src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSw_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_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_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     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  95     real             rswitch_scalar,d_scalar;
  96     __m256           dummy_mask,cutoff_mask;
  97     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  98     __m256           one     = _mm256_set1_ps(1.0);
  99     __m256           two     = _mm256_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm256_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = _mm256_set1_ps(fr->ic->k_rf);
 114     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 115     crf              = _mm256_set1_ps(fr->ic->c_rf);
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 123     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 124     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 125     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 131
 132     rswitch_scalar   = fr->rvdw_switch;
 133     rswitch          = _mm256_set1_ps(rswitch_scalar);
 134     /* Setup switch parameters */
 135     d_scalar         = rcutoff_scalar-rswitch_scalar;
 136     d                = _mm256_set1_ps(d_scalar);
 137     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 138     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 141     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 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             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 248             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 249             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 250             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 251
 252             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 253             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 254             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 255             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 256
 257             /* Load parameters for j particles */
 258             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 259                                                                  charge+jnrC+0,charge+jnrD+0,
 260                                                                  charge+jnrE+0,charge+jnrF+0,
 261                                                                  charge+jnrG+0,charge+jnrH+0);
 262             vdwjidx0A        = 2*vdwtype[jnrA+0];
 263             vdwjidx0B        = 2*vdwtype[jnrB+0];
 264             vdwjidx0C        = 2*vdwtype[jnrC+0];
 265             vdwjidx0D        = 2*vdwtype[jnrD+0];
 266             vdwjidx0E        = 2*vdwtype[jnrE+0];
 267             vdwjidx0F        = 2*vdwtype[jnrF+0];
 268             vdwjidx0G        = 2*vdwtype[jnrG+0];
 269             vdwjidx0H        = 2*vdwtype[jnrH+0];
 270
 271             fjx0             = _mm256_setzero_ps();
 272             fjy0             = _mm256_setzero_ps();
 273             fjz0             = _mm256_setzero_ps();
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 280             {
 281
 282             r00              = _mm256_mul_ps(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 286                                             vdwioffsetptr0+vdwjidx0B,
 287                                             vdwioffsetptr0+vdwjidx0C,
 288                                             vdwioffsetptr0+vdwjidx0D,
 289                                             vdwioffsetptr0+vdwjidx0E,
 290                                             vdwioffsetptr0+vdwjidx0F,
 291                                             vdwioffsetptr0+vdwjidx0G,
 292                                             vdwioffsetptr0+vdwjidx0H,
 293                                             &c6_00,&c12_00);
 294
 295             /* LENNARD-JONES DISPERSION/REPULSION */
 296
 297             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 298             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 299             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 300             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 301             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 302
 303             d                = _mm256_sub_ps(r00,rswitch);
 304             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 305             d2               = _mm256_mul_ps(d,d);
 306             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 307
 308             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 309
 310             /* Evaluate switch function */
 311             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 312             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 313             vvdw             = _mm256_mul_ps(vvdw,sw);
 314             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 318             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 319
 320             fscal            = fvdw;
 321
 322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm256_mul_ps(fscal,dx00);
 326             ty               = _mm256_mul_ps(fscal,dy00);
 327             tz               = _mm256_mul_ps(fscal,dz00);
 328
 329             /* Update vectorial force */
 330             fix0             = _mm256_add_ps(fix0,tx);
 331             fiy0             = _mm256_add_ps(fiy0,ty);
 332             fiz0             = _mm256_add_ps(fiz0,tz);
 333
 334             fjx0             = _mm256_add_ps(fjx0,tx);
 335             fjy0             = _mm256_add_ps(fjy0,ty);
 336             fjz0             = _mm256_add_ps(fjz0,tz);
 337
 338             }
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 345             {
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq10             = _mm256_mul_ps(iq1,jq0);
 349
 350             /* REACTION-FIELD ELECTROSTATICS */
 351             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 352             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 353
 354             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm256_and_ps(velec,cutoff_mask);
 358             velecsum         = _mm256_add_ps(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm256_mul_ps(fscal,dx10);
 366             ty               = _mm256_mul_ps(fscal,dy10);
 367             tz               = _mm256_mul_ps(fscal,dz10);
 368
 369             /* Update vectorial force */
 370             fix1             = _mm256_add_ps(fix1,tx);
 371             fiy1             = _mm256_add_ps(fiy1,ty);
 372             fiz1             = _mm256_add_ps(fiz1,tz);
 373
 374             fjx0             = _mm256_add_ps(fjx0,tx);
 375             fjy0             = _mm256_add_ps(fjy0,ty);
 376             fjz0             = _mm256_add_ps(fjz0,tz);
 377
 378             }
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 385             {
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq20             = _mm256_mul_ps(iq2,jq0);
