src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwCSTab_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_VdwCSTab_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_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     __m128i          vfitab;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     krf              = _mm256_set1_pd(fr->ic->k_rf);
 129     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 130     crf              = _mm256_set1_pd(fr->ic->c_rf);
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     vftab            = kernel_data->table_vdw->data;
 136     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 141     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 142     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 143     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 144
 145     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 146     rcutoff_scalar   = fr->rcoulomb;
 147     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 148     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm256_setzero_pd();
 184         fiy0             = _mm256_setzero_pd();
 185         fiz0             = _mm256_setzero_pd();
 186         fix1             = _mm256_setzero_pd();
 187         fiy1             = _mm256_setzero_pd();
 188         fiz1             = _mm256_setzero_pd();
 189         fix2             = _mm256_setzero_pd();
 190         fiy2             = _mm256_setzero_pd();
 191         fiz2             = _mm256_setzero_pd();
 192         fix3             = _mm256_setzero_pd();
 193         fiy3             = _mm256_setzero_pd();
 194         fiz3             = _mm256_setzero_pd();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm256_setzero_pd();
 198         vvdwsum          = _mm256_setzero_pd();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_pd(ix0,jx0);
 221             dy00             = _mm256_sub_pd(iy0,jy0);
 222             dz00             = _mm256_sub_pd(iz0,jz0);
 223             dx10             = _mm256_sub_pd(ix1,jx0);
 224             dy10             = _mm256_sub_pd(iy1,jy0);
 225             dz10             = _mm256_sub_pd(iz1,jz0);
 226             dx20             = _mm256_sub_pd(ix2,jx0);
 227             dy20             = _mm256_sub_pd(iy2,jy0);
 228             dz20             = _mm256_sub_pd(iz2,jz0);
 229             dx30             = _mm256_sub_pd(ix3,jx0);
 230             dy30             = _mm256_sub_pd(iy3,jy0);
 231             dz30             = _mm256_sub_pd(iz3,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 235             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 236             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 237             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 238
 239             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 240             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 241             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 242             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 243
 244             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 245             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 246             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 247
 248             /* Load parameters for j particles */
 249             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 250                                                                  charge+jnrC+0,charge+jnrD+0);
 251             vdwjidx0A        = 2*vdwtype[jnrA+0];
 252             vdwjidx0B        = 2*vdwtype[jnrB+0];
 253             vdwjidx0C        = 2*vdwtype[jnrC+0];
 254             vdwjidx0D        = 2*vdwtype[jnrD+0];
 255
 256             fjx0             = _mm256_setzero_pd();
 257             fjy0             = _mm256_setzero_pd();
 258             fjz0             = _mm256_setzero_pd();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r00              = _mm256_mul_pd(rsq00,rinv00);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 268                                             vdwioffsetptr0+vdwjidx0B,
 269                                             vdwioffsetptr0+vdwjidx0C,
 270                                             vdwioffsetptr0+vdwjidx0D,
 271                                             &c6_00,&c12_00);
 272
 273             /* Calculate table index by multiplying r with table scale and truncate to integer */
 274             rt               = _mm256_mul_pd(r00,vftabscale);
 275             vfitab           = _mm256_cvttpd_epi32(rt);
 276             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 277             vfitab           = _mm_slli_epi32(vfitab,3);
 278
 279             /* CUBIC SPLINE TABLE DISPERSION */
 280             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 281             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 282             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 283             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 284             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 285             Heps             = _mm256_mul_pd(vfeps,H);
 286             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 287             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 288             vvdw6            = _mm256_mul_pd(c6_00,VV);
