src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJEw_GeomW3W3_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_VdwLJEw_GeomW3W3_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_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     real *           vdwgridioffsetptr0;
  88     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     real *           vdwioffsetptr1;
  90     real *           vdwgridioffsetptr1;
  91     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     real *           vdwgridioffsetptr2;
  94     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  95     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  96     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  97     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  98     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  99     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 100     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 101     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 102     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 103     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 104     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 105     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 106     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 107     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 108     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 109     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 110     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 111     real             *charge;
 112     int              nvdwtype;
 113     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 114     int              *vdwtype;
 115     real             *vdwparam;
 116     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 117     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 118     __m256d           c6grid_00;
 119     __m256d           c6grid_01;
 120     __m256d           c6grid_02;
 121     __m256d           c6grid_10;
 122     __m256d           c6grid_11;
 123     __m256d           c6grid_12;
 124     __m256d           c6grid_20;
 125     __m256d           c6grid_21;
 126     __m256d           c6grid_22;
 127     real             *vdwgridparam;
 128     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 129     __m256d           one_half  = _mm256_set1_pd(0.5);
 130     __m256d           minus_one = _mm256_set1_pd(-1.0);
 131     __m128i          ewitab;
 132     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 133     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 134     real             *ewtab;
 135     __m256d          dummy_mask,cutoff_mask;
 136     __m128           tmpmask0,tmpmask1;
 137     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 138     __m256d          one     = _mm256_set1_pd(1.0);
 139     __m256d          two     = _mm256_set1_pd(2.0);
 140     x                = xx[0];
 141     f                = ff[0];
 142
 143     nri              = nlist->nri;
 144     iinr             = nlist->iinr;
 145     jindex           = nlist->jindex;
 146     jjnr             = nlist->jjnr;
 147     shiftidx         = nlist->shift;
 148     gid              = nlist->gid;
 149     shiftvec         = fr->shift_vec[0];
 150     fshift           = fr->fshift[0];
 151     facel            = _mm256_set1_pd(fr->epsfac);
 152     charge           = mdatoms->chargeA;
 153     nvdwtype         = fr->ntype;
 154     vdwparam         = fr->nbfp;
 155     vdwtype          = mdatoms->typeA;
 156     vdwgridparam     = fr->ljpme_c6grid;
 157     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 158     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 159     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 160
 161     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 162     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 163     beta2            = _mm256_mul_pd(beta,beta);
 164     beta3            = _mm256_mul_pd(beta,beta2);
 165
 166     ewtab            = fr->ic->tabq_coul_FDV0;
 167     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 168     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 169
 170     /* Setup water-specific parameters */
 171     inr              = nlist->iinr[0];
 172     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 173     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 174     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 175     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 176     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 177
 178     jq0              = _mm256_set1_pd(charge[inr+0]);
 179     jq1              = _mm256_set1_pd(charge[inr+1]);
 180     jq2              = _mm256_set1_pd(charge[inr+2]);
 181     vdwjidx0A        = 2*vdwtype[inr+0];
 182     qq00             = _mm256_mul_pd(iq0,jq0);
 183     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 184     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 185     c6grid_00        = _mm256_set1_pd(vdwgridioffsetptr0[vdwjidx0A]);
 186     qq01             = _mm256_mul_pd(iq0,jq1);
 187     qq02             = _mm256_mul_pd(iq0,jq2);
 188     qq10             = _mm256_mul_pd(iq1,jq0);
 189     qq11             = _mm256_mul_pd(iq1,jq1);
 190     qq12             = _mm256_mul_pd(iq1,jq2);
 191     qq20             = _mm256_mul_pd(iq2,jq0);
 192     qq21             = _mm256_mul_pd(iq2,jq1);
 193     qq22             = _mm256_mul_pd(iq2,jq2);
 194
 195     /* Avoid stupid compiler warnings */
 196     jnrA = jnrB = jnrC = jnrD = 0;
 197     j_coord_offsetA = 0;
 198     j_coord_offsetB = 0;
 199     j_coord_offsetC = 0;
 200     j_coord_offsetD = 0;
 201
 202     outeriter        = 0;
 203     inneriter        = 0;
 204
 205     for(iidx=0;iidx<4*DIM;iidx++)
 206     {
 207         scratch[iidx] = 0.0;
 208     }
 209
 210     /* Start outer loop over neighborlists */
 211     for(iidx=0; iidx<nri; iidx++)
 212     {
 213         /* Load shift vector for this list */
 214         i_shift_offset   = DIM*shiftidx[iidx];
 215
 216         /* Load limits for loop over neighbors */
 217         j_index_start    = jindex[iidx];
 218         j_index_end      = jindex[iidx+1];
 219
 220         /* Get outer coordinate index */
 221         inr              = iinr[iidx];
 222         i_coord_offset   = DIM*inr;
 223
 224         /* Load i particle coords and add shift vector */
 225         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 226                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 227
 228         fix0             = _mm256_setzero_pd();
 229         fiy0             = _mm256_setzero_pd();
 230         fiz0             = _mm256_setzero_pd();
 231         fix1             = _mm256_setzero_pd();
 232         fiy1             = _mm256_setzero_pd();
 233         fiz1             = _mm256_setzero_pd();
 234         fix2             = _mm256_setzero_pd();
 235         fiy2             = _mm256_setzero_pd();
 236         fiz2             = _mm256_setzero_pd();
 237
 238         /* Reset potential sums */
 239         velecsum         = _mm256_setzero_pd();
 240         vvdwsum          = _mm256_setzero_pd();
 241
 242         /* Start inner kernel loop */
 243         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 244         {
 245
 246             /* Get j neighbor index, and coordinate index */
 247             jnrA             = jjnr[jidx];
 248             jnrB             = jjnr[jidx+1];
 249             jnrC             = jjnr[jidx+2];
 250             jnrD             = jjnr[jidx+3];
 251             j_coord_offsetA  = DIM*jnrA;
 252             j_coord_offsetB  = DIM*jnrB;
 253             j_coord_offsetC  = DIM*jnrC;
 254             j_coord_offsetD  = DIM*jnrD;
 255
 256             /* load j atom coordinates */
 257             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 258                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 259                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm256_sub_pd(ix0,jx0);
 263             dy00             = _mm256_sub_pd(iy0,jy0);
 264             dz00             = _mm256_sub_pd(iz0,jz0);
 265             dx01             = _mm256_sub_pd(ix0,jx1);
 266             dy01             = _mm256_sub_pd(iy0,jy1);
 267             dz01             = _mm256_sub_pd(iz0,jz1);
 268             dx02             = _mm256_sub_pd(ix0,jx2);
 269             dy02             = _mm256_sub_pd(iy0,jy2);
 270             dz02             = _mm256_sub_pd(iz0,jz2);
 271             dx10             = _mm256_sub_pd(ix1,jx0);
 272             dy10             = _mm256_sub_pd(iy1,jy0);
 273             dz10             = _mm256_sub_pd(iz1,jz0);
 274             dx11             = _mm256_sub_pd(ix1,jx1);
 275             dy11             = _mm256_sub_pd(iy1,jy1);
 276             dz11             = _mm256_sub_pd(iz1,jz1);
 277             dx12             = _mm256_sub_pd(ix1,jx2);
 278             dy12             = _mm256_sub_pd(iy1,jy2);
 279             dz12             = _mm256_sub_pd(iz1,jz2);
 280             dx20             = _mm256_sub_pd(ix2,jx0);
 281             dy20             = _mm256_sub_pd(iy2,jy0);
 282             dz20             = _mm256_sub_pd(iz2,jz0);
 283             dx21             = _mm256_sub_pd(ix2,jx1);
 284             dy21             = _mm256_sub_pd(iy2,jy1);
 285             dz21             = _mm256_sub_pd(iz2,jz1);
 286             dx22             = _mm256_sub_pd(ix2,jx2);
 287             dy22             = _mm256_sub_pd(iy2,jy2);
 288             dz22             = _mm256_sub_pd(iz2,jz2);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 292             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 293             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 294             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 295             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 296             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 297             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 298             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 299             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 300
 301             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 302             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 303             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 304             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 305             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 306             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 307             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 308             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 309             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 310
 311             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 312             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 313             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 314             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 315             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 316             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 317             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 318             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 319             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 320
 321             fjx0             = _mm256_setzero_pd();
 322             fjy0             = _mm256_setzero_pd();
 323             fjz0             = _mm256_setzero_pd();
 324             fjx1             = _mm256_setzero_pd();
 325             fjy1             = _mm256_setzero_pd();
 326             fjz1             = _mm256_setzero_pd();
 327             fjx2             = _mm256_setzero_pd();
 328             fjy2             = _mm256_setzero_pd();
 329             fjz2             = _mm256_setzero_pd();
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r00              = _mm256_mul_pd(rsq00,rinv00);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 340             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 341             ewitab           = _mm256_cvttpd_epi32(ewrt);
 342             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 343             ewitab           = _mm_slli_epi32(ewitab,2);
 344             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 345             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 346             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 347             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 348             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 349             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 350             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 351             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 352             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 353
 354             /* Analytical LJ-PME */
 355             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 356             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 357             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 358             exponent         = gmx_simd_exp_d(ewcljrsq);
 359             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 360             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 361             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 362             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 363             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 364             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 365             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 366             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm256_add_pd(velecsum,velec);
 370             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 371
 372             fscal            = _mm256_add_pd(felec,fvdw);
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm256_mul_pd(fscal,dx00);
 376             ty               = _mm256_mul_pd(fscal,dy00);
 377             tz               = _mm256_mul_pd(fscal,dz00);
 378
 379             /* Update vectorial force */
 380             fix0             = _mm256_add_pd(fix0,tx);
 381             fiy0             = _mm256_add_pd(fiy0,ty);
 382             fiz0             = _mm256_add_pd(fiz0,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             r01              = _mm256_mul_pd(rsq01,rinv01);
 393
 394             /* EWALD ELECTROSTATICS */
 395
 396             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 397             ewrt             = _mm256_mul_pd(r01,ewtabscale);
 398             ewitab           = _mm256_cvttpd_epi32(ewrt);
 399             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 400             ewitab           = _mm_slli_epi32(ewitab,2);
