src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  84     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  86     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  93     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  94     __m256i          ewitab;
  95     __m128i          ewitab_lo,ewitab_hi;
  96     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  98     real             *ewtab;
  99     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(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_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 121     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 122     beta2            = _mm256_mul_ps(beta,beta);
 123     beta3            = _mm256_mul_ps(beta,beta2);
 124
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 132     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 133     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(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_ps(rcutoff_scalar);
 139     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 140
 141     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 142     rvdw             = _mm256_set1_ps(fr->rvdw);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_ps();
 182         fiy0             = _mm256_setzero_ps();
 183         fiz0             = _mm256_setzero_ps();
 184         fix1             = _mm256_setzero_ps();
 185         fiy1             = _mm256_setzero_ps();
 186         fiz1             = _mm256_setzero_ps();
 187         fix2             = _mm256_setzero_ps();
 188         fiy2             = _mm256_setzero_ps();
 189         fiz2             = _mm256_setzero_ps();
 190         fix3             = _mm256_setzero_ps();
 191         fiy3             = _mm256_setzero_ps();
 192         fiz3             = _mm256_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_ps();
 196         vvdwsum          = _mm256_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             jnrE             = jjnr[jidx+4];
 208             jnrF             = jjnr[jidx+5];
 209             jnrG             = jjnr[jidx+6];
 210             jnrH             = jjnr[jidx+7];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215             j_coord_offsetE  = DIM*jnrE;
 216             j_coord_offsetF  = DIM*jnrF;
 217             j_coord_offsetG  = DIM*jnrG;
 218             j_coord_offsetH  = DIM*jnrH;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 224                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_ps(ix0,jx0);
 229             dy00             = _mm256_sub_ps(iy0,jy0);
 230             dz00             = _mm256_sub_ps(iz0,jz0);
 231             dx10             = _mm256_sub_ps(ix1,jx0);
 232             dy10             = _mm256_sub_ps(iy1,jy0);
 233             dz10             = _mm256_sub_ps(iz1,jz0);
 234             dx20             = _mm256_sub_ps(ix2,jx0);
 235             dy20             = _mm256_sub_ps(iy2,jy0);
 236             dz20             = _mm256_sub_ps(iz2,jz0);
 237             dx30             = _mm256_sub_ps(ix3,jx0);
 238             dy30             = _mm256_sub_ps(iy3,jy0);
 239             dz30             = _mm256_sub_ps(iz3,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 245             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 246
 247             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 248             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 249             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 250
 251             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 252             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 253             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 254             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0,
 259                                                                  charge+jnrE+0,charge+jnrF+0,
 260                                                                  charge+jnrG+0,charge+jnrH+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265             vdwjidx0E        = 2*vdwtype[jnrE+0];
 266             vdwjidx0F        = 2*vdwtype[jnrF+0];
 267             vdwjidx0G        = 2*vdwtype[jnrG+0];
 268             vdwjidx0H        = 2*vdwtype[jnrH+0];
 269
 270             fjx0             = _mm256_setzero_ps();
 271             fjy0             = _mm256_setzero_ps();
 272             fjz0             = _mm256_setzero_ps();
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 279             {
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 283                                             vdwioffsetptr0+vdwjidx0B,
 284                                             vdwioffsetptr0+vdwjidx0C,
 285                                             vdwioffsetptr0+vdwjidx0D,
 286                                             vdwioffsetptr0+vdwjidx0E,
 287                                             vdwioffsetptr0+vdwjidx0F,
 288                                             vdwioffsetptr0+vdwjidx0G,
 289                                             vdwioffsetptr0+vdwjidx0H,
 290                                             &c6_00,&c12_00);
 291
 292             /* LENNARD-JONES DISPERSION/REPULSION */
 293
 294             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 295             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 296             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 297             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 298                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 299             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 300
 301             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 305             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 306
 307             fscal            = fvdw;
 308
 309             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm256_mul_ps(fscal,dx00);
 313             ty               = _mm256_mul_ps(fscal,dy00);
 314             tz               = _mm256_mul_ps(fscal,dz00);
 315
 316             /* Update vectorial force */
 317             fix0             = _mm256_add_ps(fix0,tx);
 318             fiy0             = _mm256_add_ps(fiy0,ty);
 319             fiz0             = _mm256_add_ps(fiz0,tz);
 320
 321             fjx0             = _mm256_add_ps(fjx0,tx);
 322             fjy0             = _mm256_add_ps(fjy0,ty);
