src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_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_ElecEw_VdwCSTab_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_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          vfitab;
  90     __m128i          ifour       = _mm_set1_epi32(4);
  91     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  92     real             *vftab;
  93     __m128i          ewitab;
  94     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  96     real             *ewtab;
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 121
 122     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 123     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 124     beta2            = _mm256_mul_pd(beta,beta);
 125     beta3            = _mm256_mul_pd(beta,beta2);
 126
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 134     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 135     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm256_setzero_pd();
 172         fiy0             = _mm256_setzero_pd();
 173         fiz0             = _mm256_setzero_pd();
 174         fix1             = _mm256_setzero_pd();
 175         fiy1             = _mm256_setzero_pd();
 176         fiz1             = _mm256_setzero_pd();
 177         fix2             = _mm256_setzero_pd();
 178         fiy2             = _mm256_setzero_pd();
 179         fiz2             = _mm256_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm256_setzero_pd();
 183         vvdwsum          = _mm256_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm256_sub_pd(ix0,jx0);
 206             dy00             = _mm256_sub_pd(iy0,jy0);
 207             dz00             = _mm256_sub_pd(iz0,jz0);
 208             dx10             = _mm256_sub_pd(ix1,jx0);
 209             dy10             = _mm256_sub_pd(iy1,jy0);
 210             dz10             = _mm256_sub_pd(iz1,jz0);
 211             dx20             = _mm256_sub_pd(ix2,jx0);
 212             dy20             = _mm256_sub_pd(iy2,jy0);
 213             dz20             = _mm256_sub_pd(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 217             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 218             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 219
 220             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 221             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 222             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 223
 224             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 225             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 226             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 230                                                                  charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             fjx0             = _mm256_setzero_pd();
 237             fjy0             = _mm256_setzero_pd();
 238             fjz0             = _mm256_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm256_mul_pd(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             qq00             = _mm256_mul_pd(iq0,jq0);
 248             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 249                                             vdwioffsetptr0+vdwjidx0B,
 250                                             vdwioffsetptr0+vdwjidx0C,
 251                                             vdwioffsetptr0+vdwjidx0D,
 252                                             &c6_00,&c12_00);
 253
 254             /* Calculate table index by multiplying r with table scale and truncate to integer */
 255             rt               = _mm256_mul_pd(r00,vftabscale);
 256             vfitab           = _mm256_cvttpd_epi32(rt);
 257             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 258             vfitab           = _mm_slli_epi32(vfitab,3);
 259
 260             /* EWALD ELECTROSTATICS */
 261
 262             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 263             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 264             ewitab           = _mm256_cvttpd_epi32(ewrt);
 265             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 266             ewitab           = _mm_slli_epi32(ewitab,2);
 267             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 268             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 269             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 270             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 271             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 272             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 273             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 274             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 275             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 276
 277             /* CUBIC SPLINE TABLE DISPERSION */
 278             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 279             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 280             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 281             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 282             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 283             Heps             = _mm256_mul_pd(vfeps,H);
 284             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 285             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 286             vvdw6            = _mm256_mul_pd(c6_00,VV);
 287             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 288             fvdw6            = _mm256_mul_pd(c6_00,FF);
 289
 290             /* CUBIC SPLINE TABLE REPULSION */
 291             vfitab           = _mm_add_epi32(vfitab,ifour);
 292             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 293             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 294             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 295             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 296             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 297             Heps             = _mm256_mul_pd(vfeps,H);
 298             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 299             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 300             vvdw12           = _mm256_mul_pd(c12_00,VV);
 301             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 302             fvdw12           = _mm256_mul_pd(c12_00,FF);
 303             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 304             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm256_add_pd(velecsum,velec);