 401             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 402             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 403             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 404             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 405             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 406             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 407             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 408             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
 409             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velecsum         = _mm256_add_pd(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             /* Calculate temporary vectorial force */
 417             tx               = _mm256_mul_pd(fscal,dx01);
 418             ty               = _mm256_mul_pd(fscal,dy01);
 419             tz               = _mm256_mul_pd(fscal,dz01);
 420
 421             /* Update vectorial force */
 422             fix0             = _mm256_add_pd(fix0,tx);
 423             fiy0             = _mm256_add_pd(fiy0,ty);
 424             fiz0             = _mm256_add_pd(fiz0,tz);
 425
 426             fjx1             = _mm256_add_pd(fjx1,tx);
 427             fjy1             = _mm256_add_pd(fjy1,ty);
 428             fjz1             = _mm256_add_pd(fjz1,tz);
 429
 430             /**************************
 431              * CALCULATE INTERACTIONS *
 432              **************************/
 433
 434             r02              = _mm256_mul_pd(rsq02,rinv02);
 435
 436             /* EWALD ELECTROSTATICS */
 437
 438             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 439             ewrt             = _mm256_mul_pd(r02,ewtabscale);
 440             ewitab           = _mm256_cvttpd_epi32(ewrt);
 441             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 442             ewitab           = _mm_slli_epi32(ewitab,2);
 443             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 444             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 445             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 446             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 447             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 448             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 449             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 450             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
 451             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velecsum         = _mm256_add_pd(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm256_mul_pd(fscal,dx02);
 460             ty               = _mm256_mul_pd(fscal,dy02);
 461             tz               = _mm256_mul_pd(fscal,dz02);
 462
 463             /* Update vectorial force */
 464             fix0             = _mm256_add_pd(fix0,tx);
 465             fiy0             = _mm256_add_pd(fiy0,ty);
 466             fiz0             = _mm256_add_pd(fiz0,tz);
 467
 468             fjx2             = _mm256_add_pd(fjx2,tx);
 469             fjy2             = _mm256_add_pd(fjy2,ty);
 470             fjz2             = _mm256_add_pd(fjz2,tz);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             r10              = _mm256_mul_pd(rsq10,rinv10);
 477
 478             /* EWALD ELECTROSTATICS */
 479
 480             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 481             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 482             ewitab           = _mm256_cvttpd_epi32(ewrt);
 483             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 484             ewitab           = _mm_slli_epi32(ewitab,2);
 485             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 486             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 487             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 488             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 489             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 490             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 491             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 492             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 493             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velecsum         = _mm256_add_pd(velecsum,velec);
 497
 498             fscal            = felec;
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm256_mul_pd(fscal,dx10);
 502             ty               = _mm256_mul_pd(fscal,dy10);
 503             tz               = _mm256_mul_pd(fscal,dz10);
 504
 505             /* Update vectorial force */
 506             fix1             = _mm256_add_pd(fix1,tx);
 507             fiy1             = _mm256_add_pd(fiy1,ty);
 508             fiz1             = _mm256_add_pd(fiz1,tz);
 509
 510             fjx0             = _mm256_add_pd(fjx0,tx);
 511             fjy0             = _mm256_add_pd(fjy0,ty);
 512             fjz0             = _mm256_add_pd(fjz0,tz);
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             r11              = _mm256_mul_pd(rsq11,rinv11);
 519
 520             /* EWALD ELECTROSTATICS */
 521
 522             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 523             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 524             ewitab           = _mm256_cvttpd_epi32(ewrt);
 525             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 526             ewitab           = _mm_slli_epi32(ewitab,2);
 527             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 528             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 529             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 530             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 531             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 532             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 533             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 534             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 535             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velecsum         = _mm256_add_pd(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm256_mul_pd(fscal,dx11);
 544             ty               = _mm256_mul_pd(fscal,dy11);
 545             tz               = _mm256_mul_pd(fscal,dz11);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm256_add_pd(fix1,tx);
 549             fiy1             = _mm256_add_pd(fiy1,ty);
 550             fiz1             = _mm256_add_pd(fiz1,tz);
 551
 552             fjx1             = _mm256_add_pd(fjx1,tx);
 553             fjy1             = _mm256_add_pd(fjy1,ty);
 554             fjz1             = _mm256_add_pd(fjz1,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r12              = _mm256_mul_pd(rsq12,rinv12);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 565             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 566             ewitab           = _mm256_cvttpd_epi32(ewrt);
 567             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 568             ewitab           = _mm_slli_epi32(ewitab,2);
 569             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 570             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 571             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 572             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 573             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 574             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 575             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 576             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 577             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velecsum         = _mm256_add_pd(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx12);
 586             ty               = _mm256_mul_pd(fscal,dy12);
 587             tz               = _mm256_mul_pd(fscal,dz12);
 588
 589             /* Update vectorial force */
 590             fix1             = _mm256_add_pd(fix1,tx);
 591             fiy1             = _mm256_add_pd(fiy1,ty);
 592             fiz1             = _mm256_add_pd(fiz1,tz);
 593
 594             fjx2             = _mm256_add_pd(fjx2,tx);
 595             fjy2             = _mm256_add_pd(fjy2,ty);
 596             fjz2             = _mm256_add_pd(fjz2,tz);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r20              = _mm256_mul_pd(rsq20,rinv20);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 607             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 608             ewitab           = _mm256_cvttpd_epi32(ewrt);
 609             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 610             ewitab           = _mm_slli_epi32(ewitab,2);
 611             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 612             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 613             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 614             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 615             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 616             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 617             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 618             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 619             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velecsum         = _mm256_add_pd(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm256_mul_pd(fscal,dx20);
 628             ty               = _mm256_mul_pd(fscal,dy20);
 629             tz               = _mm256_mul_pd(fscal,dz20);
 630
 631             /* Update vectorial force */
 632             fix2             = _mm256_add_pd(fix2,tx);
 633             fiy2             = _mm256_add_pd(fiy2,ty);
 634             fiz2             = _mm256_add_pd(fiz2,tz);
 635
 636             fjx0             = _mm256_add_pd(fjx0,tx);
 637             fjy0             = _mm256_add_pd(fjy0,ty);
 638             fjz0             = _mm256_add_pd(fjz0,tz);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             r21              = _mm256_mul_pd(rsq21,rinv21);
 645
 646             /* EWALD ELECTROSTATICS */
 647
 648             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 649             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 650             ewitab           = _mm256_cvttpd_epi32(ewrt);
 651             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 652             ewitab           = _mm_slli_epi32(ewitab,2);
 653             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 654             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 655             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 656             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 657             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 658             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 659             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 660             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 661             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 662
 663             /* Update potential sum for this i atom from the interaction with this j atom. */
 664             velecsum         = _mm256_add_pd(velecsum,velec);
 665
 666             fscal            = felec;
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_pd(fscal,dx21);
 670             ty               = _mm256_mul_pd(fscal,dy21);
 671             tz               = _mm256_mul_pd(fscal,dz21);
 672
 673             /* Update vectorial force */
 674             fix2             = _mm256_add_pd(fix2,tx);
 675             fiy2             = _mm256_add_pd(fiy2,ty);
 676             fiz2             = _mm256_add_pd(fiz2,tz);
 677
 678             fjx1             = _mm256_add_pd(fjx1,tx);
 679             fjy1             = _mm256_add_pd(fjy1,ty);
 680             fjz1             = _mm256_add_pd(fjz1,tz);
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             r22              = _mm256_mul_pd(rsq22,rinv22);
 687
 688             /* EWALD ELECTROSTATICS */
 689
 690             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 691             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 692             ewitab           = _mm256_cvttpd_epi32(ewrt);
 693             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 694             ewitab           = _mm_slli_epi32(ewitab,2);
 695             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 696             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 697             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 698             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 699             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 700             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 701             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 702             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 703             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 704
 705             /* Update potential sum for this i atom from the interaction with this j atom. */
 706             velecsum         = _mm256_add_pd(velecsum,velec);
 707
 708             fscal            = felec;
 709
 710             /* Calculate temporary vectorial force */
 711             tx               = _mm256_mul_pd(fscal,dx22);
 712             ty               = _mm256_mul_pd(fscal,dy22);
 713             tz               = _mm256_mul_pd(fscal,dz22);
 714
 715             /* Update vectorial force */
 716             fix2             = _mm256_add_pd(fix2,tx);
 717             fiy2             = _mm256_add_pd(fiy2,ty);
 718             fiz2             = _mm256_add_pd(fiz2,tz);
 719
 720             fjx2             = _mm256_add_pd(fjx2,tx);
 721             fjy2             = _mm256_add_pd(fjy2,ty);
 722             fjz2             = _mm256_add_pd(fjz2,tz);
 723
 724             fjptrA             = f+j_coord_offsetA;
 725             fjptrB             = f+j_coord_offsetB;
 726             fjptrC             = f+j_coord_offsetC;
 727             fjptrD             = f+j_coord_offsetD;
 728
 729             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 730                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 731
 732             /* Inner loop uses 400 flops */
 733         }
 734
 735         if(jidx<j_index_end)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrlistA         = jjnr[jidx];
 740             jnrlistB         = jjnr[jidx+1];
 741             jnrlistC         = jjnr[jidx+2];
 742             jnrlistD         = jjnr[jidx+3];
 743             /* Sign of each element will be negative for non-real atoms.