 323             fjz0             = _mm256_add_ps(fjz0,tz);
 324
 325             }
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 332             {
 333
 334             r10              = _mm256_mul_ps(rsq10,rinv10);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq10             = _mm256_mul_ps(iq1,jq0);
 338
 339             /* EWALD ELECTROSTATICS */
 340
 341             /* Analytical PME correction */
 342             zeta2            = _mm256_mul_ps(beta2,rsq10);
 343             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 344             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 345             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 346             felec            = _mm256_mul_ps(qq10,felec);
 347             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 348             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 349             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 350             velec            = _mm256_mul_ps(qq10,velec);
 351
 352             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm256_and_ps(velec,cutoff_mask);
 356             velecsum         = _mm256_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_ps(fscal,dx10);
 364             ty               = _mm256_mul_ps(fscal,dy10);
 365             tz               = _mm256_mul_ps(fscal,dz10);
 366
 367             /* Update vectorial force */
 368             fix1             = _mm256_add_ps(fix1,tx);
 369             fiy1             = _mm256_add_ps(fiy1,ty);
 370             fiz1             = _mm256_add_ps(fiz1,tz);
 371
 372             fjx0             = _mm256_add_ps(fjx0,tx);
 373             fjy0             = _mm256_add_ps(fjy0,ty);
 374             fjz0             = _mm256_add_ps(fjz0,tz);
 375
 376             }
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 383             {
 384
 385             r20              = _mm256_mul_ps(rsq20,rinv20);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq20             = _mm256_mul_ps(iq2,jq0);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Analytical PME correction */
 393             zeta2            = _mm256_mul_ps(beta2,rsq20);
 394             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 395             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 396             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 397             felec            = _mm256_mul_ps(qq20,felec);
 398             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 399             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 400             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 401             velec            = _mm256_mul_ps(qq20,velec);
 402
 403             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm256_and_ps(velec,cutoff_mask);
 407             velecsum         = _mm256_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_ps(fscal,dx20);
 415             ty               = _mm256_mul_ps(fscal,dy20);
 416             tz               = _mm256_mul_ps(fscal,dz20);
 417
 418             /* Update vectorial force */
 419             fix2             = _mm256_add_ps(fix2,tx);
 420             fiy2             = _mm256_add_ps(fiy2,ty);
 421             fiz2             = _mm256_add_ps(fiz2,tz);
 422
 423             fjx0             = _mm256_add_ps(fjx0,tx);
 424             fjy0             = _mm256_add_ps(fjy0,ty);
 425             fjz0             = _mm256_add_ps(fjz0,tz);
 426
 427             }
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 434             {
 435
 436             r30              = _mm256_mul_ps(rsq30,rinv30);
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq30             = _mm256_mul_ps(iq3,jq0);
 440
 441             /* EWALD ELECTROSTATICS */
 442
 443             /* Analytical PME correction */
 444             zeta2            = _mm256_mul_ps(beta2,rsq30);
 445             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 446             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 447             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 448             felec            = _mm256_mul_ps(qq30,felec);
 449             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 450             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 451             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 452             velec            = _mm256_mul_ps(qq30,velec);
 453
 454             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm256_and_ps(velec,cutoff_mask);
 458             velecsum         = _mm256_add_ps(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm256_mul_ps(fscal,dx30);
 466             ty               = _mm256_mul_ps(fscal,dy30);
 467             tz               = _mm256_mul_ps(fscal,dz30);
 468
 469             /* Update vectorial force */
 470             fix3             = _mm256_add_ps(fix3,tx);
 471             fiy3             = _mm256_add_ps(fiy3,ty);
 472             fiz3             = _mm256_add_ps(fiz3,tz);
 473
 474             fjx0             = _mm256_add_ps(fjx0,tx);
 475             fjy0             = _mm256_add_ps(fjy0,ty);
 476             fjz0             = _mm256_add_ps(fjz0,tz);
 477
 478             }
 479
 480             fjptrA             = f+j_coord_offsetA;
 481             fjptrB             = f+j_coord_offsetB;
 482             fjptrC             = f+j_coord_offsetC;
 483             fjptrD             = f+j_coord_offsetD;
 484             fjptrE             = f+j_coord_offsetE;
 485             fjptrF             = f+j_coord_offsetF;
 486             fjptrG             = f+j_coord_offsetG;
 487             fjptrH             = f+j_coord_offsetH;
 488
 489             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 490
 491             /* Inner loop uses 371 flops */
 492         }
 493
 494         if(jidx<j_index_end)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrlistA         = jjnr[jidx];
 499             jnrlistB         = jjnr[jidx+1];
 500             jnrlistC         = jjnr[jidx+2];
 501             jnrlistD         = jjnr[jidx+3];
 502             jnrlistE         = jjnr[jidx+4];
 503             jnrlistF         = jjnr[jidx+5];
 504             jnrlistG         = jjnr[jidx+6];
 505             jnrlistH         = jjnr[jidx+7];
 506             /* Sign of each element will be negative for non-real atoms.