 389
 390             /* REACTION-FIELD ELECTROSTATICS */
 391             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 392             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 393
 394             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velec            = _mm256_and_ps(velec,cutoff_mask);
 398             velecsum         = _mm256_add_ps(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_ps(fscal,dx20);
 406             ty               = _mm256_mul_ps(fscal,dy20);
 407             tz               = _mm256_mul_ps(fscal,dz20);
 408
 409             /* Update vectorial force */
 410             fix2             = _mm256_add_ps(fix2,tx);
 411             fiy2             = _mm256_add_ps(fiy2,ty);
 412             fiz2             = _mm256_add_ps(fiz2,tz);
 413
 414             fjx0             = _mm256_add_ps(fjx0,tx);
 415             fjy0             = _mm256_add_ps(fjy0,ty);
 416             fjz0             = _mm256_add_ps(fjz0,tz);
 417
 418             }
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 425             {
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq30             = _mm256_mul_ps(iq3,jq0);
 429
 430             /* REACTION-FIELD ELECTROSTATICS */
 431             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 432             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 433
 434             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm256_and_ps(velec,cutoff_mask);
 438             velecsum         = _mm256_add_ps(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm256_mul_ps(fscal,dx30);
 446             ty               = _mm256_mul_ps(fscal,dy30);
 447             tz               = _mm256_mul_ps(fscal,dz30);
 448
 449             /* Update vectorial force */
 450             fix3             = _mm256_add_ps(fix3,tx);
 451             fiy3             = _mm256_add_ps(fiy3,ty);
 452             fiz3             = _mm256_add_ps(fiz3,tz);
 453
 454             fjx0             = _mm256_add_ps(fjx0,tx);
 455             fjy0             = _mm256_add_ps(fjy0,ty);
 456             fjz0             = _mm256_add_ps(fjz0,tz);
 457
 458             }
 459
 460             fjptrA             = f+j_coord_offsetA;
 461             fjptrB             = f+j_coord_offsetB;
 462             fjptrC             = f+j_coord_offsetC;
 463             fjptrD             = f+j_coord_offsetD;
 464             fjptrE             = f+j_coord_offsetE;
 465             fjptrF             = f+j_coord_offsetF;
 466             fjptrG             = f+j_coord_offsetG;
 467             fjptrH             = f+j_coord_offsetH;
 468
 469             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 470
 471             /* Inner loop uses 170 flops */
 472         }
 473
 474         if(jidx<j_index_end)
 475         {
 476
 477             /* Get j neighbor index, and coordinate index */
 478             jnrlistA         = jjnr[jidx];
 479             jnrlistB         = jjnr[jidx+1];
 480             jnrlistC         = jjnr[jidx+2];
 481             jnrlistD         = jjnr[jidx+3];
 482             jnrlistE         = jjnr[jidx+4];
 483             jnrlistF         = jjnr[jidx+5];
 484             jnrlistG         = jjnr[jidx+6];
 485             jnrlistH         = jjnr[jidx+7];
 486             /* Sign of each element will be negative for non-real atoms.
 487              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 488              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 489              */
 490             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 491                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 492
 493             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 494             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 495             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 496             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 497             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 498             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 499             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 500             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 501             j_coord_offsetA  = DIM*jnrA;
 502             j_coord_offsetB  = DIM*jnrB;
 503             j_coord_offsetC  = DIM*jnrC;
 504             j_coord_offsetD  = DIM*jnrD;
 505             j_coord_offsetE  = DIM*jnrE;
 506             j_coord_offsetF  = DIM*jnrF;
 507             j_coord_offsetG  = DIM*jnrG;
 508             j_coord_offsetH  = DIM*jnrH;
 509
 510             /* load j atom coordinates */
 511             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 513                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 514                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 515                                                  &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm256_sub_ps(ix0,jx0);
 519             dy00             = _mm256_sub_ps(iy0,jy0);
 520             dz00             = _mm256_sub_ps(iz0,jz0);
 521             dx10             = _mm256_sub_ps(ix1,jx0);
 522             dy10             = _mm256_sub_ps(iy1,jy0);
 523             dz10             = _mm256_sub_ps(iz1,jz0);
 524             dx20             = _mm256_sub_ps(ix2,jx0);
 525             dy20             = _mm256_sub_ps(iy2,jy0);
 526             dz20             = _mm256_sub_ps(iz2,jz0);
 527             dx30             = _mm256_sub_ps(ix3,jx0);
 528             dy30             = _mm256_sub_ps(iy3,jy0);
 529             dz30             = _mm256_sub_ps(iz3,jz0);
 530
 531             /* Calculate squared distance and things based on it */
 532             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 533             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 534             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 535             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 536
 537             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 538             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 539             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 540             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 541
 542             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 543             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 544             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 545             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 546
 547             /* Load parameters for j particles */
 548             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 549                                                                  charge+jnrC+0,charge+jnrD+0,
 550                                                                  charge+jnrE+0,charge+jnrF+0,
 551                                                                  charge+jnrG+0,charge+jnrH+0);
 552             vdwjidx0A        = 2*vdwtype[jnrA+0];