 744              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 745              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 746              */
 747             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 748
 749             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 750             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 751             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 752
 753             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 754             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 755             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 756             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 757             j_coord_offsetA  = DIM*jnrA;
 758             j_coord_offsetB  = DIM*jnrB;
 759             j_coord_offsetC  = DIM*jnrC;
 760             j_coord_offsetD  = DIM*jnrD;
 761
 762             /* load j atom coordinates */
 763             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 764                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 765                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 766
 767             /* Calculate displacement vector */
 768             dx00             = _mm256_sub_pd(ix0,jx0);
 769             dy00             = _mm256_sub_pd(iy0,jy0);
 770             dz00             = _mm256_sub_pd(iz0,jz0);
 771             dx01             = _mm256_sub_pd(ix0,jx1);
 772             dy01             = _mm256_sub_pd(iy0,jy1);
 773             dz01             = _mm256_sub_pd(iz0,jz1);
 774             dx02             = _mm256_sub_pd(ix0,jx2);
 775             dy02             = _mm256_sub_pd(iy0,jy2);
 776             dz02             = _mm256_sub_pd(iz0,jz2);
 777             dx10             = _mm256_sub_pd(ix1,jx0);
 778             dy10             = _mm256_sub_pd(iy1,jy0);
 779             dz10             = _mm256_sub_pd(iz1,jz0);
 780             dx11             = _mm256_sub_pd(ix1,jx1);
 781             dy11             = _mm256_sub_pd(iy1,jy1);
 782             dz11             = _mm256_sub_pd(iz1,jz1);
 783             dx12             = _mm256_sub_pd(ix1,jx2);
 784             dy12             = _mm256_sub_pd(iy1,jy2);
 785             dz12             = _mm256_sub_pd(iz1,jz2);
 786             dx20             = _mm256_sub_pd(ix2,jx0);
 787             dy20             = _mm256_sub_pd(iy2,jy0);
 788             dz20             = _mm256_sub_pd(iz2,jz0);
 789             dx21             = _mm256_sub_pd(ix2,jx1);
 790             dy21             = _mm256_sub_pd(iy2,jy1);
 791             dz21             = _mm256_sub_pd(iz2,jz1);
 792             dx22             = _mm256_sub_pd(ix2,jx2);
 793             dy22             = _mm256_sub_pd(iy2,jy2);
 794             dz22             = _mm256_sub_pd(iz2,jz2);
 795
 796             /* Calculate squared distance and things based on it */
 797             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 798             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 799             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 800             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 801             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 802             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 803             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 804             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 805             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 806
 807             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 808             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 809             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 810             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 811             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 812             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 813             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 814             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 815             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 816
 817             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 818             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 819             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 820             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 821             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 822             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 823             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 824             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 825             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 826
 827             fjx0             = _mm256_setzero_pd();
 828             fjy0             = _mm256_setzero_pd();
 829             fjz0             = _mm256_setzero_pd();
 830             fjx1             = _mm256_setzero_pd();
 831             fjy1             = _mm256_setzero_pd();
 832             fjz1             = _mm256_setzero_pd();
 833             fjx2             = _mm256_setzero_pd();
 834             fjy2             = _mm256_setzero_pd();
 835             fjz2             = _mm256_setzero_pd();
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             r00              = _mm256_mul_pd(rsq00,rinv00);
 842             r00              = _mm256_andnot_pd(dummy_mask,r00);
 843
 844             /* EWALD ELECTROSTATICS */
 845
 846             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 847             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 848             ewitab           = _mm256_cvttpd_epi32(ewrt);
 849             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 850             ewitab           = _mm_slli_epi32(ewitab,2);
 851             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 852             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 853             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 854             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 855             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 856             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 857             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 858             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 859             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 860
 861             /* Analytical LJ-PME */
 862             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 863             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 864             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 865             exponent         = gmx_simd_exp_d(ewcljrsq);
 866             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 867             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 868             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 869             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 870             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 871             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 872             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 873             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 874
 875             /* Update potential sum for this i atom from the interaction with this j atom. */
 876             velec            = _mm256_andnot_pd(dummy_mask,velec);
 877             velecsum         = _mm256_add_pd(velecsum,velec);
 878             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 879             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 880
 881             fscal            = _mm256_add_pd(felec,fvdw);
 882
 883             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 884
 885             /* Calculate temporary vectorial force */
 886             tx               = _mm256_mul_pd(fscal,dx00);
 887             ty               = _mm256_mul_pd(fscal,dy00);
 888             tz               = _mm256_mul_pd(fscal,dz00);
 889
 890             /* Update vectorial force */
 891             fix0             = _mm256_add_pd(fix0,tx);
 892             fiy0             = _mm256_add_pd(fiy0,ty);
 893             fiz0             = _mm256_add_pd(fiz0,tz);
 894
 895             fjx0             = _mm256_add_pd(fjx0,tx);
 896             fjy0             = _mm256_add_pd(fjy0,ty);
 897             fjz0             = _mm256_add_pd(fjz0,tz);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r01              = _mm256_mul_pd(rsq01,rinv01);
 904             r01              = _mm256_andnot_pd(dummy_mask,r01);
 905
 906             /* EWALD ELECTROSTATICS */
 907
 908             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 909             ewrt             = _mm256_mul_pd(r01,ewtabscale);
 910             ewitab           = _mm256_cvttpd_epi32(ewrt);
 911             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 912             ewitab           = _mm_slli_epi32(ewitab,2);
 913             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 914             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 915             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 916             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 917             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 918             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 919             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 920             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
 921             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
 922
 923             /* Update potential sum for this i atom from the interaction with this j atom. */
 924             velec            = _mm256_andnot_pd(dummy_mask,velec);
 925             velecsum         = _mm256_add_pd(velecsum,velec);
 926
 927             fscal            = felec;
 928
 929             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 930
 931             /* Calculate temporary vectorial force */
 932             tx               = _mm256_mul_pd(fscal,dx01);
 933             ty               = _mm256_mul_pd(fscal,dy01);
 934             tz               = _mm256_mul_pd(fscal,dz01);
 935
 936             /* Update vectorial force */
 937             fix0             = _mm256_add_pd(fix0,tx);
 938             fiy0             = _mm256_add_pd(fiy0,ty);
 939             fiz0             = _mm256_add_pd(fiz0,tz);
 940
 941             fjx1             = _mm256_add_pd(fjx1,tx);
 942             fjy1             = _mm256_add_pd(fjy1,ty);
 943             fjz1             = _mm256_add_pd(fjz1,tz);
 944
 945             /**************************
 946              * CALCULATE INTERACTIONS *
 947              **************************/
 948
 949             r02              = _mm256_mul_pd(rsq02,rinv02);
 950             r02              = _mm256_andnot_pd(dummy_mask,r02);
 951
 952             /* EWALD ELECTROSTATICS */
 953
 954             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 955             ewrt             = _mm256_mul_pd(r02,ewtabscale);
 956             ewitab           = _mm256_cvttpd_epi32(ewrt);
 957             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 958             ewitab           = _mm_slli_epi32(ewitab,2);
 959             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 960             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 961             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 962             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 963             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 964             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 965             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 966             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
 967             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_pd(dummy_mask,velec);
 971             velecsum         = _mm256_add_pd(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_pd(fscal,dx02);
 979             ty               = _mm256_mul_pd(fscal,dy02);
 980             tz               = _mm256_mul_pd(fscal,dz02);
 981
 982             /* Update vectorial force */
 983             fix0             = _mm256_add_pd(fix0,tx);
 984             fiy0             = _mm256_add_pd(fiy0,ty);
 985             fiz0             = _mm256_add_pd(fiz0,tz);
 986
 987             fjx2             = _mm256_add_pd(fjx2,tx);
 988             fjy2             = _mm256_add_pd(fjy2,ty);
 989             fjz2             = _mm256_add_pd(fjz2,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             r10              = _mm256_mul_pd(rsq10,rinv10);
 996             r10              = _mm256_andnot_pd(dummy_mask,r10);
 997
 998             /* EWALD ELECTROSTATICS */
 999
1000             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1001             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1002             ewitab           = _mm256_cvttpd_epi32(ewrt);
1003             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1004             ewitab           = _mm_slli_epi32(ewitab,2);
1005             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1006             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1007             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1008             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1009             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1010             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1011             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1012             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1013             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1014