 507              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 508              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 509              */
 510             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 511                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 512
 513             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 514             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 515             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 516             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 517             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 518             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 519             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 520             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 521             j_coord_offsetA  = DIM*jnrA;
 522             j_coord_offsetB  = DIM*jnrB;
 523             j_coord_offsetC  = DIM*jnrC;
 524             j_coord_offsetD  = DIM*jnrD;
 525             j_coord_offsetE  = DIM*jnrE;
 526             j_coord_offsetF  = DIM*jnrF;
 527             j_coord_offsetG  = DIM*jnrG;
 528             j_coord_offsetH  = DIM*jnrH;
 529
 530             /* load j atom coordinates */
 531             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 533                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 534                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 535                                                  &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm256_sub_ps(ix0,jx0);
 539             dy00             = _mm256_sub_ps(iy0,jy0);
 540             dz00             = _mm256_sub_ps(iz0,jz0);
 541             dx10             = _mm256_sub_ps(ix1,jx0);
 542             dy10             = _mm256_sub_ps(iy1,jy0);
 543             dz10             = _mm256_sub_ps(iz1,jz0);
 544             dx20             = _mm256_sub_ps(ix2,jx0);
 545             dy20             = _mm256_sub_ps(iy2,jy0);
 546             dz20             = _mm256_sub_ps(iz2,jz0);
 547             dx30             = _mm256_sub_ps(ix3,jx0);
 548             dy30             = _mm256_sub_ps(iy3,jy0);
 549             dz30             = _mm256_sub_ps(iz3,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 553             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 554             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 555             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 556
 557             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 558             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 559             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 560
 561             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 562             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 563             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 564             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 565
 566             /* Load parameters for j particles */
 567             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 568                                                                  charge+jnrC+0,charge+jnrD+0,
 569                                                                  charge+jnrE+0,charge+jnrF+0,
 570                                                                  charge+jnrG+0,charge+jnrH+0);
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572             vdwjidx0B        = 2*vdwtype[jnrB+0];
 573             vdwjidx0C        = 2*vdwtype[jnrC+0];
 574             vdwjidx0D        = 2*vdwtype[jnrD+0];
 575             vdwjidx0E        = 2*vdwtype[jnrE+0];
 576             vdwjidx0F        = 2*vdwtype[jnrF+0];
 577             vdwjidx0G        = 2*vdwtype[jnrG+0];
 578             vdwjidx0H        = 2*vdwtype[jnrH+0];
 579
 580             fjx0             = _mm256_setzero_ps();
 581             fjy0             = _mm256_setzero_ps();
 582             fjz0             = _mm256_setzero_ps();
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 589             {
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 593                                             vdwioffsetptr0+vdwjidx0B,
 594                                             vdwioffsetptr0+vdwjidx0C,
 595                                             vdwioffsetptr0+vdwjidx0D,
 596                                             vdwioffsetptr0+vdwjidx0E,
 597                                             vdwioffsetptr0+vdwjidx0F,
 598                                             vdwioffsetptr0+vdwjidx0G,
 599                                             vdwioffsetptr0+vdwjidx0H,
 600                                             &c6_00,&c12_00);
 601
 602             /* LENNARD-JONES DISPERSION/REPULSION */
 603
 604             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 605             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 606             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 607             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 608                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 609             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 610
 611             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 615             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 616             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 617
 618             fscal            = fvdw;
 619
 620             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 621
 622             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm256_mul_ps(fscal,dx00);
 626             ty               = _mm256_mul_ps(fscal,dy00);
 627             tz               = _mm256_mul_ps(fscal,dz00);
 628
 629             /* Update vectorial force */
 630             fix0             = _mm256_add_ps(fix0,tx);
 631             fiy0             = _mm256_add_ps(fiy0,ty);
 632             fiz0             = _mm256_add_ps(fiz0,tz);
 633
 634             fjx0             = _mm256_add_ps(fjx0,tx);
 635             fjy0             = _mm256_add_ps(fjy0,ty);
 636             fjz0             = _mm256_add_ps(fjz0,tz);
 637
 638             }
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 645             {
 646
 647             r10              = _mm256_mul_ps(rsq10,rinv10);
 648             r10              = _mm256_andnot_ps(dummy_mask,r10);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq10             = _mm256_mul_ps(iq1,jq0);
 652
 653             /* EWALD ELECTROSTATICS */
 654
 655             /* Analytical PME correction */
 656             zeta2            = _mm256_mul_ps(beta2,rsq10);
 657             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 658             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 659             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 660             felec            = _mm256_mul_ps(qq10,felec);
 661             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 662             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 663             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 664             velec            = _mm256_mul_ps(qq10,velec);
 665
 666             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm256_and_ps(velec,cutoff_mask);
 670             velec            = _mm256_andnot_ps(dummy_mask,velec);
 671             velecsum         = _mm256_add_ps(velecsum,velec);
 672
 673             fscal            = felec;
 674
 675             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 676
 677             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm256_mul_ps(fscal,dx10);
 681             ty               = _mm256_mul_ps(fscal,dy10);
 682             tz               = _mm256_mul_ps(fscal,dz10);
 683
 684             /* Update vectorial force */
 685             fix1             = _mm256_add_ps(fix1,tx);
 686             fiy1             = _mm256_add_ps(fiy1,ty);
 687             fiz1             = _mm256_add_ps(fiz1,tz);
 688
 689             fjx0             = _mm256_add_ps(fjx0,tx);
 690             fjy0             = _mm256_add_ps(fjy0,ty);
 691             fjz0             = _mm256_add_ps(fjz0,tz);
 692
 693             }
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 700             {
 701
 702             r20              = _mm256_mul_ps(rsq20,rinv20);