 289             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 290             fvdw6            = _mm256_mul_pd(c6_00,FF);
 291
 292             /* CUBIC SPLINE TABLE REPULSION */
 293             vfitab           = _mm_add_epi32(vfitab,ifour);
 294             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 295             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 296             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 297             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 298             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 299             Heps             = _mm256_mul_pd(vfeps,H);
 300             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 301             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 302             vvdw12           = _mm256_mul_pd(c12_00,VV);
 303             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 304             fvdw12           = _mm256_mul_pd(c12_00,FF);
 305             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 306             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 310
 311             fscal            = fvdw;
 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              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 332             {
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq10             = _mm256_mul_pd(iq1,jq0);
 336
 337             /* REACTION-FIELD ELECTROSTATICS */
 338             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 339             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 340
 341             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm256_and_pd(velec,cutoff_mask);
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx10);
 353             ty               = _mm256_mul_pd(fscal,dy10);
 354             tz               = _mm256_mul_pd(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm256_add_pd(fix1,tx);
 358             fiy1             = _mm256_add_pd(fiy1,ty);
 359             fiz1             = _mm256_add_pd(fiz1,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             }
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 372             {
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq20             = _mm256_mul_pd(iq2,jq0);
 376
 377             /* REACTION-FIELD ELECTROSTATICS */
 378             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 379             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 380
 381             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm256_and_pd(velec,cutoff_mask);
 385             velecsum         = _mm256_add_pd(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_pd(fscal,dx20);
 393             ty               = _mm256_mul_pd(fscal,dy20);
 394             tz               = _mm256_mul_pd(fscal,dz20);
 395
 396             /* Update vectorial force */
 397             fix2             = _mm256_add_pd(fix2,tx);
 398             fiy2             = _mm256_add_pd(fiy2,ty);
 399             fiz2             = _mm256_add_pd(fiz2,tz);
 400
 401             fjx0             = _mm256_add_pd(fjx0,tx);
 402             fjy0             = _mm256_add_pd(fjy0,ty);
 403             fjz0             = _mm256_add_pd(fjz0,tz);
 404
 405             }
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 412             {
 413
 414             /* Compute parameters for interactions between i and j atoms */
 415             qq30             = _mm256_mul_pd(iq3,jq0);
 416
 417             /* REACTION-FIELD ELECTROSTATICS */
 418             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 419             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 420
 421             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velec            = _mm256_and_pd(velec,cutoff_mask);
 425             velecsum         = _mm256_add_pd(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 430
 431             /* Calculate temporary vectorial force */
 432             tx               = _mm256_mul_pd(fscal,dx30);
 433             ty               = _mm256_mul_pd(fscal,dy30);
 434             tz               = _mm256_mul_pd(fscal,dz30);
 435
 436             /* Update vectorial force */
 437             fix3             = _mm256_add_pd(fix3,tx);
 438             fiy3             = _mm256_add_pd(fiy3,ty);
 439             fiz3             = _mm256_add_pd(fiz3,tz);
 440
 441             fjx0             = _mm256_add_pd(fjx0,tx);
 442             fjy0             = _mm256_add_pd(fjy0,ty);
 443             fjz0             = _mm256_add_pd(fjz0,tz);
 444
 445             }
 446
 447             fjptrA             = f+j_coord_offsetA;
 448             fjptrB             = f+j_coord_offsetB;
 449             fjptrC             = f+j_coord_offsetC;
 450             fjptrD             = f+j_coord_offsetD;
 451
 452             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 453
 454             /* Inner loop uses 167 flops */
 455         }
 456
 457         if(jidx<j_index_end)
 458         {
 459
 460             /* Get j neighbor index, and coordinate index */
 461             jnrlistA         = jjnr[jidx];
 462             jnrlistB         = jjnr[jidx+1];
 463             jnrlistC         = jjnr[jidx+2];
 464             jnrlistD         = jjnr[jidx+3];
 465             /* Sign of each element will be negative for non-real atoms.
