src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJ_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m128i          ewitab;
 109     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 111     real             *ewtab;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_pd(beta,beta);
 137     beta3            = _mm256_mul_pd(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 148     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 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             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 240             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 241             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 242
 243             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 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             /* Compute parameters for interactions between i and j atoms */
 265             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 266                                             vdwioffsetptr0+vdwjidx0B,
 267                                             vdwioffsetptr0+vdwjidx0C,
 268                                             vdwioffsetptr0+vdwjidx0D,
 269                                             &c6_00,&c12_00);
 270
 271             /* LENNARD-JONES DISPERSION/REPULSION */
 272
 273             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 274             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 275             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 276             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 277             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 281
 282             fscal            = fvdw;
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm256_mul_pd(fscal,dx00);
 286             ty               = _mm256_mul_pd(fscal,dy00);
 287             tz               = _mm256_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm256_add_pd(fix0,tx);
 291             fiy0             = _mm256_add_pd(fiy0,ty);
 292             fiz0             = _mm256_add_pd(fiz0,tz);
 293
 294             fjx0             = _mm256_add_pd(fjx0,tx);
 295             fjy0             = _mm256_add_pd(fjy0,ty);
 296             fjz0             = _mm256_add_pd(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm256_mul_pd(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm256_mul_pd(iq1,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 311             ewitab           = _mm256_cvttpd_epi32(ewrt);
 312             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 313             ewitab           = _mm_slli_epi32(ewitab,2);
 314             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 315             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 316             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 317             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 318             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 319             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 320             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 321             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 322             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm256_add_pd(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm256_mul_pd(fscal,dx10);
 331             ty               = _mm256_mul_pd(fscal,dy10);
 332             tz               = _mm256_mul_pd(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm256_add_pd(fix1,tx);
 336             fiy1             = _mm256_add_pd(fiy1,ty);
 337             fiz1             = _mm256_add_pd(fiz1,tz);
 338
 339             fjx0             = _mm256_add_pd(fjx0,tx);
 340             fjy0             = _mm256_add_pd(fjy0,ty);
 341             fjz0             = _mm256_add_pd(fjz0,tz);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             r20              = _mm256_mul_pd(rsq20,rinv20);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm256_mul_pd(iq2,jq0);
 351
 352             /* EWALD ELECTROSTATICS */
 353
 354             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 355             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 356             ewitab           = _mm256_cvttpd_epi32(ewrt);
 357             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 358             ewitab           = _mm_slli_epi32(ewitab,2);
 359             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 360             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 361             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 362             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 363             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 364             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 365             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 366             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 367             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm256_add_pd(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm256_mul_pd(fscal,dx20);
 376             ty               = _mm256_mul_pd(fscal,dy20);
 377             tz               = _mm256_mul_pd(fscal,dz20);
 378
 379             /* Update vectorial force */
 380             fix2             = _mm256_add_pd(fix2,tx);
 381             fiy2             = _mm256_add_pd(fiy2,ty);
 382             fiz2             = _mm256_add_pd(fiz2,tz);
 383
 384             fjx0             = _mm256_add_pd(fjx0,tx);
 385             fjy0             = _mm256_add_pd(fjy0,ty);
 386             fjz0             = _mm256_add_pd(fjz0,tz);
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             r30              = _mm256_mul_pd(rsq30,rinv30);
 393
 394             /* Compute parameters for interactions between i and j atoms */
 395             qq30             = _mm256_mul_pd(iq3,jq0);
 396
 397             /* EWALD ELECTROSTATICS */
 398
 399             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 400             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 401             ewitab           = _mm256_cvttpd_epi32(ewrt);
 402             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 403             ewitab           = _mm_slli_epi32(ewitab,2);
 404             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 405             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 406             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 407             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 408             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 409             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 410             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 411             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 412             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velecsum         = _mm256_add_pd(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_pd(fscal,dx30);
 421             ty               = _mm256_mul_pd(fscal,dy30);
 422             tz               = _mm256_mul_pd(fscal,dz30);
 423
 424             /* Update vectorial force */
 425             fix3             = _mm256_add_pd(fix3,tx);
 426             fiy3             = _mm256_add_pd(fiy3,ty);
 427             fiz3             = _mm256_add_pd(fiz3,tz);
 428
 429             fjx0             = _mm256_add_pd(fjx0,tx);
 430             fjy0             = _mm256_add_pd(fjy0,ty);
 431             fjz0             = _mm256_add_pd(fjz0,tz);
 432
 433             fjptrA             = f+j_coord_offsetA;
 434             fjptrB             = f+j_coord_offsetB;
 435             fjptrC             = f+j_coord_offsetC;
 436             fjptrD             = f+j_coord_offsetD;
 437
 438             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 439
 440             /* Inner loop uses 158 flops */
 441         }
 442
 443         if(jidx<j_index_end)
 444         {
 445
 446             /* Get j neighbor index, and coordinate index */
 447             jnrlistA         = jjnr[jidx];
 448             jnrlistB         = jjnr[jidx+1];
 449             jnrlistC         = jjnr[jidx+2];
 450             jnrlistD         = jjnr[jidx+3];
 451             /* Sign of each element will be negative for non-real atoms.