 703             r20              = _mm256_andnot_ps(dummy_mask,r20);
 704
 705             /* Compute parameters for interactions between i and j atoms */
 706             qq20             = _mm256_mul_ps(iq2,jq0);
 707
 708             /* EWALD ELECTROSTATICS */
 709
 710             /* Analytical PME correction */
 711             zeta2            = _mm256_mul_ps(beta2,rsq20);
 712             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 713             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 714             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 715             felec            = _mm256_mul_ps(qq20,felec);
 716             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 717             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 718             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 719             velec            = _mm256_mul_ps(qq20,velec);
 720
 721             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 722
 723             /* Update potential sum for this i atom from the interaction with this j atom. */
 724             velec            = _mm256_and_ps(velec,cutoff_mask);
 725             velec            = _mm256_andnot_ps(dummy_mask,velec);
 726             velecsum         = _mm256_add_ps(velecsum,velec);
 727
 728             fscal            = felec;
 729
 730             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 731
 732             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 733
 734             /* Calculate temporary vectorial force */
 735             tx               = _mm256_mul_ps(fscal,dx20);
 736             ty               = _mm256_mul_ps(fscal,dy20);
 737             tz               = _mm256_mul_ps(fscal,dz20);
 738
 739             /* Update vectorial force */
 740             fix2             = _mm256_add_ps(fix2,tx);
 741             fiy2             = _mm256_add_ps(fiy2,ty);
 742             fiz2             = _mm256_add_ps(fiz2,tz);
 743
 744             fjx0             = _mm256_add_ps(fjx0,tx);
 745             fjy0             = _mm256_add_ps(fjy0,ty);
 746             fjz0             = _mm256_add_ps(fjz0,tz);
 747
 748             }
 749
 750             /**************************
 751              * CALCULATE INTERACTIONS *
 752              **************************/
 753
 754             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 755             {
 756
 757             r30              = _mm256_mul_ps(rsq30,rinv30);
 758             r30              = _mm256_andnot_ps(dummy_mask,r30);
 759
 760             /* Compute parameters for interactions between i and j atoms */
 761             qq30             = _mm256_mul_ps(iq3,jq0);
 762
 763             /* EWALD ELECTROSTATICS */
 764
 765             /* Analytical PME correction */
 766             zeta2            = _mm256_mul_ps(beta2,rsq30);
 767             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 768             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 769             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 770             felec            = _mm256_mul_ps(qq30,felec);
 771             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 772             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 773             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 774             velec            = _mm256_mul_ps(qq30,velec);
 775
 776             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 777
 778             /* Update potential sum for this i atom from the interaction with this j atom. */
 779             velec            = _mm256_and_ps(velec,cutoff_mask);
 780             velec            = _mm256_andnot_ps(dummy_mask,velec);
 781             velecsum         = _mm256_add_ps(velecsum,velec);
 782
 783             fscal            = felec;
 784
 785             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 786
 787             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 788
 789             /* Calculate temporary vectorial force */
 790             tx               = _mm256_mul_ps(fscal,dx30);
 791             ty               = _mm256_mul_ps(fscal,dy30);
 792             tz               = _mm256_mul_ps(fscal,dz30);
 793
 794             /* Update vectorial force */
 795             fix3             = _mm256_add_ps(fix3,tx);
 796             fiy3             = _mm256_add_ps(fiy3,ty);
 797             fiz3             = _mm256_add_ps(fiz3,tz);
 798
 799             fjx0             = _mm256_add_ps(fjx0,tx);
 800             fjy0             = _mm256_add_ps(fjy0,ty);
 801             fjz0             = _mm256_add_ps(fjz0,tz);
 802
 803             }
 804
 805             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 806             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 807             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 808             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 809             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 810             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 811             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 812             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 813
 814             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 815
 816             /* Inner loop uses 374 flops */
 817         }
 818
 819         /* End of innermost loop */
 820
 821         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 822                                                  f+i_coord_offset,fshift+i_shift_offset);
 823
 824         ggid                        = gid[iidx];
 825         /* Update potential energies */
 826         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 827         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 828
 829         /* Increment number of inner iterations */
 830         inneriter                  += j_index_end - j_index_start;
 831
 832         /* Outer loop uses 26 flops */
 833     }
 834
 835     /* Increment number of outer iterations */
 836     outeriter        += nri;
 837
 838     /* Update outer/inner flops */
 839
 840     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*374);
 841 }
 842 /*
 843  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
 844  * Electrostatics interaction: Ewald
 845  * VdW interaction:            LennardJones
 846  * Geometry:                   Water4-Particle
 847  * Calculate force/pot:        Force
 848  */
 849 void
 850 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
 851                     (t_nblist * gmx_restrict                nlist,
 852                      rvec * gmx_restrict                    xx,
 853                      rvec * gmx_restrict                    ff,
 854                      t_forcerec * gmx_restrict              fr,
 855                      t_mdatoms * gmx_restrict               mdatoms,
 856                      nb_kernel_data_t * gmx_restrict        kernel_data,
 857                      t_nrnb * gmx_restrict                  nrnb)
 858 {
 859     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 860      * just 0 for non-waters.
 861      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 862      * jnr indices corresponding to data put in the four positions in the SIMD register.
 863      */
 864     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 865     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 866     int              jnrA,jnrB,jnrC,jnrD;
 867     int              jnrE,jnrF,jnrG,jnrH;
 868     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 869     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 870     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 871     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 872     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 873     real             rcutoff_scalar;
 874     real             *shiftvec,*fshift,*x,*f;
 875     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 876     real             scratch[4*DIM];
 877     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 878     real *           vdwioffsetptr0;
 879     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 880     real *           vdwioffsetptr1;
 881     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 882     real *           vdwioffsetptr2;
