src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  84     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  91     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  92     __m128i          ewitab;
  93     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  96     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     __m256d          dummy_mask,cutoff_mask;
  99     __m128           tmpmask0,tmpmask1;
 100     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 101     __m256d          one     = _mm256_set1_pd(1.0);
 102     __m256d          two     = _mm256_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 121     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 122     beta2            = _mm256_mul_pd(beta,beta);
 123     beta3            = _mm256_mul_pd(beta,beta2);
 124
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 132     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 133     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 134     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 139     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 140
 141     rswitch_scalar   = fr->rcoulomb_switch;
 142     rswitch          = _mm256_set1_pd(rswitch_scalar);
 143     /* Setup switch parameters */
 144     d_scalar         = rcutoff_scalar-rswitch_scalar;
 145     d                = _mm256_set1_pd(d_scalar);
 146     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 147     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 148     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 150     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = jnrC = jnrD = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157     j_coord_offsetC = 0;
 158     j_coord_offsetD = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     for(iidx=0;iidx<4*DIM;iidx++)
 164     {
 165         scratch[iidx] = 0.0;
 166     }
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 185
 186         fix0             = _mm256_setzero_pd();
 187         fiy0             = _mm256_setzero_pd();
 188         fiz0             = _mm256_setzero_pd();
 189         fix1             = _mm256_setzero_pd();
 190         fiy1             = _mm256_setzero_pd();
 191         fiz1             = _mm256_setzero_pd();
 192         fix2             = _mm256_setzero_pd();
 193         fiy2             = _mm256_setzero_pd();
 194         fiz2             = _mm256_setzero_pd();
 195         fix3             = _mm256_setzero_pd();
 196         fiy3             = _mm256_setzero_pd();
 197         fiz3             = _mm256_setzero_pd();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm256_setzero_pd();
 201         vvdwsum          = _mm256_setzero_pd();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             jnrC             = jjnr[jidx+2];
 211             jnrD             = jjnr[jidx+3];
 212             j_coord_offsetA  = DIM*jnrA;
 213             j_coord_offsetB  = DIM*jnrB;
 214             j_coord_offsetC  = DIM*jnrC;
 215             j_coord_offsetD  = DIM*jnrD;
 216
 217             /* load j atom coordinates */
 218             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 219                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 220                                                  &jx0,&jy0,&jz0);
 221
 222             /* Calculate displacement vector */
 223             dx00             = _mm256_sub_pd(ix0,jx0);
 224             dy00             = _mm256_sub_pd(iy0,jy0);
 225             dz00             = _mm256_sub_pd(iz0,jz0);
 226             dx10             = _mm256_sub_pd(ix1,jx0);
 227             dy10             = _mm256_sub_pd(iy1,jy0);
 228             dz10             = _mm256_sub_pd(iz1,jz0);
 229             dx20             = _mm256_sub_pd(ix2,jx0);
 230             dy20             = _mm256_sub_pd(iy2,jy0);
 231             dz20             = _mm256_sub_pd(iz2,jz0);
 232             dx30             = _mm256_sub_pd(ix3,jx0);
 233             dy30             = _mm256_sub_pd(iy3,jy0);
 234             dz30             = _mm256_sub_pd(iz3,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 238             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 239             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 240             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 241
 242             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 243             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 244             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 245             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 246
 247             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 248             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 249             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 250             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 254                                                                  charge+jnrC+0,charge+jnrD+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259
 260             fjx0             = _mm256_setzero_pd();
 261             fjy0             = _mm256_setzero_pd();
 262             fjz0             = _mm256_setzero_pd();
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 269             {
 270
 271             r00              = _mm256_mul_pd(rsq00,rinv00);
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             &c6_00,&c12_00);
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 284             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 285             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 286             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 287
 288             d                = _mm256_sub_pd(r00,rswitch);
 289             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 290             d2               = _mm256_mul_pd(d,d);
 291             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 292
 293             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 294
 295             /* Evaluate switch function */
 296             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 297             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 298             vvdw             = _mm256_mul_pd(vvdw,sw);
 299             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 303             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 304
 305             fscal            = fvdw;
 306
 307             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_pd(fscal,dx00);
 311             ty               = _mm256_mul_pd(fscal,dy00);
 312             tz               = _mm256_mul_pd(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm256_add_pd(fix0,tx);
 316             fiy0             = _mm256_add_pd(fiy0,ty);
 317             fiz0             = _mm256_add_pd(fiz0,tz);
 318
 319             fjx0             = _mm256_add_pd(fjx0,tx);
 320             fjy0             = _mm256_add_pd(fjy0,ty);
 321             fjz0             = _mm256_add_pd(fjz0,tz);
 322
 323             }
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 330             {
 331
 332             r10              = _mm256_mul_pd(rsq10,rinv10);
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq10             = _mm256_mul_pd(iq1,jq0);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 340             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 341             ewitab           = _mm256_cvttpd_epi32(ewrt);
 342             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 343             ewitab           = _mm_slli_epi32(ewitab,2);
 344             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 345             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 346             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 347             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 348             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 349             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 350             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 351             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 352             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 353