 308             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 309
 310             fscal            = _mm256_add_pd(felec,fvdw);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm256_mul_pd(fscal,dx00);
 314             ty               = _mm256_mul_pd(fscal,dy00);
 315             tz               = _mm256_mul_pd(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm256_add_pd(fix0,tx);
 319             fiy0             = _mm256_add_pd(fiy0,ty);
 320             fiz0             = _mm256_add_pd(fiz0,tz);
 321
 322             fjx0             = _mm256_add_pd(fjx0,tx);
 323             fjy0             = _mm256_add_pd(fjy0,ty);
 324             fjz0             = _mm256_add_pd(fjz0,tz);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             r10              = _mm256_mul_pd(rsq10,rinv10);
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq10             = _mm256_mul_pd(iq1,jq0);
 334
 335             /* EWALD ELECTROSTATICS */
 336
 337             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 338             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 339             ewitab           = _mm256_cvttpd_epi32(ewrt);
 340             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 341             ewitab           = _mm_slli_epi32(ewitab,2);
 342             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 343             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 344             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 345             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 346             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 347             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 348             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 349             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 350             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm256_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm256_mul_pd(fscal,dx10);
 359             ty               = _mm256_mul_pd(fscal,dy10);
 360             tz               = _mm256_mul_pd(fscal,dz10);
 361
 362             /* Update vectorial force */
 363             fix1             = _mm256_add_pd(fix1,tx);
 364             fiy1             = _mm256_add_pd(fiy1,ty);
 365             fiz1             = _mm256_add_pd(fiz1,tz);
 366
 367             fjx0             = _mm256_add_pd(fjx0,tx);
 368             fjy0             = _mm256_add_pd(fjy0,ty);
 369             fjz0             = _mm256_add_pd(fjz0,tz);
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             r20              = _mm256_mul_pd(rsq20,rinv20);
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq20             = _mm256_mul_pd(iq2,jq0);
 379
 380             /* EWALD ELECTROSTATICS */
 381
 382             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 383             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 384             ewitab           = _mm256_cvttpd_epi32(ewrt);
 385             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 386             ewitab           = _mm_slli_epi32(ewitab,2);
 387             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 388             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 389             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 390             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 391             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 392             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 393             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 394             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 395             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velecsum         = _mm256_add_pd(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             /* Calculate temporary vectorial force */
 403             tx               = _mm256_mul_pd(fscal,dx20);
 404             ty               = _mm256_mul_pd(fscal,dy20);
 405             tz               = _mm256_mul_pd(fscal,dz20);
 406
 407             /* Update vectorial force */
 408             fix2             = _mm256_add_pd(fix2,tx);
 409             fiy2             = _mm256_add_pd(fiy2,ty);
 410             fiz2             = _mm256_add_pd(fiz2,tz);
 411
 412             fjx0             = _mm256_add_pd(fjx0,tx);
 413             fjy0             = _mm256_add_pd(fjy0,ty);
 414             fjz0             = _mm256_add_pd(fjz0,tz);
 415
 416             fjptrA             = f+j_coord_offsetA;
 417             fjptrB             = f+j_coord_offsetB;
 418             fjptrC             = f+j_coord_offsetC;
 419             fjptrD             = f+j_coord_offsetD;
 420
 421             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 422
 423             /* Inner loop uses 160 flops */
 424         }
 425
 426         if(jidx<j_index_end)
 427         {
 428
 429             /* Get j neighbor index, and coordinate index */
 430             jnrlistA         = jjnr[jidx];
 431             jnrlistB         = jjnr[jidx+1];
 432             jnrlistC         = jjnr[jidx+2];
 433             jnrlistD         = jjnr[jidx+3];
 434             /* Sign of each element will be negative for non-real atoms.
 435              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 436              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 437              */
 438             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 439
 440             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 441             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 442             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 443
 444             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 445             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 446             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 447             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 448             j_coord_offsetA  = DIM*jnrA;
 449             j_coord_offsetB  = DIM*jnrB;
 450             j_coord_offsetC  = DIM*jnrC;
 451             j_coord_offsetD  = DIM*jnrD;
 452
 453             /* load j atom coordinates */
 454             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 455                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 456                                                  &jx0,&jy0,&jz0);
 457
 458             /* Calculate displacement vector */
 459             dx00             = _mm256_sub_pd(ix0,jx0);