 883     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 884     real *           vdwioffsetptr3;
 885     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 886     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 887     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 888     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 889     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 890     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 891     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 892     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 893     real             *charge;
 894     int              nvdwtype;
 895     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 896     int              *vdwtype;
 897     real             *vdwparam;
 898     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 899     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 900     __m256i          ewitab;
 901     __m128i          ewitab_lo,ewitab_hi;
 902     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 903     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 904     real             *ewtab;
 905     __m256           dummy_mask,cutoff_mask;
 906     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 907     __m256           one     = _mm256_set1_ps(1.0);
 908     __m256           two     = _mm256_set1_ps(2.0);
 909     x                = xx[0];
 910     f                = ff[0];
 911
 912     nri              = nlist->nri;
 913     iinr             = nlist->iinr;
 914     jindex           = nlist->jindex;
 915     jjnr             = nlist->jjnr;
 916     shiftidx         = nlist->shift;
 917     gid              = nlist->gid;
 918     shiftvec         = fr->shift_vec[0];
 919     fshift           = fr->fshift[0];
 920     facel            = _mm256_set1_ps(fr->epsfac);
 921     charge           = mdatoms->chargeA;
 922     nvdwtype         = fr->ntype;
 923     vdwparam         = fr->nbfp;
 924     vdwtype          = mdatoms->typeA;
 925
 926     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 927     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 928     beta2            = _mm256_mul_ps(beta,beta);
 929     beta3            = _mm256_mul_ps(beta,beta2);
 930
 931     ewtab            = fr->ic->tabq_coul_F;
 932     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 933     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 934
 935     /* Setup water-specific parameters */
 936     inr              = nlist->iinr[0];
 937     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 938     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 939     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 940     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 941
 942     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 943     rcutoff_scalar   = fr->rcoulomb;
 944     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 945     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 946
 947     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 948     rvdw             = _mm256_set1_ps(fr->rvdw);
 949
 950     /* Avoid stupid compiler warnings */
 951     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 952     j_coord_offsetA = 0;
 953     j_coord_offsetB = 0;
 954     j_coord_offsetC = 0;
 955     j_coord_offsetD = 0;
 956     j_coord_offsetE = 0;
 957     j_coord_offsetF = 0;
 958     j_coord_offsetG = 0;
 959     j_coord_offsetH = 0;
 960
 961     outeriter        = 0;
 962     inneriter        = 0;
 963
 964     for(iidx=0;iidx<4*DIM;iidx++)
 965     {
 966         scratch[iidx] = 0.0;
 967     }
 968
 969     /* Start outer loop over neighborlists */
 970     for(iidx=0; iidx<nri; iidx++)
 971     {
 972         /* Load shift vector for this list */
 973         i_shift_offset   = DIM*shiftidx[iidx];
 974
 975         /* Load limits for loop over neighbors */
 976         j_index_start    = jindex[iidx];
 977         j_index_end      = jindex[iidx+1];
 978
 979         /* Get outer coordinate index */
 980         inr              = iinr[iidx];
 981         i_coord_offset   = DIM*inr;
 982
 983         /* Load i particle coords and add shift vector */
 984         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 985                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 986
 987         fix0             = _mm256_setzero_ps();
 988         fiy0             = _mm256_setzero_ps();
 989         fiz0             = _mm256_setzero_ps();
 990         fix1             = _mm256_setzero_ps();
 991         fiy1             = _mm256_setzero_ps();
 992         fiz1             = _mm256_setzero_ps();
 993         fix2             = _mm256_setzero_ps();
 994         fiy2             = _mm256_setzero_ps();
 995         fiz2             = _mm256_setzero_ps();
 996         fix3             = _mm256_setzero_ps();
 997         fiy3             = _mm256_setzero_ps();
 998         fiz3             = _mm256_setzero_ps();
 999
1000         /* Start inner kernel loop */
1001         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1002         {
1003
1004             /* Get j neighbor index, and coordinate index */
1005             jnrA             = jjnr[jidx];
1006             jnrB             = jjnr[jidx+1];
1007             jnrC             = jjnr[jidx+2];
1008             jnrD             = jjnr[jidx+3];
1009             jnrE             = jjnr[jidx+4];
1010             jnrF             = jjnr[jidx+5];
1011             jnrG             = jjnr[jidx+6];
1012             jnrH             = jjnr[jidx+7];
1013             j_coord_offsetA  = DIM*jnrA;
1014             j_coord_offsetB  = DIM*jnrB;
1015             j_coord_offsetC  = DIM*jnrC;
1016             j_coord_offsetD  = DIM*jnrD;
1017             j_coord_offsetE  = DIM*jnrE;
1018             j_coord_offsetF  = DIM*jnrF;
1019             j_coord_offsetG  = DIM*jnrG;
1020             j_coord_offsetH  = DIM*jnrH;
1021
1022             /* load j atom coordinates */
1023             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1024                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1025                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1026                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1027                                                  &jx0,&jy0,&jz0);
1028
1029             /* Calculate displacement vector */
1030             dx00             = _mm256_sub_ps(ix0,jx0);
1031             dy00             = _mm256_sub_ps(iy0,jy0);
1032             dz00             = _mm256_sub_ps(iz0,jz0);
1033             dx10             = _mm256_sub_ps(ix1,jx0);
1034             dy10             = _mm256_sub_ps(iy1,jy0);
1035             dz10             = _mm256_sub_ps(iz1,jz0);
1036             dx20             = _mm256_sub_ps(ix2,jx0);
1037             dy20             = _mm256_sub_ps(iy2,jy0);
1038             dz20             = _mm256_sub_ps(iz2,jz0);
1039             dx30             = _mm256_sub_ps(ix3,jx0);
1040             dy30             = _mm256_sub_ps(iy3,jy0);
1041             dz30             = _mm256_sub_ps(iz3,jz0);
1042
1043             /* Calculate squared distance and things based on it */
1044             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1045             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1046             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1047             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1048
1049             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1050             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1051             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1052
1053             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1054             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1055             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1056             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1057
1058             /* Load parameters for j particles */
1059             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1060                                                                  charge+jnrC+0,charge+jnrD+0,
1061                                                                  charge+jnrE+0,charge+jnrF+0,
1062                                                                  charge+jnrG+0,charge+jnrH+0);
1063             vdwjidx0A        = 2*vdwtype[jnrA+0];
