src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 107     real             rswitch_scalar,d_scalar;
 108     __m256d          dummy_mask,cutoff_mask;
 109     __m128           tmpmask0,tmpmask1;
 110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 111     __m256d          one     = _mm256_set1_pd(1.0);
 112     __m256d          two     = _mm256_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm256_set1_pd(fr->ic->k_rf);
 127     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 128     crf              = _mm256_set1_pd(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 136     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 137     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 144
 145     rswitch_scalar   = fr->rvdw_switch;
 146     rswitch          = _mm256_set1_pd(rswitch_scalar);
 147     /* Setup switch parameters */
 148     d_scalar         = rcutoff_scalar-rswitch_scalar;
 149     d                = _mm256_set1_pd(d_scalar);
 150     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 151     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 154     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 188                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 189
 190         fix0             = _mm256_setzero_pd();
 191         fiy0             = _mm256_setzero_pd();
 192         fiz0             = _mm256_setzero_pd();
 193         fix1             = _mm256_setzero_pd();
 194         fiy1             = _mm256_setzero_pd();
 195         fiz1             = _mm256_setzero_pd();
 196         fix2             = _mm256_setzero_pd();
 197         fiy2             = _mm256_setzero_pd();
 198         fiz2             = _mm256_setzero_pd();
 199         fix3             = _mm256_setzero_pd();
 200         fiy3             = _mm256_setzero_pd();
 201         fiz3             = _mm256_setzero_pd();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_pd();
 205         vvdwsum          = _mm256_setzero_pd();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220
 221             /* load j atom coordinates */
 222             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_pd(ix0,jx0);
 228             dy00             = _mm256_sub_pd(iy0,jy0);
 229             dz00             = _mm256_sub_pd(iz0,jz0);
 230             dx10             = _mm256_sub_pd(ix1,jx0);
 231             dy10             = _mm256_sub_pd(iy1,jy0);
 232             dz10             = _mm256_sub_pd(iz1,jz0);
 233             dx20             = _mm256_sub_pd(ix2,jx0);
 234             dy20             = _mm256_sub_pd(iy2,jy0);
 235             dz20             = _mm256_sub_pd(iz2,jz0);
 236             dx30             = _mm256_sub_pd(ix3,jx0);
 237             dy30             = _mm256_sub_pd(iy3,jy0);
 238             dz30             = _mm256_sub_pd(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 242             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 243             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 244             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 245
 246             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 247             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 248             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 249             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 250
 251             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 252             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 253             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 254             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0);
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260             vdwjidx0B        = 2*vdwtype[jnrB+0];
 261             vdwjidx0C        = 2*vdwtype[jnrC+0];
 262             vdwjidx0D        = 2*vdwtype[jnrD+0];
 263
 264             fjx0             = _mm256_setzero_pd();
 265             fjy0             = _mm256_setzero_pd();
 266             fjz0             = _mm256_setzero_pd();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 273             {
 274
 275             r00              = _mm256_mul_pd(rsq00,rinv00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 279                                             vdwioffsetptr0+vdwjidx0B,
 280                                             vdwioffsetptr0+vdwjidx0C,
 281                                             vdwioffsetptr0+vdwjidx0D,
 282                                             &c6_00,&c12_00);
 283
 284             /* LENNARD-JONES DISPERSION/REPULSION */
 285
 286             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 287             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 288             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 289             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 290             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 291
 292             d                = _mm256_sub_pd(r00,rswitch);
 293             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 294             d2               = _mm256_mul_pd(d,d);
 295             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 296