 553             vdwjidx0B        = 2*vdwtype[jnrB+0];
 554             vdwjidx0C        = 2*vdwtype[jnrC+0];
 555             vdwjidx0D        = 2*vdwtype[jnrD+0];
 556             vdwjidx0E        = 2*vdwtype[jnrE+0];
 557             vdwjidx0F        = 2*vdwtype[jnrF+0];
 558             vdwjidx0G        = 2*vdwtype[jnrG+0];
 559             vdwjidx0H        = 2*vdwtype[jnrH+0];
 560
 561             fjx0             = _mm256_setzero_ps();
 562             fjy0             = _mm256_setzero_ps();
 563             fjz0             = _mm256_setzero_ps();
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 570             {
 571
 572             r00              = _mm256_mul_ps(rsq00,rinv00);
 573             r00              = _mm256_andnot_ps(dummy_mask,r00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 577                                             vdwioffsetptr0+vdwjidx0B,
 578                                             vdwioffsetptr0+vdwjidx0C,
 579                                             vdwioffsetptr0+vdwjidx0D,
 580                                             vdwioffsetptr0+vdwjidx0E,
 581                                             vdwioffsetptr0+vdwjidx0F,
 582                                             vdwioffsetptr0+vdwjidx0G,
 583                                             vdwioffsetptr0+vdwjidx0H,
 584                                             &c6_00,&c12_00);
 585
 586             /* LENNARD-JONES DISPERSION/REPULSION */
 587
 588             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 589             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 590             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 591             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 592             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 593
 594             d                = _mm256_sub_ps(r00,rswitch);
 595             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 596             d2               = _mm256_mul_ps(d,d);
 597             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 598
 599             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 600
 601             /* Evaluate switch function */
 602             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 603             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 604             vvdw             = _mm256_mul_ps(vvdw,sw);
 605             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 609             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 610             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 611
 612             fscal            = fvdw;
 613
 614             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 615
 616             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm256_mul_ps(fscal,dx00);
 620             ty               = _mm256_mul_ps(fscal,dy00);
 621             tz               = _mm256_mul_ps(fscal,dz00);
 622
 623             /* Update vectorial force */
 624             fix0             = _mm256_add_ps(fix0,tx);
 625             fiy0             = _mm256_add_ps(fiy0,ty);
 626             fiz0             = _mm256_add_ps(fiz0,tz);
 627
 628             fjx0             = _mm256_add_ps(fjx0,tx);
 629             fjy0             = _mm256_add_ps(fjy0,ty);
 630             fjz0             = _mm256_add_ps(fjz0,tz);
 631
 632             }
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 639             {
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq10             = _mm256_mul_ps(iq1,jq0);
 643
 644             /* REACTION-FIELD ELECTROSTATICS */
 645             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 646             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 647
 648             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm256_and_ps(velec,cutoff_mask);
 652             velec            = _mm256_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm256_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 658
 659             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_ps(fscal,dx10);
 663             ty               = _mm256_mul_ps(fscal,dy10);
 664             tz               = _mm256_mul_ps(fscal,dz10);
 665
 666             /* Update vectorial force */
 667             fix1             = _mm256_add_ps(fix1,tx);
 668             fiy1             = _mm256_add_ps(fiy1,ty);
 669             fiz1             = _mm256_add_ps(fiz1,tz);
 670
 671             fjx0             = _mm256_add_ps(fjx0,tx);
 672             fjy0             = _mm256_add_ps(fjy0,ty);
 673             fjz0             = _mm256_add_ps(fjz0,tz);
 674
 675             }
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 682             {
 683
 684             /* Compute parameters for interactions between i and j atoms */
 685             qq20             = _mm256_mul_ps(iq2,jq0);
 686
 687             /* REACTION-FIELD ELECTROSTATICS */
 688             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 689             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 690
 691             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 692
 693             /* Update potential sum for this i atom from the interaction with this j atom. */
 694             velec            = _mm256_and_ps(velec,cutoff_mask);
 695             velec            = _mm256_andnot_ps(dummy_mask,velec);
 696             velecsum         = _mm256_add_ps(velecsum,velec);
 697
 698             fscal            = felec;
 699
 700             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 701
 702             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 703
 704             /* Calculate temporary vectorial force */
 705             tx               = _mm256_mul_ps(fscal,dx20);
 706             ty               = _mm256_mul_ps(fscal,dy20);
 707             tz               = _mm256_mul_ps(fscal,dz20);
 708
 709             /* Update vectorial force */
 710             fix2             = _mm256_add_ps(fix2,tx);
 711             fiy2             = _mm256_add_ps(fiy2,ty);
 712             fiz2             = _mm256_add_ps(fiz2,tz);
 713
 714             fjx0             = _mm256_add_ps(fjx0,tx);
 715             fjy0             = _mm256_add_ps(fjy0,ty);
 716             fjz0             = _mm256_add_ps(fjz0,tz);
 717
 718             }
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 725             {
 726
 727             /* Compute parameters for interactions between i and j atoms */
 728             qq30             = _mm256_mul_ps(iq3,jq0);
 729
 730             /* REACTION-FIELD ELECTROSTATICS */
 731             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 732             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 733
 734             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 735
 736             /* Update potential sum for this i atom from the interaction with this j atom. */
 737             velec            = _mm256_and_ps(velec,cutoff_mask);