 452              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 453              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 454              */
 455             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 456
 457             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 458             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 459             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 460
 461             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 462             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 463             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 464             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 465             j_coord_offsetA  = DIM*jnrA;
 466             j_coord_offsetB  = DIM*jnrB;
 467             j_coord_offsetC  = DIM*jnrC;
 468             j_coord_offsetD  = DIM*jnrD;
 469
 470             /* load j atom coordinates */
 471             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 473                                                  &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm256_sub_pd(ix0,jx0);
 477             dy00             = _mm256_sub_pd(iy0,jy0);
 478             dz00             = _mm256_sub_pd(iz0,jz0);
 479             dx10             = _mm256_sub_pd(ix1,jx0);
 480             dy10             = _mm256_sub_pd(iy1,jy0);
 481             dz10             = _mm256_sub_pd(iz1,jz0);
 482             dx20             = _mm256_sub_pd(ix2,jx0);
 483             dy20             = _mm256_sub_pd(iy2,jy0);
 484             dz20             = _mm256_sub_pd(iz2,jz0);
 485             dx30             = _mm256_sub_pd(ix3,jx0);
 486             dy30             = _mm256_sub_pd(iy3,jy0);
 487             dz30             = _mm256_sub_pd(iz3,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 491             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 492             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 493             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 494
 495             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 496             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 497             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 498
 499             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 500             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 501             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 502             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 503
 504             /* Load parameters for j particles */
 505             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 506                                                                  charge+jnrC+0,charge+jnrD+0);
 507             vdwjidx0A        = 2*vdwtype[jnrA+0];
 508             vdwjidx0B        = 2*vdwtype[jnrB+0];
 509             vdwjidx0C        = 2*vdwtype[jnrC+0];
 510             vdwjidx0D        = 2*vdwtype[jnrD+0];
 511
 512             fjx0             = _mm256_setzero_pd();
 513             fjy0             = _mm256_setzero_pd();
 514             fjz0             = _mm256_setzero_pd();
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 522                                             vdwioffsetptr0+vdwjidx0B,
 523                                             vdwioffsetptr0+vdwjidx0C,
 524                                             vdwioffsetptr0+vdwjidx0D,
 525                                             &c6_00,&c12_00);
 526
 527             /* LENNARD-JONES DISPERSION/REPULSION */
 528
 529             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 530             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 531             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 532             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 533             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 537             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 538
 539             fscal            = fvdw;
 540
 541             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm256_mul_pd(fscal,dx00);
 545             ty               = _mm256_mul_pd(fscal,dy00);
 546             tz               = _mm256_mul_pd(fscal,dz00);
 547
 548             /* Update vectorial force */
 549             fix0             = _mm256_add_pd(fix0,tx);
 550             fiy0             = _mm256_add_pd(fiy0,ty);
 551             fiz0             = _mm256_add_pd(fiz0,tz);
 552
 553             fjx0             = _mm256_add_pd(fjx0,tx);
 554             fjy0             = _mm256_add_pd(fjy0,ty);
 555             fjz0             = _mm256_add_pd(fjz0,tz);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             r10              = _mm256_mul_pd(rsq10,rinv10);
 562             r10              = _mm256_andnot_pd(dummy_mask,r10);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm256_mul_pd(iq1,jq0);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 571             ewitab           = _mm256_cvttpd_epi32(ewrt);
 572             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 573             ewitab           = _mm_slli_epi32(ewitab,2);
 574             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 575             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 576             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 577             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 578             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 579             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 580             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 581             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 582             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm256_andnot_pd(dummy_mask,velec);