 297             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 298
 299             /* Evaluate switch function */
 300             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 301             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 302             vvdw             = _mm256_mul_pd(vvdw,sw);
 303             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 307             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 308
 309             fscal            = fvdw;
 310
 311             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm256_mul_pd(fscal,dx00);
 315             ty               = _mm256_mul_pd(fscal,dy00);
 316             tz               = _mm256_mul_pd(fscal,dz00);
 317
 318             /* Update vectorial force */
 319             fix0             = _mm256_add_pd(fix0,tx);
 320             fiy0             = _mm256_add_pd(fiy0,ty);
 321             fiz0             = _mm256_add_pd(fiz0,tz);
 322
 323             fjx0             = _mm256_add_pd(fjx0,tx);
 324             fjy0             = _mm256_add_pd(fjy0,ty);
 325             fjz0             = _mm256_add_pd(fjz0,tz);
 326
 327             }
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 334             {
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq10             = _mm256_mul_pd(iq1,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 341             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 342
 343             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm256_and_pd(velec,cutoff_mask);
 347             velecsum         = _mm256_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_pd(fscal,dx10);
 355             ty               = _mm256_mul_pd(fscal,dy10);
 356             tz               = _mm256_mul_pd(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm256_add_pd(fix1,tx);
 360             fiy1             = _mm256_add_pd(fiy1,ty);
 361             fiz1             = _mm256_add_pd(fiz1,tz);
 362
 363             fjx0             = _mm256_add_pd(fjx0,tx);
 364             fjy0             = _mm256_add_pd(fjy0,ty);
 365             fjz0             = _mm256_add_pd(fjz0,tz);
 366
 367             }
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 374             {
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq20             = _mm256_mul_pd(iq2,jq0);
 378
 379             /* REACTION-FIELD ELECTROSTATICS */
 380             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 381             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 382
 383             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm256_and_pd(velec,cutoff_mask);
 387             velecsum         = _mm256_add_pd(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm256_mul_pd(fscal,dx20);
 395             ty               = _mm256_mul_pd(fscal,dy20);
 396             tz               = _mm256_mul_pd(fscal,dz20);
 397
 398             /* Update vectorial force */
 399             fix2             = _mm256_add_pd(fix2,tx);
 400             fiy2             = _mm256_add_pd(fiy2,ty);
 401             fiz2             = _mm256_add_pd(fiz2,tz);
 402
 403             fjx0             = _mm256_add_pd(fjx0,tx);
 404             fjy0             = _mm256_add_pd(fjy0,ty);
 405             fjz0             = _mm256_add_pd(fjz0,tz);
 406
 407             }
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 414             {
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq30             = _mm256_mul_pd(iq3,jq0);
 418
 419             /* REACTION-FIELD ELECTROSTATICS */
 420             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 421             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 422
 423             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm256_and_pd(velec,cutoff_mask);
 427             velecsum         = _mm256_add_pd(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_pd(fscal,dx30);
 435             ty               = _mm256_mul_pd(fscal,dy30);
 436             tz               = _mm256_mul_pd(fscal,dz30);
 437
 438             /* Update vectorial force */
 439             fix3             = _mm256_add_pd(fix3,tx);
 440             fiy3             = _mm256_add_pd(fiy3,ty);
 441             fiz3             = _mm256_add_pd(fiz3,tz);
 442
 443             fjx0             = _mm256_add_pd(fjx0,tx);
 444             fjy0             = _mm256_add_pd(fjy0,ty);
 445             fjz0             = _mm256_add_pd(fjz0,tz);
 446
 447             }
 448
 449             fjptrA             = f+j_coord_offsetA;
 450             fjptrB             = f+j_coord_offsetB;
 451             fjptrC             = f+j_coord_offsetC;
 452             fjptrD             = f+j_coord_offsetD;
 453
 454             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 455
 456             /* Inner loop uses 170 flops */
 457         }
 458
 459         if(jidx<j_index_end)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrlistA         = jjnr[jidx];
 464             jnrlistB         = jjnr[jidx+1];
 465             jnrlistC         = jjnr[jidx+2];
 466             jnrlistD         = jjnr[jidx+3];
 467             /* Sign of each element will be negative for non-real atoms.