 738             velec            = _mm256_andnot_ps(dummy_mask,velec);
 739             velecsum         = _mm256_add_ps(velecsum,velec);
 740
 741             fscal            = felec;
 742
 743             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 744
 745             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_ps(fscal,dx30);
 749             ty               = _mm256_mul_ps(fscal,dy30);
 750             tz               = _mm256_mul_ps(fscal,dz30);
 751
 752             /* Update vectorial force */
 753             fix3             = _mm256_add_ps(fix3,tx);
 754             fiy3             = _mm256_add_ps(fiy3,ty);
 755             fiz3             = _mm256_add_ps(fiz3,tz);
 756
 757             fjx0             = _mm256_add_ps(fjx0,tx);
 758             fjy0             = _mm256_add_ps(fjy0,ty);
 759             fjz0             = _mm256_add_ps(fjz0,tz);
 760
 761             }
 762
 763             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 764             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 765             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 766             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 767             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 768             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 769             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 770             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 771
 772             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 773
 774             /* Inner loop uses 171 flops */
 775         }
 776
 777         /* End of innermost loop */
 778
 779         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 780                                                  f+i_coord_offset,fshift+i_shift_offset);
 781
 782         ggid                        = gid[iidx];
 783         /* Update potential energies */
 784         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 785         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 786
 787         /* Increment number of inner iterations */
 788         inneriter                  += j_index_end - j_index_start;
 789
 790         /* Outer loop uses 26 flops */
 791     }
 792
 793     /* Increment number of outer iterations */
 794     outeriter        += nri;
 795
 796     /* Update outer/inner flops */
 797
 798     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
 799 }
 800 /*
 801  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_single
 802  * Electrostatics interaction: ReactionField
 803  * VdW interaction:            LennardJones
 804  * Geometry:                   Water4-Particle
 805  * Calculate force/pot:        Force
 806  */
 807 void
 808 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_single
 809                     (t_nblist * gmx_restrict                nlist,
 810                      rvec * gmx_restrict                    xx,
 811                      rvec * gmx_restrict                    ff,
 812                      t_forcerec * gmx_restrict              fr,
 813                      t_mdatoms * gmx_restrict               mdatoms,
 814                      nb_kernel_data_t * gmx_restrict        kernel_data,
 815                      t_nrnb * gmx_restrict                  nrnb)
 816 {
 817     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 818      * just 0 for non-waters.
 819      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 820      * jnr indices corresponding to data put in the four positions in the SIMD register.
 821      */
 822     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 823     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 824     int              jnrA,jnrB,jnrC,jnrD;
 825     int              jnrE,jnrF,jnrG,jnrH;
 826     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 827     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 828     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 829     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 830     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 831     real             rcutoff_scalar;
 832     real             *shiftvec,*fshift,*x,*f;
 833     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 834     real             scratch[4*DIM];
 835     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 836     real *           vdwioffsetptr0;
 837     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 838     real *           vdwioffsetptr1;
 839     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 840     real *           vdwioffsetptr2;
 841     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 842     real *           vdwioffsetptr3;
 843     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 844     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 845     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 846     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 847     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 848     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 849     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 850     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 851     real             *charge;
 852     int              nvdwtype;
 853     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 854     int              *vdwtype;
 855     real             *vdwparam;
 856     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 857     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 858     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 859     real             rswitch_scalar,d_scalar;
 860     __m256           dummy_mask,cutoff_mask;
 861     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 862     __m256           one     = _mm256_set1_ps(1.0);
 863     __m256           two     = _mm256_set1_ps(2.0);
 864     x                = xx[0];
 865     f                = ff[0];
 866
 867     nri              = nlist->nri;
 868     iinr             = nlist->iinr;
 869     jindex           = nlist->jindex;
 870     jjnr             = nlist->jjnr;
 871     shiftidx         = nlist->shift;
 872     gid              = nlist->gid;
 873     shiftvec         = fr->shift_vec[0];
 874     fshift           = fr->fshift[0];
 875     facel            = _mm256_set1_ps(fr->epsfac);
 876     charge           = mdatoms->chargeA;
 877     krf              = _mm256_set1_ps(fr->ic->k_rf);
 878     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 879     crf              = _mm256_set1_ps(fr->ic->c_rf);
 880     nvdwtype         = fr->ntype;
 881     vdwparam         = fr->nbfp;
 882     vdwtype          = mdatoms->typeA;
 883
 884     /* Setup water-specific parameters */
 885     inr              = nlist->iinr[0];
 886     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 887     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 888     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 889     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 890