 354             d                = _mm256_sub_pd(r10,rswitch);
 355             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 356             d2               = _mm256_mul_pd(d,d);
 357             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 358
 359             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 360
 361             /* Evaluate switch function */
 362             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 363             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 364             velec            = _mm256_mul_pd(velec,sw);
 365             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm256_and_pd(velec,cutoff_mask);
 369             velecsum         = _mm256_add_pd(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm256_mul_pd(fscal,dx10);
 377             ty               = _mm256_mul_pd(fscal,dy10);
 378             tz               = _mm256_mul_pd(fscal,dz10);
 379
 380             /* Update vectorial force */
 381             fix1             = _mm256_add_pd(fix1,tx);
 382             fiy1             = _mm256_add_pd(fiy1,ty);
 383             fiz1             = _mm256_add_pd(fiz1,tz);
 384
 385             fjx0             = _mm256_add_pd(fjx0,tx);
 386             fjy0             = _mm256_add_pd(fjy0,ty);
 387             fjz0             = _mm256_add_pd(fjz0,tz);
 388
 389             }
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 396             {
 397
 398             r20              = _mm256_mul_pd(rsq20,rinv20);
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq20             = _mm256_mul_pd(iq2,jq0);
 402
 403             /* EWALD ELECTROSTATICS */
 404
 405             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 406             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 407             ewitab           = _mm256_cvttpd_epi32(ewrt);
 408             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 409             ewitab           = _mm_slli_epi32(ewitab,2);
 410             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 411             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 412             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 413             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 414             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 415             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 416             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 417             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 418             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 419
 420             d                = _mm256_sub_pd(r20,rswitch);
 421             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 422             d2               = _mm256_mul_pd(d,d);
 423             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 424
 425             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 426
 427             /* Evaluate switch function */
 428             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 429             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 430             velec            = _mm256_mul_pd(velec,sw);
 431             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm256_and_pd(velec,cutoff_mask);
 435             velecsum         = _mm256_add_pd(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_pd(fscal,dx20);
 443             ty               = _mm256_mul_pd(fscal,dy20);
 444             tz               = _mm256_mul_pd(fscal,dz20);
 445
 446             /* Update vectorial force */
 447             fix2             = _mm256_add_pd(fix2,tx);
 448             fiy2             = _mm256_add_pd(fiy2,ty);
 449             fiz2             = _mm256_add_pd(fiz2,tz);
 450
 451             fjx0             = _mm256_add_pd(fjx0,tx);
 452             fjy0             = _mm256_add_pd(fjy0,ty);
 453             fjz0             = _mm256_add_pd(fjz0,tz);
 454
 455             }
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 462             {
 463
 464             r30              = _mm256_mul_pd(rsq30,rinv30);
 465
 466             /* Compute parameters for interactions between i and j atoms */
 467             qq30             = _mm256_mul_pd(iq3,jq0);
 468
 469             /* EWALD ELECTROSTATICS */
 470
 471             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 472             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 473             ewitab           = _mm256_cvttpd_epi32(ewrt);
 474             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 475             ewitab           = _mm_slli_epi32(ewitab,2);
 476             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 477             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 478             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 479             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 480             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 481             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 482             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 483             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 484             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 485
 486             d                = _mm256_sub_pd(r30,rswitch);
 487             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 488             d2               = _mm256_mul_pd(d,d);
 489             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 490
 491             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 492
 493             /* Evaluate switch function */
 494             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 495             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
 496             velec            = _mm256_mul_pd(velec,sw);
 497             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             velec            = _mm256_and_pd(velec,cutoff_mask);
 501             velecsum         = _mm256_add_pd(velecsum,velec);
 502
 503             fscal            = felec;
 504
 505             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 506
 507             /* Calculate temporary vectorial force */
 508             tx               = _mm256_mul_pd(fscal,dx30);
 509             ty               = _mm256_mul_pd(fscal,dy30);
 510             tz               = _mm256_mul_pd(fscal,dz30);
 511
 512             /* Update vectorial force */
 513             fix3             = _mm256_add_pd(fix3,tx);
 514             fiy3             = _mm256_add_pd(fiy3,ty);
 515             fiz3             = _mm256_add_pd(fiz3,tz);
 516
 517             fjx0             = _mm256_add_pd(fjx0,tx);
 518             fjy0             = _mm256_add_pd(fjy0,ty);
 519             fjz0             = _mm256_add_pd(fjz0,tz);
 520
 521             }
 522
 523             fjptrA             = f+j_coord_offsetA;
 524             fjptrB             = f+j_coord_offsetB;
 525             fjptrC             = f+j_coord_offsetC;
 526             fjptrD             = f+j_coord_offsetD;
 527
 528             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 529
 530             /* Inner loop uses 257 flops */
 531         }
 532
 533         if(jidx<j_index_end)
 534         {
 535
 536             /* Get j neighbor index, and coordinate index */
 537             jnrlistA         = jjnr[jidx];
 538             jnrlistB         = jjnr[jidx+1];
 539             jnrlistC         = jjnr[jidx+2];
 540             jnrlistD         = jjnr[jidx+3];
 541             /* Sign of each element will be negative for non-real atoms.