 460             dy00             = _mm256_sub_pd(iy0,jy0);
 461             dz00             = _mm256_sub_pd(iz0,jz0);
 462             dx10             = _mm256_sub_pd(ix1,jx0);
 463             dy10             = _mm256_sub_pd(iy1,jy0);
 464             dz10             = _mm256_sub_pd(iz1,jz0);
 465             dx20             = _mm256_sub_pd(ix2,jx0);
 466             dy20             = _mm256_sub_pd(iy2,jy0);
 467             dz20             = _mm256_sub_pd(iz2,jz0);
 468
 469             /* Calculate squared distance and things based on it */
 470             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 471             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 472             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 473
 474             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 475             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 476             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 477
 478             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 479             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 480             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 481
 482             /* Load parameters for j particles */
 483             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 484                                                                  charge+jnrC+0,charge+jnrD+0);
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487             vdwjidx0C        = 2*vdwtype[jnrC+0];
 488             vdwjidx0D        = 2*vdwtype[jnrD+0];
 489
 490             fjx0             = _mm256_setzero_pd();
 491             fjy0             = _mm256_setzero_pd();
 492             fjz0             = _mm256_setzero_pd();
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r00              = _mm256_mul_pd(rsq00,rinv00);
 499             r00              = _mm256_andnot_pd(dummy_mask,r00);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq00             = _mm256_mul_pd(iq0,jq0);
 503             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 504                                             vdwioffsetptr0+vdwjidx0B,
 505                                             vdwioffsetptr0+vdwjidx0C,
 506                                             vdwioffsetptr0+vdwjidx0D,
 507                                             &c6_00,&c12_00);
 508
 509             /* Calculate table index by multiplying r with table scale and truncate to integer */
 510             rt               = _mm256_mul_pd(r00,vftabscale);
 511             vfitab           = _mm256_cvttpd_epi32(rt);
 512             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 513             vfitab           = _mm_slli_epi32(vfitab,3);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 518             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 519             ewitab           = _mm256_cvttpd_epi32(ewrt);
 520             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 521             ewitab           = _mm_slli_epi32(ewitab,2);
 522             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 523             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 524             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 525             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 526             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 527             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 528             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 529             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 530             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 531
 532             /* CUBIC SPLINE TABLE DISPERSION */
 533             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 534             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 535             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 536             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 537             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 538             Heps             = _mm256_mul_pd(vfeps,H);
 539             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 540             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 541             vvdw6            = _mm256_mul_pd(c6_00,VV);
 542             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 543             fvdw6            = _mm256_mul_pd(c6_00,FF);
 544
 545             /* CUBIC SPLINE TABLE REPULSION */
 546             vfitab           = _mm_add_epi32(vfitab,ifour);
 547             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 548             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 549             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 550             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 551             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 552             Heps             = _mm256_mul_pd(vfeps,H);
 553             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 554             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 555             vvdw12           = _mm256_mul_pd(c12_00,VV);
 556             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 557             fvdw12           = _mm256_mul_pd(c12_00,FF);
 558             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 559             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm256_andnot_pd(dummy_mask,velec);
 563             velecsum         = _mm256_add_pd(velecsum,velec);
 564             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 565             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 566
 567             fscal            = _mm256_add_pd(felec,fvdw);
 568
 569             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm256_mul_pd(fscal,dx00);
 573             ty               = _mm256_mul_pd(fscal,dy00);
 574             tz               = _mm256_mul_pd(fscal,dz00);
 575
 576             /* Update vectorial force */
 577             fix0             = _mm256_add_pd(fix0,tx);
 578             fiy0             = _mm256_add_pd(fiy0,ty);
 579             fiz0             = _mm256_add_pd(fiz0,tz);
 580
 581             fjx0             = _mm256_add_pd(fjx0,tx);
 582             fjy0             = _mm256_add_pd(fjy0,ty);
 583             fjz0             = _mm256_add_pd(fjz0,tz);
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r10              = _mm256_mul_pd(rsq10,rinv10);
 590             r10              = _mm256_andnot_pd(dummy_mask,r10);
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             qq10             = _mm256_mul_pd(iq1,jq0);