 468              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 469              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 470              */
 471             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 472
 473             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 474             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 475             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 476
 477             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 478             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 479             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 480             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 481             j_coord_offsetA  = DIM*jnrA;
 482             j_coord_offsetB  = DIM*jnrB;
 483             j_coord_offsetC  = DIM*jnrC;
 484             j_coord_offsetD  = DIM*jnrD;
 485
 486             /* load j atom coordinates */
 487             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 488                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 489                                                  &jx0,&jy0,&jz0);
 490
 491             /* Calculate displacement vector */
 492             dx00             = _mm256_sub_pd(ix0,jx0);
 493             dy00             = _mm256_sub_pd(iy0,jy0);
 494             dz00             = _mm256_sub_pd(iz0,jz0);
 495             dx10             = _mm256_sub_pd(ix1,jx0);
 496             dy10             = _mm256_sub_pd(iy1,jy0);
 497             dz10             = _mm256_sub_pd(iz1,jz0);
 498             dx20             = _mm256_sub_pd(ix2,jx0);
 499             dy20             = _mm256_sub_pd(iy2,jy0);
 500             dz20             = _mm256_sub_pd(iz2,jz0);
 501             dx30             = _mm256_sub_pd(ix3,jx0);
 502             dy30             = _mm256_sub_pd(iy3,jy0);
 503             dz30             = _mm256_sub_pd(iz3,jz0);
 504
 505             /* Calculate squared distance and things based on it */
 506             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 507             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 508             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 509             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 510
 511             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 512             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 513             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 514             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 515
 516             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 517             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 518             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 519             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 523                                                                  charge+jnrC+0,charge+jnrD+0);
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525             vdwjidx0B        = 2*vdwtype[jnrB+0];
 526             vdwjidx0C        = 2*vdwtype[jnrC+0];
 527             vdwjidx0D        = 2*vdwtype[jnrD+0];
 528
 529             fjx0             = _mm256_setzero_pd();
 530             fjy0             = _mm256_setzero_pd();
 531             fjz0             = _mm256_setzero_pd();
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 538             {
 539
 540             r00              = _mm256_mul_pd(rsq00,rinv00);
 541             r00              = _mm256_andnot_pd(dummy_mask,r00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 545                                             vdwioffsetptr0+vdwjidx0B,
 546                                             vdwioffsetptr0+vdwjidx0C,
 547                                             vdwioffsetptr0+vdwjidx0D,
 548                                             &c6_00,&c12_00);
 549
 550             /* LENNARD-JONES DISPERSION/REPULSION */
 551
 552             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 553             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 554             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 555             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 556             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 557
 558             d                = _mm256_sub_pd(r00,rswitch);
 559             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 560             d2               = _mm256_mul_pd(d,d);
 561             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 562
 563             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 564
 565             /* Evaluate switch function */
 566             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 567             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 568             vvdw             = _mm256_mul_pd(vvdw,sw);
 569             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 573             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 574             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 575
 576             fscal            = fvdw;
 577
 578             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 579
 580             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx00);
 584             ty               = _mm256_mul_pd(fscal,dy00);
 585             tz               = _mm256_mul_pd(fscal,dz00);
 586
 587             /* Update vectorial force */
 588             fix0             = _mm256_add_pd(fix0,tx);
 589             fiy0             = _mm256_add_pd(fiy0,ty);
 590             fiz0             = _mm256_add_pd(fiz0,tz);
 591
 592             fjx0             = _mm256_add_pd(fjx0,tx);
 593             fjy0             = _mm256_add_pd(fjy0,ty);
 594             fjz0             = _mm256_add_pd(fjz0,tz);
 595
 596             }
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 603             {
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq10             = _mm256_mul_pd(iq1,jq0);
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 610             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 611
