src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     int              nvdwtype;
  84     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  85     int              *vdwtype;
  86     real             *vdwparam;
  87     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  88     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  89     __m128i          ewitab;
  90     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  92     real             *ewtab;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 116     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 117     beta2            = _mm256_mul_pd(beta,beta);
 118     beta3            = _mm256_mul_pd(beta,beta2);
 119
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 127     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 128     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 129     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->rcoulomb;
 133     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 134     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 135
 136     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 137     rvdw             = _mm256_set1_pd(fr->rvdw);
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm256_setzero_pd();
 173         fiy0             = _mm256_setzero_pd();
 174         fiz0             = _mm256_setzero_pd();
 175         fix1             = _mm256_setzero_pd();
 176         fiy1             = _mm256_setzero_pd();
 177         fiz1             = _mm256_setzero_pd();
 178         fix2             = _mm256_setzero_pd();
 179         fiy2             = _mm256_setzero_pd();
 180         fiz2             = _mm256_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_pd();
 184         vvdwsum          = _mm256_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_pd(ix0,jx0);
 207             dy00             = _mm256_sub_pd(iy0,jy0);
 208             dz00             = _mm256_sub_pd(iz0,jz0);
 209             dx10             = _mm256_sub_pd(ix1,jx0);
 210             dy10             = _mm256_sub_pd(iy1,jy0);
 211             dz10             = _mm256_sub_pd(iz1,jz0);
 212             dx20             = _mm256_sub_pd(ix2,jx0);
 213             dy20             = _mm256_sub_pd(iy2,jy0);
 214             dz20             = _mm256_sub_pd(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 222             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 223             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 224
 225             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 226             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 227             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 231                                                                  charge+jnrC+0,charge+jnrD+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236
 237             fjx0             = _mm256_setzero_pd();
 238             fjy0             = _mm256_setzero_pd();
 239             fjz0             = _mm256_setzero_pd();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             r00              = _mm256_mul_pd(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm256_mul_pd(iq0,jq0);
 252             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 253                                             vdwioffsetptr0+vdwjidx0B,
 254                                             vdwioffsetptr0+vdwjidx0C,
 255                                             vdwioffsetptr0+vdwjidx0D,
 256                                             &c6_00,&c12_00);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 261             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 262             ewitab           = _mm256_cvttpd_epi32(ewrt);
 263             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 264             ewitab           = _mm_slli_epi32(ewitab,2);
 265             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 266             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 267             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 268             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 269             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 270             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 271             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 272             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 273             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 279             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 280             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 281                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 282             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 283
 284             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm256_and_pd(velec,cutoff_mask);
 288             velecsum         = _mm256_add_pd(velecsum,velec);
 289             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 290             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = _mm256_add_pd(felec,fvdw);
 293
 294             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm256_mul_pd(fscal,dx00);
 298             ty               = _mm256_mul_pd(fscal,dy00);
 299             tz               = _mm256_mul_pd(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm256_add_pd(fix0,tx);
 303             fiy0             = _mm256_add_pd(fiy0,ty);
 304             fiz0             = _mm256_add_pd(fiz0,tz);
 305
 306             fjx0             = _mm256_add_pd(fjx0,tx);
 307             fjy0             = _mm256_add_pd(fjy0,ty);
 308             fjz0             = _mm256_add_pd(fjz0,tz);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 317             {
 318
 319             r10              = _mm256_mul_pd(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm256_mul_pd(iq1,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 328             ewitab           = _mm256_cvttpd_epi32(ewrt);