 594
 595             /* EWALD ELECTROSTATICS */
 596
 597             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 598             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 599             ewitab           = _mm256_cvttpd_epi32(ewrt);
 600             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 601             ewitab           = _mm_slli_epi32(ewitab,2);
 602             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 603             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 604             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 605             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 606             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 607             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 608             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 609             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 610             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm256_andnot_pd(dummy_mask,velec);
 614             velecsum         = _mm256_add_pd(velecsum,velec);
 615
 616             fscal            = felec;
 617
 618             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 619
 620             /* Calculate temporary vectorial force */
 621             tx               = _mm256_mul_pd(fscal,dx10);
 622             ty               = _mm256_mul_pd(fscal,dy10);
 623             tz               = _mm256_mul_pd(fscal,dz10);
 624
 625             /* Update vectorial force */
 626             fix1             = _mm256_add_pd(fix1,tx);
 627             fiy1             = _mm256_add_pd(fiy1,ty);
 628             fiz1             = _mm256_add_pd(fiz1,tz);
 629
 630             fjx0             = _mm256_add_pd(fjx0,tx);
 631             fjy0             = _mm256_add_pd(fjy0,ty);
 632             fjz0             = _mm256_add_pd(fjz0,tz);
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             r20              = _mm256_mul_pd(rsq20,rinv20);
 639             r20              = _mm256_andnot_pd(dummy_mask,r20);
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq20             = _mm256_mul_pd(iq2,jq0);
 643
 644             /* EWALD ELECTROSTATICS */
 645
 646             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 647             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 648             ewitab           = _mm256_cvttpd_epi32(ewrt);
 649             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 650             ewitab           = _mm_slli_epi32(ewitab,2);
 651             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 652             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 653             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 654             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 655             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 656             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 657             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 658             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 659             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 660
 661             /* Update potential sum for this i atom from the interaction with this j atom. */
 662             velec            = _mm256_andnot_pd(dummy_mask,velec);
 663             velecsum         = _mm256_add_pd(velecsum,velec);
 664
 665             fscal            = felec;
 666
 667             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 668
 669             /* Calculate temporary vectorial force */
 670             tx               = _mm256_mul_pd(fscal,dx20);
 671             ty               = _mm256_mul_pd(fscal,dy20);
 672             tz               = _mm256_mul_pd(fscal,dz20);
 673
 674             /* Update vectorial force */
 675             fix2             = _mm256_add_pd(fix2,tx);
 676             fiy2             = _mm256_add_pd(fiy2,ty);
 677             fiz2             = _mm256_add_pd(fiz2,tz);
 678
 679             fjx0             = _mm256_add_pd(fjx0,tx);
 680             fjy0             = _mm256_add_pd(fjy0,ty);
 681             fjz0             = _mm256_add_pd(fjz0,tz);
 682
 683             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 684             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 685             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 686             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 687
 688             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 689
 690             /* Inner loop uses 163 flops */
 691         }
 692
 693         /* End of innermost loop */
 694
 695         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 696                                                  f+i_coord_offset,fshift+i_shift_offset);
 697
 698         ggid                        = gid[iidx];
 699         /* Update potential energies */
 700         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 701         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 702
 703         /* Increment number of inner iterations */
 704         inneriter                  += j_index_end - j_index_start;
 705
 706         /* Outer loop uses 20 flops */
 707     }
 708
 709     /* Increment number of outer iterations */
 710     outeriter        += nri;
 711
 712     /* Update outer/inner flops */
 713
 714     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*163);
 715 }
 716 /*
 717  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_double
 718  * Electrostatics interaction: Ewald
 719  * VdW interaction:            CubicSplineTable
 720  * Geometry:                   Water3-Particle
 721  * Calculate force/pot:        Force
 722  */
 723 void
 724 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_double
 725                     (t_nblist * gmx_restrict                nlist,
 726                      rvec * gmx_restrict                    xx,
 727                      rvec * gmx_restrict                    ff,
 728                      t_forcerec * gmx_restrict              fr,
 729                      t_mdatoms * gmx_restrict               mdatoms,
 730                      nb_kernel_data_t * gmx_restrict        kernel_data,
 731                      t_nrnb * gmx_restrict                  nrnb)
 732 {
 733     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 734      * just 0 for non-waters.
 735      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 736      * jnr indices corresponding to data put in the four positions in the SIMD register.