 586             velecsum         = _mm256_add_pd(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm256_mul_pd(fscal,dx10);
 594             ty               = _mm256_mul_pd(fscal,dy10);
 595             tz               = _mm256_mul_pd(fscal,dz10);
 596
 597             /* Update vectorial force */
 598             fix1             = _mm256_add_pd(fix1,tx);
 599             fiy1             = _mm256_add_pd(fiy1,ty);
 600             fiz1             = _mm256_add_pd(fiz1,tz);
 601
 602             fjx0             = _mm256_add_pd(fjx0,tx);
 603             fjy0             = _mm256_add_pd(fjy0,ty);
 604             fjz0             = _mm256_add_pd(fjz0,tz);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             r20              = _mm256_mul_pd(rsq20,rinv20);
 611             r20              = _mm256_andnot_pd(dummy_mask,r20);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq20             = _mm256_mul_pd(iq2,jq0);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 619             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 620             ewitab           = _mm256_cvttpd_epi32(ewrt);
 621             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 622             ewitab           = _mm_slli_epi32(ewitab,2);
 623             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 624             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 625             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 626             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 627             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 628             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 629             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 630             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 631             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm256_andnot_pd(dummy_mask,velec);
 635             velecsum         = _mm256_add_pd(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm256_mul_pd(fscal,dx20);
 643             ty               = _mm256_mul_pd(fscal,dy20);
 644             tz               = _mm256_mul_pd(fscal,dz20);
 645
 646             /* Update vectorial force */
 647             fix2             = _mm256_add_pd(fix2,tx);
 648             fiy2             = _mm256_add_pd(fiy2,ty);
 649             fiz2             = _mm256_add_pd(fiz2,tz);
 650
 651             fjx0             = _mm256_add_pd(fjx0,tx);
 652             fjy0             = _mm256_add_pd(fjy0,ty);
 653             fjz0             = _mm256_add_pd(fjz0,tz);
 654
 655             /**************************
 656              * CALCULATE INTERACTIONS *
 657              **************************/
 658
 659             r30              = _mm256_mul_pd(rsq30,rinv30);
 660             r30              = _mm256_andnot_pd(dummy_mask,r30);
 661
 662             /* Compute parameters for interactions between i and j atoms */
 663             qq30             = _mm256_mul_pd(iq3,jq0);
 664
 665             /* EWALD ELECTROSTATICS */
 666
 667             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 668             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 669             ewitab           = _mm256_cvttpd_epi32(ewrt);
 670             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 671             ewitab           = _mm_slli_epi32(ewitab,2);
 672             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 673             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 674             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 675             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 676             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 677             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 678             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 679             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 680             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 681
 682             /* Update potential sum for this i atom from the interaction with this j atom. */
 683             velec            = _mm256_andnot_pd(dummy_mask,velec);
 684             velecsum         = _mm256_add_pd(velecsum,velec);
 685
 686             fscal            = felec;
 687
 688             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 689
 690             /* Calculate temporary vectorial force */
 691             tx               = _mm256_mul_pd(fscal,dx30);
 692             ty               = _mm256_mul_pd(fscal,dy30);
 693             tz               = _mm256_mul_pd(fscal,dz30);
 694
 695             /* Update vectorial force */
 696             fix3             = _mm256_add_pd(fix3,tx);
 697             fiy3             = _mm256_add_pd(fiy3,ty);
 698             fiz3             = _mm256_add_pd(fiz3,tz);
 699
 700             fjx0             = _mm256_add_pd(fjx0,tx);
 701             fjy0             = _mm256_add_pd(fjy0,ty);
 702             fjz0             = _mm256_add_pd(fjz0,tz);
 703
 704             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 705             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 706             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 707             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 708
 709             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 710
 711             /* Inner loop uses 161 flops */
 712         }
 713
 714         /* End of innermost loop */
 715