 329             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 333             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 334             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 335             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 336             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 337             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 338             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 339             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 340
 341             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm256_and_pd(velec,cutoff_mask);
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx10);
 353             ty               = _mm256_mul_pd(fscal,dy10);
 354             tz               = _mm256_mul_pd(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm256_add_pd(fix1,tx);
 358             fiy1             = _mm256_add_pd(fiy1,ty);
 359             fiz1             = _mm256_add_pd(fiz1,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             }
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 372             {
 373
 374             r20              = _mm256_mul_pd(rsq20,rinv20);
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq20             = _mm256_mul_pd(iq2,jq0);
 378
 379             /* EWALD ELECTROSTATICS */
 380
 381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 382             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 383             ewitab           = _mm256_cvttpd_epi32(ewrt);
 384             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 385             ewitab           = _mm_slli_epi32(ewitab,2);
 386             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 387             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 388             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 389             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 390             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 391             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 392             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 393             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 394             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 395
 396             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velec            = _mm256_and_pd(velec,cutoff_mask);
 400             velecsum         = _mm256_add_pd(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm256_mul_pd(fscal,dx20);
 408             ty               = _mm256_mul_pd(fscal,dy20);
 409             tz               = _mm256_mul_pd(fscal,dz20);
 410
 411             /* Update vectorial force */
 412             fix2             = _mm256_add_pd(fix2,tx);
 413             fiy2             = _mm256_add_pd(fiy2,ty);
 414             fiz2             = _mm256_add_pd(fiz2,tz);
 415
 416             fjx0             = _mm256_add_pd(fjx0,tx);
 417             fjy0             = _mm256_add_pd(fjy0,ty);
 418             fjz0             = _mm256_add_pd(fjz0,tz);
 419
 420             }
 421
 422             fjptrA             = f+j_coord_offsetA;
 423             fjptrB             = f+j_coord_offsetB;
 424             fjptrC             = f+j_coord_offsetC;
 425             fjptrD             = f+j_coord_offsetD;
 426
 427             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 428
 429             /* Inner loop uses 159 flops */
 430         }
 431
 432         if(jidx<j_index_end)
 433         {
 434
 435             /* Get j neighbor index, and coordinate index */
 436             jnrlistA         = jjnr[jidx];
 437             jnrlistB         = jjnr[jidx+1];
 438             jnrlistC         = jjnr[jidx+2];
 439             jnrlistD         = jjnr[jidx+3];
 440             /* Sign of each element will be negative for non-real atoms.
 441              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 442              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 443              */
 444             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 445
 446             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 447             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 448             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 449
 450             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 451             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 452             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 453             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456             j_coord_offsetC  = DIM*jnrC;
 457             j_coord_offsetD  = DIM*jnrD;
 458
 459             /* load j atom coordinates */
 460             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 461                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 462                                                  &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm256_sub_pd(ix0,jx0);
 466             dy00             = _mm256_sub_pd(iy0,jy0);
 467             dz00             = _mm256_sub_pd(iz0,jz0);
 468             dx10             = _mm256_sub_pd(ix1,jx0);
 469             dy10             = _mm256_sub_pd(iy1,jy0);
 470             dz10             = _mm256_sub_pd(iz1,jz0);
 471             dx20             = _mm256_sub_pd(ix2,jx0);
 472             dy20             = _mm256_sub_pd(iy2,jy0);
 473             dz20             = _mm256_sub_pd(iz2,jz0);
 474
 475             /* Calculate squared distance and things based on it */
 476             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 477             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 478             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 479
 480             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 481             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 482             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 483
 484             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 485             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 486             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 490                                                                  charge+jnrC+0,charge+jnrD+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495
 496             fjx0             = _mm256_setzero_pd();
 497             fjy0             = _mm256_setzero_pd();
 498             fjz0             = _mm256_setzero_pd();