 737      */
 738     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 739     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 740     int              jnrA,jnrB,jnrC,jnrD;
 741     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 742     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 743     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 744     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 745     real             rcutoff_scalar;
 746     real             *shiftvec,*fshift,*x,*f;
 747     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 748     real             scratch[4*DIM];
 749     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 750     real *           vdwioffsetptr0;
 751     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 752     real *           vdwioffsetptr1;
 753     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 754     real *           vdwioffsetptr2;
 755     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 756     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 757     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 758     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 759     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 760     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 761     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 762     real             *charge;
 763     int              nvdwtype;
 764     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 765     int              *vdwtype;
 766     real             *vdwparam;
 767     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 768     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 769     __m128i          vfitab;
 770     __m128i          ifour       = _mm_set1_epi32(4);
 771     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 772     real             *vftab;
 773     __m128i          ewitab;
 774     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 775     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 776     real             *ewtab;
 777     __m256d          dummy_mask,cutoff_mask;
 778     __m128           tmpmask0,tmpmask1;
 779     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 780     __m256d          one     = _mm256_set1_pd(1.0);
 781     __m256d          two     = _mm256_set1_pd(2.0);
 782     x                = xx[0];
 783     f                = ff[0];
 784
 785     nri              = nlist->nri;
 786     iinr             = nlist->iinr;
 787     jindex           = nlist->jindex;
 788     jjnr             = nlist->jjnr;
 789     shiftidx         = nlist->shift;
 790     gid              = nlist->gid;
 791     shiftvec         = fr->shift_vec[0];
 792     fshift           = fr->fshift[0];
 793     facel            = _mm256_set1_pd(fr->epsfac);
 794     charge           = mdatoms->chargeA;
 795     nvdwtype         = fr->ntype;
 796     vdwparam         = fr->nbfp;
 797     vdwtype          = mdatoms->typeA;
 798
 799     vftab            = kernel_data->table_vdw->data;
 800     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 801
 802     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 803     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 804     beta2            = _mm256_mul_pd(beta,beta);
 805     beta3            = _mm256_mul_pd(beta,beta2);
 806
 807     ewtab            = fr->ic->tabq_coul_F;
 808     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 809     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 810
 811     /* Setup water-specific parameters */
 812     inr              = nlist->iinr[0];
 813     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 814     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 815     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 816     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 817
 818     /* Avoid stupid compiler warnings */
 819     jnrA = jnrB = jnrC = jnrD = 0;
 820     j_coord_offsetA = 0;
 821     j_coord_offsetB = 0;
 822     j_coord_offsetC = 0;
 823     j_coord_offsetD = 0;
 824
 825     outeriter        = 0;
 826     inneriter        = 0;
 827
 828     for(iidx=0;iidx<4*DIM;iidx++)
 829     {
 830         scratch[iidx] = 0.0;
 831     }
 832
 833     /* Start outer loop over neighborlists */
 834     for(iidx=0; iidx<nri; iidx++)
 835     {
 836         /* Load shift vector for this list */
 837         i_shift_offset   = DIM*shiftidx[iidx];
 838
 839         /* Load limits for loop over neighbors */
 840         j_index_start    = jindex[iidx];
 841         j_index_end      = jindex[iidx+1];
 842
 843         /* Get outer coordinate index */
 844         inr              = iinr[iidx];
 845         i_coord_offset   = DIM*inr;
 846
 847         /* Load i particle coords and add shift vector */
 848         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 849                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 850
 851         fix0             = _mm256_setzero_pd();
 852         fiy0             = _mm256_setzero_pd();
 853         fiz0             = _mm256_setzero_pd();
 854         fix1             = _mm256_setzero_pd();
 855         fiy1             = _mm256_setzero_pd();
 856         fiz1             = _mm256_setzero_pd();
 857         fix2             = _mm256_setzero_pd();
 858         fiy2             = _mm256_setzero_pd();
 859         fiz2             = _mm256_setzero_pd();
 860
 861         /* Start inner kernel loop */
 862         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 863         {
 864
 865             /* Get j neighbor index, and coordinate index */
 866             jnrA             = jjnr[jidx];
 867             jnrB             = jjnr[jidx+1];
 868             jnrC             = jjnr[jidx+2];
 869             jnrD             = jjnr[jidx+3];
 870             j_coord_offsetA  = DIM*jnrA;
 871             j_coord_offsetB  = DIM*jnrB;
 872             j_coord_offsetC  = DIM*jnrC;
 873             j_coord_offsetD  = DIM*jnrD;
 874
 875             /* load j atom coordinates */
 876             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 877                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 878                                                  &jx0,&jy0,&jz0);
 879
 880             /* Calculate displacement vector */
 881             dx00             = _mm256_sub_pd(ix0,jx0);
 882             dy00             = _mm256_sub_pd(iy0,jy0);
 883             dz00             = _mm256_sub_pd(iz0,jz0);
 884             dx10             = _mm256_sub_pd(ix1,jx0);
 885             dy10             = _mm256_sub_pd(iy1,jy0);
 886             dz10             = _mm256_sub_pd(iz1,jz0);
 887             dx20             = _mm256_sub_pd(ix2,jx0);
 888             dy20             = _mm256_sub_pd(iy2,jy0);
 889             dz20             = _mm256_sub_pd(iz2,jz0);
 890
 891             /* Calculate squared distance and things based on it */
 892             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 893             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 894             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 895
 896             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 897             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 898             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 899
 900             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 901             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 902             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 903
 904             /* Load parameters for j particles */
 905             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 906                                                                  charge+jnrC+0,charge+jnrD+0);
 907             vdwjidx0A        = 2*vdwtype[jnrA+0];