 542              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 543              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 544              */
 545             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 546
 547             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 548             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 549             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 550
 551             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 552             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 553             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 554             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 555             j_coord_offsetA  = DIM*jnrA;
 556             j_coord_offsetB  = DIM*jnrB;
 557             j_coord_offsetC  = DIM*jnrC;
 558             j_coord_offsetD  = DIM*jnrD;
 559
 560             /* load j atom coordinates */
 561             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 562                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 563                                                  &jx0,&jy0,&jz0);
 564
 565             /* Calculate displacement vector */
 566             dx00             = _mm256_sub_pd(ix0,jx0);
 567             dy00             = _mm256_sub_pd(iy0,jy0);
 568             dz00             = _mm256_sub_pd(iz0,jz0);
 569             dx10             = _mm256_sub_pd(ix1,jx0);
 570             dy10             = _mm256_sub_pd(iy1,jy0);
 571             dz10             = _mm256_sub_pd(iz1,jz0);
 572             dx20             = _mm256_sub_pd(ix2,jx0);
 573             dy20             = _mm256_sub_pd(iy2,jy0);
 574             dz20             = _mm256_sub_pd(iz2,jz0);
 575             dx30             = _mm256_sub_pd(ix3,jx0);
 576             dy30             = _mm256_sub_pd(iy3,jy0);
 577             dz30             = _mm256_sub_pd(iz3,jz0);
 578
 579             /* Calculate squared distance and things based on it */
 580             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 581             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 582             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 583             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 584
 585             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 586             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 587             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 588             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 589
 590             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 591             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 592             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 593             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 594
 595             /* Load parameters for j particles */
 596             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 597                                                                  charge+jnrC+0,charge+jnrD+0);
 598             vdwjidx0A        = 2*vdwtype[jnrA+0];
 599             vdwjidx0B        = 2*vdwtype[jnrB+0];
 600             vdwjidx0C        = 2*vdwtype[jnrC+0];
 601             vdwjidx0D        = 2*vdwtype[jnrD+0];
 602
 603             fjx0             = _mm256_setzero_pd();
 604             fjy0             = _mm256_setzero_pd();
 605             fjz0             = _mm256_setzero_pd();
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 612             {
 613
 614             r00              = _mm256_mul_pd(rsq00,rinv00);
 615             r00              = _mm256_andnot_pd(dummy_mask,r00);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 619                                             vdwioffsetptr0+vdwjidx0B,
 620                                             vdwioffsetptr0+vdwjidx0C,
 621                                             vdwioffsetptr0+vdwjidx0D,
 622                                             &c6_00,&c12_00);
 623
 624             /* LENNARD-JONES DISPERSION/REPULSION */
 625
 626             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 627             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 628             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 629             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 630             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 631
 632             d                = _mm256_sub_pd(r00,rswitch);
 633             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 634             d2               = _mm256_mul_pd(d,d);
 635             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 636
 637             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 638
 639             /* Evaluate switch function */
 640             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 641             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 642             vvdw             = _mm256_mul_pd(vvdw,sw);
 643             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 647             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 648             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 649
 650             fscal            = fvdw;
 651
 652             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 653
 654             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm256_mul_pd(fscal,dx00);
 658             ty               = _mm256_mul_pd(fscal,dy00);
 659             tz               = _mm256_mul_pd(fscal,dz00);
 660
 661             /* Update vectorial force */
 662             fix0             = _mm256_add_pd(fix0,tx);
 663             fiy0             = _mm256_add_pd(fiy0,ty);
 664             fiz0             = _mm256_add_pd(fiz0,tz);
 665
 666             fjx0             = _mm256_add_pd(fjx0,tx);
 667             fjy0             = _mm256_add_pd(fjy0,ty);
 668             fjz0             = _mm256_add_pd(fjz0,tz);
 669
 670             }
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 677             {
 678
 679             r10              = _mm256_mul_pd(rsq10,rinv10);
 680             r10              = _mm256_andnot_pd(dummy_mask,r10);
 681
 682             /* Compute parameters for interactions between i and j atoms */
 683             qq10             = _mm256_mul_pd(iq1,jq0);
 684
 685             /* EWALD ELECTROSTATICS */
 686
 687             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 688             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 689             ewitab           = _mm256_cvttpd_epi32(ewrt);
 690             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 691             ewitab           = _mm_slli_epi32(ewitab,2);
 692             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 693             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 694             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 695             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 696             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 697             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 698             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 699             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 700             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 701
 702             d                = _mm256_sub_pd(r10,rswitch);
 703             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 704             d2               = _mm256_mul_pd(d,d);
 705             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 706
 707             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 708
 709             /* Evaluate switch function */
 710             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 711             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 712             velec            = _mm256_mul_pd(velec,sw);
 713             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 714
 715             /* Update potential sum for this i atom from the interaction with this j atom. */
 716             velec            = _mm256_and_pd(velec,cutoff_mask);
 717             velec            = _mm256_andnot_pd(dummy_mask,velec);
 718             velecsum         = _mm256_add_pd(velecsum,velec);
 719
 720             fscal            = felec;
 721
 722             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 723
 724             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 725
 726             /* Calculate temporary vectorial force */
 727             tx               = _mm256_mul_pd(fscal,dx10);
 728             ty               = _mm256_mul_pd(fscal,dy10);
 729             tz               = _mm256_mul_pd(fscal,dz10);
 730
 731             /* Update vectorial force */
 732             fix1             = _mm256_add_pd(fix1,tx);
 733             fiy1             = _mm256_add_pd(fiy1,ty);