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 505             {
 506
 507             r00              = _mm256_mul_pd(rsq00,rinv00);
 508             r00              = _mm256_andnot_pd(dummy_mask,r00);
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq00             = _mm256_mul_pd(iq0,jq0);
 512             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 513                                             vdwioffsetptr0+vdwjidx0B,
 514                                             vdwioffsetptr0+vdwjidx0C,
 515                                             vdwioffsetptr0+vdwjidx0D,
 516                                             &c6_00,&c12_00);
 517
 518             /* EWALD ELECTROSTATICS */
 519
 520             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 521             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 522             ewitab           = _mm256_cvttpd_epi32(ewrt);
 523             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 524             ewitab           = _mm_slli_epi32(ewitab,2);
 525             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 526             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 527             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 528             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 529             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 530             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 531             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 532             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 533             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 534
 535             /* LENNARD-JONES DISPERSION/REPULSION */
 536
 537             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 538             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 539             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 540             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 541                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 542             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 543
 544             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm256_and_pd(velec,cutoff_mask);
 548             velec            = _mm256_andnot_pd(dummy_mask,velec);
 549             velecsum         = _mm256_add_pd(velecsum,velec);
 550             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 551             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 552             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 553
 554             fscal            = _mm256_add_pd(felec,fvdw);
 555
 556             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 557
 558             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm256_mul_pd(fscal,dx00);
 562             ty               = _mm256_mul_pd(fscal,dy00);
 563             tz               = _mm256_mul_pd(fscal,dz00);
 564
 565             /* Update vectorial force */
 566             fix0             = _mm256_add_pd(fix0,tx);
 567             fiy0             = _mm256_add_pd(fiy0,ty);
 568             fiz0             = _mm256_add_pd(fiz0,tz);
 569
 570             fjx0             = _mm256_add_pd(fjx0,tx);
 571             fjy0             = _mm256_add_pd(fjy0,ty);
 572             fjz0             = _mm256_add_pd(fjz0,tz);
 573
 574             }
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 581             {
 582
 583             r10              = _mm256_mul_pd(rsq10,rinv10);
 584             r10              = _mm256_andnot_pd(dummy_mask,r10);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq10             = _mm256_mul_pd(iq1,jq0);
 588
 589             /* EWALD ELECTROSTATICS */
 590
 591             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 592             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 593             ewitab           = _mm256_cvttpd_epi32(ewrt);
 594             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 595             ewitab           = _mm_slli_epi32(ewitab,2);
 596             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 597             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 598             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 599             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 600             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 601             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 602             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 603             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 604             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 605
 606             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm256_and_pd(velec,cutoff_mask);
 610             velec            = _mm256_andnot_pd(dummy_mask,velec);
 611             velecsum         = _mm256_add_pd(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 616
 617             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm256_mul_pd(fscal,dx10);
 621             ty               = _mm256_mul_pd(fscal,dy10);
 622             tz               = _mm256_mul_pd(fscal,dz10);
 623
 624             /* Update vectorial force */
 625             fix1             = _mm256_add_pd(fix1,tx);
 626             fiy1             = _mm256_add_pd(fiy1,ty);
 627             fiz1             = _mm256_add_pd(fiz1,tz);
 628
 629             fjx0             = _mm256_add_pd(fjx0,tx);
 630             fjy0             = _mm256_add_pd(fjy0,ty);
 631             fjz0             = _mm256_add_pd(fjz0,tz);
 632
 633             }
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 640             {
 641
 642             r20              = _mm256_mul_pd(rsq20,rinv20);
 643             r20              = _mm256_andnot_pd(dummy_mask,r20);
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq20             = _mm256_mul_pd(iq2,jq0);
 647
 648             /* EWALD ELECTROSTATICS */
 649
 650             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 651             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 652             ewitab           = _mm256_cvttpd_epi32(ewrt);