 908             vdwjidx0B        = 2*vdwtype[jnrB+0];
 909             vdwjidx0C        = 2*vdwtype[jnrC+0];
 910             vdwjidx0D        = 2*vdwtype[jnrD+0];
 911
 912             fjx0             = _mm256_setzero_pd();
 913             fjy0             = _mm256_setzero_pd();
 914             fjz0             = _mm256_setzero_pd();
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r00              = _mm256_mul_pd(rsq00,rinv00);
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             qq00             = _mm256_mul_pd(iq0,jq0);
 924             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 925                                             vdwioffsetptr0+vdwjidx0B,
 926                                             vdwioffsetptr0+vdwjidx0C,
 927                                             vdwioffsetptr0+vdwjidx0D,
 928                                             &c6_00,&c12_00);
 929
 930             /* Calculate table index by multiplying r with table scale and truncate to integer */
 931             rt               = _mm256_mul_pd(r00,vftabscale);
 932             vfitab           = _mm256_cvttpd_epi32(rt);
 933             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 934             vfitab           = _mm_slli_epi32(vfitab,3);
 935
 936             /* EWALD ELECTROSTATICS */
 937
 938             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 939             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 940             ewitab           = _mm256_cvttpd_epi32(ewrt);
 941             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 942             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 943                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 944                                             &ewtabF,&ewtabFn);
 945             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 946             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 947
 948             /* CUBIC SPLINE TABLE DISPERSION */
 949             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 950             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 951             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 952             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 953             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 954             Heps             = _mm256_mul_pd(vfeps,H);
 955             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 956             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 957             fvdw6            = _mm256_mul_pd(c6_00,FF);
 958
 959             /* CUBIC SPLINE TABLE REPULSION */
 960             vfitab           = _mm_add_epi32(vfitab,ifour);
 961             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 962             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 963             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 964             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 965             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 966             Heps             = _mm256_mul_pd(vfeps,H);
 967             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 968             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 969             fvdw12           = _mm256_mul_pd(c12_00,FF);
 970             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 971
 972             fscal            = _mm256_add_pd(felec,fvdw);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm256_mul_pd(fscal,dx00);
 976             ty               = _mm256_mul_pd(fscal,dy00);
 977             tz               = _mm256_mul_pd(fscal,dz00);
 978
 979             /* Update vectorial force */
 980             fix0             = _mm256_add_pd(fix0,tx);
 981             fiy0             = _mm256_add_pd(fiy0,ty);
 982             fiz0             = _mm256_add_pd(fiz0,tz);
 983
 984             fjx0             = _mm256_add_pd(fjx0,tx);
 985             fjy0             = _mm256_add_pd(fjy0,ty);
 986             fjz0             = _mm256_add_pd(fjz0,tz);
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             r10              = _mm256_mul_pd(rsq10,rinv10);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq10             = _mm256_mul_pd(iq1,jq0);
 996
 997             /* EWALD ELECTROSTATICS */
 998
 999             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1001             ewitab           = _mm256_cvttpd_epi32(ewrt);
1002             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1003             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1004                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1005                                             &ewtabF,&ewtabFn);
1006             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1007             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1008
1009             fscal            = felec;
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm256_mul_pd(fscal,dx10);
1013             ty               = _mm256_mul_pd(fscal,dy10);
1014             tz               = _mm256_mul_pd(fscal,dz10);
1015
1016             /* Update vectorial force */
1017             fix1             = _mm256_add_pd(fix1,tx);
1018             fiy1             = _mm256_add_pd(fiy1,ty);
1019             fiz1             = _mm256_add_pd(fiz1,tz);
1020
1021             fjx0             = _mm256_add_pd(fjx0,tx);
1022             fjy0             = _mm256_add_pd(fjy0,ty);
1023             fjz0             = _mm256_add_pd(fjz0,tz);
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             r20              = _mm256_mul_pd(rsq20,rinv20);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq20             = _mm256_mul_pd(iq2,jq0);
1033
1034             /* EWALD ELECTROSTATICS */
1035
1036             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1037             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1038             ewitab           = _mm256_cvttpd_epi32(ewrt);
1039             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1040             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1041                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1042                                             &ewtabF,&ewtabFn);
1043             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1044             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1045
1046             fscal            = felec;
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_pd(fscal,dx20);
1050             ty               = _mm256_mul_pd(fscal,dy20);
1051             tz               = _mm256_mul_pd(fscal,dz20);
1052
1053             /* Update vectorial force */
1054             fix2             = _mm256_add_pd(fix2,tx);
1055             fiy2             = _mm256_add_pd(fiy2,ty);
1056             fiz2             = _mm256_add_pd(fiz2,tz);
1057
1058             fjx0             = _mm256_add_pd(fjx0,tx);
1059             fjy0             = _mm256_add_pd(fjy0,ty);
1060             fjz0             = _mm256_add_pd(fjz0,tz);
1061
1062             fjptrA             = f+j_coord_offsetA;
1063             fjptrB             = f+j_coord_offsetB;
1064             fjptrC             = f+j_coord_offsetC;
1065             fjptrD             = f+j_coord_offsetD;
1066
1067             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1068
1069             /* Inner loop uses 137 flops */
1070         }
1071
1072         if(jidx<j_index_end)
1073         {
1074
1075             /* Get j neighbor index, and coordinate index */
1076             jnrlistA         = jjnr[jidx];
1077             jnrlistB         = jjnr[jidx+1];
1078             jnrlistC         = jjnr[jidx+2];
1079             jnrlistD         = jjnr[jidx+3];
1080             /* Sign of each element will be negative for non-real atoms.