 734             fiz1             = _mm256_add_pd(fiz1,tz);
 735
 736             fjx0             = _mm256_add_pd(fjx0,tx);
 737             fjy0             = _mm256_add_pd(fjy0,ty);
 738             fjz0             = _mm256_add_pd(fjz0,tz);
 739
 740             }
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 747             {
 748
 749             r20              = _mm256_mul_pd(rsq20,rinv20);
 750             r20              = _mm256_andnot_pd(dummy_mask,r20);
 751
 752             /* Compute parameters for interactions between i and j atoms */
 753             qq20             = _mm256_mul_pd(iq2,jq0);
 754
 755             /* EWALD ELECTROSTATICS */
 756
 757             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 758             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 759             ewitab           = _mm256_cvttpd_epi32(ewrt);
 760             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 761             ewitab           = _mm_slli_epi32(ewitab,2);
 762             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 763             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 764             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 765             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 766             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 767             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 768             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 769             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 770             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 771
 772             d                = _mm256_sub_pd(r20,rswitch);
 773             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 774             d2               = _mm256_mul_pd(d,d);
 775             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 776
 777             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 778
 779             /* Evaluate switch function */
 780             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 781             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 782             velec            = _mm256_mul_pd(velec,sw);
 783             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 784
 785             /* Update potential sum for this i atom from the interaction with this j atom. */
 786             velec            = _mm256_and_pd(velec,cutoff_mask);
 787             velec            = _mm256_andnot_pd(dummy_mask,velec);
 788             velecsum         = _mm256_add_pd(velecsum,velec);
 789
 790             fscal            = felec;
 791
 792             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 793
 794             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 795
 796             /* Calculate temporary vectorial force */
 797             tx               = _mm256_mul_pd(fscal,dx20);
 798             ty               = _mm256_mul_pd(fscal,dy20);
 799             tz               = _mm256_mul_pd(fscal,dz20);
 800
 801             /* Update vectorial force */
 802             fix2             = _mm256_add_pd(fix2,tx);
 803             fiy2             = _mm256_add_pd(fiy2,ty);
 804             fiz2             = _mm256_add_pd(fiz2,tz);
 805
 806             fjx0             = _mm256_add_pd(fjx0,tx);
 807             fjy0             = _mm256_add_pd(fjy0,ty);
 808             fjz0             = _mm256_add_pd(fjz0,tz);
 809
 810             }
 811
 812             /**************************
 813              * CALCULATE INTERACTIONS *
 814              **************************/
 815
 816             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 817             {
 818
 819             r30              = _mm256_mul_pd(rsq30,rinv30);
 820             r30              = _mm256_andnot_pd(dummy_mask,r30);
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq30             = _mm256_mul_pd(iq3,jq0);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 828             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 829             ewitab           = _mm256_cvttpd_epi32(ewrt);
 830             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 831             ewitab           = _mm_slli_epi32(ewitab,2);
 832             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 833             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 834             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 835             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 836             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 837             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 838             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 839             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 840             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 841
 842             d                = _mm256_sub_pd(r30,rswitch);
 843             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 844             d2               = _mm256_mul_pd(d,d);
 845             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 846
 847             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 848
 849             /* Evaluate switch function */
 850             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 851             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
 852             velec            = _mm256_mul_pd(velec,sw);
 853             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 854
 855             /* Update potential sum for this i atom from the interaction with this j atom. */
 856             velec            = _mm256_and_pd(velec,cutoff_mask);
 857             velec            = _mm256_andnot_pd(dummy_mask,velec);
 858             velecsum         = _mm256_add_pd(velecsum,velec);
 859
 860             fscal            = felec;
 861
 862             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 863
 864             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm256_mul_pd(fscal,dx30);
 868             ty               = _mm256_mul_pd(fscal,dy30);
 869             tz               = _mm256_mul_pd(fscal,dz30);
 870
 871             /* Update vectorial force */
 872             fix3             = _mm256_add_pd(fix3,tx);
 873             fiy3             = _mm256_add_pd(fiy3,ty);
 874             fiz3             = _mm256_add_pd(fiz3,tz);
 875
 876             fjx0             = _mm256_add_pd(fjx0,tx);
 877             fjy0             = _mm256_add_pd(fjy0,ty);
 878             fjz0             = _mm256_add_pd(fjz0,tz);
 879
 880             }
 881
 882             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 883             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 884             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 885             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 886
 887             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 888
 889             /* Inner loop uses 261 flops */
 890         }
 891
 892         /* End of innermost loop */
 893
 894         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 895                                                  f+i_coord_offset,fshift+i_shift_offset);
 896
 897         ggid                        = gid[iidx];
 898         /* Update potential energies */
 899         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 900         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 901
 902         /* Increment number of inner iterations */
 903         inneriter                  += j_index_end - j_index_start;
 904
 905         /* Outer loop uses 26 flops */
 906     }
 907
 908     /* Increment number of outer iterations */
 909     outeriter        += nri;
 910
 911     /* Update outer/inner flops */
 912
 913     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*261);
 914 }
 915 /*
 916  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_256_double
 917  * Electrostatics interaction: Ewald
 918  * VdW interaction:            LennardJones
 919  * Geometry:                   Water4-Particle
 920  * Calculate force/pot:        Force
 921  */
 922 void
 923 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_256_double
 924                     (t_nblist * gmx_restrict                nlist,
 925                      rvec * gmx_restrict                    xx,
 926                      rvec * gmx_restrict                    ff,
 927                      t_forcerec * gmx_restrict              fr,
 928                      t_mdatoms * gmx_restrict               mdatoms,
 929                      nb_kernel_data_t * gmx_restrict        kernel_data,
 930                      t_nrnb * gmx_restrict                  nrnb)
 931 {
 932     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 933      * just 0 for non-waters.
 934      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 935      * jnr indices corresponding to data put in the four positions in the SIMD register.