 891     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 892     rcutoff_scalar   = fr->rcoulomb;
 893     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 894     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 895
 896     rswitch_scalar   = fr->rvdw_switch;
 897     rswitch          = _mm256_set1_ps(rswitch_scalar);
 898     /* Setup switch parameters */
 899     d_scalar         = rcutoff_scalar-rswitch_scalar;
 900     d                = _mm256_set1_ps(d_scalar);
 901     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 902     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 903     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 904     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 905     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 906     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 907
 908     /* Avoid stupid compiler warnings */
 909     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 910     j_coord_offsetA = 0;
 911     j_coord_offsetB = 0;
 912     j_coord_offsetC = 0;
 913     j_coord_offsetD = 0;
 914     j_coord_offsetE = 0;
 915     j_coord_offsetF = 0;
 916     j_coord_offsetG = 0;
 917     j_coord_offsetH = 0;
 918
 919     outeriter        = 0;
 920     inneriter        = 0;
 921
 922     for(iidx=0;iidx<4*DIM;iidx++)
 923     {
 924         scratch[iidx] = 0.0;
 925     }
 926
 927     /* Start outer loop over neighborlists */
 928     for(iidx=0; iidx<nri; iidx++)
 929     {
 930         /* Load shift vector for this list */
 931         i_shift_offset   = DIM*shiftidx[iidx];
 932
 933         /* Load limits for loop over neighbors */
 934         j_index_start    = jindex[iidx];
 935         j_index_end      = jindex[iidx+1];
 936
 937         /* Get outer coordinate index */
 938         inr              = iinr[iidx];
 939         i_coord_offset   = DIM*inr;
 940
 941         /* Load i particle coords and add shift vector */
 942         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 943                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 944
 945         fix0             = _mm256_setzero_ps();
 946         fiy0             = _mm256_setzero_ps();
 947         fiz0             = _mm256_setzero_ps();
 948         fix1             = _mm256_setzero_ps();
 949         fiy1             = _mm256_setzero_ps();
 950         fiz1             = _mm256_setzero_ps();
 951         fix2             = _mm256_setzero_ps();
 952         fiy2             = _mm256_setzero_ps();
 953         fiz2             = _mm256_setzero_ps();
 954         fix3             = _mm256_setzero_ps();
 955         fiy3             = _mm256_setzero_ps();
 956         fiz3             = _mm256_setzero_ps();
 957
 958         /* Start inner kernel loop */
 959         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 960         {
 961
 962             /* Get j neighbor index, and coordinate index */
 963             jnrA             = jjnr[jidx];
 964             jnrB             = jjnr[jidx+1];
 965             jnrC             = jjnr[jidx+2];
 966             jnrD             = jjnr[jidx+3];
 967             jnrE             = jjnr[jidx+4];
 968             jnrF             = jjnr[jidx+5];
 969             jnrG             = jjnr[jidx+6];
 970             jnrH             = jjnr[jidx+7];
 971             j_coord_offsetA  = DIM*jnrA;
 972             j_coord_offsetB  = DIM*jnrB;
 973             j_coord_offsetC  = DIM*jnrC;
 974             j_coord_offsetD  = DIM*jnrD;
 975             j_coord_offsetE  = DIM*jnrE;
 976             j_coord_offsetF  = DIM*jnrF;
 977             j_coord_offsetG  = DIM*jnrG;
 978             j_coord_offsetH  = DIM*jnrH;
 979
 980             /* load j atom coordinates */
 981             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 982                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 983                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 984                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 985                                                  &jx0,&jy0,&jz0);
 986
 987             /* Calculate displacement vector */
 988             dx00             = _mm256_sub_ps(ix0,jx0);
 989             dy00             = _mm256_sub_ps(iy0,jy0);
 990             dz00             = _mm256_sub_ps(iz0,jz0);
 991             dx10             = _mm256_sub_ps(ix1,jx0);
 992             dy10             = _mm256_sub_ps(iy1,jy0);
 993             dz10             = _mm256_sub_ps(iz1,jz0);
 994             dx20             = _mm256_sub_ps(ix2,jx0);
 995             dy20             = _mm256_sub_ps(iy2,jy0);
 996             dz20             = _mm256_sub_ps(iz2,jz0);
 997             dx30             = _mm256_sub_ps(ix3,jx0);
 998             dy30             = _mm256_sub_ps(iy3,jy0);
 999             dz30             = _mm256_sub_ps(iz3,jz0);
1000
1001             /* Calculate squared distance and things based on it */
1002             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1003             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1004             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1005             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1006
1007             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1008             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1009             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1010             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1011
1012             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1013             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1014             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1015             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1016
1017             /* Load parameters for j particles */
1018             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1019                                                                  charge+jnrC+0,charge+jnrD+0,
1020                                                                  charge+jnrE+0,charge+jnrF+0,
1021                                                                  charge+jnrG+0,charge+jnrH+0);
1022             vdwjidx0A        = 2*vdwtype[jnrA+0];
1023             vdwjidx0B        = 2*vdwtype[jnrB+0];
1024             vdwjidx0C        = 2*vdwtype[jnrC+0];
1025             vdwjidx0D        = 2*vdwtype[jnrD+0];
1026             vdwjidx0E        = 2*vdwtype[jnrE+0];
1027             vdwjidx0F        = 2*vdwtype[jnrF+0];
1028             vdwjidx0G        = 2*vdwtype[jnrG+0];
1029             vdwjidx0H        = 2*vdwtype[jnrH+0];
1030
1031             fjx0             = _mm256_setzero_ps();
1032             fjy0             = _mm256_setzero_ps();
1033             fjz0             = _mm256_setzero_ps();
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1040             {
1041
1042             r00              = _mm256_mul_ps(rsq00,rinv00);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1046                                             vdwioffsetptr0+vdwjidx0B,