1064             vdwjidx0B        = 2*vdwtype[jnrB+0];
1065             vdwjidx0C        = 2*vdwtype[jnrC+0];
1066             vdwjidx0D        = 2*vdwtype[jnrD+0];
1067             vdwjidx0E        = 2*vdwtype[jnrE+0];
1068             vdwjidx0F        = 2*vdwtype[jnrF+0];
1069             vdwjidx0G        = 2*vdwtype[jnrG+0];
1070             vdwjidx0H        = 2*vdwtype[jnrH+0];
1071
1072             fjx0             = _mm256_setzero_ps();
1073             fjy0             = _mm256_setzero_ps();
1074             fjz0             = _mm256_setzero_ps();
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1081             {
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1085                                             vdwioffsetptr0+vdwjidx0B,
1086                                             vdwioffsetptr0+vdwjidx0C,
1087                                             vdwioffsetptr0+vdwjidx0D,
1088                                             vdwioffsetptr0+vdwjidx0E,
1089                                             vdwioffsetptr0+vdwjidx0F,
1090                                             vdwioffsetptr0+vdwjidx0G,
1091                                             vdwioffsetptr0+vdwjidx0H,
1092                                             &c6_00,&c12_00);
1093
1094             /* LENNARD-JONES DISPERSION/REPULSION */
1095
1096             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1097             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1098
1099             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1100
1101             fscal            = fvdw;
1102
1103             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1104
1105             /* Calculate temporary vectorial force */
1106             tx               = _mm256_mul_ps(fscal,dx00);
1107             ty               = _mm256_mul_ps(fscal,dy00);
1108             tz               = _mm256_mul_ps(fscal,dz00);
1109
1110             /* Update vectorial force */
1111             fix0             = _mm256_add_ps(fix0,tx);
1112             fiy0             = _mm256_add_ps(fiy0,ty);
1113             fiz0             = _mm256_add_ps(fiz0,tz);
1114
1115             fjx0             = _mm256_add_ps(fjx0,tx);
1116             fjy0             = _mm256_add_ps(fjy0,ty);
1117             fjz0             = _mm256_add_ps(fjz0,tz);
1118
1119             }
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1126             {
1127
1128             r10              = _mm256_mul_ps(rsq10,rinv10);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq10             = _mm256_mul_ps(iq1,jq0);
1132
1133             /* EWALD ELECTROSTATICS */
1134
1135             /* Analytical PME correction */
1136             zeta2            = _mm256_mul_ps(beta2,rsq10);
1137             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1138             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1139             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1140             felec            = _mm256_mul_ps(qq10,felec);
1141
1142             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm256_mul_ps(fscal,dx10);
1150             ty               = _mm256_mul_ps(fscal,dy10);
1151             tz               = _mm256_mul_ps(fscal,dz10);
1152
1153             /* Update vectorial force */
1154             fix1             = _mm256_add_ps(fix1,tx);
1155             fiy1             = _mm256_add_ps(fiy1,ty);
1156             fiz1             = _mm256_add_ps(fiz1,tz);
1157
1158             fjx0             = _mm256_add_ps(fjx0,tx);
1159             fjy0             = _mm256_add_ps(fjy0,ty);
1160             fjz0             = _mm256_add_ps(fjz0,tz);
1161
1162             }
1163
1164             /**************************
1165              * CALCULATE INTERACTIONS *
1166              **************************/
1167
1168             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1169             {
1170
1171             r20              = _mm256_mul_ps(rsq20,rinv20);
1172
1173             /* Compute parameters for interactions between i and j atoms */
1174             qq20             = _mm256_mul_ps(iq2,jq0);
1175
1176             /* EWALD ELECTROSTATICS */
1177
1178             /* Analytical PME correction */
1179             zeta2            = _mm256_mul_ps(beta2,rsq20);
1180             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1181             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1182             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1183             felec            = _mm256_mul_ps(qq20,felec);
1184
1185             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1186
1187             fscal            = felec;
1188
1189             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1190
1191             /* Calculate temporary vectorial force */
1192             tx               = _mm256_mul_ps(fscal,dx20);
1193             ty               = _mm256_mul_ps(fscal,dy20);
1194             tz               = _mm256_mul_ps(fscal,dz20);
1195
1196             /* Update vectorial force */
1197             fix2             = _mm256_add_ps(fix2,tx);
1198             fiy2             = _mm256_add_ps(fiy2,ty);
1199             fiz2             = _mm256_add_ps(fiz2,tz);
1200
1201             fjx0             = _mm256_add_ps(fjx0,tx);
1202             fjy0             = _mm256_add_ps(fjy0,ty);
1203             fjz0             = _mm256_add_ps(fjz0,tz);
1204
1205             }
1206
1207             /**************************
1208              * CALCULATE INTERACTIONS *
1209              **************************/
1210
1211             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1212             {
1213
1214             r30              = _mm256_mul_ps(rsq30,rinv30);
1215
1216             /* Compute parameters for interactions between i and j atoms */
1217             qq30             = _mm256_mul_ps(iq3,jq0);
1218
1219             /* EWALD ELECTROSTATICS */
1220
1221             /* Analytical PME correction */
1222             zeta2            = _mm256_mul_ps(beta2,rsq30);
1223             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1224             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1225             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1226             felec            = _mm256_mul_ps(qq30,felec);
1227
1228             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1229
1230             fscal            = felec;
1231
1232             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1233
1234             /* Calculate temporary vectorial force */
1235             tx               = _mm256_mul_ps(fscal,dx30);
1236             ty               = _mm256_mul_ps(fscal,dy30);
1237             tz               = _mm256_mul_ps(fscal,dz30);
1238
1239             /* Update vectorial force */
1240             fix3             = _mm256_add_ps(fix3,tx);
1241             fiy3             = _mm256_add_ps(fiy3,ty);
1242             fiz3             = _mm256_add_ps(fiz3,tz);
1243
1244             fjx0             = _mm256_add_ps(fjx0,tx);
1245             fjy0             = _mm256_add_ps(fjy0,ty);
1246             fjz0             = _mm256_add_ps(fjz0,tz);
1247
1248             }
1249
1250             fjptrA             = f+j_coord_offsetA;
1251             fjptrB             = f+j_coord_offsetB;
1252             fjptrC             = f+j_coord_offsetC;
1253             fjptrD             = f+j_coord_offsetD;
1254             fjptrE             = f+j_coord_offsetE;
1255             fjptrF             = f+j_coord_offsetF;
1256             fjptrG             = f+j_coord_offsetG;
1257             fjptrH             = f+j_coord_offsetH;
1258
1259             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1260
1261             /* Inner loop uses 210 flops */
1262         }
1263
1264         if(jidx<j_index_end)
1265         {
1266
1267             /* Get j neighbor index, and coordinate index */
1268             jnrlistA         = jjnr[jidx];
1269             jnrlistB         = jjnr[jidx+1];
1270             jnrlistC         = jjnr[jidx+2];
1271             jnrlistD         = jjnr[jidx+3];
1272             jnrlistE         = jjnr[jidx+4];
1273             jnrlistF         = jjnr[jidx+5];
1274             jnrlistG         = jjnr[jidx+6];
1275             jnrlistH         = jjnr[jidx+7];
1276             /* Sign of each element will be negative for non-real atoms.