1015             /* Update potential sum for this i atom from the interaction with this j atom. */
1016             velec            = _mm256_andnot_pd(dummy_mask,velec);
1017             velecsum         = _mm256_add_pd(velecsum,velec);
1018
1019             fscal            = felec;
1020
1021             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm256_mul_pd(fscal,dx10);
1025             ty               = _mm256_mul_pd(fscal,dy10);
1026             tz               = _mm256_mul_pd(fscal,dz10);
1027
1028             /* Update vectorial force */
1029             fix1             = _mm256_add_pd(fix1,tx);
1030             fiy1             = _mm256_add_pd(fiy1,ty);
1031             fiz1             = _mm256_add_pd(fiz1,tz);
1032
1033             fjx0             = _mm256_add_pd(fjx0,tx);
1034             fjy0             = _mm256_add_pd(fjy0,ty);
1035             fjz0             = _mm256_add_pd(fjz0,tz);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r11              = _mm256_mul_pd(rsq11,rinv11);
1042             r11              = _mm256_andnot_pd(dummy_mask,r11);
1043
1044             /* EWALD ELECTROSTATICS */
1045
1046             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1048             ewitab           = _mm256_cvttpd_epi32(ewrt);
1049             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1050             ewitab           = _mm_slli_epi32(ewitab,2);
1051             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1052             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1053             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1054             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1055             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1056             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1057             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1058             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
1059             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1060
1061             /* Update potential sum for this i atom from the interaction with this j atom. */
1062             velec            = _mm256_andnot_pd(dummy_mask,velec);
1063             velecsum         = _mm256_add_pd(velecsum,velec);
1064
1065             fscal            = felec;
1066
1067             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_pd(fscal,dx11);
1071             ty               = _mm256_mul_pd(fscal,dy11);
1072             tz               = _mm256_mul_pd(fscal,dz11);
1073
1074             /* Update vectorial force */
1075             fix1             = _mm256_add_pd(fix1,tx);
1076             fiy1             = _mm256_add_pd(fiy1,ty);
1077             fiz1             = _mm256_add_pd(fiz1,tz);
1078
1079             fjx1             = _mm256_add_pd(fjx1,tx);
1080             fjy1             = _mm256_add_pd(fjy1,ty);
1081             fjz1             = _mm256_add_pd(fjz1,tz);
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             r12              = _mm256_mul_pd(rsq12,rinv12);
1088             r12              = _mm256_andnot_pd(dummy_mask,r12);
1089
1090             /* EWALD ELECTROSTATICS */
1091
1092             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1093             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1094             ewitab           = _mm256_cvttpd_epi32(ewrt);
1095             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1096             ewitab           = _mm_slli_epi32(ewitab,2);
1097             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1098             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1099             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1100             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1101             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1102             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1103             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1104             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1105             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1106
1107             /* Update potential sum for this i atom from the interaction with this j atom. */
1108             velec            = _mm256_andnot_pd(dummy_mask,velec);
1109             velecsum         = _mm256_add_pd(velecsum,velec);
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm256_mul_pd(fscal,dx12);
1117             ty               = _mm256_mul_pd(fscal,dy12);
1118             tz               = _mm256_mul_pd(fscal,dz12);
1119
1120             /* Update vectorial force */
1121             fix1             = _mm256_add_pd(fix1,tx);
1122             fiy1             = _mm256_add_pd(fiy1,ty);
1123             fiz1             = _mm256_add_pd(fiz1,tz);
1124
1125             fjx2             = _mm256_add_pd(fjx2,tx);
1126             fjy2             = _mm256_add_pd(fjy2,ty);
1127             fjz2             = _mm256_add_pd(fjz2,tz);
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r20              = _mm256_mul_pd(rsq20,rinv20);
1134             r20              = _mm256_andnot_pd(dummy_mask,r20);
1135
1136             /* EWALD ELECTROSTATICS */
1137
1138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1139             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1140             ewitab           = _mm256_cvttpd_epi32(ewrt);
1141             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1142             ewitab           = _mm_slli_epi32(ewitab,2);
1143             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1144             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1145             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1146             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1147             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1148             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1149             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1150             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1151             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1152
1153             /* Update potential sum for this i atom from the interaction with this j atom. */
1154             velec            = _mm256_andnot_pd(dummy_mask,velec);
1155             velecsum         = _mm256_add_pd(velecsum,velec);
1156
1157             fscal            = felec;
1158
1159             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1160
1161             /* Calculate temporary vectorial force */
1162             tx               = _mm256_mul_pd(fscal,dx20);
1163             ty               = _mm256_mul_pd(fscal,dy20);
1164             tz               = _mm256_mul_pd(fscal,dz20);
1165
1166             /* Update vectorial force */
1167             fix2             = _mm256_add_pd(fix2,tx);
1168             fiy2             = _mm256_add_pd(fiy2,ty);
1169             fiz2             = _mm256_add_pd(fiz2,tz);
1170
1171             fjx0             = _mm256_add_pd(fjx0,tx);
1172             fjy0             = _mm256_add_pd(fjy0,ty);
1173             fjz0             = _mm256_add_pd(fjz0,tz);
1174
1175             /**************************
1176              * CALCULATE INTERACTIONS *
1177              **************************/
1178
1179             r21              = _mm256_mul_pd(rsq21,rinv21);
1180             r21              = _mm256_andnot_pd(dummy_mask,r21);
1181
1182             /* EWALD ELECTROSTATICS */
1183
1184             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1185             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1186             ewitab           = _mm256_cvttpd_epi32(ewrt);
1187             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1188             ewitab           = _mm_slli_epi32(ewitab,2);
1189             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1190             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1191             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1192             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1193             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1194             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1195             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1196             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
1197             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1198
1199             /* Update potential sum for this i atom from the interaction with this j atom. */
1200             velec            = _mm256_andnot_pd(dummy_mask,velec);
1201             velecsum         = _mm256_add_pd(velecsum,velec);
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm256_mul_pd(fscal,dx21);
1209             ty               = _mm256_mul_pd(fscal,dy21);
1210             tz               = _mm256_mul_pd(fscal,dz21);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm256_add_pd(fix2,tx);
1214             fiy2             = _mm256_add_pd(fiy2,ty);
1215             fiz2             = _mm256_add_pd(fiz2,tz);
1216
1217             fjx1             = _mm256_add_pd(fjx1,tx);
1218             fjy1             = _mm256_add_pd(fjy1,ty);
1219             fjz1             = _mm256_add_pd(fjz1,tz);
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             r22              = _mm256_mul_pd(rsq22,rinv22);
1226             r22              = _mm256_andnot_pd(dummy_mask,r22);
1227
1228             /* EWALD ELECTROSTATICS */
1229
1230             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1231             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1232             ewitab           = _mm256_cvttpd_epi32(ewrt);
1233             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1234             ewitab           = _mm_slli_epi32(ewitab,2);
1235             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1236             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1237             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1238             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1239             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1240             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1241             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1242             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
1243             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1244
1245             /* Update potential sum for this i atom from the interaction with this j atom. */
1246             velec            = _mm256_andnot_pd(dummy_mask,velec);
1247             velecsum         = _mm256_add_pd(velecsum,velec);
1248
1249             fscal            = felec;
1250
1251             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm256_mul_pd(fscal,dx22);
1255             ty               = _mm256_mul_pd(fscal,dy22);
1256             tz               = _mm256_mul_pd(fscal,dz22);
1257
1258             /* Update vectorial force */
1259             fix2             = _mm256_add_pd(fix2,tx);
1260             fiy2             = _mm256_add_pd(fiy2,ty);
1261             fiz2             = _mm256_add_pd(fiz2,tz);
1262
1263             fjx2             = _mm256_add_pd(fjx2,tx);
1264             fjy2             = _mm256_add_pd(fjy2,ty);
1265             fjz2             = _mm256_add_pd(fjz2,tz);
1266
1267             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1268             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1269             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1270             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1271
1272             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1273                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1274
1275             /* Inner loop uses 409 flops */
1276         }
1277
1278         /* End of innermost loop */
1279
1280         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1281                                                  f+i_coord_offset,fshift+i_shift_offset);
1282
1283         ggid                        = gid[iidx];
1284         /* Update potential energies */
1285         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1286         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1287
1288         /* Increment number of inner iterations */
1289         inneriter                  += j_index_end - j_index_start;
1290
1291         /* Outer loop uses 20 flops */
1292     }
1293
1294     /* Increment number of outer iterations */
1295     outeriter        += nri;
1296
1297     /* Update outer/inner flops */
1298
1299     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*409);
1300 }
1301 /*
1302  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_avx_256_double
1303  * Electrostatics interaction: Ewald
1304  * VdW interaction:            LJEwald
1305  * Geometry:                   Water3-Water3
1306  * Calculate force/pot:        Force
1307  */
1308 void
1309 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_avx_256_double
1310                     (t_nblist                    * gmx_restrict       nlist,
1311                      rvec                        * gmx_restrict          xx,
1312                      rvec                        * gmx_restrict          ff,
1313                      t_forcerec                  * gmx_restrict          fr,
1314                      t_mdatoms                   * gmx_restrict     mdatoms,
1315                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1316                      t_nrnb                      * gmx_restrict        nrnb)
1317 {
1318     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1319      * just 0 for non-waters.