 653             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 654             ewitab           = _mm_slli_epi32(ewitab,2);
 655             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 656             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 657             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 658             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 659             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 660             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 661             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 662             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 663             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 664
 665             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             velec            = _mm256_and_pd(velec,cutoff_mask);
 669             velec            = _mm256_andnot_pd(dummy_mask,velec);
 670             velecsum         = _mm256_add_pd(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 675
 676             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 677
 678             /* Calculate temporary vectorial force */
 679             tx               = _mm256_mul_pd(fscal,dx20);
 680             ty               = _mm256_mul_pd(fscal,dy20);
 681             tz               = _mm256_mul_pd(fscal,dz20);
 682
 683             /* Update vectorial force */
 684             fix2             = _mm256_add_pd(fix2,tx);
 685             fiy2             = _mm256_add_pd(fiy2,ty);
 686             fiz2             = _mm256_add_pd(fiz2,tz);
 687
 688             fjx0             = _mm256_add_pd(fjx0,tx);
 689             fjy0             = _mm256_add_pd(fjy0,ty);
 690             fjz0             = _mm256_add_pd(fjz0,tz);
 691
 692             }
 693
 694             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 695             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 696             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 697             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 698
 699             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 700
 701             /* Inner loop uses 162 flops */
 702         }
 703
 704         /* End of innermost loop */
 705
 706         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 707                                                  f+i_coord_offset,fshift+i_shift_offset);
 708
 709         ggid                        = gid[iidx];
 710         /* Update potential energies */
 711         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 712         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 713
 714         /* Increment number of inner iterations */
 715         inneriter                  += j_index_end - j_index_start;
 716
 717         /* Outer loop uses 20 flops */
 718     }
 719
 720     /* Increment number of outer iterations */
 721     outeriter        += nri;
 722
 723     /* Update outer/inner flops */
 724
 725     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 726 }
 727 /*
 728  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_double
 729  * Electrostatics interaction: Ewald
 730  * VdW interaction:            LennardJones
 731  * Geometry:                   Water3-Particle
 732  * Calculate force/pot:        Force
 733  */
 734 void
 735 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_double
 736                     (t_nblist * gmx_restrict                nlist,
 737                      rvec * gmx_restrict                    xx,
 738                      rvec * gmx_restrict                    ff,
 739                      t_forcerec * gmx_restrict              fr,
 740                      t_mdatoms * gmx_restrict               mdatoms,
 741                      nb_kernel_data_t * gmx_restrict        kernel_data,
 742                      t_nrnb * gmx_restrict                  nrnb)
 743 {
 744     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 745      * just 0 for non-waters.
 746      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 747      * jnr indices corresponding to data put in the four positions in the SIMD register.
 748      */
 749     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 750     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 751     int              jnrA,jnrB,jnrC,jnrD;
 752     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 753     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 754     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 755     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 756     real             rcutoff_scalar;
 757     real             *shiftvec,*fshift,*x,*f;
 758     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 759     real             scratch[4*DIM];
 760     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 761     real *           vdwioffsetptr0;
 762     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 763     real *           vdwioffsetptr1;
 764     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 765     real *           vdwioffsetptr2;
 766     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 767     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 768     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 769     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 770     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 771     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 772     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 773     real             *charge;
 774     int              nvdwtype;
 775     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 776     int              *vdwtype;
 777     real             *vdwparam;
 778     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 779     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 780     __m128i          ewitab;
 781     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 782     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 783     real             *ewtab;
 784     __m256d          dummy_mask,cutoff_mask;
 785     __m128           tmpmask0,tmpmask1;
 786     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 787     __m256d          one     = _mm256_set1_pd(1.0);