 466              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 467              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 468              */
 469             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 470
 471             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 472             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 473             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 474
 475             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 476             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 477             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 478             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 479             j_coord_offsetA  = DIM*jnrA;
 480             j_coord_offsetB  = DIM*jnrB;
 481             j_coord_offsetC  = DIM*jnrC;
 482             j_coord_offsetD  = DIM*jnrD;
 483
 484             /* load j atom coordinates */
 485             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 486                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 487                                                  &jx0,&jy0,&jz0);
 488
 489             /* Calculate displacement vector */
 490             dx00             = _mm256_sub_pd(ix0,jx0);
 491             dy00             = _mm256_sub_pd(iy0,jy0);
 492             dz00             = _mm256_sub_pd(iz0,jz0);
 493             dx10             = _mm256_sub_pd(ix1,jx0);
 494             dy10             = _mm256_sub_pd(iy1,jy0);
 495             dz10             = _mm256_sub_pd(iz1,jz0);
 496             dx20             = _mm256_sub_pd(ix2,jx0);
 497             dy20             = _mm256_sub_pd(iy2,jy0);
 498             dz20             = _mm256_sub_pd(iz2,jz0);
 499             dx30             = _mm256_sub_pd(ix3,jx0);
 500             dy30             = _mm256_sub_pd(iy3,jy0);
 501             dz30             = _mm256_sub_pd(iz3,jz0);
 502
 503             /* Calculate squared distance and things based on it */
 504             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 505             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 506             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 507             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 508
 509             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 510             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 511             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 512             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 513
 514             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 515             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 516             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 517
 518             /* Load parameters for j particles */
 519             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 520                                                                  charge+jnrC+0,charge+jnrD+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523             vdwjidx0C        = 2*vdwtype[jnrC+0];
 524             vdwjidx0D        = 2*vdwtype[jnrD+0];
 525
 526             fjx0             = _mm256_setzero_pd();
 527             fjy0             = _mm256_setzero_pd();
 528             fjz0             = _mm256_setzero_pd();
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _mm256_mul_pd(rsq00,rinv00);
 535             r00              = _mm256_andnot_pd(dummy_mask,r00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 539                                             vdwioffsetptr0+vdwjidx0B,
 540                                             vdwioffsetptr0+vdwjidx0C,
 541                                             vdwioffsetptr0+vdwjidx0D,
 542                                             &c6_00,&c12_00);
 543
 544             /* Calculate table index by multiplying r with table scale and truncate to integer */
 545             rt               = _mm256_mul_pd(r00,vftabscale);
 546             vfitab           = _mm256_cvttpd_epi32(rt);
 547             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 548             vfitab           = _mm_slli_epi32(vfitab,3);
 549
 550             /* CUBIC SPLINE TABLE DISPERSION */
 551             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 552             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 553             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 554             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 555             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 556             Heps             = _mm256_mul_pd(vfeps,H);
 557             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 558             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 559             vvdw6            = _mm256_mul_pd(c6_00,VV);
 560             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 561             fvdw6            = _mm256_mul_pd(c6_00,FF);
 562
 563             /* CUBIC SPLINE TABLE REPULSION */
 564             vfitab           = _mm_add_epi32(vfitab,ifour);
 565             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 566             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 567             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 568             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 569             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 570             Heps             = _mm256_mul_pd(vfeps,H);
 571             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 572             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 573             vvdw12           = _mm256_mul_pd(c12_00,VV);
 574             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 575             fvdw12           = _mm256_mul_pd(c12_00,FF);
 576             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 577             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 581             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 582
 583             fscal            = fvdw;
 584
 585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_pd(fscal,dx00);
 589             ty               = _mm256_mul_pd(fscal,dy00);
 590             tz               = _mm256_mul_pd(fscal,dz00);
 591
 592             /* Update vectorial force */
 593             fix0             = _mm256_add_pd(fix0,tx);
 594             fiy0             = _mm256_add_pd(fiy0,ty);
 595             fiz0             = _mm256_add_pd(fiz0,tz);
 596