 936      */
 937     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 938     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 939     int              jnrA,jnrB,jnrC,jnrD;
 940     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 941     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 942     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 943     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 944     real             rcutoff_scalar;
 945     real             *shiftvec,*fshift,*x,*f;
 946     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 947     real             scratch[4*DIM];
 948     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 949     real *           vdwioffsetptr0;
 950     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 951     real *           vdwioffsetptr1;
 952     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 953     real *           vdwioffsetptr2;
 954     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 955     real *           vdwioffsetptr3;
 956     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 957     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 958     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 959     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 960     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 961     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 962     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 963     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 964     real             *charge;
 965     int              nvdwtype;
 966     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 967     int              *vdwtype;
 968     real             *vdwparam;
 969     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 970     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 971     __m128i          ewitab;
 972     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 973     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 974     real             *ewtab;
 975     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 976     real             rswitch_scalar,d_scalar;
 977     __m256d          dummy_mask,cutoff_mask;
 978     __m128           tmpmask0,tmpmask1;
 979     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 980     __m256d          one     = _mm256_set1_pd(1.0);
 981     __m256d          two     = _mm256_set1_pd(2.0);
 982     x                = xx[0];
 983     f                = ff[0];
 984
 985     nri              = nlist->nri;
 986     iinr             = nlist->iinr;
 987     jindex           = nlist->jindex;
 988     jjnr             = nlist->jjnr;
 989     shiftidx         = nlist->shift;
 990     gid              = nlist->gid;
 991     shiftvec         = fr->shift_vec[0];
 992     fshift           = fr->fshift[0];
 993     facel            = _mm256_set1_pd(fr->epsfac);
 994     charge           = mdatoms->chargeA;
 995     nvdwtype         = fr->ntype;
 996     vdwparam         = fr->nbfp;
 997     vdwtype          = mdatoms->typeA;
 998
 999     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1000     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
1001     beta2            = _mm256_mul_pd(beta,beta);
1002     beta3            = _mm256_mul_pd(beta,beta2);
1003
1004     ewtab            = fr->ic->tabq_coul_FDV0;
1005     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1006     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1007
1008     /* Setup water-specific parameters */
1009     inr              = nlist->iinr[0];
1010     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1011     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1012     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1013     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1014
1015     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1016     rcutoff_scalar   = fr->rcoulomb;
1017     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1018     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1019
1020     rswitch_scalar   = fr->rcoulomb_switch;
1021     rswitch          = _mm256_set1_pd(rswitch_scalar);
1022     /* Setup switch parameters */
1023     d_scalar         = rcutoff_scalar-rswitch_scalar;
1024     d                = _mm256_set1_pd(d_scalar);
1025     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1026     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1027     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1028     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1029     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1030     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1031
1032     /* Avoid stupid compiler warnings */
1033     jnrA = jnrB = jnrC = jnrD = 0;
1034     j_coord_offsetA = 0;
1035     j_coord_offsetB = 0;
1036     j_coord_offsetC = 0;
1037     j_coord_offsetD = 0;
1038
1039     outeriter        = 0;
1040     inneriter        = 0;
1041
1042     for(iidx=0;iidx<4*DIM;iidx++)
1043     {
1044         scratch[iidx] = 0.0;
1045     }
1046
1047     /* Start outer loop over neighborlists */
1048     for(iidx=0; iidx<nri; iidx++)
1049     {
1050         /* Load shift vector for this list */
1051         i_shift_offset   = DIM*shiftidx[iidx];
1052
1053         /* Load limits for loop over neighbors */
1054         j_index_start    = jindex[iidx];
1055         j_index_end      = jindex[iidx+1];
1056
1057         /* Get outer coordinate index */
1058         inr              = iinr[iidx];
1059         i_coord_offset   = DIM*inr;
1060
1061         /* Load i particle coords and add shift vector */
1062         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1063                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1064
1065         fix0             = _mm256_setzero_pd();
1066         fiy0             = _mm256_setzero_pd();
1067         fiz0             = _mm256_setzero_pd();
1068         fix1             = _mm256_setzero_pd();
1069         fiy1             = _mm256_setzero_pd();
1070         fiz1             = _mm256_setzero_pd();
1071         fix2             = _mm256_setzero_pd();
1072         fiy2             = _mm256_setzero_pd();
1073         fiz2             = _mm256_setzero_pd();
1074         fix3             = _mm256_setzero_pd();
1075         fiy3             = _mm256_setzero_pd();
1076         fiz3             = _mm256_setzero_pd();
1077
1078         /* Start inner kernel loop */
1079         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1080         {
1081
1082             /* Get j neighbor index, and coordinate index */
1083             jnrA             = jjnr[jidx];
1084             jnrB             = jjnr[jidx+1];
1085             jnrC             = jjnr[jidx+2];
1086             jnrD             = jjnr[jidx+3];
1087             j_coord_offsetA  = DIM*jnrA;
1088             j_coord_offsetB  = DIM*jnrB;
1089             j_coord_offsetC  = DIM*jnrC;
1090             j_coord_offsetD  = DIM*jnrD;
1091
1092             /* load j atom coordinates */
1093             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1094                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1095                                                  &jx0,&jy0,&jz0);
1096
1097             /* Calculate displacement vector */
1098             dx00             = _mm256_sub_pd(ix0,jx0);
1099             dy00             = _mm256_sub_pd(iy0,jy0);
1100             dz00             = _mm256_sub_pd(iz0,jz0);
1101             dx10             = _mm256_sub_pd(ix1,jx0);
1102             dy10             = _mm256_sub_pd(iy1,jy0);
1103             dz10             = _mm256_sub_pd(iz1,jz0);
1104             dx20             = _mm256_sub_pd(ix2,jx0);
1105             dy20             = _mm256_sub_pd(iy2,jy0);
1106             dz20             = _mm256_sub_pd(iz2,jz0);
1107             dx30             = _mm256_sub_pd(ix3,jx0);
1108             dy30             = _mm256_sub_pd(iy3,jy0);
1109             dz30             = _mm256_sub_pd(iz3,jz0);
1110
1111             /* Calculate squared distance and things based on it */
1112             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1113             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1114             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1115             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1116
1117             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1118             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1119             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1120             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1121
1122             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1123             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1124             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1125             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1126
1127             /* Load parameters for j particles */
1128             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1129                                                                  charge+jnrC+0,charge+jnrD+0);
1130             vdwjidx0A        = 2*vdwtype[jnrA+0];
1131             vdwjidx0B        = 2*vdwtype[jnrB+0];
1132             vdwjidx0C        = 2*vdwtype[jnrC+0];
1133             vdwjidx0D        = 2*vdwtype[jnrD+0];
1134
1135             fjx0             = _mm256_setzero_pd();
1136             fjy0             = _mm256_setzero_pd();
1137             fjz0             = _mm256_setzero_pd();
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1144             {
1145
1146             r00              = _mm256_mul_pd(rsq00,rinv00);
1147
1148             /* Compute parameters for interactions between i and j atoms */
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             /* LENNARD-JONES DISPERSION/REPULSION */
1156
1157             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1158             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1159             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1160             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1161             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1162
1163             d                = _mm256_sub_pd(r00,rswitch);