 788     __m256d          two     = _mm256_set1_pd(2.0);
 789     x                = xx[0];
 790     f                = ff[0];
 791
 792     nri              = nlist->nri;
 793     iinr             = nlist->iinr;
 794     jindex           = nlist->jindex;
 795     jjnr             = nlist->jjnr;
 796     shiftidx         = nlist->shift;
 797     gid              = nlist->gid;
 798     shiftvec         = fr->shift_vec[0];
 799     fshift           = fr->fshift[0];
 800     facel            = _mm256_set1_pd(fr->epsfac);
 801     charge           = mdatoms->chargeA;
 802     nvdwtype         = fr->ntype;
 803     vdwparam         = fr->nbfp;
 804     vdwtype          = mdatoms->typeA;
 805
 806     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 807     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 808     beta2            = _mm256_mul_pd(beta,beta);
 809     beta3            = _mm256_mul_pd(beta,beta2);
 810
 811     ewtab            = fr->ic->tabq_coul_F;
 812     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 813     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 814
 815     /* Setup water-specific parameters */
 816     inr              = nlist->iinr[0];
 817     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 818     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 819     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 820     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 821
 822     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 823     rcutoff_scalar   = fr->rcoulomb;
 824     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 825     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 826
 827     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 828     rvdw             = _mm256_set1_pd(fr->rvdw);
 829
 830     /* Avoid stupid compiler warnings */
 831     jnrA = jnrB = jnrC = jnrD = 0;
 832     j_coord_offsetA = 0;
 833     j_coord_offsetB = 0;
 834     j_coord_offsetC = 0;
 835     j_coord_offsetD = 0;
 836
 837     outeriter        = 0;
 838     inneriter        = 0;
 839
 840     for(iidx=0;iidx<4*DIM;iidx++)
 841     {
 842         scratch[iidx] = 0.0;
 843     }
 844
 845     /* Start outer loop over neighborlists */
 846     for(iidx=0; iidx<nri; iidx++)
 847     {
 848         /* Load shift vector for this list */
 849         i_shift_offset   = DIM*shiftidx[iidx];
 850
 851         /* Load limits for loop over neighbors */
 852         j_index_start    = jindex[iidx];
 853         j_index_end      = jindex[iidx+1];
 854
 855         /* Get outer coordinate index */
 856         inr              = iinr[iidx];
 857         i_coord_offset   = DIM*inr;
 858
 859         /* Load i particle coords and add shift vector */
 860         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 861                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 862
 863         fix0             = _mm256_setzero_pd();
 864         fiy0             = _mm256_setzero_pd();
 865         fiz0             = _mm256_setzero_pd();
 866         fix1             = _mm256_setzero_pd();
 867         fiy1             = _mm256_setzero_pd();
 868         fiz1             = _mm256_setzero_pd();
 869         fix2             = _mm256_setzero_pd();
 870         fiy2             = _mm256_setzero_pd();
 871         fiz2             = _mm256_setzero_pd();
 872
 873         /* Start inner kernel loop */
 874         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 875         {
 876
 877             /* Get j neighbor index, and coordinate index */
 878             jnrA             = jjnr[jidx];
 879             jnrB             = jjnr[jidx+1];
 880             jnrC             = jjnr[jidx+2];
 881             jnrD             = jjnr[jidx+3];
 882             j_coord_offsetA  = DIM*jnrA;
 883             j_coord_offsetB  = DIM*jnrB;
 884             j_coord_offsetC  = DIM*jnrC;
 885             j_coord_offsetD  = DIM*jnrD;
 886
 887             /* load j atom coordinates */
 888             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 889                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 890                                                  &jx0,&jy0,&jz0);
 891
 892             /* Calculate displacement vector */
 893             dx00             = _mm256_sub_pd(ix0,jx0);
 894             dy00             = _mm256_sub_pd(iy0,jy0);
 895             dz00             = _mm256_sub_pd(iz0,jz0);
 896             dx10             = _mm256_sub_pd(ix1,jx0);
 897             dy10             = _mm256_sub_pd(iy1,jy0);
 898             dz10             = _mm256_sub_pd(iz1,jz0);
 899             dx20             = _mm256_sub_pd(ix2,jx0);
 900             dy20             = _mm256_sub_pd(iy2,jy0);
 901             dz20             = _mm256_sub_pd(iz2,jz0);
 902
 903             /* Calculate squared distance and things based on it */
 904             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 905             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 906             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 907
 908             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 909             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 910             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 911
 912             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 913             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 914             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 915
 916             /* Load parameters for j particles */
 917             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 918                                                                  charge+jnrC+0,charge+jnrD+0);
 919             vdwjidx0A        = 2*vdwtype[jnrA+0];
 920             vdwjidx0B        = 2*vdwtype[jnrB+0];
 921             vdwjidx0C        = 2*vdwtype[jnrC+0];
 922             vdwjidx0D        = 2*vdwtype[jnrD+0];
 923
 924             fjx0             = _mm256_setzero_pd();
 925             fjy0             = _mm256_setzero_pd();
 926             fjz0             = _mm256_setzero_pd();
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 933             {
 934
 935             r00              = _mm256_mul_pd(rsq00,rinv00);
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq00             = _mm256_mul_pd(iq0,jq0);