 597             fjx0             = _mm256_add_pd(fjx0,tx);
 598             fjy0             = _mm256_add_pd(fjy0,ty);
 599             fjz0             = _mm256_add_pd(fjz0,tz);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 606             {
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq10             = _mm256_mul_pd(iq1,jq0);
 610
 611             /* REACTION-FIELD ELECTROSTATICS */
 612             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 613             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 614
 615             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm256_and_pd(velec,cutoff_mask);
 619             velec            = _mm256_andnot_pd(dummy_mask,velec);
 620             velecsum         = _mm256_add_pd(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 625
 626             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm256_mul_pd(fscal,dx10);
 630             ty               = _mm256_mul_pd(fscal,dy10);
 631             tz               = _mm256_mul_pd(fscal,dz10);
 632
 633             /* Update vectorial force */
 634             fix1             = _mm256_add_pd(fix1,tx);
 635             fiy1             = _mm256_add_pd(fiy1,ty);
 636             fiz1             = _mm256_add_pd(fiz1,tz);
 637
 638             fjx0             = _mm256_add_pd(fjx0,tx);
 639             fjy0             = _mm256_add_pd(fjy0,ty);
 640             fjz0             = _mm256_add_pd(fjz0,tz);
 641
 642             }
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 649             {
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq20             = _mm256_mul_pd(iq2,jq0);
 653
 654             /* REACTION-FIELD ELECTROSTATICS */
 655             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 656             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 657
 658             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm256_and_pd(velec,cutoff_mask);
 662             velec            = _mm256_andnot_pd(dummy_mask,velec);
 663             velecsum         = _mm256_add_pd(velecsum,velec);
 664
 665             fscal            = felec;
 666
 667             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 668
 669             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm256_mul_pd(fscal,dx20);
 673             ty               = _mm256_mul_pd(fscal,dy20);
 674             tz               = _mm256_mul_pd(fscal,dz20);
 675
 676             /* Update vectorial force */
 677             fix2             = _mm256_add_pd(fix2,tx);
 678             fiy2             = _mm256_add_pd(fiy2,ty);
 679             fiz2             = _mm256_add_pd(fiz2,tz);
 680
 681             fjx0             = _mm256_add_pd(fjx0,tx);
 682             fjy0             = _mm256_add_pd(fjy0,ty);
 683             fjz0             = _mm256_add_pd(fjz0,tz);
 684
 685             }
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 692             {
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq30             = _mm256_mul_pd(iq3,jq0);
 696
 697             /* REACTION-FIELD ELECTROSTATICS */
 698             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 699             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 700
 701             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velec            = _mm256_and_pd(velec,cutoff_mask);
 705             velec            = _mm256_andnot_pd(dummy_mask,velec);
 706             velecsum         = _mm256_add_pd(velecsum,velec);
 707
 708             fscal            = felec;
 709
 710             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 711
 712             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm256_mul_pd(fscal,dx30);
 716             ty               = _mm256_mul_pd(fscal,dy30);
 717             tz               = _mm256_mul_pd(fscal,dz30);
 718
 719             /* Update vectorial force */
 720             fix3             = _mm256_add_pd(fix3,tx);
 721             fiy3             = _mm256_add_pd(fiy3,ty);
 722             fiz3             = _mm256_add_pd(fiz3,tz);
 723
 724             fjx0             = _mm256_add_pd(fjx0,tx);
 725             fjy0             = _mm256_add_pd(fjy0,ty);
 726             fjz0             = _mm256_add_pd(fjz0,tz);
 727
 728             }
 729
 730             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 731             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 732             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 733             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 734
 735             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 736
 737             /* Inner loop uses 168 flops */
 738         }
 739
 740         /* End of innermost loop */
 741
 742         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 743                                                  f+i_coord_offset,fshift+i_shift_offset);
 744
 745         ggid                        = gid[iidx];
 746         /* Update potential energies */
 747         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 748         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 749
 750         /* Increment number of inner iterations */
 751         inneriter                  += j_index_end - j_index_start;
 752
 753         /* Outer loop uses 26 flops */
 754     }
 755
 756     /* Increment number of outer iterations */
 757     outeriter        += nri;
 758
 759     /* Update outer/inner flops */
 760
 761     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
 762 }
 763 /*
 764  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_double
 765  * Electrostatics interaction: ReactionField
 766  * VdW interaction:            CubicSplineTable
 767  * Geometry:                   Water4-Particle
 768  * Calculate force/pot:        Force
 769  */
 770 void
 771 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_double
 772                     (t_nblist                    * gmx_restrict       nlist,
 773                      rvec                        * gmx_restrict          xx,
 774                      rvec                        * gmx_restrict          ff,
 775                      t_forcerec                  * gmx_restrict          fr,
 776                      t_mdatoms                   * gmx_restrict     mdatoms,
 777                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 778                      t_nrnb                      * gmx_restrict        nrnb)
 779 {
 780     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 781      * just 0 for non-waters.