1277              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1278              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1279              */
1280             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1281                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1282
1283             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1284             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1285             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1286             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1287             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1288             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1289             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1290             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1291             j_coord_offsetA  = DIM*jnrA;
1292             j_coord_offsetB  = DIM*jnrB;
1293             j_coord_offsetC  = DIM*jnrC;
1294             j_coord_offsetD  = DIM*jnrD;
1295             j_coord_offsetE  = DIM*jnrE;
1296             j_coord_offsetF  = DIM*jnrF;
1297             j_coord_offsetG  = DIM*jnrG;
1298             j_coord_offsetH  = DIM*jnrH;
1299
1300             /* load j atom coordinates */
1301             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1302                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1303                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1304                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1305                                                  &jx0,&jy0,&jz0);
1306
1307             /* Calculate displacement vector */
1308             dx00             = _mm256_sub_ps(ix0,jx0);
1309             dy00             = _mm256_sub_ps(iy0,jy0);
1310             dz00             = _mm256_sub_ps(iz0,jz0);
1311             dx10             = _mm256_sub_ps(ix1,jx0);
1312             dy10             = _mm256_sub_ps(iy1,jy0);
1313             dz10             = _mm256_sub_ps(iz1,jz0);
1314             dx20             = _mm256_sub_ps(ix2,jx0);
1315             dy20             = _mm256_sub_ps(iy2,jy0);
1316             dz20             = _mm256_sub_ps(iz2,jz0);
1317             dx30             = _mm256_sub_ps(ix3,jx0);
1318             dy30             = _mm256_sub_ps(iy3,jy0);
1319             dz30             = _mm256_sub_ps(iz3,jz0);
1320
1321             /* Calculate squared distance and things based on it */
1322             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1323             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1324             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1325             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1326
1327             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1328             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1329             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1330
1331             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1332             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1333             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1334             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1335
1336             /* Load parameters for j particles */
1337             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1338                                                                  charge+jnrC+0,charge+jnrD+0,
1339                                                                  charge+jnrE+0,charge+jnrF+0,
1340                                                                  charge+jnrG+0,charge+jnrH+0);
1341             vdwjidx0A        = 2*vdwtype[jnrA+0];
1342             vdwjidx0B        = 2*vdwtype[jnrB+0];
1343             vdwjidx0C        = 2*vdwtype[jnrC+0];
1344             vdwjidx0D        = 2*vdwtype[jnrD+0];
1345             vdwjidx0E        = 2*vdwtype[jnrE+0];
1346             vdwjidx0F        = 2*vdwtype[jnrF+0];
1347             vdwjidx0G        = 2*vdwtype[jnrG+0];
1348             vdwjidx0H        = 2*vdwtype[jnrH+0];
1349
1350             fjx0             = _mm256_setzero_ps();
1351             fjy0             = _mm256_setzero_ps();
1352             fjz0             = _mm256_setzero_ps();
1353
1354             /**************************
1355              * CALCULATE INTERACTIONS *
1356              **************************/
1357
1358             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1359             {
1360
1361             /* Compute parameters for interactions between i and j atoms */
1362             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1363                                             vdwioffsetptr0+vdwjidx0B,
1364                                             vdwioffsetptr0+vdwjidx0C,
1365                                             vdwioffsetptr0+vdwjidx0D,
1366                                             vdwioffsetptr0+vdwjidx0E,
1367                                             vdwioffsetptr0+vdwjidx0F,
1368                                             vdwioffsetptr0+vdwjidx0G,
1369                                             vdwioffsetptr0+vdwjidx0H,
1370                                             &c6_00,&c12_00);
1371
1372             /* LENNARD-JONES DISPERSION/REPULSION */
1373
1374             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1375             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1376
1377             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1378
1379             fscal            = fvdw;
1380
1381             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1382
1383             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1384
1385             /* Calculate temporary vectorial force */
1386             tx               = _mm256_mul_ps(fscal,dx00);
1387             ty               = _mm256_mul_ps(fscal,dy00);
1388             tz               = _mm256_mul_ps(fscal,dz00);
1389
1390             /* Update vectorial force */
1391             fix0             = _mm256_add_ps(fix0,tx);
1392             fiy0             = _mm256_add_ps(fiy0,ty);
1393             fiz0             = _mm256_add_ps(fiz0,tz);
1394
1395             fjx0             = _mm256_add_ps(fjx0,tx);
1396             fjy0             = _mm256_add_ps(fjy0,ty);
1397             fjz0             = _mm256_add_ps(fjz0,tz);
1398
1399             }
1400
1401             /**************************
1402              * CALCULATE INTERACTIONS *
1403              **************************/
1404
1405             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1406             {
1407
1408             r10              = _mm256_mul_ps(rsq10,rinv10);