 939             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 940                                             vdwioffsetptr0+vdwjidx0B,
 941                                             vdwioffsetptr0+vdwjidx0C,
 942                                             vdwioffsetptr0+vdwjidx0D,
 943                                             &c6_00,&c12_00);
 944
 945             /* EWALD ELECTROSTATICS */
 946
 947             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 948             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 949             ewitab           = _mm256_cvttpd_epi32(ewrt);
 950             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 951             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 952                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 953                                             &ewtabF,&ewtabFn);
 954             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 955             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 956
 957             /* LENNARD-JONES DISPERSION/REPULSION */
 958
 959             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 960             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 961
 962             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 963
 964             fscal            = _mm256_add_pd(felec,fvdw);
 965
 966             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm256_mul_pd(fscal,dx00);
 970             ty               = _mm256_mul_pd(fscal,dy00);
 971             tz               = _mm256_mul_pd(fscal,dz00);
 972
 973             /* Update vectorial force */
 974             fix0             = _mm256_add_pd(fix0,tx);
 975             fiy0             = _mm256_add_pd(fiy0,ty);
 976             fiz0             = _mm256_add_pd(fiz0,tz);
 977
 978             fjx0             = _mm256_add_pd(fjx0,tx);
 979             fjy0             = _mm256_add_pd(fjy0,ty);
 980             fjz0             = _mm256_add_pd(fjz0,tz);
 981
 982             }
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 989             {
 990
 991             r10              = _mm256_mul_pd(rsq10,rinv10);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq10             = _mm256_mul_pd(iq1,jq0);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 999             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1000             ewitab           = _mm256_cvttpd_epi32(ewrt);
1001             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1002             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1003                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1004                                             &ewtabF,&ewtabFn);
1005             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1006             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1007
1008             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1009
1010             fscal            = felec;
1011
1012             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_pd(fscal,dx10);
1016             ty               = _mm256_mul_pd(fscal,dy10);
1017             tz               = _mm256_mul_pd(fscal,dz10);
1018
1019             /* Update vectorial force */
1020             fix1             = _mm256_add_pd(fix1,tx);
1021             fiy1             = _mm256_add_pd(fiy1,ty);
1022             fiz1             = _mm256_add_pd(fiz1,tz);
1023
1024             fjx0             = _mm256_add_pd(fjx0,tx);
1025             fjy0             = _mm256_add_pd(fjy0,ty);
1026             fjz0             = _mm256_add_pd(fjz0,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1035             {
1036
1037             r20              = _mm256_mul_pd(rsq20,rinv20);
1038
1039             /* Compute parameters for interactions between i and j atoms */
1040             qq20             = _mm256_mul_pd(iq2,jq0);
1041
1042             /* EWALD ELECTROSTATICS */
1043
1044             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1045             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1046             ewitab           = _mm256_cvttpd_epi32(ewrt);
1047             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1048             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1049                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1050                                             &ewtabF,&ewtabFn);
1051             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1052             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1053
1054             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1055
1056             fscal            = felec;
1057
1058             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1059
1060             /* Calculate temporary vectorial force */
1061             tx               = _mm256_mul_pd(fscal,dx20);
1062             ty               = _mm256_mul_pd(fscal,dy20);
1063             tz               = _mm256_mul_pd(fscal,dz20);
1064
1065             /* Update vectorial force */
1066             fix2             = _mm256_add_pd(fix2,tx);
1067             fiy2             = _mm256_add_pd(fiy2,ty);
1068             fiz2             = _mm256_add_pd(fiz2,tz);
1069
1070             fjx0             = _mm256_add_pd(fjx0,tx);
1071             fjy0             = _mm256_add_pd(fjy0,ty);
1072             fjz0             = _mm256_add_pd(fjz0,tz);
1073
1074             }
1075
1076             fjptrA             = f+j_coord_offsetA;
1077             fjptrB             = f+j_coord_offsetB;
1078             fjptrC             = f+j_coord_offsetC;
1079             fjptrD             = f+j_coord_offsetD;
1080
1081             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1082
1083             /* Inner loop uses 127 flops */
1084         }
1085
1086         if(jidx<j_index_end)
1087         {
1088
1089             /* Get j neighbor index, and coordinate index */
1090             jnrlistA         = jjnr[jidx];
1091             jnrlistB         = jjnr[jidx+1];
1092             jnrlistC         = jjnr[jidx+2];
1093             jnrlistD         = jjnr[jidx+3];
1094             /* Sign of each element will be negative for non-real atoms.