 612             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm256_and_pd(velec,cutoff_mask);
 616             velec            = _mm256_andnot_pd(dummy_mask,velec);
 617             velecsum         = _mm256_add_pd(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 622
 623             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm256_mul_pd(fscal,dx10);
 627             ty               = _mm256_mul_pd(fscal,dy10);
 628             tz               = _mm256_mul_pd(fscal,dz10);
 629
 630             /* Update vectorial force */
 631             fix1             = _mm256_add_pd(fix1,tx);
 632             fiy1             = _mm256_add_pd(fiy1,ty);
 633             fiz1             = _mm256_add_pd(fiz1,tz);
 634
 635             fjx0             = _mm256_add_pd(fjx0,tx);
 636             fjy0             = _mm256_add_pd(fjy0,ty);
 637             fjz0             = _mm256_add_pd(fjz0,tz);
 638
 639             }
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 646             {
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq20             = _mm256_mul_pd(iq2,jq0);
 650
 651             /* REACTION-FIELD ELECTROSTATICS */
 652             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 653             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 654
 655             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm256_and_pd(velec,cutoff_mask);
 659             velec            = _mm256_andnot_pd(dummy_mask,velec);
 660             velecsum         = _mm256_add_pd(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 665
 666             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_pd(fscal,dx20);
 670             ty               = _mm256_mul_pd(fscal,dy20);
 671             tz               = _mm256_mul_pd(fscal,dz20);
 672
 673             /* Update vectorial force */
 674             fix2             = _mm256_add_pd(fix2,tx);
 675             fiy2             = _mm256_add_pd(fiy2,ty);
 676             fiz2             = _mm256_add_pd(fiz2,tz);
 677
 678             fjx0             = _mm256_add_pd(fjx0,tx);
 679             fjy0             = _mm256_add_pd(fjy0,ty);
 680             fjz0             = _mm256_add_pd(fjz0,tz);
 681
 682             }
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 689             {
 690
 691             /* Compute parameters for interactions between i and j atoms */
 692             qq30             = _mm256_mul_pd(iq3,jq0);
 693
 694             /* REACTION-FIELD ELECTROSTATICS */
 695             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 696             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 697
 698             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 699
 700             /* Update potential sum for this i atom from the interaction with this j atom. */
 701             velec            = _mm256_and_pd(velec,cutoff_mask);
 702             velec            = _mm256_andnot_pd(dummy_mask,velec);
 703             velecsum         = _mm256_add_pd(velecsum,velec);
 704
 705             fscal            = felec;
 706
 707             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 708
 709             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm256_mul_pd(fscal,dx30);
 713             ty               = _mm256_mul_pd(fscal,dy30);
 714             tz               = _mm256_mul_pd(fscal,dz30);
 715
 716             /* Update vectorial force */
 717             fix3             = _mm256_add_pd(fix3,tx);
 718             fiy3             = _mm256_add_pd(fiy3,ty);
 719             fiz3             = _mm256_add_pd(fiz3,tz);
 720
 721             fjx0             = _mm256_add_pd(fjx0,tx);
 722             fjy0             = _mm256_add_pd(fjy0,ty);
 723             fjz0             = _mm256_add_pd(fjz0,tz);
 724
 725             }
 726
 727             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 728             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 729             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 730             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 731
 732             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 733
 734             /* Inner loop uses 171 flops */
 735         }
 736
 737         /* End of innermost loop */
 738
 739         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 740                                                  f+i_coord_offset,fshift+i_shift_offset);
 741
 742         ggid                        = gid[iidx];
 743         /* Update potential energies */
 744         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 745         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 746
 747         /* Increment number of inner iterations */
 748         inneriter                  += j_index_end - j_index_start;
 749
 750         /* Outer loop uses 26 flops */
 751     }
 752
 753     /* Increment number of outer iterations */
 754     outeriter        += nri;
 755
 756     /* Update outer/inner flops */
 757
 758     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
 759 }
 760 /*
 761  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 762  * Electrostatics interaction: ReactionField
 763  * VdW interaction:            LennardJones
 764  * Geometry:                   Water4-Particle
 765  * Calculate force/pot:        Force
 766  */
 767 void
 768 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 769                     (t_nblist                    * gmx_restrict       nlist,
 770                      rvec                        * gmx_restrict          xx,
 771                      rvec                        * gmx_restrict          ff,
 772                      t_forcerec                  * gmx_restrict          fr,
 773                      t_mdatoms                   * gmx_restrict     mdatoms,
 774                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 775                      t_nrnb                      * gmx_restrict        nrnb)
 776 {
 777     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 778      * just 0 for non-waters.
 779      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 780      * jnr indices corresponding to data put in the four positions in the SIMD register.