 716         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 717                                                  f+i_coord_offset,fshift+i_shift_offset);
 718
 719         ggid                        = gid[iidx];
 720         /* Update potential energies */
 721         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 722         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 723
 724         /* Increment number of inner iterations */
 725         inneriter                  += j_index_end - j_index_start;
 726
 727         /* Outer loop uses 26 flops */
 728     }
 729
 730     /* Increment number of outer iterations */
 731     outeriter        += nri;
 732
 733     /* Update outer/inner flops */
 734
 735     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*161);
 736 }
 737 /*
 738  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_double
 739  * Electrostatics interaction: Ewald
 740  * VdW interaction:            LennardJones
 741  * Geometry:                   Water4-Particle
 742  * Calculate force/pot:        Force
 743  */
 744 void
 745 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_double
 746                     (t_nblist                    * gmx_restrict       nlist,
 747                      rvec                        * gmx_restrict          xx,
 748                      rvec                        * gmx_restrict          ff,
 749                      t_forcerec                  * gmx_restrict          fr,
 750                      t_mdatoms                   * gmx_restrict     mdatoms,
 751                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 752                      t_nrnb                      * gmx_restrict        nrnb)
 753 {
 754     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 755      * just 0 for non-waters.
 756      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 757      * jnr indices corresponding to data put in the four positions in the SIMD register.
 758      */
 759     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 760     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 761     int              jnrA,jnrB,jnrC,jnrD;
 762     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 763     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 764     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 765     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 766     real             rcutoff_scalar;
 767     real             *shiftvec,*fshift,*x,*f;
 768     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 769     real             scratch[4*DIM];
 770     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 771     real *           vdwioffsetptr0;
 772     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 773     real *           vdwioffsetptr1;
 774     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 775     real *           vdwioffsetptr2;
 776     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 777     real *           vdwioffsetptr3;
 778     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 779     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 780     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 781     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 782     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 783     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 784     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 785     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 786     real             *charge;
 787     int              nvdwtype;
 788     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 789     int              *vdwtype;
 790     real             *vdwparam;
 791     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 792     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 793     __m128i          ewitab;
 794     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 795     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 796     real             *ewtab;
 797     __m256d          dummy_mask,cutoff_mask;
 798     __m128           tmpmask0,tmpmask1;
 799     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 800     __m256d          one     = _mm256_set1_pd(1.0);
 801     __m256d          two     = _mm256_set1_pd(2.0);
 802     x                = xx[0];
 803     f                = ff[0];
 804
 805     nri              = nlist->nri;
 806     iinr             = nlist->iinr;
 807     jindex           = nlist->jindex;
 808     jjnr             = nlist->jjnr;
 809     shiftidx         = nlist->shift;
 810     gid              = nlist->gid;
 811     shiftvec         = fr->shift_vec[0];
 812     fshift           = fr->fshift[0];
 813     facel            = _mm256_set1_pd(fr->epsfac);
 814     charge           = mdatoms->chargeA;
 815     nvdwtype         = fr->ntype;
 816     vdwparam         = fr->nbfp;
 817     vdwtype          = mdatoms->typeA;
 818
 819     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 820     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 821     beta2            = _mm256_mul_pd(beta,beta);
 822     beta3            = _mm256_mul_pd(beta,beta2);
 823
 824     ewtab            = fr->ic->tabq_coul_F;
 825     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 826     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 827
 828     /* Setup water-specific parameters */
 829     inr              = nlist->iinr[0];
 830     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 831     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 832     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 833     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 834
 835     /* Avoid stupid compiler warnings */
 836     jnrA = jnrB = jnrC = jnrD = 0;
 837     j_coord_offsetA = 0;