1320      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1321      * jnr indices corresponding to data put in the four positions in the SIMD register.
1322      */
1323     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1324     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1325     int              jnrA,jnrB,jnrC,jnrD;
1326     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1327     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1328     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1329     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1330     real             rcutoff_scalar;
1331     real             *shiftvec,*fshift,*x,*f;
1332     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1333     real             scratch[4*DIM];
1334     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1335     real *           vdwioffsetptr0;
1336     real *           vdwgridioffsetptr0;
1337     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1338     real *           vdwioffsetptr1;
1339     real *           vdwgridioffsetptr1;
1340     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1341     real *           vdwioffsetptr2;
1342     real *           vdwgridioffsetptr2;
1343     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1344     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1345     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1346     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1347     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1348     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1349     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1350     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1351     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1352     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1353     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1354     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1355     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1356     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1357     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1358     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1359     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1360     real             *charge;
1361     int              nvdwtype;
1362     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1363     int              *vdwtype;
1364     real             *vdwparam;
1365     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1366     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1367     __m256d           c6grid_00;
1368     __m256d           c6grid_01;
1369     __m256d           c6grid_02;
1370     __m256d           c6grid_10;
1371     __m256d           c6grid_11;
1372     __m256d           c6grid_12;
1373     __m256d           c6grid_20;
1374     __m256d           c6grid_21;
1375     __m256d           c6grid_22;
1376     real             *vdwgridparam;
1377     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1378     __m256d           one_half  = _mm256_set1_pd(0.5);
1379     __m256d           minus_one = _mm256_set1_pd(-1.0);
1380     __m128i          ewitab;
1381     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1382     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1383     real             *ewtab;
1384     __m256d          dummy_mask,cutoff_mask;
1385     __m128           tmpmask0,tmpmask1;
1386     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1387     __m256d          one     = _mm256_set1_pd(1.0);
1388     __m256d          two     = _mm256_set1_pd(2.0);
1389     x                = xx[0];
1390     f                = ff[0];
1391
1392     nri              = nlist->nri;
1393     iinr             = nlist->iinr;
1394     jindex           = nlist->jindex;
1395     jjnr             = nlist->jjnr;
1396     shiftidx         = nlist->shift;
1397     gid              = nlist->gid;
1398     shiftvec         = fr->shift_vec[0];
1399     fshift           = fr->fshift[0];
1400     facel            = _mm256_set1_pd(fr->epsfac);
1401     charge           = mdatoms->chargeA;
1402     nvdwtype         = fr->ntype;
1403     vdwparam         = fr->nbfp;
1404     vdwtype          = mdatoms->typeA;
1405     vdwgridparam     = fr->ljpme_c6grid;
1406     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
1407     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
1408     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
1409
1410     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1411     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1412     beta2            = _mm256_mul_pd(beta,beta);
1413     beta3            = _mm256_mul_pd(beta,beta2);
1414
1415     ewtab            = fr->ic->tabq_coul_F;
1416     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1417     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1418
1419     /* Setup water-specific parameters */
1420     inr              = nlist->iinr[0];
1421     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1422     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1423     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1424     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1425     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
1426
1427     jq0              = _mm256_set1_pd(charge[inr+0]);
1428     jq1              = _mm256_set1_pd(charge[inr+1]);
1429     jq2              = _mm256_set1_pd(charge[inr+2]);
1430     vdwjidx0A        = 2*vdwtype[inr+0];
1431     qq00             = _mm256_mul_pd(iq0,jq0);
1432     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1433     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1434     c6grid_00        = _mm256_set1_pd(vdwgridioffsetptr0[vdwjidx0A]);
1435     qq01             = _mm256_mul_pd(iq0,jq1);
1436     qq02             = _mm256_mul_pd(iq0,jq2);
1437     qq10             = _mm256_mul_pd(iq1,jq0);
1438     qq11             = _mm256_mul_pd(iq1,jq1);
1439     qq12             = _mm256_mul_pd(iq1,jq2);
1440     qq20             = _mm256_mul_pd(iq2,jq0);
1441     qq21             = _mm256_mul_pd(iq2,jq1);
1442     qq22             = _mm256_mul_pd(iq2,jq2);
1443
1444     /* Avoid stupid compiler warnings */
1445     jnrA = jnrB = jnrC = jnrD = 0;
1446     j_coord_offsetA = 0;
1447     j_coord_offsetB = 0;
1448     j_coord_offsetC = 0;
1449     j_coord_offsetD = 0;
1450
1451     outeriter        = 0;
1452     inneriter        = 0;
1453
1454     for(iidx=0;iidx<4*DIM;iidx++)
1455     {
1456         scratch[iidx] = 0.0;
1457     }
1458
1459     /* Start outer loop over neighborlists */
1460     for(iidx=0; iidx<nri; iidx++)
1461     {
1462         /* Load shift vector for this list */
1463         i_shift_offset   = DIM*shiftidx[iidx];
1464
1465         /* Load limits for loop over neighbors */
1466         j_index_start    = jindex[iidx];
1467         j_index_end      = jindex[iidx+1];
1468
1469         /* Get outer coordinate index */
1470         inr              = iinr[iidx];
1471         i_coord_offset   = DIM*inr;
1472
1473         /* Load i particle coords and add shift vector */
1474         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1475                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1476
1477         fix0             = _mm256_setzero_pd();
1478         fiy0             = _mm256_setzero_pd();
1479         fiz0             = _mm256_setzero_pd();
1480         fix1             = _mm256_setzero_pd();
1481         fiy1             = _mm256_setzero_pd();
1482         fiz1             = _mm256_setzero_pd();
1483         fix2             = _mm256_setzero_pd();
1484         fiy2             = _mm256_setzero_pd();
1485         fiz2             = _mm256_setzero_pd();
1486
1487         /* Start inner kernel loop */
1488         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1489         {
1490
1491             /* Get j neighbor index, and coordinate index */
1492             jnrA             = jjnr[jidx];
1493             jnrB             = jjnr[jidx+1];
1494             jnrC             = jjnr[jidx+2];
1495             jnrD             = jjnr[jidx+3];
1496             j_coord_offsetA  = DIM*jnrA;
1497             j_coord_offsetB  = DIM*jnrB;
1498             j_coord_offsetC  = DIM*jnrC;
1499             j_coord_offsetD  = DIM*jnrD;
1500
1501             /* load j atom coordinates */
1502             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1503                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1504                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1505
1506             /* Calculate displacement vector */
1507             dx00             = _mm256_sub_pd(ix0,jx0);
1508             dy00             = _mm256_sub_pd(iy0,jy0);
1509             dz00             = _mm256_sub_pd(iz0,jz0);
1510             dx01             = _mm256_sub_pd(ix0,jx1);
1511             dy01             = _mm256_sub_pd(iy0,jy1);
1512             dz01             = _mm256_sub_pd(iz0,jz1);
1513             dx02             = _mm256_sub_pd(ix0,jx2);
1514             dy02             = _mm256_sub_pd(iy0,jy2);
1515             dz02             = _mm256_sub_pd(iz0,jz2);
1516             dx10             = _mm256_sub_pd(ix1,jx0);
1517             dy10             = _mm256_sub_pd(iy1,jy0);
1518             dz10             = _mm256_sub_pd(iz1,jz0);
1519             dx11             = _mm256_sub_pd(ix1,jx1);
1520             dy11             = _mm256_sub_pd(iy1,jy1);
1521             dz11             = _mm256_sub_pd(iz1,jz1);
1522             dx12             = _mm256_sub_pd(ix1,jx2);
1523             dy12             = _mm256_sub_pd(iy1,jy2);
1524             dz12             = _mm256_sub_pd(iz1,jz2);
1525             dx20             = _mm256_sub_pd(ix2,jx0);
1526             dy20             = _mm256_sub_pd(iy2,jy0);
1527             dz20             = _mm256_sub_pd(iz2,jz0);
1528             dx21             = _mm256_sub_pd(ix2,jx1);
1529             dy21             = _mm256_sub_pd(iy2,jy1);
1530             dz21             = _mm256_sub_pd(iz2,jz1);
1531             dx22             = _mm256_sub_pd(ix2,jx2);
1532             dy22             = _mm256_sub_pd(iy2,jy2);
1533             dz22             = _mm256_sub_pd(iz2,jz2);
1534
1535             /* Calculate squared distance and things based on it */
1536             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1537             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1538             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1539             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1540             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1541             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1542             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1543             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1544             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1545
1546             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1547             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1548             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1549             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1550             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1551             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1552             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1553             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1554             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1555
1556             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1557             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1558             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1559             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1560             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1561             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1562             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1563             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1564             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1565
1566             fjx0             = _mm256_setzero_pd();
1567             fjy0             = _mm256_setzero_pd();
1568             fjz0             = _mm256_setzero_pd();
1569             fjx1             = _mm256_setzero_pd();
1570             fjy1             = _mm256_setzero_pd();
1571             fjz1             = _mm256_setzero_pd();
1572             fjx2             = _mm256_setzero_pd();
1573             fjy2             = _mm256_setzero_pd();
1574             fjz2             = _mm256_setzero_pd();
1575
1576             /**************************
1577              * CALCULATE INTERACTIONS *
1578              **************************/
1579
1580             r00              = _mm256_mul_pd(rsq00,rinv00);
1581
1582             /* EWALD ELECTROSTATICS */
1583
1584             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1585             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1586             ewitab           = _mm256_cvttpd_epi32(ewrt);
1587             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1588             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1589                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1590                                             &ewtabF,&ewtabFn);
1591             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1592             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1593
1594             /* Analytical LJ-PME */
1595             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1596             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1597             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1598             exponent         = gmx_simd_exp_d(ewcljrsq);
1599             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1600             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1601             /* f6A = 6 * C6grid * (1 - poly) */
1602             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1603             /* f6B = C6grid * exponent * beta^6 */
1604             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1605             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1606             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1607