1047                                             vdwioffsetptr0+vdwjidx0C,
1048                                             vdwioffsetptr0+vdwjidx0D,
1049                                             vdwioffsetptr0+vdwjidx0E,
1050                                             vdwioffsetptr0+vdwjidx0F,
1051                                             vdwioffsetptr0+vdwjidx0G,
1052                                             vdwioffsetptr0+vdwjidx0H,
1053                                             &c6_00,&c12_00);
1054
1055             /* LENNARD-JONES DISPERSION/REPULSION */
1056
1057             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1058             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1059             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1060             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1061             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1062
1063             d                = _mm256_sub_ps(r00,rswitch);
1064             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1065             d2               = _mm256_mul_ps(d,d);
1066             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1067
1068             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1069
1070             /* Evaluate switch function */
1071             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1072             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1073             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1074
1075             fscal            = fvdw;
1076
1077             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1078
1079             /* Calculate temporary vectorial force */
1080             tx               = _mm256_mul_ps(fscal,dx00);
1081             ty               = _mm256_mul_ps(fscal,dy00);
1082             tz               = _mm256_mul_ps(fscal,dz00);
1083
1084             /* Update vectorial force */
1085             fix0             = _mm256_add_ps(fix0,tx);
1086             fiy0             = _mm256_add_ps(fiy0,ty);
1087             fiz0             = _mm256_add_ps(fiz0,tz);
1088
1089             fjx0             = _mm256_add_ps(fjx0,tx);
1090             fjy0             = _mm256_add_ps(fjy0,ty);
1091             fjz0             = _mm256_add_ps(fjz0,tz);
1092
1093             }
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1100             {
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq10             = _mm256_mul_ps(iq1,jq0);
1104
1105             /* REACTION-FIELD ELECTROSTATICS */
1106             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1107
1108             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1109
1110             fscal            = felec;
1111
1112             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1113
1114             /* Calculate temporary vectorial force */
1115             tx               = _mm256_mul_ps(fscal,dx10);
1116             ty               = _mm256_mul_ps(fscal,dy10);
1117             tz               = _mm256_mul_ps(fscal,dz10);
1118
1119             /* Update vectorial force */
1120             fix1             = _mm256_add_ps(fix1,tx);
1121             fiy1             = _mm256_add_ps(fiy1,ty);
1122             fiz1             = _mm256_add_ps(fiz1,tz);
1123
1124             fjx0             = _mm256_add_ps(fjx0,tx);
1125             fjy0             = _mm256_add_ps(fjy0,ty);
1126             fjz0             = _mm256_add_ps(fjz0,tz);
1127
1128             }
1129
1130             /**************************
1131              * CALCULATE INTERACTIONS *
1132              **************************/
1133
1134             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1135             {
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq20             = _mm256_mul_ps(iq2,jq0);
1139
1140             /* REACTION-FIELD ELECTROSTATICS */
1141             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1142
1143             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm256_mul_ps(fscal,dx20);
1151             ty               = _mm256_mul_ps(fscal,dy20);
1152             tz               = _mm256_mul_ps(fscal,dz20);
1153
1154             /* Update vectorial force */
1155             fix2             = _mm256_add_ps(fix2,tx);
1156             fiy2             = _mm256_add_ps(fiy2,ty);
1157             fiz2             = _mm256_add_ps(fiz2,tz);
1158
1159             fjx0             = _mm256_add_ps(fjx0,tx);
1160             fjy0             = _mm256_add_ps(fjy0,ty);
1161             fjz0             = _mm256_add_ps(fjz0,tz);
1162
1163             }
1164
1165             /**************************
1166              * CALCULATE INTERACTIONS *
1167              **************************/
1168
1169             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1170             {
1171
1172             /* Compute parameters for interactions between i and j atoms */
1173             qq30             = _mm256_mul_ps(iq3,jq0);
1174
1175             /* REACTION-FIELD ELECTROSTATICS */
1176             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1177
1178             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm256_mul_ps(fscal,dx30);
1186             ty               = _mm256_mul_ps(fscal,dy30);
1187             tz               = _mm256_mul_ps(fscal,dz30);
1188
1189             /* Update vectorial force */
1190             fix3             = _mm256_add_ps(fix3,tx);
1191             fiy3             = _mm256_add_ps(fiy3,ty);
1192             fiz3             = _mm256_add_ps(fiz3,tz);
1193
1194             fjx0             = _mm256_add_ps(fjx0,tx);
1195             fjy0             = _mm256_add_ps(fjy0,ty);
1196             fjz0             = _mm256_add_ps(fjz0,tz);
1197
1198             }
1199
1200             fjptrA             = f+j_coord_offsetA;
1201             fjptrB             = f+j_coord_offsetB;
1202             fjptrC             = f+j_coord_offsetC;
1203             fjptrD             = f+j_coord_offsetD;
1204             fjptrE             = f+j_coord_offsetE;
1205             fjptrF             = f+j_coord_offsetF;
1206             fjptrG             = f+j_coord_offsetG;
1207             fjptrH             = f+j_coord_offsetH;
1208
1209             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1210
1211             /* Inner loop uses 149 flops */
1212         }
1213
1214         if(jidx<j_index_end)
1215         {
1216
1217             /* Get j neighbor index, and coordinate index */
1218             jnrlistA         = jjnr[jidx];
1219             jnrlistB         = jjnr[jidx+1];
1220             jnrlistC         = jjnr[jidx+2];
1221             jnrlistD         = jjnr[jidx+3];
1222             jnrlistE         = jjnr[jidx+4];
1223             jnrlistF         = jjnr[jidx+5];
1224             jnrlistG         = jjnr[jidx+6];
1225             jnrlistH         = jjnr[jidx+7];
1226             /* Sign of each element will be negative for non-real atoms.