 838     j_coord_offsetB = 0;
 839     j_coord_offsetC = 0;
 840     j_coord_offsetD = 0;
 841
 842     outeriter        = 0;
 843     inneriter        = 0;
 844
 845     for(iidx=0;iidx<4*DIM;iidx++)
 846     {
 847         scratch[iidx] = 0.0;
 848     }
 849
 850     /* Start outer loop over neighborlists */
 851     for(iidx=0; iidx<nri; iidx++)
 852     {
 853         /* Load shift vector for this list */
 854         i_shift_offset   = DIM*shiftidx[iidx];
 855
 856         /* Load limits for loop over neighbors */
 857         j_index_start    = jindex[iidx];
 858         j_index_end      = jindex[iidx+1];
 859
 860         /* Get outer coordinate index */
 861         inr              = iinr[iidx];
 862         i_coord_offset   = DIM*inr;
 863
 864         /* Load i particle coords and add shift vector */
 865         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 866                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 867
 868         fix0             = _mm256_setzero_pd();
 869         fiy0             = _mm256_setzero_pd();
 870         fiz0             = _mm256_setzero_pd();
 871         fix1             = _mm256_setzero_pd();
 872         fiy1             = _mm256_setzero_pd();
 873         fiz1             = _mm256_setzero_pd();
 874         fix2             = _mm256_setzero_pd();
 875         fiy2             = _mm256_setzero_pd();
 876         fiz2             = _mm256_setzero_pd();
 877         fix3             = _mm256_setzero_pd();
 878         fiy3             = _mm256_setzero_pd();
 879         fiz3             = _mm256_setzero_pd();
 880
 881         /* Start inner kernel loop */
 882         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 883         {
 884
 885             /* Get j neighbor index, and coordinate index */
 886             jnrA             = jjnr[jidx];
 887             jnrB             = jjnr[jidx+1];
 888             jnrC             = jjnr[jidx+2];
 889             jnrD             = jjnr[jidx+3];
 890             j_coord_offsetA  = DIM*jnrA;
 891             j_coord_offsetB  = DIM*jnrB;
 892             j_coord_offsetC  = DIM*jnrC;
 893             j_coord_offsetD  = DIM*jnrD;
 894
 895             /* load j atom coordinates */
 896             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 897                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 898                                                  &jx0,&jy0,&jz0);
 899
 900             /* Calculate displacement vector */
 901             dx00             = _mm256_sub_pd(ix0,jx0);
 902             dy00             = _mm256_sub_pd(iy0,jy0);
 903             dz00             = _mm256_sub_pd(iz0,jz0);
 904             dx10             = _mm256_sub_pd(ix1,jx0);
 905             dy10             = _mm256_sub_pd(iy1,jy0);
 906             dz10             = _mm256_sub_pd(iz1,jz0);
 907             dx20             = _mm256_sub_pd(ix2,jx0);
 908             dy20             = _mm256_sub_pd(iy2,jy0);
 909             dz20             = _mm256_sub_pd(iz2,jz0);
 910             dx30             = _mm256_sub_pd(ix3,jx0);
 911             dy30             = _mm256_sub_pd(iy3,jy0);
 912             dz30             = _mm256_sub_pd(iz3,jz0);
 913
 914             /* Calculate squared distance and things based on it */
 915             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 916             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 917             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 918             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 919
 920             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 921             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 922             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 923
 924             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 925             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 926             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 927             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 928
 929             /* Load parameters for j particles */
 930             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 931                                                                  charge+jnrC+0,charge+jnrD+0);
 932             vdwjidx0A        = 2*vdwtype[jnrA+0];
 933             vdwjidx0B        = 2*vdwtype[jnrB+0];
 934             vdwjidx0C        = 2*vdwtype[jnrC+0];
 935             vdwjidx0D        = 2*vdwtype[jnrD+0];
 936
 937             fjx0             = _mm256_setzero_pd();
 938             fjy0             = _mm256_setzero_pd();
 939             fjz0             = _mm256_setzero_pd();
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 947                                             vdwioffsetptr0+vdwjidx0B,
 948                                             vdwioffsetptr0+vdwjidx0C,
 949                                             vdwioffsetptr0+vdwjidx0D,
 950                                             &c6_00,&c12_00);
 951
 952             /* LENNARD-JONES DISPERSION/REPULSION */
 953
 954             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 955             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 956
 957             fscal            = fvdw;
 958
 959             /* Calculate temporary vectorial force */
 960             tx               = _mm256_mul_pd(fscal,dx00);
 961             ty               = _mm256_mul_pd(fscal,dy00);
 962             tz               = _mm256_mul_pd(fscal,dz00);
 963
 964             /* Update vectorial force */
 965             fix0             = _mm256_add_pd(fix0,tx);
 966             fiy0             = _mm256_add_pd(fiy0,ty);