1608             fscal            = _mm256_add_pd(felec,fvdw);
1609
1610             /* Calculate temporary vectorial force */
1611             tx               = _mm256_mul_pd(fscal,dx00);
1612             ty               = _mm256_mul_pd(fscal,dy00);
1613             tz               = _mm256_mul_pd(fscal,dz00);
1614
1615             /* Update vectorial force */
1616             fix0             = _mm256_add_pd(fix0,tx);
1617             fiy0             = _mm256_add_pd(fiy0,ty);
1618             fiz0             = _mm256_add_pd(fiz0,tz);
1619
1620             fjx0             = _mm256_add_pd(fjx0,tx);
1621             fjy0             = _mm256_add_pd(fjy0,ty);
1622             fjz0             = _mm256_add_pd(fjz0,tz);
1623
1624             /**************************
1625              * CALCULATE INTERACTIONS *
1626              **************************/
1627
1628             r01              = _mm256_mul_pd(rsq01,rinv01);
1629
1630             /* EWALD ELECTROSTATICS */
1631
1632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1633             ewrt             = _mm256_mul_pd(r01,ewtabscale);
1634             ewitab           = _mm256_cvttpd_epi32(ewrt);
1635             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1636             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1637                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1638                                             &ewtabF,&ewtabFn);
1639             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1640             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1641
1642             fscal            = felec;
1643
1644             /* Calculate temporary vectorial force */
1645             tx               = _mm256_mul_pd(fscal,dx01);
1646             ty               = _mm256_mul_pd(fscal,dy01);
1647             tz               = _mm256_mul_pd(fscal,dz01);
1648
1649             /* Update vectorial force */
1650             fix0             = _mm256_add_pd(fix0,tx);
1651             fiy0             = _mm256_add_pd(fiy0,ty);
1652             fiz0             = _mm256_add_pd(fiz0,tz);
1653
1654             fjx1             = _mm256_add_pd(fjx1,tx);
1655             fjy1             = _mm256_add_pd(fjy1,ty);
1656             fjz1             = _mm256_add_pd(fjz1,tz);
1657
1658             /**************************
1659              * CALCULATE INTERACTIONS *
1660              **************************/
1661
1662             r02              = _mm256_mul_pd(rsq02,rinv02);
1663
1664             /* EWALD ELECTROSTATICS */
1665
1666             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1667             ewrt             = _mm256_mul_pd(r02,ewtabscale);
1668             ewitab           = _mm256_cvttpd_epi32(ewrt);
1669             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1670             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1671                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1672                                             &ewtabF,&ewtabFn);
1673             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1674             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1675
1676             fscal            = felec;
1677
1678             /* Calculate temporary vectorial force */
1679             tx               = _mm256_mul_pd(fscal,dx02);
1680             ty               = _mm256_mul_pd(fscal,dy02);
1681             tz               = _mm256_mul_pd(fscal,dz02);
1682
1683             /* Update vectorial force */
1684             fix0             = _mm256_add_pd(fix0,tx);
1685             fiy0             = _mm256_add_pd(fiy0,ty);
1686             fiz0             = _mm256_add_pd(fiz0,tz);
1687
1688             fjx2             = _mm256_add_pd(fjx2,tx);
1689             fjy2             = _mm256_add_pd(fjy2,ty);
1690             fjz2             = _mm256_add_pd(fjz2,tz);
1691
1692             /**************************
1693              * CALCULATE INTERACTIONS *
1694              **************************/
1695
1696             r10              = _mm256_mul_pd(rsq10,rinv10);
1697
1698             /* EWALD ELECTROSTATICS */
1699
1700             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1701             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1702             ewitab           = _mm256_cvttpd_epi32(ewrt);
1703             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1704             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1705                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1706                                             &ewtabF,&ewtabFn);
1707             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1708             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1709
1710             fscal            = felec;
1711
1712             /* Calculate temporary vectorial force */
1713             tx               = _mm256_mul_pd(fscal,dx10);
1714             ty               = _mm256_mul_pd(fscal,dy10);
1715             tz               = _mm256_mul_pd(fscal,dz10);
1716
1717             /* Update vectorial force */
1718             fix1             = _mm256_add_pd(fix1,tx);
1719             fiy1             = _mm256_add_pd(fiy1,ty);
1720             fiz1             = _mm256_add_pd(fiz1,tz);
1721
1722             fjx0             = _mm256_add_pd(fjx0,tx);
1723             fjy0             = _mm256_add_pd(fjy0,ty);
1724             fjz0             = _mm256_add_pd(fjz0,tz);
1725
1726             /**************************
1727              * CALCULATE INTERACTIONS *
1728              **************************/
1729
1730             r11              = _mm256_mul_pd(rsq11,rinv11);
1731
1732             /* EWALD ELECTROSTATICS */
1733
1734             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1735             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1736             ewitab           = _mm256_cvttpd_epi32(ewrt);
1737             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1738             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1739                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1740                                             &ewtabF,&ewtabFn);
1741             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1742             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1743
1744             fscal            = felec;
1745
1746             /* Calculate temporary vectorial force */
1747             tx               = _mm256_mul_pd(fscal,dx11);
1748             ty               = _mm256_mul_pd(fscal,dy11);
1749             tz               = _mm256_mul_pd(fscal,dz11);
1750
1751             /* Update vectorial force */
1752             fix1             = _mm256_add_pd(fix1,tx);
1753             fiy1             = _mm256_add_pd(fiy1,ty);
1754             fiz1             = _mm256_add_pd(fiz1,tz);
1755
1756             fjx1             = _mm256_add_pd(fjx1,tx);
1757             fjy1             = _mm256_add_pd(fjy1,ty);
1758             fjz1             = _mm256_add_pd(fjz1,tz);
1759
1760             /**************************
1761              * CALCULATE INTERACTIONS *
1762              **************************/
1763
1764             r12              = _mm256_mul_pd(rsq12,rinv12);
1765
1766             /* EWALD ELECTROSTATICS */
1767
1768             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1769             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1770             ewitab           = _mm256_cvttpd_epi32(ewrt);
1771             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1772             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1773                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1774                                             &ewtabF,&ewtabFn);
1775             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1776             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1777
1778             fscal            = felec;
1779
1780             /* Calculate temporary vectorial force */
1781             tx               = _mm256_mul_pd(fscal,dx12);
1782             ty               = _mm256_mul_pd(fscal,dy12);
1783             tz               = _mm256_mul_pd(fscal,dz12);
1784
1785             /* Update vectorial force */
1786             fix1             = _mm256_add_pd(fix1,tx);
1787             fiy1             = _mm256_add_pd(fiy1,ty);
1788             fiz1             = _mm256_add_pd(fiz1,tz);
1789
1790             fjx2             = _mm256_add_pd(fjx2,tx);
1791             fjy2             = _mm256_add_pd(fjy2,ty);
1792             fjz2             = _mm256_add_pd(fjz2,tz);
1793
1794             /**************************
1795              * CALCULATE INTERACTIONS *
1796              **************************/
1797
1798             r20              = _mm256_mul_pd(rsq20,rinv20);
1799
1800             /* EWALD ELECTROSTATICS */
1801
1802             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1803             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1804             ewitab           = _mm256_cvttpd_epi32(ewrt);
1805             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1806             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1807                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1808                                             &ewtabF,&ewtabFn);
1809             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1810             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1811
1812             fscal            = felec;
1813
1814             /* Calculate temporary vectorial force */
1815             tx               = _mm256_mul_pd(fscal,dx20);
1816             ty               = _mm256_mul_pd(fscal,dy20);
1817             tz               = _mm256_mul_pd(fscal,dz20);
1818
1819             /* Update vectorial force */
1820             fix2             = _mm256_add_pd(fix2,tx);
1821             fiy2             = _mm256_add_pd(fiy2,ty);
1822             fiz2             = _mm256_add_pd(fiz2,tz);
1823
1824             fjx0             = _mm256_add_pd(fjx0,tx);
1825             fjy0             = _mm256_add_pd(fjy0,ty);
1826             fjz0             = _mm256_add_pd(fjz0,tz);
1827
1828             /**************************
1829              * CALCULATE INTERACTIONS *
1830              **************************/
1831
1832             r21              = _mm256_mul_pd(rsq21,rinv21);
1833
1834             /* EWALD ELECTROSTATICS */
1835
1836             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1837             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1838             ewitab           = _mm256_cvttpd_epi32(ewrt);
1839             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1840             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1841                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1842                                             &ewtabF,&ewtabFn);
1843             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1844             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1845
1846             fscal            = felec;
1847
1848             /* Calculate temporary vectorial force */
1849             tx               = _mm256_mul_pd(fscal,dx21);
1850             ty               = _mm256_mul_pd(fscal,dy21);
1851             tz               = _mm256_mul_pd(fscal,dz21);
1852
1853             /* Update vectorial force */
1854             fix2             = _mm256_add_pd(fix2,tx);
1855             fiy2             = _mm256_add_pd(fiy2,ty);
1856             fiz2             = _mm256_add_pd(fiz2,tz);
1857
1858             fjx1             = _mm256_add_pd(fjx1,tx);
1859             fjy1             = _mm256_add_pd(fjy1,ty);
1860             fjz1             = _mm256_add_pd(fjz1,tz);
1861
1862             /**************************
1863              * CALCULATE INTERACTIONS *
1864              **************************/
1865
1866             r22              = _mm256_mul_pd(rsq22,rinv22);
1867
1868             /* EWALD ELECTROSTATICS */
1869
1870             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1871             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1872             ewitab           = _mm256_cvttpd_epi32(ewrt);
1873             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1874             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1875                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1876                                             &ewtabF,&ewtabFn);
1877             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1878             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1879
1880             fscal            = felec;
1881
1882             /* Calculate temporary vectorial force */
1883             tx               = _mm256_mul_pd(fscal,dx22);
1884             ty               = _mm256_mul_pd(fscal,dy22);
1885             tz               = _mm256_mul_pd(fscal,dz22);
1886
1887             /* Update vectorial force */
1888             fix2             = _mm256_add_pd(fix2,tx);
1889             fiy2             = _mm256_add_pd(fiy2,ty);
1890             fiz2             = _mm256_add_pd(fiz2,tz);
1891
1892             fjx2             = _mm256_add_pd(fjx2,tx);
1893             fjy2             = _mm256_add_pd(fjy2,ty);
1894             fjz2             = _mm256_add_pd(fjz2,tz);
1895
1896             fjptrA             = f+j_coord_offsetA;
1897             fjptrB             = f+j_coord_offsetB;
1898             fjptrC             = f+j_coord_offsetC;
1899             fjptrD             = f+j_coord_offsetD;
1900
1901             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1902                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1903
1904             /* Inner loop uses 347 flops */
1905         }
1906
1907         if(jidx<j_index_end)
1908         {
1909
1910             /* Get j neighbor index, and coordinate index */
1911             jnrlistA         = jjnr[jidx];
1912             jnrlistB         = jjnr[jidx+1];
1913             jnrlistC         = jjnr[jidx+2];
1914             jnrlistD         = jjnr[jidx+3];
1915             /* Sign of each element will be negative for non-real atoms.