 781      */
 782     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 783     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 784     int              jnrA,jnrB,jnrC,jnrD;
 785     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 786     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 787     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 788     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 789     real             rcutoff_scalar;
 790     real             *shiftvec,*fshift,*x,*f;
 791     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 792     real             scratch[4*DIM];
 793     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 794     real *           vdwioffsetptr0;
 795     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 796     real *           vdwioffsetptr1;
 797     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 798     real *           vdwioffsetptr2;
 799     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 800     real *           vdwioffsetptr3;
 801     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 802     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 803     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 804     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 805     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 806     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 807     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 808     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 809     real             *charge;
 810     int              nvdwtype;
 811     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 812     int              *vdwtype;
 813     real             *vdwparam;
 814     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 815     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 816     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 817     real             rswitch_scalar,d_scalar;
 818     __m256d          dummy_mask,cutoff_mask;
 819     __m128           tmpmask0,tmpmask1;
 820     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 821     __m256d          one     = _mm256_set1_pd(1.0);
 822     __m256d          two     = _mm256_set1_pd(2.0);
 823     x                = xx[0];
 824     f                = ff[0];
 825
 826     nri              = nlist->nri;
 827     iinr             = nlist->iinr;
 828     jindex           = nlist->jindex;
 829     jjnr             = nlist->jjnr;
 830     shiftidx         = nlist->shift;
 831     gid              = nlist->gid;
 832     shiftvec         = fr->shift_vec[0];
 833     fshift           = fr->fshift[0];
 834     facel            = _mm256_set1_pd(fr->epsfac);
 835     charge           = mdatoms->chargeA;
 836     krf              = _mm256_set1_pd(fr->ic->k_rf);
 837     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 838     crf              = _mm256_set1_pd(fr->ic->c_rf);
 839     nvdwtype         = fr->ntype;
 840     vdwparam         = fr->nbfp;
 841     vdwtype          = mdatoms->typeA;
 842
 843     /* Setup water-specific parameters */
 844     inr              = nlist->iinr[0];
 845     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 846     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 847     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 848     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 849
 850     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 851     rcutoff_scalar   = fr->rcoulomb;
 852     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 853     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 854
 855     rswitch_scalar   = fr->rvdw_switch;
 856     rswitch          = _mm256_set1_pd(rswitch_scalar);
 857     /* Setup switch parameters */
 858     d_scalar         = rcutoff_scalar-rswitch_scalar;
 859     d                = _mm256_set1_pd(d_scalar);
 860     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 861     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 862     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 863     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 864     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 865     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 899
 900         fix0             = _mm256_setzero_pd();
 901         fiy0             = _mm256_setzero_pd();
 902         fiz0             = _mm256_setzero_pd();
 903         fix1             = _mm256_setzero_pd();
 904         fiy1             = _mm256_setzero_pd();
 905         fiz1             = _mm256_setzero_pd();
 906         fix2             = _mm256_setzero_pd();
 907         fiy2             = _mm256_setzero_pd();
 908         fiz2             = _mm256_setzero_pd();
 909         fix3             = _mm256_setzero_pd();
 910         fiy3             = _mm256_setzero_pd();
 911         fiz3             = _mm256_setzero_pd();
 912
 913         /* Start inner kernel loop */
 914         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrA             = jjnr[jidx];
 919             jnrB             = jjnr[jidx+1];
 920             jnrC             = jjnr[jidx+2];
 921             jnrD             = jjnr[jidx+3];
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926
 927             /* load j atom coordinates */
 928             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 929                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 930                                                  &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm256_sub_pd(ix0,jx0);
 934             dy00             = _mm256_sub_pd(iy0,jy0);
 935             dz00             = _mm256_sub_pd(iz0,jz0);
 936             dx10             = _mm256_sub_pd(ix1,jx0);
 937             dy10             = _mm256_sub_pd(iy1,jy0);
 938             dz10             = _mm256_sub_pd(iz1,jz0);
 939             dx20             = _mm256_sub_pd(ix2,jx0);
 940             dy20             = _mm256_sub_pd(iy2,jy0);
 941             dz20             = _mm256_sub_pd(iz2,jz0);
 942             dx30             = _mm256_sub_pd(ix3,jx0);
 943             dy30             = _mm256_sub_pd(iy3,jy0);
 944             dz30             = _mm256_sub_pd(iz3,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 948             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 949             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 950             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 951
 952             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 953             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 954             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 955             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 956