1164             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1165             d2               = _mm256_mul_pd(d,d);
1166             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1167
1168             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1169
1170             /* Evaluate switch function */
1171             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1172             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1173             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1174
1175             fscal            = fvdw;
1176
1177             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_pd(fscal,dx00);
1181             ty               = _mm256_mul_pd(fscal,dy00);
1182             tz               = _mm256_mul_pd(fscal,dz00);
1183
1184             /* Update vectorial force */
1185             fix0             = _mm256_add_pd(fix0,tx);
1186             fiy0             = _mm256_add_pd(fiy0,ty);
1187             fiz0             = _mm256_add_pd(fiz0,tz);
1188
1189             fjx0             = _mm256_add_pd(fjx0,tx);
1190             fjy0             = _mm256_add_pd(fjy0,ty);
1191             fjz0             = _mm256_add_pd(fjz0,tz);
1192
1193             }
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1200             {
1201
1202             r10              = _mm256_mul_pd(rsq10,rinv10);
1203
1204             /* Compute parameters for interactions between i and j atoms */
1205             qq10             = _mm256_mul_pd(iq1,jq0);
1206
1207             /* EWALD ELECTROSTATICS */
1208
1209             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1210             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1211             ewitab           = _mm256_cvttpd_epi32(ewrt);
1212             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1213             ewitab           = _mm_slli_epi32(ewitab,2);
1214             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1215             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1216             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1217             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1218             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1219             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1220             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1221             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1222             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1223
1224             d                = _mm256_sub_pd(r10,rswitch);
1225             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1226             d2               = _mm256_mul_pd(d,d);
1227             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1228
1229             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1230
1231             /* Evaluate switch function */
1232             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1233             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1234             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1235
1236             fscal            = felec;
1237
1238             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1239
1240             /* Calculate temporary vectorial force */
1241             tx               = _mm256_mul_pd(fscal,dx10);
1242             ty               = _mm256_mul_pd(fscal,dy10);
1243             tz               = _mm256_mul_pd(fscal,dz10);
1244
1245             /* Update vectorial force */
1246             fix1             = _mm256_add_pd(fix1,tx);
1247             fiy1             = _mm256_add_pd(fiy1,ty);
1248             fiz1             = _mm256_add_pd(fiz1,tz);
1249
1250             fjx0             = _mm256_add_pd(fjx0,tx);
1251             fjy0             = _mm256_add_pd(fjy0,ty);
1252             fjz0             = _mm256_add_pd(fjz0,tz);
1253
1254             }
1255
1256             /**************************
1257              * CALCULATE INTERACTIONS *
1258              **************************/
1259
1260             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1261             {
1262
1263             r20              = _mm256_mul_pd(rsq20,rinv20);
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             qq20             = _mm256_mul_pd(iq2,jq0);
1267
1268             /* EWALD ELECTROSTATICS */
1269
1270             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1271             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1272             ewitab           = _mm256_cvttpd_epi32(ewrt);
1273             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1274             ewitab           = _mm_slli_epi32(ewitab,2);
1275             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1276             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1277             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1278             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1279             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1280             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1281             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1282             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1283             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1284
1285             d                = _mm256_sub_pd(r20,rswitch);
1286             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1287             d2               = _mm256_mul_pd(d,d);
1288             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1289
1290             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1291
1292             /* Evaluate switch function */
1293             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1294             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1295             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1296
1297             fscal            = felec;
1298
1299             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1300
1301             /* Calculate temporary vectorial force */
1302             tx               = _mm256_mul_pd(fscal,dx20);
1303             ty               = _mm256_mul_pd(fscal,dy20);
1304             tz               = _mm256_mul_pd(fscal,dz20);
1305
1306             /* Update vectorial force */
1307             fix2             = _mm256_add_pd(fix2,tx);
1308             fiy2             = _mm256_add_pd(fiy2,ty);
1309             fiz2             = _mm256_add_pd(fiz2,tz);
1310
1311             fjx0             = _mm256_add_pd(fjx0,tx);
1312             fjy0             = _mm256_add_pd(fjy0,ty);
1313             fjz0             = _mm256_add_pd(fjz0,tz);
1314
1315             }
1316
1317             /**************************
1318              * CALCULATE INTERACTIONS *
1319              **************************/
1320
1321             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1322             {
1323
1324             r30              = _mm256_mul_pd(rsq30,rinv30);
1325
1326             /* Compute parameters for interactions between i and j atoms */
1327             qq30             = _mm256_mul_pd(iq3,jq0);
1328
1329             /* EWALD ELECTROSTATICS */
1330
1331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1332             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1333             ewitab           = _mm256_cvttpd_epi32(ewrt);
1334             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1335             ewitab           = _mm_slli_epi32(ewitab,2);
1336             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1337             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1338             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1339             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1340             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1341             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1342             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1343             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
1344             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1345
1346             d                = _mm256_sub_pd(r30,rswitch);
1347             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1348             d2               = _mm256_mul_pd(d,d);
1349             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1350
1351             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1352
1353             /* Evaluate switch function */
1354             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1355             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
1356             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1357
1358             fscal            = felec;
1359
1360             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1361
1362             /* Calculate temporary vectorial force */
1363             tx               = _mm256_mul_pd(fscal,dx30);
1364             ty               = _mm256_mul_pd(fscal,dy30);
1365             tz               = _mm256_mul_pd(fscal,dz30);
1366
1367             /* Update vectorial force */
1368             fix3             = _mm256_add_pd(fix3,tx);
1369             fiy3             = _mm256_add_pd(fiy3,ty);
1370             fiz3             = _mm256_add_pd(fiz3,tz);
1371
1372             fjx0             = _mm256_add_pd(fjx0,tx);
1373             fjy0             = _mm256_add_pd(fjy0,ty);
1374             fjz0             = _mm256_add_pd(fjz0,tz);
1375
1376             }
1377
1378             fjptrA             = f+j_coord_offsetA;
1379             fjptrB             = f+j_coord_offsetB;
1380             fjptrC             = f+j_coord_offsetC;
1381             fjptrD             = f+j_coord_offsetD;
1382
1383             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1384
1385             /* Inner loop uses 245 flops */
1386         }
1387
1388         if(jidx<j_index_end)
1389         {
1390
1391             /* Get j neighbor index, and coordinate index */
1392             jnrlistA         = jjnr[jidx];
1393             jnrlistB         = jjnr[jidx+1];
1394             jnrlistC         = jjnr[jidx+2];
1395             jnrlistD         = jjnr[jidx+3];
1396             /* Sign of each element will be negative for non-real atoms.
