _mm256_add_pd(fjx0,tx);
1081             fjy0             = _mm256_add_pd(fjy0,ty);
1082             fjz0             = _mm256_add_pd(fjz0,tz);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1091             {
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq30             = _mm256_mul_pd(iq3,jq0);
1095
1096             /* REACTION-FIELD ELECTROSTATICS */
1097             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1098
1099             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1100
1101             fscal            = felec;
1102
1103             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1104
1105             /* Calculate temporary vectorial force */
1106             tx               = _mm256_mul_pd(fscal,dx30);
1107             ty               = _mm256_mul_pd(fscal,dy30);
1108             tz               = _mm256_mul_pd(fscal,dz30);
1109
1110             /* Update vectorial force */
1111             fix3             = _mm256_add_pd(fix3,tx);
1112             fiy3             = _mm256_add_pd(fiy3,ty);
1113             fiz3             = _mm256_add_pd(fiz3,tz);
1114
1115             fjx0             = _mm256_add_pd(fjx0,tx);
1116             fjy0             = _mm256_add_pd(fjy0,ty);
1117             fjz0             = _mm256_add_pd(fjz0,tz);
1118
1119             }
1120
1121             fjptrA             = f+j_coord_offsetA;
1122             fjptrB             = f+j_coord_offsetB;
1123             fjptrC             = f+j_coord_offsetC;
1124             fjptrD             = f+j_coord_offsetD;
1125
1126             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1127
1128             /* Inner loop uses 149 flops */
1129         }
1130
1131         if(jidx<j_index_end)
1132         {
1133
1134             /* Get j neighbor index, and coordinate index */
1135             jnrlistA         = jjnr[jidx];
1136             jnrlistB         = jjnr[jidx+1];
1137             jnrlistC         = jjnr[jidx+2];
1138             jnrlistD         = jjnr[jidx+3];
1139             /* Sign of each element will be negative for non-real atoms.
1140              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1141              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1142              */
1143             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1144
1145             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1146             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1147             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1148
1149             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1150             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1151             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1152             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1153             j_coord_offsetA  = DIM*jnrA;
1154             j_coord_offsetB  = DIM*jnrB;
1155             j_coord_offsetC  = DIM*jnrC;
1156             j_coord_offsetD  = DIM*jnrD;
1157
1158             /* load j atom coordinates */
1159             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1160                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1161                                                  &jx0,&jy0,&jz0);
1162
1163             /* Calculate displacement vector */
1164             dx00             = _mm256_sub_pd(ix0,jx0);
1165             dy00             = _mm256_sub_pd(iy0,jy0);
1166             dz00             = _mm256_sub_pd(iz0,jz0);
1167             dx10             = _mm256_sub_pd(ix1,jx0);
1168             dy10             = _mm256_sub_pd(iy1,jy0);
1169             dz10             = _mm256_sub_pd(iz1,jz0);
1170             dx20             = _mm256_sub_pd(ix2,jx0);
1171             dy20             = _mm256_sub_pd(iy2,jy0);
1172             dz20             = _mm256_sub_pd(iz2,jz0);
1173             dx30             = _mm256_sub_pd(ix3,jx0);
1174             dy30             = _mm256_sub_pd(iy3,jy0);
1175             dz30             = _mm256_sub_pd(iz3,jz0);
1176
1177             /* Calculate squared distance and things based on it */
1178             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1179             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1180             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1181             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1182
1183             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1184             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1185             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1186             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1187
1188             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1189             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1190             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1191             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1192
1193             /* Load parameters for j particles */
1194             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1195                                                                  charge+jnrC+0,charge+jnrD+0);
1196             vdwjidx0A        = 2*vdwtype[jnrA+0];
1197             vdwjidx0B        = 2*vdwtype[jnrB+0];
1198             vdwjidx0C        = 2*vdwtype[jnrC+0];
1199             vdwjidx0D        = 2*vdwtype[jnrD+0];
1200
1201             fjx0             = _mm256_setzero_pd();
1202             fjy0             = _mm256_setzero_pd();
1203             fjz0             = _mm256_setzero_pd();
1204
1205             /**************************
1206              * CALCULATE INTERACTIONS *
1207              **************************/
1208
1209             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1210             {
1211
1212             r00              = _mm256_mul_pd(rsq00,rinv00);
1213             r00              = _mm256_andnot_pd(dummy_mask,r00);
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1217                                             vdwioffsetptr0+vdwjidx0B,
1218                                             vdwioffsetptr0+vdwjidx0C,
1219                                             vdwioffsetptr0+vdwjidx0D,
1220                                             &c6_00,&c12_00);
1221
1222             /* LENNARD-JONES DISPERSION/REPULSION */
1223
1224             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1225             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1226             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1227             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1228             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1229
1230             d                = _mm256_sub_pd(r00,rswitch);