 782      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 783      * jnr indices corresponding to data put in the four positions in the SIMD register.
 784      */
 785     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 786     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 787     int              jnrA,jnrB,jnrC,jnrD;
 788     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 789     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 790     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 791     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 792     real             rcutoff_scalar;
 793     real             *shiftvec,*fshift,*x,*f;
 794     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 795     real             scratch[4*DIM];
 796     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 797     real *           vdwioffsetptr0;
 798     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 799     real *           vdwioffsetptr1;
 800     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 801     real *           vdwioffsetptr2;
 802     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 803     real *           vdwioffsetptr3;
 804     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 805     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 806     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 807     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 808     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 809     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 810     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 811     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 812     real             *charge;
 813     int              nvdwtype;
 814     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 815     int              *vdwtype;
 816     real             *vdwparam;
 817     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 818     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 819     __m128i          vfitab;
 820     __m128i          ifour       = _mm_set1_epi32(4);
 821     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 822     real             *vftab;
 823     __m256d          dummy_mask,cutoff_mask;
 824     __m128           tmpmask0,tmpmask1;
 825     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 826     __m256d          one     = _mm256_set1_pd(1.0);
 827     __m256d          two     = _mm256_set1_pd(2.0);
 828     x                = xx[0];
 829     f                = ff[0];
 830
 831     nri              = nlist->nri;
 832     iinr             = nlist->iinr;
 833     jindex           = nlist->jindex;
 834     jjnr             = nlist->jjnr;
 835     shiftidx         = nlist->shift;
 836     gid              = nlist->gid;
 837     shiftvec         = fr->shift_vec[0];
 838     fshift           = fr->fshift[0];
 839     facel            = _mm256_set1_pd(fr->epsfac);
 840     charge           = mdatoms->chargeA;
 841     krf              = _mm256_set1_pd(fr->ic->k_rf);
 842     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 843     crf              = _mm256_set1_pd(fr->ic->c_rf);
 844     nvdwtype         = fr->ntype;
 845     vdwparam         = fr->nbfp;
 846     vdwtype          = mdatoms->typeA;
 847
 848     vftab            = kernel_data->table_vdw->data;
 849     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 850
 851     /* Setup water-specific parameters */
 852     inr              = nlist->iinr[0];
 853     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 854     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 855     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 856     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 857
 858     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 859     rcutoff_scalar   = fr->rcoulomb;
 860     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 861     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 862
 863     /* Avoid stupid compiler warnings */
 864     jnrA = jnrB = jnrC = jnrD = 0;
 865     j_coord_offsetA = 0;
 866     j_coord_offsetB = 0;
 867     j_coord_offsetC = 0;
 868     j_coord_offsetD = 0;
 869
 870     outeriter        = 0;
 871     inneriter        = 0;
 872
 873     for(iidx=0;iidx<4*DIM;iidx++)
 874     {
 875         scratch[iidx] = 0.0;
 876     }
 877
 878     /* Start outer loop over neighborlists */
 879     for(iidx=0; iidx<nri; iidx++)
 880     {
 881         /* Load shift vector for this list */
 882         i_shift_offset   = DIM*shiftidx[iidx];
 883
 884         /* Load limits for loop over neighbors */
 885         j_index_start    = jindex[iidx];
 886         j_index_end      = jindex[iidx+1];
 887
 888         /* Get outer coordinate index */
 889         inr              = iinr[iidx];
 890         i_coord_offset   = DIM*inr;
 891
 892         /* Load i particle coords and add shift vector */