 967             fiz0             = _mm256_add_pd(fiz0,tz);
 968
 969             fjx0             = _mm256_add_pd(fjx0,tx);
 970             fjy0             = _mm256_add_pd(fjy0,ty);
 971             fjz0             = _mm256_add_pd(fjz0,tz);
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             r10              = _mm256_mul_pd(rsq10,rinv10);
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             qq10             = _mm256_mul_pd(iq1,jq0);
 981
 982             /* EWALD ELECTROSTATICS */
 983
 984             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 985             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 986             ewitab           = _mm256_cvttpd_epi32(ewrt);
 987             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 988             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 989                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 990                                             &ewtabF,&ewtabFn);
 991             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 992             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 993
 994             fscal            = felec;
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm256_mul_pd(fscal,dx10);
 998             ty               = _mm256_mul_pd(fscal,dy10);
 999             tz               = _mm256_mul_pd(fscal,dz10);
1000
1001             /* Update vectorial force */
1002             fix1             = _mm256_add_pd(fix1,tx);
1003             fiy1             = _mm256_add_pd(fiy1,ty);
1004             fiz1             = _mm256_add_pd(fiz1,tz);
1005
1006             fjx0             = _mm256_add_pd(fjx0,tx);
1007             fjy0             = _mm256_add_pd(fjy0,ty);
1008             fjz0             = _mm256_add_pd(fjz0,tz);
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             r20              = _mm256_mul_pd(rsq20,rinv20);
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq20             = _mm256_mul_pd(iq2,jq0);
1018
1019             /* EWALD ELECTROSTATICS */
1020
1021             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1022             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1023             ewitab           = _mm256_cvttpd_epi32(ewrt);
1024             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1025             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1026                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1027                                             &ewtabF,&ewtabFn);
1028             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1029             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1030
1031             fscal            = felec;
1032
1033             /* Calculate temporary vectorial force */
1034             tx               = _mm256_mul_pd(fscal,dx20);
1035             ty               = _mm256_mul_pd(fscal,dy20);
1036             tz               = _mm256_mul_pd(fscal,dz20);
1037
1038             /* Update vectorial force */
1039             fix2             = _mm256_add_pd(fix2,tx);
1040             fiy2             = _mm256_add_pd(fiy2,ty);
1041             fiz2             = _mm256_add_pd(fiz2,tz);
1042
1043             fjx0             = _mm256_add_pd(fjx0,tx);
1044             fjy0             = _mm256_add_pd(fjy0,ty);
1045             fjz0             = _mm256_add_pd(fjz0,tz);
1046
1047             /**************************
1048              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             r30              = _mm256_mul_pd(rsq30,rinv30);
1052
1053             /* Compute parameters for interactions between i and j atoms */
1054             qq30             = _mm256_mul_pd(iq3,jq0);
1055
1056             /* EWALD ELECTROSTATICS */
1057
1058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1060             ewitab           = _mm256_cvttpd_epi32(ewrt);
1061             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1062             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1063                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1064                                             &ewtabF,&ewtabFn);
1065             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1066             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1067
1068             fscal            = felec;
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_pd(fscal,dx30);
1072             ty               = _mm256_mul_pd(fscal,dy30);
1073             tz               = _mm256_mul_pd(fscal,dz30);
1074
1075             /* Update vectorial force */
1076             fix3             = _mm256_add_pd(fix3,tx);
1077             fiy3             = _mm256_add_pd(fiy3,ty);
1078             fiz3             = _mm256_add_pd(fiz3,tz);
1079
1080             fjx0             = _mm256_add_pd(fjx0,tx);
1081             fjy0             = _mm256_add_pd(fjy0,ty);
1082             fjz0             = _mm256_add_pd(fjz0,tz);
1083
1084             fjptrA             = f+j_coord_offsetA;
1085             fjptrB             = f+j_coord_offsetB;
1086             fjptrC             = f+j_coord_offsetC;
1087             fjptrD             = f+j_coord_offsetD;
1088
1089             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1090
1091             /* Inner loop uses 138 flops */
1092         }
1093
1094         if(jidx<j_index_end)
1095         {
1096
1097             /* Get j neighbor index, and coordinate index */
1098             jnrlistA         = jjnr[jidx];
1099             jnrlistB         = jjnr[jidx+1];
1100             jnrlistC         = jjnr[jidx+2];
1101             jnrlistD         = jjnr[jidx+3];
1102             /* Sign of each element will be negative for non-real atoms.