1231             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1232             d2               = _mm256_mul_pd(d,d);
1233             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1234
1235             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1236
1237             /* Evaluate switch function */
1238             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1239             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1240             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1241
1242             fscal            = fvdw;
1243
1244             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1245
1246             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1247
1248             /* Calculate temporary vectorial force */
1249             tx               = _mm256_mul_pd(fscal,dx00);
1250             ty               = _mm256_mul_pd(fscal,dy00);
1251             tz               = _mm256_mul_pd(fscal,dz00);
1252
1253             /* Update vectorial force */
1254             fix0             = _mm256_add_pd(fix0,tx);
1255             fiy0             = _mm256_add_pd(fiy0,ty);
1256             fiz0             = _mm256_add_pd(fiz0,tz);
1257
1258             fjx0             = _mm256_add_pd(fjx0,tx);
1259             fjy0             = _mm256_add_pd(fjy0,ty);
1260             fjz0             = _mm256_add_pd(fjz0,tz);
1261
1262             }
1263
1264             /**************************
1265              * CALCULATE INTERACTIONS *
1266              **************************/
1267
1268             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1269             {
1270
1271             /* Compute parameters for interactions between i and j atoms */
1272             qq10             = _mm256_mul_pd(iq1,jq0);
1273
1274             /* REACTION-FIELD ELECTROSTATICS */
1275             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1276
1277             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1278
1279             fscal            = felec;
1280
1281             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1282
1283             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1284
1285             /* Calculate temporary vectorial force */
1286             tx               = _mm256_mul_pd(fscal,dx10);
1287             ty               = _mm256_mul_pd(fscal,dy10);
1288             tz               = _mm256_mul_pd(fscal,dz10);
1289
1290             /* Update vectorial force */
1291             fix1             = _mm256_add_pd(fix1,tx);
1292             fiy1             = _mm256_add_pd(fiy1,ty);
1293             fiz1             = _mm256_add_pd(fiz1,tz);
1294
1295             fjx0             = _mm256_add_pd(fjx0,tx);
1296             fjy0             = _mm256_add_pd(fjy0,ty);
1297             fjz0             = _mm256_add_pd(fjz0,tz);
1298
1299             }
1300
1301             /**************************
1302              * CALCULATE INTERACTIONS *
1303              **************************/
1304
1305             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1306             {
1307
1308             /* Compute parameters for interactions between i and j atoms */
1309             qq20             = _mm256_mul_pd(iq2,jq0);
1310
1311             /* REACTION-FIELD ELECTROSTATICS */
1312             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1313
1314             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1315
1316             fscal            = felec;
1317
1318             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1319
1320             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1321
1322             /* Calculate temporary vectorial force */
1323             tx               = _mm256_mul_pd(fscal,dx20);
1324             ty               = _mm256_mul_pd(fscal,dy20);
1325             tz               = _mm256_mul_pd(fscal,dz20);
1326
1327             /* Update vectorial force */
1328             fix2             = _mm256_add_pd(fix2,tx);
1329             fiy2             = _mm256_add_pd(fiy2,ty);
1330             fiz2             = _mm256_add_pd(fiz2,tz);
1331
1332             fjx0             = _mm256_add_pd(fjx0,tx);
1333             fjy0             = _mm256_add_pd(fjy0,ty);
1334             fjz0             = _mm256_add_pd(fjz0,tz);
1335
1336             }
1337
1338             /**************************
1339              * CALCULATE INTERACTIONS *
1340              **************************/
1341
1342             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1343             {
1344
1345             /* Compute parameters for interactions between i and j atoms */
1346             qq30             = _mm256_mul_pd(iq3,jq0);
1347
1348             /* REACTION-FIELD ELECTROSTATICS */
1349             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1350
1351             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1352
1353             fscal            = felec;
1354
1355             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1356
1357             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1358
1359             /* Calculate temporary vectorial force */
1360             tx               = _mm256_mul_pd(fscal,dx30);
1361             ty               = _mm256_mul_pd(fscal,dy30);
1362             tz               = _mm256_mul_pd(fscal,dz30);
1363
1364             /* Update vectorial force */
1365             fix3             = _mm256_add_pd(fix3,tx);
1366             fiy3             = _mm256_add_pd(fiy3,ty);
1367             fiz3             = _mm256_add_pd(fiz3,tz);
1368
1369             fjx0             = _mm256_add_pd(fjx0,tx);
1370             fjy0             = _mm256_add_pd(fjy0,ty);
1371             fjz0             = _mm256_add_pd(fjz0,tz);
1372
1373             }
1374
1375             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1376             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1377             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1378             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1379
1380             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1381
1382             /* Inner loop uses 150 flops */
1383         }
1384
1385         /* End of innermost loop */
1386
1387         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1388                                                  f+i_coord_offset,fshift+i_shift_offset);
1389
1390         /* Increment number of inner iterations */
1391         inneriter                  += j_index_end - j_index_start;
1392
1393         /* Outer loop uses 24 flops */
1394     }
1395
1396     /* Increment number of outer iterations */
1397     outeriter        += nri;
1398
1399     /* Update outer/inner flops */
1400
1401     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1402 }