1095              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1096              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1097              */
1098             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1099
1100             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1101             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1102             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
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             j_coord_offsetA  = DIM*jnrA;
1109             j_coord_offsetB  = DIM*jnrB;
1110             j_coord_offsetC  = DIM*jnrC;
1111             j_coord_offsetD  = DIM*jnrD;
1112
1113             /* load j atom coordinates */
1114             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1115                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1116                                                  &jx0,&jy0,&jz0);
1117
1118             /* Calculate displacement vector */
1119             dx00             = _mm256_sub_pd(ix0,jx0);
1120             dy00             = _mm256_sub_pd(iy0,jy0);
1121             dz00             = _mm256_sub_pd(iz0,jz0);
1122             dx10             = _mm256_sub_pd(ix1,jx0);
1123             dy10             = _mm256_sub_pd(iy1,jy0);
1124             dz10             = _mm256_sub_pd(iz1,jz0);
1125             dx20             = _mm256_sub_pd(ix2,jx0);
1126             dy20             = _mm256_sub_pd(iy2,jy0);
1127             dz20             = _mm256_sub_pd(iz2,jz0);
1128
1129             /* Calculate squared distance and things based on it */
1130             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1131             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1132             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1133
1134             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1135             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1136             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1137
1138             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1139             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1140             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1141
1142             /* Load parameters for j particles */
1143             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1144                                                                  charge+jnrC+0,charge+jnrD+0);
1145             vdwjidx0A        = 2*vdwtype[jnrA+0];
1146             vdwjidx0B        = 2*vdwtype[jnrB+0];
1147             vdwjidx0C        = 2*vdwtype[jnrC+0];
1148             vdwjidx0D        = 2*vdwtype[jnrD+0];
1149
1150             fjx0             = _mm256_setzero_pd();
1151             fjy0             = _mm256_setzero_pd();
1152             fjz0             = _mm256_setzero_pd();
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1159             {
1160
1161             r00              = _mm256_mul_pd(rsq00,rinv00);
1162             r00              = _mm256_andnot_pd(dummy_mask,r00);
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq00             = _mm256_mul_pd(iq0,jq0);
1166             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1167                                             vdwioffsetptr0+vdwjidx0B,
1168                                             vdwioffsetptr0+vdwjidx0C,
1169                                             vdwioffsetptr0+vdwjidx0D,
1170                                             &c6_00,&c12_00);
1171
1172             /* EWALD ELECTROSTATICS */
1173
1174             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1175             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1176             ewitab           = _mm256_cvttpd_epi32(ewrt);
1177             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1178             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1179                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1180                                             &ewtabF,&ewtabFn);
1181             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1182             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1183
1184             /* LENNARD-JONES DISPERSION/REPULSION */
1185
1186             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1187             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1188
1189             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1190
1191             fscal            = _mm256_add_pd(felec,fvdw);
1192
1193             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1194
1195             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1196
1197             /* Calculate temporary vectorial force */
1198             tx               = _mm256_mul_pd(fscal,dx00);
1199             ty               = _mm256_mul_pd(fscal,dy00);
1200             tz               = _mm256_mul_pd(fscal,dz00);
1201