1081              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1082              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1083              */
1084             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1085
1086             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1087             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1088             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1089
1090             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1091             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1092             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1093             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1094             j_coord_offsetA  = DIM*jnrA;
1095             j_coord_offsetB  = DIM*jnrB;
1096             j_coord_offsetC  = DIM*jnrC;
1097             j_coord_offsetD  = DIM*jnrD;
1098
1099             /* load j atom coordinates */
1100             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1101                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1102                                                  &jx0,&jy0,&jz0);
1103
1104             /* Calculate displacement vector */
1105             dx00             = _mm256_sub_pd(ix0,jx0);
1106             dy00             = _mm256_sub_pd(iy0,jy0);
1107             dz00             = _mm256_sub_pd(iz0,jz0);
1108             dx10             = _mm256_sub_pd(ix1,jx0);
1109             dy10             = _mm256_sub_pd(iy1,jy0);
1110             dz10             = _mm256_sub_pd(iz1,jz0);
1111             dx20             = _mm256_sub_pd(ix2,jx0);
1112             dy20             = _mm256_sub_pd(iy2,jy0);
1113             dz20             = _mm256_sub_pd(iz2,jz0);
1114
1115             /* Calculate squared distance and things based on it */
1116             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1117             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1118             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1119
1120             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1121             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1122             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1123
1124             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1125             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1126             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1127
1128             /* Load parameters for j particles */
1129             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1130                                                                  charge+jnrC+0,charge+jnrD+0);
1131             vdwjidx0A        = 2*vdwtype[jnrA+0];
1132             vdwjidx0B        = 2*vdwtype[jnrB+0];
1133             vdwjidx0C        = 2*vdwtype[jnrC+0];
1134             vdwjidx0D        = 2*vdwtype[jnrD+0];
1135
1136             fjx0             = _mm256_setzero_pd();
1137             fjy0             = _mm256_setzero_pd();
1138             fjz0             = _mm256_setzero_pd();
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             r00              = _mm256_mul_pd(rsq00,rinv00);
1145             r00              = _mm256_andnot_pd(dummy_mask,r00);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             qq00             = _mm256_mul_pd(iq0,jq0);
1149             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1150                                             vdwioffsetptr0+vdwjidx0B,
1151                                             vdwioffsetptr0+vdwjidx0C,
1152                                             vdwioffsetptr0+vdwjidx0D,
1153                                             &c6_00,&c12_00);
1154
1155             /* Calculate table index by multiplying r with table scale and truncate to integer */
1156             rt               = _mm256_mul_pd(r00,vftabscale);
1157             vfitab           = _mm256_cvttpd_epi32(rt);
1158             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1159             vfitab           = _mm_slli_epi32(vfitab,3);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1165             ewitab           = _mm256_cvttpd_epi32(ewrt);
1166             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1167             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1168                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1169                                             &ewtabF,&ewtabFn);
1170             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1171             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1172
1173             /* CUBIC SPLINE TABLE DISPERSION */
1174             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1175             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1176             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1177             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1178             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1179             Heps             = _mm256_mul_pd(vfeps,H);
1180             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1181             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1182             fvdw6            = _mm256_mul_pd(c6_00,FF);
1183
1184             /* CUBIC SPLINE TABLE REPULSION */
1185             vfitab           = _mm_add_epi32(vfitab,ifour);