 893         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 894                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 895
 896         fix0             = _mm256_setzero_pd();
 897         fiy0             = _mm256_setzero_pd();
 898         fiz0             = _mm256_setzero_pd();
 899         fix1             = _mm256_setzero_pd();
 900         fiy1             = _mm256_setzero_pd();
 901         fiz1             = _mm256_setzero_pd();
 902         fix2             = _mm256_setzero_pd();
 903         fiy2             = _mm256_setzero_pd();
 904         fiz2             = _mm256_setzero_pd();
 905         fix3             = _mm256_setzero_pd();
 906         fiy3             = _mm256_setzero_pd();
 907         fiz3             = _mm256_setzero_pd();
 908
 909         /* Start inner kernel loop */
 910         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 911         {
 912
 913             /* Get j neighbor index, and coordinate index */
 914             jnrA             = jjnr[jidx];
 915             jnrB             = jjnr[jidx+1];
 916             jnrC             = jjnr[jidx+2];
 917             jnrD             = jjnr[jidx+3];
 918             j_coord_offsetA  = DIM*jnrA;
 919             j_coord_offsetB  = DIM*jnrB;
 920             j_coord_offsetC  = DIM*jnrC;
 921             j_coord_offsetD  = DIM*jnrD;
 922
 923             /* load j atom coordinates */
 924             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 925                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 926                                                  &jx0,&jy0,&jz0);
 927
 928             /* Calculate displacement vector */
 929             dx00             = _mm256_sub_pd(ix0,jx0);
 930             dy00             = _mm256_sub_pd(iy0,jy0);
 931             dz00             = _mm256_sub_pd(iz0,jz0);
 932             dx10             = _mm256_sub_pd(ix1,jx0);
 933             dy10             = _mm256_sub_pd(iy1,jy0);
 934             dz10             = _mm256_sub_pd(iz1,jz0);
 935             dx20             = _mm256_sub_pd(ix2,jx0);
 936             dy20             = _mm256_sub_pd(iy2,jy0);
 937             dz20             = _mm256_sub_pd(iz2,jz0);
 938             dx30             = _mm256_sub_pd(ix3,jx0);
 939             dy30             = _mm256_sub_pd(iy3,jy0);
 940             dz30             = _mm256_sub_pd(iz3,jz0);
 941
 942             /* Calculate squared distance and things based on it */
 943             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 944             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 945             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 946             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 947
 948             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 949             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 950             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 951             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 952
 953             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 954             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 955             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 956
 957             /* Load parameters for j particles */
 958             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 959                                                                  charge+jnrC+0,charge+jnrD+0);
 960             vdwjidx0A        = 2*vdwtype[jnrA+0];
 961             vdwjidx0B        = 2*vdwtype[jnrB+0];
 962             vdwjidx0C        = 2*vdwtype[jnrC+0];
 963             vdwjidx0D        = 2*vdwtype[jnrD+0];
 964
 965             fjx0             = _mm256_setzero_pd();
 966             fjy0             = _mm256_setzero_pd();
 967             fjz0             = _mm256_setzero_pd();
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r00              = _mm256_mul_pd(rsq00,rinv00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 977                                             vdwioffsetptr0+vdwjidx0B,
 978                                             vdwioffsetptr0+vdwjidx0C,
 979                                             vdwioffsetptr0+vdwjidx0D,
 980                                             &c6_00,&c12_00);
 981
 982             /* Calculate table index by multiplying r with table scale and truncate to integer */
 983             rt               = _mm256_mul_pd(r00,vftabscale);
 984             vfitab           = _mm256_cvttpd_epi32(rt);
 985             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 986             vfitab           = _mm_slli_epi32(vfitab,3);
 987
 988             /* CUBIC SPLINE TABLE DISPERSION */
 989             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 990             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 991             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 992             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 993             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 994             Heps             = _mm256_mul_pd(vfeps,H);