1397              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1398              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1399              */
1400             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1401
1402             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1403             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1404             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1405
1406             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1407             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1408             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1409             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1410             j_coord_offsetA  = DIM*jnrA;
1411             j_coord_offsetB  = DIM*jnrB;
1412             j_coord_offsetC  = DIM*jnrC;
1413             j_coord_offsetD  = DIM*jnrD;
1414
1415             /* load j atom coordinates */
1416             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1417                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1418                                                  &jx0,&jy0,&jz0);
1419
1420             /* Calculate displacement vector */
1421             dx00             = _mm256_sub_pd(ix0,jx0);
1422             dy00             = _mm256_sub_pd(iy0,jy0);
1423             dz00             = _mm256_sub_pd(iz0,jz0);
1424             dx10             = _mm256_sub_pd(ix1,jx0);
1425             dy10             = _mm256_sub_pd(iy1,jy0);
1426             dz10             = _mm256_sub_pd(iz1,jz0);
1427             dx20             = _mm256_sub_pd(ix2,jx0);
1428             dy20             = _mm256_sub_pd(iy2,jy0);
1429             dz20             = _mm256_sub_pd(iz2,jz0);
1430             dx30             = _mm256_sub_pd(ix3,jx0);
1431             dy30             = _mm256_sub_pd(iy3,jy0);
1432             dz30             = _mm256_sub_pd(iz3,jz0);
1433
1434             /* Calculate squared distance and things based on it */
1435             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1436             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1437             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1438             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1439
1440             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1441             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1442             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1443             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1444
1445             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1446             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1447             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1448             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1449
1450             /* Load parameters for j particles */
1451             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1452                                                                  charge+jnrC+0,charge+jnrD+0);
1453             vdwjidx0A        = 2*vdwtype[jnrA+0];
1454             vdwjidx0B        = 2*vdwtype[jnrB+0];
1455             vdwjidx0C        = 2*vdwtype[jnrC+0];
1456             vdwjidx0D        = 2*vdwtype[jnrD+0];
1457
1458             fjx0             = _mm256_setzero_pd();
1459             fjy0             = _mm256_setzero_pd();
1460             fjz0             = _mm256_setzero_pd();
1461
1462             /**************************
1463              * CALCULATE INTERACTIONS *
1464              **************************/
1465
1466             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1467             {
1468
1469             r00              = _mm256_mul_pd(rsq00,rinv00);
1470             r00              = _mm256_andnot_pd(dummy_mask,r00);
1471
1472             /* Compute parameters for interactions between i and j atoms */
1473             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1474                                             vdwioffsetptr0+vdwjidx0B,
1475                                             vdwioffsetptr0+vdwjidx0C,
1476                                             vdwioffsetptr0+vdwjidx0D,
1477                                             &c6_00,&c12_00);
1478
1479             /* LENNARD-JONES DISPERSION/REPULSION */
1480
1481             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1482             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1483             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1484             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1485             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1486
1487             d                = _mm256_sub_pd(r00,rswitch);
1488             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1489             d2               = _mm256_mul_pd(d,d);
1490             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1491
1492             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1493
1494             /* Evaluate switch function */
1495             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1496             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1497             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1498
1499             fscal            = fvdw;
1500
1501             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1502
1503             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1504
1505             /* Calculate temporary vectorial force */
1506             tx               = _mm256_mul_pd(fscal,dx00);
1507             ty               = _mm256_mul_pd(fscal,dy00);
1508             tz               = _mm256_mul_pd(fscal,dz00);
1509
1510             /* Update vectorial force */
1511             fix0             = _mm256_add_pd(fix0,tx);
1512             fiy0             = _mm256_add_pd(fiy0,ty);
1513             fiz0             = _mm256_add_pd(fiz0,tz);
1514
1515             fjx0             = _mm256_add_pd(fjx0,tx);
1516             fjy0             = _mm256_add_pd(fjy0,ty);
1517             fjz0             = _mm256_add_pd(fjz0,tz);
1518
1519             }
1520
1521             /**************************
1522              * CALCULATE INTERACTIONS *
1523              **************************/
1524
1525             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1526             {
1527
1528             r10              = _mm256_mul_pd(rsq10,rinv10);
1529             r10              = _mm256_andnot_pd(dummy_mask,r10);
1530
1531             /* Compute parameters for interactions between i and j atoms */
1532             qq10             = _mm256_mul_pd(iq1,jq0);
1533
1534             /* EWALD ELECTROSTATICS */
1535
1536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1537             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1538             ewitab           = _mm256_cvttpd_epi32(ewrt);
1539             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1540             ewitab           = _mm_slli_epi32(ewitab,2);
1541             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1542             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1543             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1544             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1545             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1546             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1547             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1548             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1549             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1550
1551             d                = _mm256_sub_pd(r10,rswitch);
1552             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1553             d2               = _mm256_mul_pd(d,d);
1554             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1555
1556             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1557
1558             /* Evaluate switch function */