1916              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1917              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1918              */
1919             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1920
1921             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1922             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1923             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1924
1925             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1926             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1927             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1928             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1929             j_coord_offsetA  = DIM*jnrA;
1930             j_coord_offsetB  = DIM*jnrB;
1931             j_coord_offsetC  = DIM*jnrC;
1932             j_coord_offsetD  = DIM*jnrD;
1933
1934             /* load j atom coordinates */
1935             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1936                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1937                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1938
1939             /* Calculate displacement vector */
1940             dx00             = _mm256_sub_pd(ix0,jx0);
1941             dy00             = _mm256_sub_pd(iy0,jy0);
1942             dz00             = _mm256_sub_pd(iz0,jz0);
1943             dx01             = _mm256_sub_pd(ix0,jx1);
1944             dy01             = _mm256_sub_pd(iy0,jy1);
1945             dz01             = _mm256_sub_pd(iz0,jz1);
1946             dx02             = _mm256_sub_pd(ix0,jx2);
1947             dy02             = _mm256_sub_pd(iy0,jy2);
1948             dz02             = _mm256_sub_pd(iz0,jz2);
1949             dx10             = _mm256_sub_pd(ix1,jx0);
1950             dy10             = _mm256_sub_pd(iy1,jy0);
1951             dz10             = _mm256_sub_pd(iz1,jz0);
1952             dx11             = _mm256_sub_pd(ix1,jx1);
1953             dy11             = _mm256_sub_pd(iy1,jy1);
1954             dz11             = _mm256_sub_pd(iz1,jz1);
1955             dx12             = _mm256_sub_pd(ix1,jx2);
1956             dy12             = _mm256_sub_pd(iy1,jy2);
1957             dz12             = _mm256_sub_pd(iz1,jz2);
1958             dx20             = _mm256_sub_pd(ix2,jx0);
1959             dy20             = _mm256_sub_pd(iy2,jy0);
1960             dz20             = _mm256_sub_pd(iz2,jz0);
1961             dx21             = _mm256_sub_pd(ix2,jx1);
1962             dy21             = _mm256_sub_pd(iy2,jy1);
1963             dz21             = _mm256_sub_pd(iz2,jz1);
1964             dx22             = _mm256_sub_pd(ix2,jx2);
1965             dy22             = _mm256_sub_pd(iy2,jy2);
1966             dz22             = _mm256_sub_pd(iz2,jz2);
1967
1968             /* Calculate squared distance and things based on it */
1969             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1970             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1971             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1972             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1973             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1974             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1975             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1976             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1977             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1978
1979             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1980             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1981             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1982             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1983             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1984             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1985             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1986             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1987             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1988
1989             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1990             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1991             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1992             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1993             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1994             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1995             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1996             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1997             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1998
1999             fjx0             = _mm256_setzero_pd();
2000             fjy0             = _mm256_setzero_pd();
2001             fjz0             = _mm256_setzero_pd();
2002             fjx1             = _mm256_setzero_pd();
2003             fjy1             = _mm256_setzero_pd();
2004             fjz1             = _mm256_setzero_pd();
2005             fjx2             = _mm256_setzero_pd();
2006             fjy2             = _mm256_setzero_pd();
2007             fjz2             = _mm256_setzero_pd();
2008
2009             /**************************
2010              * CALCULATE INTERACTIONS *
2011              **************************/
2012
2013             r00              = _mm256_mul_pd(rsq00,rinv00);
2014             r00              = _mm256_andnot_pd(dummy_mask,r00);
2015
2016             /* EWALD ELECTROSTATICS */
2017
2018             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2019             ewrt             = _mm256_mul_pd(r00,ewtabscale);
2020             ewitab           = _mm256_cvttpd_epi32(ewrt);
2021             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2022             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2023                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2024                                             &ewtabF,&ewtabFn);
2025             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2026             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
2027
2028             /* Analytical LJ-PME */
2029             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2030             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
2031             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
2032             exponent         = gmx_simd_exp_d(ewcljrsq);
2033             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2034             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
2035             /* f6A = 6 * C6grid * (1 - poly) */
2036             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
2037             /* f6B = C6grid * exponent * beta^6 */
2038             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
2039             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2040             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
2041
2042             fscal            = _mm256_add_pd(felec,fvdw);
2043
2044             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2045
2046             /* Calculate temporary vectorial force */
2047             tx               = _mm256_mul_pd(fscal,dx00);
2048             ty               = _mm256_mul_pd(fscal,dy00);
2049             tz               = _mm256_mul_pd(fscal,dz00);
2050
2051             /* Update vectorial force */
2052             fix0             = _mm256_add_pd(fix0,tx);
2053             fiy0             = _mm256_add_pd(fiy0,ty);
2054             fiz0             = _mm256_add_pd(fiz0,tz);
2055
2056             fjx0             = _mm256_add_pd(fjx0,tx);
2057             fjy0             = _mm256_add_pd(fjy0,ty);
2058             fjz0             = _mm256_add_pd(fjz0,tz);
2059
2060             /**************************
2061              * CALCULATE INTERACTIONS *
2062              **************************/
2063
2064             r01              = _mm256_mul_pd(rsq01,rinv01);
2065             r01              = _mm256_andnot_pd(dummy_mask,r01);
2066
2067             /* EWALD ELECTROSTATICS */
2068
2069             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2070             ewrt             = _mm256_mul_pd(r01,ewtabscale);
2071             ewitab           = _mm256_cvttpd_epi32(ewrt);
2072             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2073             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2074                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2075                                             &ewtabF,&ewtabFn);
2076             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2077             felec            = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2082
2083             /* Calculate temporary vectorial force */
2084             tx               = _mm256_mul_pd(fscal,dx01);
2085             ty               = _mm256_mul_pd(fscal,dy01);
2086             tz               = _mm256_mul_pd(fscal,dz01);
2087
2088             /* Update vectorial force */
2089             fix0             = _mm256_add_pd(fix0,tx);
2090             fiy0             = _mm256_add_pd(fiy0,ty);
2091             fiz0             = _mm256_add_pd(fiz0,tz);
2092
2093             fjx1             = _mm256_add_pd(fjx1,tx);
2094             fjy1             = _mm256_add_pd(fjy1,ty);
2095             fjz1             = _mm256_add_pd(fjz1,tz);
2096
2097             /**************************
2098              * CALCULATE INTERACTIONS *
2099              **************************/
2100
2101             r02              = _mm256_mul_pd(rsq02,rinv02);
2102             r02              = _mm256_andnot_pd(dummy_mask,r02);
2103
2104             /* EWALD ELECTROSTATICS */
2105
2106             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2107             ewrt             = _mm256_mul_pd(r02,ewtabscale);
2108             ewitab           = _mm256_cvttpd_epi32(ewrt);
2109             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2110             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2111                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2112                                             &ewtabF,&ewtabFn);
2113             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2114             felec            = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
2115
2116             fscal            = felec;
2117
2118             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2119
2120             /* Calculate temporary vectorial force */
2121             tx               = _mm256_mul_pd(fscal,dx02);
2122             ty               = _mm256_mul_pd(fscal,dy02);
2123             tz               = _mm256_mul_pd(fscal,dz02);
2124
2125             /* Update vectorial force */
2126             fix0             = _mm256_add_pd(fix0,tx);
2127             fiy0             = _mm256_add_pd(fiy0,ty);
2128             fiz0             = _mm256_add_pd(fiz0,tz);
2129
2130             fjx2             = _mm256_add_pd(fjx2,tx);
2131             fjy2             = _mm256_add_pd(fjy2,ty);
2132             fjz2             = _mm256_add_pd(fjz2,tz);
2133
2134             /**************************
2135              * CALCULATE INTERACTIONS *