1186             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1187             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1188             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1189             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1190             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1191             Heps             = _mm256_mul_pd(vfeps,H);
1192             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1193             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1194             fvdw12           = _mm256_mul_pd(c12_00,FF);
1195             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1196
1197             fscal            = _mm256_add_pd(felec,fvdw);
1198
1199             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1200
1201             /* Calculate temporary vectorial force */
1202             tx               = _mm256_mul_pd(fscal,dx00);
1203             ty               = _mm256_mul_pd(fscal,dy00);
1204             tz               = _mm256_mul_pd(fscal,dz00);
1205
1206             /* Update vectorial force */
1207             fix0             = _mm256_add_pd(fix0,tx);
1208             fiy0             = _mm256_add_pd(fiy0,ty);
1209             fiz0             = _mm256_add_pd(fiz0,tz);
1210
1211             fjx0             = _mm256_add_pd(fjx0,tx);
1212             fjy0             = _mm256_add_pd(fjy0,ty);
1213             fjz0             = _mm256_add_pd(fjz0,tz);
1214
1215             /**************************
1216              * CALCULATE INTERACTIONS *
1217              **************************/
1218
1219             r10              = _mm256_mul_pd(rsq10,rinv10);
1220             r10              = _mm256_andnot_pd(dummy_mask,r10);
1221
1222             /* Compute parameters for interactions between i and j atoms */
1223             qq10             = _mm256_mul_pd(iq1,jq0);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1229             ewitab           = _mm256_cvttpd_epi32(ewrt);
1230             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1231             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1232                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1233                                             &ewtabF,&ewtabFn);
1234             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1235             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm256_mul_pd(fscal,dx10);
1243             ty               = _mm256_mul_pd(fscal,dy10);
1244             tz               = _mm256_mul_pd(fscal,dz10);
1245
1246             /* Update vectorial force */
1247             fix1             = _mm256_add_pd(fix1,tx);
1248             fiy1             = _mm256_add_pd(fiy1,ty);
1249             fiz1             = _mm256_add_pd(fiz1,tz);
1250
1251             fjx0             = _mm256_add_pd(fjx0,tx);
1252             fjy0             = _mm256_add_pd(fjy0,ty);
1253             fjz0             = _mm256_add_pd(fjz0,tz);
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             r20              = _mm256_mul_pd(rsq20,rinv20);
1260             r20              = _mm256_andnot_pd(dummy_mask,r20);
1261
1262             /* Compute parameters for interactions between i and j atoms */
1263             qq20             = _mm256_mul_pd(iq2,jq0);
1264
1265             /* EWALD ELECTROSTATICS */
1266
1267             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1268             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1269             ewitab           = _mm256_cvttpd_epi32(ewrt);
1270             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1271             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1272                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1273                                             &ewtabF,&ewtabFn);
1274             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1275             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1276
1277             fscal            = felec;
1278
1279             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1280
1281             /* Calculate temporary vectorial force */
1282             tx               = _mm256_mul_pd(fscal,dx20);
1283             ty               = _mm256_mul_pd(fscal,dy20);
1284             tz               = _mm256_mul_pd(fscal,dz20);
1285
1286             /* Update vectorial force */
1287             fix2             = _mm256_add_pd(fix2,tx);
1288             fiy2             = _mm256_add_pd(fiy2,ty);
1289             fiz2             = _mm256_add_pd(fiz2,tz);
1290
1291             fjx0             = _mm256_add_pd(fjx0,tx);
1292             fjy0             = _mm256_add_pd(fjy0,ty);
1293             fjz0             = _mm256_add_pd(fjz0,tz);
1294
1295             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1296             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1297             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1298             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1299
1300             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1301
1302             /* Inner loop uses 140 flops */
1303         }
1304
1305         /* End of innermost loop */
1306
1307         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1308                                                  f+i_coord_offset,fshift+i_shift_offset);
1309
1310         /* Increment number of inner iterations */
1311         inneriter                  += j_index_end - j_index_start;
1312
1313         /* Outer loop uses 18 flops */
1314     }
1315
1316     /* Increment number of outer iterations */
1317     outeriter        += nri;
1318
1319     /* Update outer/inner flops */
1320
1321     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*140);
1322 }