1409             r10              = _mm256_andnot_ps(dummy_mask,r10);
1410
1411             /* Compute parameters for interactions between i and j atoms */
1412             qq10             = _mm256_mul_ps(iq1,jq0);
1413
1414             /* EWALD ELECTROSTATICS */
1415
1416             /* Analytical PME correction */
1417             zeta2            = _mm256_mul_ps(beta2,rsq10);
1418             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1419             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1420             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1421             felec            = _mm256_mul_ps(qq10,felec);
1422
1423             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1424
1425             fscal            = felec;
1426
1427             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1428
1429             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1430
1431             /* Calculate temporary vectorial force */
1432             tx               = _mm256_mul_ps(fscal,dx10);
1433             ty               = _mm256_mul_ps(fscal,dy10);
1434             tz               = _mm256_mul_ps(fscal,dz10);
1435
1436             /* Update vectorial force */
1437             fix1             = _mm256_add_ps(fix1,tx);
1438             fiy1             = _mm256_add_ps(fiy1,ty);
1439             fiz1             = _mm256_add_ps(fiz1,tz);
1440
1441             fjx0             = _mm256_add_ps(fjx0,tx);
1442             fjy0             = _mm256_add_ps(fjy0,ty);
1443             fjz0             = _mm256_add_ps(fjz0,tz);
1444
1445             }
1446
1447             /**************************
1448              * CALCULATE INTERACTIONS *
1449              **************************/
1450
1451             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1452             {
1453
1454             r20              = _mm256_mul_ps(rsq20,rinv20);
1455             r20              = _mm256_andnot_ps(dummy_mask,r20);
1456
1457             /* Compute parameters for interactions between i and j atoms */
1458             qq20             = _mm256_mul_ps(iq2,jq0);
1459
1460             /* EWALD ELECTROSTATICS */
1461
1462             /* Analytical PME correction */
1463             zeta2            = _mm256_mul_ps(beta2,rsq20);
1464             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1465             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1466             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1467             felec            = _mm256_mul_ps(qq20,felec);
1468
1469             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1470
1471             fscal            = felec;
1472
1473             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1474
1475             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1476
1477             /* Calculate temporary vectorial force */
1478             tx               = _mm256_mul_ps(fscal,dx20);
1479             ty               = _mm256_mul_ps(fscal,dy20);
1480             tz               = _mm256_mul_ps(fscal,dz20);
1481
1482             /* Update vectorial force */
1483             fix2             = _mm256_add_ps(fix2,tx);
1484             fiy2             = _mm256_add_ps(fiy2,ty);
1485             fiz2             = _mm256_add_ps(fiz2,tz);
1486
1487             fjx0             = _mm256_add_ps(fjx0,tx);
1488             fjy0             = _mm256_add_ps(fjy0,ty);
1489             fjz0             = _mm256_add_ps(fjz0,tz);
1490
1491             }
1492
1493             /**************************
1494              * CALCULATE INTERACTIONS *
1495              **************************/
1496
1497             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1498             {
1499
1500             r30              = _mm256_mul_ps(rsq30,rinv30);
1501             r30              = _mm256_andnot_ps(dummy_mask,r30);
1502
1503             /* Compute parameters for interactions between i and j atoms */
1504             qq30             = _mm256_mul_ps(iq3,jq0);
1505
1506             /* EWALD ELECTROSTATICS */
1507
1508             /* Analytical PME correction */
1509             zeta2            = _mm256_mul_ps(beta2,rsq30);
1510             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1511             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1512             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1513             felec            = _mm256_mul_ps(qq30,felec);
1514
1515             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1516
1517             fscal            = felec;
1518
1519             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1520
1521             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1522
1523             /* Calculate temporary vectorial force */
1524             tx               = _mm256_mul_ps(fscal,dx30);
1525             ty               = _mm256_mul_ps(fscal,dy30);
1526             tz               = _mm256_mul_ps(fscal,dz30);
1527
1528             /* Update vectorial force */
1529             fix3             = _mm256_add_ps(fix3,tx);
1530             fiy3             = _mm256_add_ps(fiy3,ty);
1531             fiz3             = _mm256_add_ps(fiz3,tz);
1532
1533             fjx0             = _mm256_add_ps(fjx0,tx);
1534             fjy0             = _mm256_add_ps(fjy0,ty);
1535             fjz0             = _mm256_add_ps(fjz0,tz);
1536
1537             }
1538
1539             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1540             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1541             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1542             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1543             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1544             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1545             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1546             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1547
1548             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1549
1550             /* Inner loop uses 213 flops */
1551         }
1552
1553         /* End of innermost loop */
1554
1555         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1556                                                  f+i_coord_offset,fshift+i_shift_offset);
1557
1558         /* Increment number of inner iterations */
1559         inneriter                  += j_index_end - j_index_start;
1560
1561         /* Outer loop uses 24 flops */
1562     }
1563
1564     /* Increment number of outer iterations */
1565     outeriter        += nri;
1566
1567     /* Update outer/inner flops */
1568
1569     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*213);
1570 }