1202             /* Update vectorial force */
1203             fix0             = _mm256_add_pd(fix0,tx);
1204             fiy0             = _mm256_add_pd(fiy0,ty);
1205             fiz0             = _mm256_add_pd(fiz0,tz);
1206
1207             fjx0             = _mm256_add_pd(fjx0,tx);
1208             fjy0             = _mm256_add_pd(fjy0,ty);
1209             fjz0             = _mm256_add_pd(fjz0,tz);
1210
1211             }
1212
1213             /**************************
1214              * CALCULATE INTERACTIONS *
1215              **************************/
1216
1217             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1218             {
1219
1220             r10              = _mm256_mul_pd(rsq10,rinv10);
1221             r10              = _mm256_andnot_pd(dummy_mask,r10);
1222
1223             /* Compute parameters for interactions between i and j atoms */
1224             qq10             = _mm256_mul_pd(iq1,jq0);
1225
1226             /* EWALD ELECTROSTATICS */
1227
1228             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1229             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1230             ewitab           = _mm256_cvttpd_epi32(ewrt);
1231             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1232             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1233                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1234                                             &ewtabF,&ewtabFn);
1235             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1236             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1237
1238             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1243
1244             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm256_mul_pd(fscal,dx10);
1248             ty               = _mm256_mul_pd(fscal,dy10);
1249             tz               = _mm256_mul_pd(fscal,dz10);
1250
1251             /* Update vectorial force */
1252             fix1             = _mm256_add_pd(fix1,tx);
1253             fiy1             = _mm256_add_pd(fiy1,ty);
1254             fiz1             = _mm256_add_pd(fiz1,tz);
1255
1256             fjx0             = _mm256_add_pd(fjx0,tx);
1257             fjy0             = _mm256_add_pd(fjy0,ty);
1258             fjz0             = _mm256_add_pd(fjz0,tz);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1267             {
1268
1269             r20              = _mm256_mul_pd(rsq20,rinv20);
1270             r20              = _mm256_andnot_pd(dummy_mask,r20);
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             qq20             = _mm256_mul_pd(iq2,jq0);
1274
1275             /* EWALD ELECTROSTATICS */
1276
1277             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1278             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1279             ewitab           = _mm256_cvttpd_epi32(ewrt);
1280             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1281             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1282                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1283                                             &ewtabF,&ewtabFn);
1284             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1285             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1286
1287             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1288
1289             fscal            = felec;
1290
1291             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1292
1293             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm256_mul_pd(fscal,dx20);
1297             ty               = _mm256_mul_pd(fscal,dy20);
1298             tz               = _mm256_mul_pd(fscal,dz20);
1299
1300             /* Update vectorial force */
1301             fix2             = _mm256_add_pd(fix2,tx);
1302             fiy2             = _mm256_add_pd(fiy2,ty);
1303             fiz2             = _mm256_add_pd(fiz2,tz);
1304
1305             fjx0             = _mm256_add_pd(fjx0,tx);
1306             fjy0             = _mm256_add_pd(fjy0,ty);
1307             fjz0             = _mm256_add_pd(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
1316             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1317
1318             /* Inner loop uses 130 flops */
1319         }
1320
1321         /* End of innermost loop */
1322
1323         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1324                                                  f+i_coord_offset,fshift+i_shift_offset);
1325
1326         /* Increment number of inner iterations */
1327         inneriter                  += j_index_end - j_index_start;
1328
1329         /* Outer loop uses 18 flops */
1330     }
1331
1332     /* Increment number of outer iterations */
1333     outeriter        += nri;
1334
1335     /* Update outer/inner flops */
1336
1337     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*130);
1338 }