 995             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 996             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 997             fvdw6            = _mm256_mul_pd(c6_00,FF);
 998
 999             /* CUBIC SPLINE TABLE REPULSION */
1000             vfitab           = _mm_add_epi32(vfitab,ifour);
1001             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1002             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1003             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1004             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1005             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1006             Heps             = _mm256_mul_pd(vfeps,H);
1007             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1008             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1009             fvdw12           = _mm256_mul_pd(c12_00,FF);
1010             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1011
1012             fscal            = fvdw;
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              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1033             {
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq10             = _mm256_mul_pd(iq1,jq0);
1037
1038             /* REACTION-FIELD ELECTROSTATICS */
1039             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1040
1041             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm256_mul_pd(fscal,dx10);
1049             ty               = _mm256_mul_pd(fscal,dy10);
1050             tz               = _mm256_mul_pd(fscal,dz10);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm256_add_pd(fix1,tx);
1054             fiy1             = _mm256_add_pd(fiy1,ty);
1055             fiz1             = _mm256_add_pd(fiz1,tz);
1056
1057             fjx0             = _mm256_add_pd(fjx0,tx);
1058             fjy0             = _mm256_add_pd(fjy0,ty);
1059             fjz0             = _mm256_add_pd(fjz0,tz);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1068             {
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq20             = _mm256_mul_pd(iq2,jq0);
1072
1073             /* REACTION-FIELD ELECTROSTATICS */
1074             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1075
1076             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm256_mul_pd(fscal,dx20);
1084             ty               = _mm256_mul_pd(fscal,dy20);
1085             tz               = _mm256_mul_pd(fscal,dz20);
1086
1087             /* Update vectorial force */
1088             fix2             = _mm256_add_pd(fix2,tx);
1089             fiy2             = _mm256_add_pd(fiy2,ty);
1090             fiz2             = _mm256_add_pd(fiz2,tz);
1091
1092             fjx0             = _mm256_add_pd(fjx0,tx);
1093             fjy0             = _mm256_add_pd(fjy0,ty);
1094             fjz0             = _mm256_add_pd(fjz0,tz);
1095
1096             }
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1103             {
1104
1105             /* Compute parameters for interactions between i and j atoms */
1106             qq30             = _mm256_mul_pd(iq3,jq0);
1107
1108             /* REACTION-FIELD ELECTROSTATICS */
1109             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1110
1111             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1112
1113             fscal            = felec;
1114
1115             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm256_mul_pd(fscal,dx30);
1119             ty               = _mm256_mul_pd(fscal,dy30);
1120             tz               = _mm256_mul_pd(fscal,dz30);
1121
1122             /* Update vectorial force */
1123             fix3             = _mm256_add_pd(fix3,tx);
1124             fiy3             = _mm256_add_pd(fiy3,ty);
1125             fiz3             = _mm256_add_pd(fiz3,tz);
1126
1127             fjx0             = _mm256_add_pd(fjx0,tx);
1128             fjy0             = _mm256_add_pd(fjy0,ty);
1129             fjz0             = _mm256_add_pd(fjz0,tz);
1130
1131             }
1132
1133             fjptrA             = f+j_coord_offsetA;
1134             fjptrB             = f+j_coord_offsetB;
1135             fjptrC             = f+j_coord_offsetC;
1136             fjptrD             = f+j_coord_offsetD;
1137
1138             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1139
1140             /* Inner loop uses 141 flops */
1141         }
1142
1143         if(jidx<j_index_end)
1144         {
1145
1146             /* Get j neighbor index, and coordinate index */
1147             jnrlistA         = jjnr[jidx];
1148             jnrlistB         = jjnr[jidx+1];
1149             jnrlistC         = jjnr[jidx+2];
1150             jnrlistD         = jjnr[jidx+3];
1151             /* Sign of each element will be negative for non-real atoms.