 957             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 958             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 959             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 960             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 961
 962             /* Load parameters for j particles */
 963             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 964                                                                  charge+jnrC+0,charge+jnrD+0);
 965             vdwjidx0A        = 2*vdwtype[jnrA+0];
 966             vdwjidx0B        = 2*vdwtype[jnrB+0];
 967             vdwjidx0C        = 2*vdwtype[jnrC+0];
 968             vdwjidx0D        = 2*vdwtype[jnrD+0];
 969
 970             fjx0             = _mm256_setzero_pd();
 971             fjy0             = _mm256_setzero_pd();
 972             fjz0             = _mm256_setzero_pd();
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 979             {
 980
 981             r00              = _mm256_mul_pd(rsq00,rinv00);
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 985                                             vdwioffsetptr0+vdwjidx0B,
 986                                             vdwioffsetptr0+vdwjidx0C,
 987                                             vdwioffsetptr0+vdwjidx0D,
 988                                             &c6_00,&c12_00);
 989
 990             /* LENNARD-JONES DISPERSION/REPULSION */
 991
 992             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 993             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 994             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 995             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 996             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 997
 998             d                = _mm256_sub_pd(r00,rswitch);
 999             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1000             d2               = _mm256_mul_pd(d,d);
1001             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1002
1003             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1004
1005             /* Evaluate switch function */
1006             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1007             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1008             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1009
1010             fscal            = fvdw;
1011
1012             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_pd(fscal,dx00);
1016             ty               = _mm256_mul_pd(fscal,dy00);
1017             tz               = _mm256_mul_pd(fscal,dz00);
1018
1019             /* Update vectorial force */
1020             fix0             = _mm256_add_pd(fix0,tx);
1021             fiy0             = _mm256_add_pd(fiy0,ty);
1022             fiz0             = _mm256_add_pd(fiz0,tz);
1023
1024             fjx0             = _mm256_add_pd(fjx0,tx);
1025             fjy0             = _mm256_add_pd(fjy0,ty);
1026             fjz0             = _mm256_add_pd(fjz0,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1035             {
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq10             = _mm256_mul_pd(iq1,jq0);
1039
1040             /* REACTION-FIELD ELECTROSTATICS */
1041             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1042
1043             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm256_mul_pd(fscal,dx10);
1051             ty               = _mm256_mul_pd(fscal,dy10);
1052             tz               = _mm256_mul_pd(fscal,dz10);
1053
1054             /* Update vectorial force */
1055             fix1             = _mm256_add_pd(fix1,tx);
1056             fiy1             = _mm256_add_pd(fiy1,ty);
1057             fiz1             = _mm256_add_pd(fiz1,tz);
1058
1059             fjx0             = _mm256_add_pd(fjx0,tx);
1060             fjy0             = _mm256_add_pd(fjy0,ty);
1061             fjz0             = _mm256_add_pd(fjz0,tz);
1062
1063             }
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1070             {
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq20             = _mm256_mul_pd(iq2,jq0);
1074
1075             /* REACTION-FIELD ELECTROSTATICS */
1076             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1077
1078             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1079
1080             fscal            = felec;
1081
1082             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm256_mul_pd(fscal,dx20);
1086             ty               = _mm256_mul_pd(fscal,dy20);
1087             tz               = _mm256_mul_pd(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm256_add_pd(fix2,tx);
1091             fiy2             = _mm256_add_pd(fiy2,ty);
1092             fiz2             = _mm256_add_pd(fiz2,tz);
1093
1094             fjx0             = _mm256_add_pd(fjx0,tx);
1095             fjy0             = _mm256_add_pd(fjy0,ty);
1096             fjz0             = _mm256_add_pd(fjz0,tz);
1097
1098             }
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1105             {
1106
1107             /* Compute parameters for interactions between i and j atoms */
1108             qq30             = _mm256_mul_pd(iq3,jq0);
1109
1110             /* REACTION-FIELD ELECTROSTATICS */
1111             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1112
1113             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm256_mul_pd(fscal,dx30);
1121             ty               = _mm256_mul_pd(fscal,dy30);
1122             tz               = _mm256_mul_pd(fscal,dz30);
1123
1124             /* Update vectorial force */
1125             fix3             = _mm256_add_pd(fix3,tx);
1126             fiy3             = _mm256_add_pd(fiy3,ty);
1127             fiz3             = _mm256_add_pd(fiz3,tz);
1128
1129             fjx0             = _mm256_add_pd(fjx0,tx);
1130             fjy0             = _mm256_add_pd(fjy0,ty);
1131             fjz0             = _mm256_add_pd(fjz0,tz);
1132
1133             }
1134
1135             fjptrA             = f+j_coord_offsetA;
1136             fjptrB             = f+j_coord_offsetB;
1137             fjptrC             = f+j_coord_offsetC;
1138             fjptrD             = f+j_coord_offsetD;
1139
1140             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1141
1142             /* Inner loop uses 149 flops */
1143         }
1144
1145         if(jidx<j_index_end)
1146         {
1147
1148             /* Get j neighbor index, and coordinate index */
1149             jnrlistA         = jjnr[jidx];
1150             jnrlistB         = jjnr[jidx+1];
1151             jnrlistC         = jjnr[jidx+2];
1152             jnrlistD         = jjnr[jidx+3];
1153             /* Sign of each element will be negative for non-real atoms.