1103              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1104              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1105              */
1106             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1107
1108             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1109             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1110             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1111
1112             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1113             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1114             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1115             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1116             j_coord_offsetA  = DIM*jnrA;
1117             j_coord_offsetB  = DIM*jnrB;
1118             j_coord_offsetC  = DIM*jnrC;
1119             j_coord_offsetD  = DIM*jnrD;
1120
1121             /* load j atom coordinates */
1122             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1123                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1124                                                  &jx0,&jy0,&jz0);
1125
1126             /* Calculate displacement vector */
1127             dx00             = _mm256_sub_pd(ix0,jx0);
1128             dy00             = _mm256_sub_pd(iy0,jy0);
1129             dz00             = _mm256_sub_pd(iz0,jz0);
1130             dx10             = _mm256_sub_pd(ix1,jx0);
1131             dy10             = _mm256_sub_pd(iy1,jy0);
1132             dz10             = _mm256_sub_pd(iz1,jz0);
1133             dx20             = _mm256_sub_pd(ix2,jx0);
1134             dy20             = _mm256_sub_pd(iy2,jy0);
1135             dz20             = _mm256_sub_pd(iz2,jz0);
1136             dx30             = _mm256_sub_pd(ix3,jx0);
1137             dy30             = _mm256_sub_pd(iy3,jy0);
1138             dz30             = _mm256_sub_pd(iz3,jz0);
1139
1140             /* Calculate squared distance and things based on it */
1141             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1142             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1143             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1144             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1145
1146             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1147             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1148             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1149
1150             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1151             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1152             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1153             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1154
1155             /* Load parameters for j particles */
1156             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1157                                                                  charge+jnrC+0,charge+jnrD+0);
1158             vdwjidx0A        = 2*vdwtype[jnrA+0];
1159             vdwjidx0B        = 2*vdwtype[jnrB+0];
1160             vdwjidx0C        = 2*vdwtype[jnrC+0];
1161             vdwjidx0D        = 2*vdwtype[jnrD+0];
1162
1163             fjx0             = _mm256_setzero_pd();
1164             fjy0             = _mm256_setzero_pd();
1165             fjz0             = _mm256_setzero_pd();
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             /* Compute parameters for interactions between i and j atoms */
1172             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1173                                             vdwioffsetptr0+vdwjidx0B,
1174                                             vdwioffsetptr0+vdwjidx0C,
1175                                             vdwioffsetptr0+vdwjidx0D,
1176                                             &c6_00,&c12_00);
1177
1178             /* LENNARD-JONES DISPERSION/REPULSION */
1179
1180             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1181             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1182
1183             fscal            = fvdw;
1184
1185             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm256_mul_pd(fscal,dx00);
1189             ty               = _mm256_mul_pd(fscal,dy00);
1190             tz               = _mm256_mul_pd(fscal,dz00);
1191
1192             /* Update vectorial force */
1193             fix0             = _mm256_add_pd(fix0,tx);
1194             fiy0             = _mm256_add_pd(fiy0,ty);
1195             fiz0             = _mm256_add_pd(fiz0,tz);
1196
1197             fjx0             = _mm256_add_pd(fjx0,tx);
1198             fjy0             = _mm256_add_pd(fjy0,ty);
1199             fjz0             = _mm256_add_pd(fjz0,tz);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r10              = _mm256_mul_pd(rsq10,rinv10);
1206             r10              = _mm256_andnot_pd(dummy_mask,r10);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq10             = _mm256_mul_pd(iq1,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1214             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1215             ewitab           = _mm256_cvttpd_epi32(ewrt);
1216             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1217             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1218                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1219                                             &ewtabF,&ewtabFn);