1152              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1153              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1154              */
1155             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1156
1157             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1158             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1159             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1160
1161             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1162             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1163             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1164             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1165             j_coord_offsetA  = DIM*jnrA;
1166             j_coord_offsetB  = DIM*jnrB;
1167             j_coord_offsetC  = DIM*jnrC;
1168             j_coord_offsetD  = DIM*jnrD;
1169
1170             /* load j atom coordinates */
1171             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1172                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1173                                                  &jx0,&jy0,&jz0);
1174
1175             /* Calculate displacement vector */
1176             dx00             = _mm256_sub_pd(ix0,jx0);
1177             dy00             = _mm256_sub_pd(iy0,jy0);
1178             dz00             = _mm256_sub_pd(iz0,jz0);
1179             dx10             = _mm256_sub_pd(ix1,jx0);
1180             dy10             = _mm256_sub_pd(iy1,jy0);
1181             dz10             = _mm256_sub_pd(iz1,jz0);
1182             dx20             = _mm256_sub_pd(ix2,jx0);
1183             dy20             = _mm256_sub_pd(iy2,jy0);
1184             dz20             = _mm256_sub_pd(iz2,jz0);
1185             dx30             = _mm256_sub_pd(ix3,jx0);
1186             dy30             = _mm256_sub_pd(iy3,jy0);
1187             dz30             = _mm256_sub_pd(iz3,jz0);
1188
1189             /* Calculate squared distance and things based on it */
1190             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1191             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1192             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1193             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1194
1195             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1196             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1197             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1198             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1199
1200             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1201             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1202             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1203
1204             /* Load parameters for j particles */
1205             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1206                                                                  charge+jnrC+0,charge+jnrD+0);
1207             vdwjidx0A        = 2*vdwtype[jnrA+0];
1208             vdwjidx0B        = 2*vdwtype[jnrB+0];
1209             vdwjidx0C        = 2*vdwtype[jnrC+0];
1210             vdwjidx0D        = 2*vdwtype[jnrD+0];
1211
1212             fjx0             = _mm256_setzero_pd();
1213             fjy0             = _mm256_setzero_pd();
1214             fjz0             = _mm256_setzero_pd();
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             r00              = _mm256_mul_pd(rsq00,rinv00);
1221             r00              = _mm256_andnot_pd(dummy_mask,r00);
1222
1223             /* Compute parameters for interactions between i and j atoms */
1224             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1225                                             vdwioffsetptr0+vdwjidx0B,
1226                                             vdwioffsetptr0+vdwjidx0C,
1227                                             vdwioffsetptr0+vdwjidx0D,
1228                                             &c6_00,&c12_00);
1229
1230             /* Calculate table index by multiplying r with table scale and truncate to integer */
1231             rt               = _mm256_mul_pd(r00,vftabscale);
1232             vfitab           = _mm256_cvttpd_epi32(rt);
1233             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1234             vfitab           = _mm_slli_epi32(vfitab,3);
1235
1236             /* CUBIC SPLINE TABLE DISPERSION */
1237             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1238             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1239             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1240             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1241             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1242             Heps             = _mm256_mul_pd(vfeps,H);
1243             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1244             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1245             fvdw6            = _mm256_mul_pd(c6_00,FF);
1246
1247             /* CUBIC SPLINE TABLE REPULSION */
1248             vfitab           = _mm_add_epi32(vfitab,ifour);
1249             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1250             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1251             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1252             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1253             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1254             Heps             = _mm256_mul_pd(vfeps,H);
1255             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1256             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1257             fvdw12           = _mm256_mul_pd(c12_00,FF);
1258             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1259
1260             fscal            = fvdw;
1261
1262             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1263
1264             /* Calculate temporary vectorial force */
1265             tx               = _mm256_mul_pd(fscal,dx00);
1266             ty               = _mm256_mul_pd(fscal,dy00);
1267             tz               = _mm256_mul_pd(fscal,dz00);
1268
1269             /* Update vectorial force */
1270             fix0             = _mm256_add_pd(fix0,tx);
1271             fiy0             = _mm256_add_pd(fiy0,ty);
1272             fiz0             = _mm256_add_pd(fiz0,tz);
1273
1274             fjx0             = _mm256_add_pd(fjx0,tx);
1275             fjy0             = _mm256_add_pd(fjy0,ty);
1276             fjz0             = _mm256_add_pd(fjz0,tz);
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 142 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*142);
1416 }