1559             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1560             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1561             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1562
1563             fscal            = felec;
1564
1565             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1566
1567             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1568
1569             /* Calculate temporary vectorial force */
1570             tx               = _mm256_mul_pd(fscal,dx10);
1571             ty               = _mm256_mul_pd(fscal,dy10);
1572             tz               = _mm256_mul_pd(fscal,dz10);
1573
1574             /* Update vectorial force */
1575             fix1             = _mm256_add_pd(fix1,tx);
1576             fiy1             = _mm256_add_pd(fiy1,ty);
1577             fiz1             = _mm256_add_pd(fiz1,tz);
1578
1579             fjx0             = _mm256_add_pd(fjx0,tx);
1580             fjy0             = _mm256_add_pd(fjy0,ty);
1581             fjz0             = _mm256_add_pd(fjz0,tz);
1582
1583             }
1584
1585             /**************************
1586              * CALCULATE INTERACTIONS *
1587              **************************/
1588
1589             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1590             {
1591
1592             r20              = _mm256_mul_pd(rsq20,rinv20);
1593             r20              = _mm256_andnot_pd(dummy_mask,r20);
1594
1595             /* Compute parameters for interactions between i and j atoms */
1596             qq20             = _mm256_mul_pd(iq2,jq0);
1597
1598             /* EWALD ELECTROSTATICS */
1599
1600             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1601             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1602             ewitab           = _mm256_cvttpd_epi32(ewrt);
1603             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1604             ewitab           = _mm_slli_epi32(ewitab,2);
1605             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1606             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1607             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1608             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1609             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1610             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1611             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1612             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1613             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1614
1615             d                = _mm256_sub_pd(r20,rswitch);
1616             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1617             d2               = _mm256_mul_pd(d,d);
1618             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1619
1620             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1621
1622             /* Evaluate switch function */
1623             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1624             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1625             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1626
1627             fscal            = felec;
1628
1629             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1630
1631             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1632
1633             /* Calculate temporary vectorial force */
1634             tx               = _mm256_mul_pd(fscal,dx20);
1635             ty               = _mm256_mul_pd(fscal,dy20);
1636             tz               = _mm256_mul_pd(fscal,dz20);
1637
1638             /* Update vectorial force */
1639             fix2             = _mm256_add_pd(fix2,tx);
1640             fiy2             = _mm256_add_pd(fiy2,ty);
1641             fiz2             = _mm256_add_pd(fiz2,tz);
1642
1643             fjx0             = _mm256_add_pd(fjx0,tx);
1644             fjy0             = _mm256_add_pd(fjy0,ty);
1645             fjz0             = _mm256_add_pd(fjz0,tz);
1646
1647             }
1648
1649             /**************************
1650              * CALCULATE INTERACTIONS *
1651              **************************/
1652
1653             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1654             {
1655
1656             r30              = _mm256_mul_pd(rsq30,rinv30);
1657             r30              = _mm256_andnot_pd(dummy_mask,r30);
1658
1659             /* Compute parameters for interactions between i and j atoms */
1660             qq30             = _mm256_mul_pd(iq3,jq0);
1661
1662             /* EWALD ELECTROSTATICS */
1663
1664             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1665             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1666             ewitab           = _mm256_cvttpd_epi32(ewrt);
1667             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1668             ewitab           = _mm_slli_epi32(ewitab,2);
1669             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1670             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1671             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1672             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1673             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1674             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1675             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1676             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
1677             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1678
1679             d                = _mm256_sub_pd(r30,rswitch);
1680             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1681             d2               = _mm256_mul_pd(d,d);
1682             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1683
1684             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1685
1686             /* Evaluate switch function */
1687             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1688             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
1689             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1690
1691             fscal            = felec;
1692
1693             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1694
1695             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1696
1697             /* Calculate temporary vectorial force */
1698             tx               = _mm256_mul_pd(fscal,dx30);
1699             ty               = _mm256_mul_pd(fscal,dy30);
1700             tz               = _mm256_mul_pd(fscal,dz30);
1701
1702             /* Update vectorial force */
1703             fix3             = _mm256_add_pd(fix3,tx);
1704             fiy3             = _mm256_add_pd(fiy3,ty);
1705             fiz3             = _mm256_add_pd(fiz3,tz);
1706
1707             fjx0             = _mm256_add_pd(fjx0,tx);
1708             fjy0             = _mm256_add_pd(fjy0,ty);
1709             fjz0             = _mm256_add_pd(fjz0,tz);
1710
1711             }
1712
1713             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1714             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1715             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1716             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1717
1718             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1719
1720             /* Inner loop uses 249 flops */
1721         }
1722
1723         /* End of innermost loop */
1724
1725         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1726                                                  f+i_coord_offset,fshift+i_shift_offset);
1727
1728         /* Increment number of inner iterations */
1729         inneriter                  += j_index_end - j_index_start;
1730
1731         /* Outer loop uses 24 flops */
1732     }
1733
1734     /* Increment number of outer iterations */
1735     outeriter        += nri;
1736
1737     /* Update outer/inner flops */
1738
1739     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*249);
1740 }