1220             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1221             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm256_mul_pd(fscal,dx10);
1229             ty               = _mm256_mul_pd(fscal,dy10);
1230             tz               = _mm256_mul_pd(fscal,dz10);
1231
1232             /* Update vectorial force */
1233             fix1             = _mm256_add_pd(fix1,tx);
1234             fiy1             = _mm256_add_pd(fiy1,ty);
1235             fiz1             = _mm256_add_pd(fiz1,tz);
1236
1237             fjx0             = _mm256_add_pd(fjx0,tx);
1238             fjy0             = _mm256_add_pd(fjy0,ty);
1239             fjz0             = _mm256_add_pd(fjz0,tz);
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             r20              = _mm256_mul_pd(rsq20,rinv20);
1246             r20              = _mm256_andnot_pd(dummy_mask,r20);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             qq20             = _mm256_mul_pd(iq2,jq0);
1250
1251             /* EWALD ELECTROSTATICS */
1252
1253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1254             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1255             ewitab           = _mm256_cvttpd_epi32(ewrt);
1256             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1257             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1258                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1259                                             &ewtabF,&ewtabFn);
1260             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1261             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1262
1263             fscal            = felec;
1264
1265             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1266
1267             /* Calculate temporary vectorial force */
1268             tx               = _mm256_mul_pd(fscal,dx20);
1269             ty               = _mm256_mul_pd(fscal,dy20);
1270             tz               = _mm256_mul_pd(fscal,dz20);
1271
1272             /* Update vectorial force */
1273             fix2             = _mm256_add_pd(fix2,tx);
1274             fiy2             = _mm256_add_pd(fiy2,ty);
1275             fiz2             = _mm256_add_pd(fiz2,tz);
1276
1277             fjx0             = _mm256_add_pd(fjx0,tx);
1278             fjy0             = _mm256_add_pd(fjy0,ty);
1279             fjz0             = _mm256_add_pd(fjz0,tz);
1280
1281             /**************************
1282              * CALCULATE INTERACTIONS *
1283              **************************/
1284
1285             r30              = _mm256_mul_pd(rsq30,rinv30);
1286             r30              = _mm256_andnot_pd(dummy_mask,r30);
1287
1288             /* Compute parameters for interactions between i and j atoms */
1289             qq30             = _mm256_mul_pd(iq3,jq0);
1290
1291             /* EWALD ELECTROSTATICS */
1292
1293             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1294             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1295             ewitab           = _mm256_cvttpd_epi32(ewrt);
1296             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1297             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1298                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1299                                             &ewtabF,&ewtabFn);
1300             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1301             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1302
1303             fscal            = felec;
1304
1305             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1306
1307             /* Calculate temporary vectorial force */
1308             tx               = _mm256_mul_pd(fscal,dx30);
1309             ty               = _mm256_mul_pd(fscal,dy30);
1310             tz               = _mm256_mul_pd(fscal,dz30);
1311
1312             /* Update vectorial force */
1313             fix3             = _mm256_add_pd(fix3,tx);
1314             fiy3             = _mm256_add_pd(fiy3,ty);
1315             fiz3             = _mm256_add_pd(fiz3,tz);
1316
1317             fjx0             = _mm256_add_pd(fjx0,tx);
1318             fjy0             = _mm256_add_pd(fjy0,ty);
1319             fjz0             = _mm256_add_pd(fjz0,tz);
1320
1321             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1322             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1323             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1324             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1325
1326             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1327
1328             /* Inner loop uses 141 flops */
1329         }
1330
1331         /* End of innermost loop */
1332
1333         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1334                                                  f+i_coord_offset,fshift+i_shift_offset);
1335
1336         /* Increment number of inner iterations */
1337         inneriter                  += j_index_end - j_index_start;
1338
1339         /* Outer loop uses 24 flops */
1340     }
1341
1342     /* Increment number of outer iterations */
1343     outeriter        += nri;
1344
1345     /* Update outer/inner flops */
1346
1347     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*141);
1348 }