1154              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1155              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1156              */
1157             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1158
1159             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1160             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1161             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1162
1163             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1164             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1165             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1166             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1167             j_coord_offsetA  = DIM*jnrA;
1168             j_coord_offsetB  = DIM*jnrB;
1169             j_coord_offsetC  = DIM*jnrC;
1170             j_coord_offsetD  = DIM*jnrD;
1171
1172             /* load j atom coordinates */
1173             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1174                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1175                                                  &jx0,&jy0,&jz0);
1176
1177             /* Calculate displacement vector */
1178             dx00             = _mm256_sub_pd(ix0,jx0);
1179             dy00             = _mm256_sub_pd(iy0,jy0);
1180             dz00             = _mm256_sub_pd(iz0,jz0);
1181             dx10             = _mm256_sub_pd(ix1,jx0);
1182             dy10             = _mm256_sub_pd(iy1,jy0);
1183             dz10             = _mm256_sub_pd(iz1,jz0);
1184             dx20             = _mm256_sub_pd(ix2,jx0);
1185             dy20             = _mm256_sub_pd(iy2,jy0);
1186             dz20             = _mm256_sub_pd(iz2,jz0);
1187             dx30             = _mm256_sub_pd(ix3,jx0);
1188             dy30             = _mm256_sub_pd(iy3,jy0);
1189             dz30             = _mm256_sub_pd(iz3,jz0);
1190
1191             /* Calculate squared distance and things based on it */
1192             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1193             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1194             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1195             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1196
1197             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1198             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1199             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1200             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1201
1202             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1203             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1204             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1205             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1206
1207             /* Load parameters for j particles */
1208             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1209                                                                  charge+jnrC+0,charge+jnrD+0);
1210             vdwjidx0A        = 2*vdwtype[jnrA+0];
1211             vdwjidx0B        = 2*vdwtype[jnrB+0];
1212             vdwjidx0C        = 2*vdwtype[jnrC+0];
1213             vdwjidx0D        = 2*vdwtype[jnrD+0];
1214
1215             fjx0             = _mm256_setzero_pd();
1216             fjy0             = _mm256_setzero_pd();
1217             fjz0             = _mm256_setzero_pd();
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1224             {
1225
1226             r00              = _mm256_mul_pd(rsq00,rinv00);
1227             r00              = _mm256_andnot_pd(dummy_mask,r00);
1228
1229             /* Compute parameters for interactions between i and j atoms */
1230             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1231                                             vdwioffsetptr0+vdwjidx0B,
1232                                             vdwioffsetptr0+vdwjidx0C,
1233                                             vdwioffsetptr0+vdwjidx0D,
1234                                             &c6_00,&c12_00);
1235
1236             /* LENNARD-JONES DISPERSION/REPULSION */
1237
1238             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1239             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1240             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1241             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1242             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1243
1244             d                = _mm256_sub_pd(r00,rswitch);
1245             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1246             d2               = _mm256_mul_pd(d,d);
1247             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1248
1249             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1250
1251             /* Evaluate switch function */
1252             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1253             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1254             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1255
1256             fscal            = fvdw;
1257
1258             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1259
1260             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1261
1262             /* Calculate temporary vectorial force */
1263             tx               = _mm256_mul_pd(fscal,dx00);
1264             ty               = _mm256_mul_pd(fscal,dy00);
1265             tz               = _mm256_mul_pd(fscal,dz00);
1266
1267             /* Update vectorial force */