1227              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1228              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1229              */
1230             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1231                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1232
1233             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1234             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1235             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1236             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1237             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1238             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1239             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1240             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1241             j_coord_offsetA  = DIM*jnrA;
1242             j_coord_offsetB  = DIM*jnrB;
1243             j_coord_offsetC  = DIM*jnrC;
1244             j_coord_offsetD  = DIM*jnrD;
1245             j_coord_offsetE  = DIM*jnrE;
1246             j_coord_offsetF  = DIM*jnrF;
1247             j_coord_offsetG  = DIM*jnrG;
1248             j_coord_offsetH  = DIM*jnrH;
1249
1250             /* load j atom coordinates */
1251             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1252                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1253                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1254                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1255                                                  &jx0,&jy0,&jz0);
1256
1257             /* Calculate displacement vector */
1258             dx00             = _mm256_sub_ps(ix0,jx0);
1259             dy00             = _mm256_sub_ps(iy0,jy0);
1260             dz00             = _mm256_sub_ps(iz0,jz0);
1261             dx10             = _mm256_sub_ps(ix1,jx0);
1262             dy10             = _mm256_sub_ps(iy1,jy0);
1263             dz10             = _mm256_sub_ps(iz1,jz0);
1264             dx20             = _mm256_sub_ps(ix2,jx0);
1265             dy20             = _mm256_sub_ps(iy2,jy0);
1266             dz20             = _mm256_sub_ps(iz2,jz0);
1267             dx30             = _mm256_sub_ps(ix3,jx0);
1268             dy30             = _mm256_sub_ps(iy3,jy0);
1269             dz30             = _mm256_sub_ps(iz3,jz0);
1270
1271             /* Calculate squared distance and things based on it */
1272             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1273             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1274             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1275             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1276
1277             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1278             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1279             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1280             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1281
1282             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1283             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1284             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1285             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1286
1287             /* Load parameters for j particles */
1288             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1289                                                                  charge+jnrC+0,charge+jnrD+0,
1290                                                                  charge+jnrE+0,charge+jnrF+0,
1291                                                                  charge+jnrG+0,charge+jnrH+0);
1292             vdwjidx0A        = 2*vdwtype[jnrA+0];
1293             vdwjidx0B        = 2*vdwtype[jnrB+0];
1294             vdwjidx0C        = 2*vdwtype[jnrC+0];
1295             vdwjidx0D        = 2*vdwtype[jnrD+0];
1296             vdwjidx0E        = 2*vdwtype[jnrE+0];
1297             vdwjidx0F        = 2*vdwtype[jnrF+0];
1298             vdwjidx0G        = 2*vdwtype[jnrG+0];
1299             vdwjidx0H        = 2*vdwtype[jnrH+0];
1300
1301             fjx0             = _mm256_setzero_ps();
1302             fjy0             = _mm256_setzero_ps();
1303             fjz0             = _mm256_setzero_ps();
1304
1305             /**************************
1306              * CALCULATE INTERACTIONS *
1307              **************************/
1308
1309             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1310             {
1311
1312             r00              = _mm256_mul_ps(rsq00,rinv00);
1313             r00              = _mm256_andnot_ps(dummy_mask,r00);
1314
1315             /* Compute parameters for interactions between i and j atoms */
1316             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1317                                             vdwioffsetptr0+vdwjidx0B,
1318                                             vdwioffsetptr0+vdwjidx0C,
1319                                             vdwioffsetptr0+vdwjidx0D,
1320                                             vdwioffsetptr0+vdwjidx0E,
1321                                             vdwioffsetptr0+vdwjidx0F,
1322                                             vdwioffsetptr0+vdwjidx0G,
1323                                             vdwioffsetptr0+vdwjidx0H,
1324                                             &c6_00,&c12_00);
1325
1326             /* LENNARD-JONES DISPERSION/REPULSION */
1327
1328             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1329             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1330             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1331             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1332             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1333
1334             d                = _mm256_sub_ps(r00,rswitch);
1335             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1336             d2               = _mm256_mul_ps(d,d);
1337             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1338
1339             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1340
1341             /* Evaluate switch function */
1342             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1343             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1344             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1345
1346             fscal            = fvdw;