2136              **************************/
2137
2138             r10              = _mm256_mul_pd(rsq10,rinv10);
2139             r10              = _mm256_andnot_pd(dummy_mask,r10);
2140
2141             /* EWALD ELECTROSTATICS */
2142
2143             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2144             ewrt             = _mm256_mul_pd(r10,ewtabscale);
2145             ewitab           = _mm256_cvttpd_epi32(ewrt);
2146             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2147             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2148                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2149                                             &ewtabF,&ewtabFn);
2150             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2151             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
2152
2153             fscal            = felec;
2154
2155             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2156
2157             /* Calculate temporary vectorial force */
2158             tx               = _mm256_mul_pd(fscal,dx10);
2159             ty               = _mm256_mul_pd(fscal,dy10);
2160             tz               = _mm256_mul_pd(fscal,dz10);
2161
2162             /* Update vectorial force */
2163             fix1             = _mm256_add_pd(fix1,tx);
2164             fiy1             = _mm256_add_pd(fiy1,ty);
2165             fiz1             = _mm256_add_pd(fiz1,tz);
2166
2167             fjx0             = _mm256_add_pd(fjx0,tx);
2168             fjy0             = _mm256_add_pd(fjy0,ty);
2169             fjz0             = _mm256_add_pd(fjz0,tz);
2170
2171             /**************************
2172              * CALCULATE INTERACTIONS *
2173              **************************/
2174
2175             r11              = _mm256_mul_pd(rsq11,rinv11);
2176             r11              = _mm256_andnot_pd(dummy_mask,r11);
2177
2178             /* EWALD ELECTROSTATICS */
2179
2180             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2181             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2182             ewitab           = _mm256_cvttpd_epi32(ewrt);
2183             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2184             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2185                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2186                                             &ewtabF,&ewtabFn);
2187             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2188             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2189
2190             fscal            = felec;
2191
2192             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2193
2194             /* Calculate temporary vectorial force */
2195             tx               = _mm256_mul_pd(fscal,dx11);
2196             ty               = _mm256_mul_pd(fscal,dy11);
2197             tz               = _mm256_mul_pd(fscal,dz11);
2198
2199             /* Update vectorial force */
2200             fix1             = _mm256_add_pd(fix1,tx);
2201             fiy1             = _mm256_add_pd(fiy1,ty);
2202             fiz1             = _mm256_add_pd(fiz1,tz);
2203
2204             fjx1             = _mm256_add_pd(fjx1,tx);
2205             fjy1             = _mm256_add_pd(fjy1,ty);
2206             fjz1             = _mm256_add_pd(fjz1,tz);
2207
2208             /**************************
2209              * CALCULATE INTERACTIONS *
2210              **************************/
2211
2212             r12              = _mm256_mul_pd(rsq12,rinv12);
2213             r12              = _mm256_andnot_pd(dummy_mask,r12);
2214
2215             /* EWALD ELECTROSTATICS */
2216
2217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2218             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2219             ewitab           = _mm256_cvttpd_epi32(ewrt);
2220             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2221             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2222                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2223                                             &ewtabF,&ewtabFn);
2224             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2225             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2226
2227             fscal            = felec;
2228
2229             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2230
2231             /* Calculate temporary vectorial force */
2232             tx               = _mm256_mul_pd(fscal,dx12);
2233             ty               = _mm256_mul_pd(fscal,dy12);
2234             tz               = _mm256_mul_pd(fscal,dz12);
2235
2236             /* Update vectorial force */
2237             fix1             = _mm256_add_pd(fix1,tx);
2238             fiy1             = _mm256_add_pd(fiy1,ty);
2239             fiz1             = _mm256_add_pd(fiz1,tz);
2240
2241             fjx2             = _mm256_add_pd(fjx2,tx);
2242             fjy2             = _mm256_add_pd(fjy2,ty);
2243             fjz2             = _mm256_add_pd(fjz2,tz);
2244
2245             /**************************
2246              * CALCULATE INTERACTIONS *
2247              **************************/
2248
2249             r20              = _mm256_mul_pd(rsq20,rinv20);
2250             r20              = _mm256_andnot_pd(dummy_mask,r20);
2251
2252             /* EWALD ELECTROSTATICS */
2253
2254             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2255             ewrt             = _mm256_mul_pd(r20,ewtabscale);
2256             ewitab           = _mm256_cvttpd_epi32(ewrt);
2257             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2258             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2259                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2260                                             &ewtabF,&ewtabFn);
2261             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2262             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
2263
2264             fscal            = felec;
2265
2266             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2267
2268             /* Calculate temporary vectorial force */
2269             tx               = _mm256_mul_pd(fscal,dx20);
2270             ty               = _mm256_mul_pd(fscal,dy20);
2271             tz               = _mm256_mul_pd(fscal,dz20);
2272
2273             /* Update vectorial force */
2274             fix2             = _mm256_add_pd(fix2,tx);
2275             fiy2             = _mm256_add_pd(fiy2,ty);
2276             fiz2             = _mm256_add_pd(fiz2,tz);
2277
2278             fjx0             = _mm256_add_pd(fjx0,tx);
2279             fjy0             = _mm256_add_pd(fjy0,ty);
2280             fjz0             = _mm256_add_pd(fjz0,tz);
2281
2282             /**************************
2283              * CALCULATE INTERACTIONS *
2284              **************************/
2285
2286             r21              = _mm256_mul_pd(rsq21,rinv21);
2287             r21              = _mm256_andnot_pd(dummy_mask,r21);
2288
2289             /* EWALD ELECTROSTATICS */
2290
2291             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2292             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2293             ewitab           = _mm256_cvttpd_epi32(ewrt);
2294             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2295             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2296                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2297                                             &ewtabF,&ewtabFn);
2298             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2299             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2300
2301             fscal            = felec;
2302
2303             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2304
2305             /* Calculate temporary vectorial force */
2306             tx               = _mm256_mul_pd(fscal,dx21);
2307             ty               = _mm256_mul_pd(fscal,dy21);
2308             tz               = _mm256_mul_pd(fscal,dz21);
2309
2310             /* Update vectorial force */
2311             fix2             = _mm256_add_pd(fix2,tx);
2312             fiy2             = _mm256_add_pd(fiy2,ty);
2313             fiz2             = _mm256_add_pd(fiz2,tz);
2314
2315             fjx1             = _mm256_add_pd(fjx1,tx);
2316             fjy1             = _mm256_add_pd(fjy1,ty);
2317             fjz1             = _mm256_add_pd(fjz1,tz);
2318
2319             /**************************
2320              * CALCULATE INTERACTIONS *
2321              **************************/
2322
2323             r22              = _mm256_mul_pd(rsq22,rinv22);
2324             r22              = _mm256_andnot_pd(dummy_mask,r22);
2325
2326             /* EWALD ELECTROSTATICS */
2327
2328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2329             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2330             ewitab           = _mm256_cvttpd_epi32(ewrt);
2331             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2332             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2333                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2334                                             &ewtabF,&ewtabFn);
2335             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2336             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2337
2338             fscal            = felec;
2339
2340             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2341
2342             /* Calculate temporary vectorial force */
2343             tx               = _mm256_mul_pd(fscal,dx22);
2344             ty               = _mm256_mul_pd(fscal,dy22);
2345             tz               = _mm256_mul_pd(fscal,dz22);
2346
2347             /* Update vectorial force */
2348             fix2             = _mm256_add_pd(fix2,tx);
2349             fiy2             = _mm256_add_pd(fiy2,ty);
2350             fiz2             = _mm256_add_pd(fiz2,tz);
2351
2352             fjx2             = _mm256_add_pd(fjx2,tx);
2353             fjy2             = _mm256_add_pd(fjy2,ty);
2354             fjz2             = _mm256_add_pd(fjz2,tz);
2355
2356             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2357             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2358             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2359             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2360
2361             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2362                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2363
2364             /* Inner loop uses 356 flops */
2365         }
2366
2367         /* End of innermost loop */
2368
2369         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2370                                                  f+i_coord_offset,fshift+i_shift_offset);
2371
2372         /* Increment number of inner iterations */
2373         inneriter                  += j_index_end - j_index_start;
2374
2375         /* Outer loop uses 18 flops */
2376     }
2377
2378     /* Increment number of outer iterations */
2379     outeriter        += nri;
2380
2381     /* Update outer/inner flops */
2382
2383     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*356);
2384 }