1268             fix0             = _mm256_add_pd(fix0,tx);
1269             fiy0             = _mm256_add_pd(fiy0,ty);
1270             fiz0             = _mm256_add_pd(fiz0,tz);
1271
1272             fjx0             = _mm256_add_pd(fjx0,tx);
1273             fjy0             = _mm256_add_pd(fjy0,ty);
1274             fjz0             = _mm256_add_pd(fjz0,tz);
1275
1276             }
1277
1278             /**************************
1279              * CALCULATE INTERACTIONS *
1280              **************************/
1281
1282             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1283             {
1284
1285             /* Compute parameters for interactions between i and j atoms */
1286             qq10             = _mm256_mul_pd(iq1,jq0);
1287
1288             /* REACTION-FIELD ELECTROSTATICS */
1289             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1290
1291             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1296
1297             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1298
1299             /* Calculate temporary vectorial force */
1300             tx               = _mm256_mul_pd(fscal,dx10);
1301             ty               = _mm256_mul_pd(fscal,dy10);
1302             tz               = _mm256_mul_pd(fscal,dz10);
1303
1304             /* Update vectorial force */
1305             fix1             = _mm256_add_pd(fix1,tx);
1306             fiy1             = _mm256_add_pd(fiy1,ty);
1307             fiz1             = _mm256_add_pd(fiz1,tz);
1308
1309             fjx0             = _mm256_add_pd(fjx0,tx);
1310             fjy0             = _mm256_add_pd(fjy0,ty);
1311             fjz0             = _mm256_add_pd(fjz0,tz);
1312
1313             }
1314
1315             /**************************
1316              * CALCULATE INTERACTIONS *
1317              **************************/
1318
1319             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1320             {
1321
1322             /* Compute parameters for interactions between i and j atoms */
1323             qq20             = _mm256_mul_pd(iq2,jq0);
1324
1325             /* REACTION-FIELD ELECTROSTATICS */
1326             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1327
1328             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1329
1330             fscal            = felec;
1331
1332             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1333
1334             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1335
1336             /* Calculate temporary vectorial force */
1337             tx               = _mm256_mul_pd(fscal,dx20);
1338             ty               = _mm256_mul_pd(fscal,dy20);
1339             tz               = _mm256_mul_pd(fscal,dz20);
1340
1341             /* Update vectorial force */
1342             fix2             = _mm256_add_pd(fix2,tx);
1343             fiy2             = _mm256_add_pd(fiy2,ty);
1344             fiz2             = _mm256_add_pd(fiz2,tz);
1345
1346             fjx0             = _mm256_add_pd(fjx0,tx);
1347             fjy0             = _mm256_add_pd(fjy0,ty);
1348             fjz0             = _mm256_add_pd(fjz0,tz);
1349
1350             }
1351
1352             /**************************
1353              * CALCULATE INTERACTIONS *
1354              **************************/
1355
1356             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1357             {
1358
1359             /* Compute parameters for interactions between i and j atoms */
1360             qq30             = _mm256_mul_pd(iq3,jq0);
1361
1362             /* REACTION-FIELD ELECTROSTATICS */
1363             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1364
1365             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1366
1367             fscal            = felec;
1368
1369             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1370
1371             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1372
1373             /* Calculate temporary vectorial force */
1374             tx               = _mm256_mul_pd(fscal,dx30);
1375             ty               = _mm256_mul_pd(fscal,dy30);
1376             tz               = _mm256_mul_pd(fscal,dz30);
1377
1378             /* Update vectorial force */
1379             fix3             = _mm256_add_pd(fix3,tx);
1380             fiy3             = _mm256_add_pd(fiy3,ty);
1381             fiz3             = _mm256_add_pd(fiz3,tz);
1382
1383             fjx0             = _mm256_add_pd(fjx0,tx);
1384             fjy0             = _mm256_add_pd(fjy0,ty);
1385             fjz0             = _mm256_add_pd(fjz0,tz);
1386
1387             }
1388
1389             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1390             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1391             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1392             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1393
1394             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1395
1396             /* Inner loop uses 150 flops */
1397         }
1398
1399         /* End of innermost loop */
1400
1401         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1402                                                  f+i_coord_offset,fshift+i_shift_offset);
1403
1404         /* Increment number of inner iterations */
1405         inneriter                  += j_index_end - j_index_start;
1406
1407         /* Outer loop uses 24 flops */
1408     }
1409
1410     /* Increment number of outer iterations */
1411     outeriter        += nri;
1412
1413     /* Update outer/inner flops */
1414
1415     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1416 }