1347
1348             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1349
1350             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1351
1352             /* Calculate temporary vectorial force */
1353             tx               = _mm256_mul_ps(fscal,dx00);
1354             ty               = _mm256_mul_ps(fscal,dy00);
1355             tz               = _mm256_mul_ps(fscal,dz00);
1356
1357             /* Update vectorial force */
1358             fix0             = _mm256_add_ps(fix0,tx);
1359             fiy0             = _mm256_add_ps(fiy0,ty);
1360             fiz0             = _mm256_add_ps(fiz0,tz);
1361
1362             fjx0             = _mm256_add_ps(fjx0,tx);
1363             fjy0             = _mm256_add_ps(fjy0,ty);
1364             fjz0             = _mm256_add_ps(fjz0,tz);
1365
1366             }
1367
1368             /**************************
1369              * CALCULATE INTERACTIONS *
1370              **************************/
1371
1372             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1373             {
1374
1375             /* Compute parameters for interactions between i and j atoms */
1376             qq10             = _mm256_mul_ps(iq1,jq0);
1377
1378             /* REACTION-FIELD ELECTROSTATICS */
1379             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1380
1381             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1382
1383             fscal            = felec;
1384
1385             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1386
1387             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1388
1389             /* Calculate temporary vectorial force */
1390             tx               = _mm256_mul_ps(fscal,dx10);
1391             ty               = _mm256_mul_ps(fscal,dy10);
1392             tz               = _mm256_mul_ps(fscal,dz10);
1393
1394             /* Update vectorial force */
1395             fix1             = _mm256_add_ps(fix1,tx);
1396             fiy1             = _mm256_add_ps(fiy1,ty);
1397             fiz1             = _mm256_add_ps(fiz1,tz);
1398
1399             fjx0             = _mm256_add_ps(fjx0,tx);
1400             fjy0             = _mm256_add_ps(fjy0,ty);
1401             fjz0             = _mm256_add_ps(fjz0,tz);
1402
1403             }
1404
1405             /**************************
1406              * CALCULATE INTERACTIONS *
1407              **************************/
1408
1409             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1410             {
1411
1412             /* Compute parameters for interactions between i and j atoms */
1413             qq20             = _mm256_mul_ps(iq2,jq0);
1414
1415             /* REACTION-FIELD ELECTROSTATICS */
1416             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1417
1418             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1419
1420             fscal            = felec;
1421
1422             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1423
1424             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1425
1426             /* Calculate temporary vectorial force */
1427             tx               = _mm256_mul_ps(fscal,dx20);
1428             ty               = _mm256_mul_ps(fscal,dy20);
1429             tz               = _mm256_mul_ps(fscal,dz20);
1430
1431             /* Update vectorial force */
1432             fix2             = _mm256_add_ps(fix2,tx);
1433             fiy2             = _mm256_add_ps(fiy2,ty);
1434             fiz2             = _mm256_add_ps(fiz2,tz);
1435
1436             fjx0             = _mm256_add_ps(fjx0,tx);
1437             fjy0             = _mm256_add_ps(fjy0,ty);
1438             fjz0             = _mm256_add_ps(fjz0,tz);
1439
1440             }
1441
1442             /**************************
1443              * CALCULATE INTERACTIONS *
1444              **************************/
1445
1446             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1447             {
1448
1449             /* Compute parameters for interactions between i and j atoms */
1450             qq30             = _mm256_mul_ps(iq3,jq0);
1451
1452             /* REACTION-FIELD ELECTROSTATICS */
1453             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1454
1455             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1456
1457             fscal            = felec;
1458
1459             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1460
1461             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1462
1463             /* Calculate temporary vectorial force */
1464             tx               = _mm256_mul_ps(fscal,dx30);
1465             ty               = _mm256_mul_ps(fscal,dy30);
1466             tz               = _mm256_mul_ps(fscal,dz30);
1467
1468             /* Update vectorial force */
1469             fix3             = _mm256_add_ps(fix3,tx);
1470             fiy3             = _mm256_add_ps(fiy3,ty);
1471             fiz3             = _mm256_add_ps(fiz3,tz);
1472
1473             fjx0             = _mm256_add_ps(fjx0,tx);
1474             fjy0             = _mm256_add_ps(fjy0,ty);
1475             fjz0             = _mm256_add_ps(fjz0,tz);
1476
1477             }
1478
1479             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1480             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1481             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1482             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1483             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1484             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1485             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1486             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1487
1488             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1489
1490             /* Inner loop uses 150 flops */
1491         }
1492
1493         /* End of innermost loop */
1494
1495         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1496                                                  f+i_coord_offset,fshift+i_shift_offset);
1497
1498         /* Increment number of inner iterations */
1499         inneriter                  += j_index_end - j_index_start;
1500
1501         /* Outer loop uses 24 flops */
1502     }
1503
1504     /* Increment number of outer iterations */
1505     outeriter        += nri;
1506
1507     /* Update outer/inner flops */
1508
1509     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1510 }