src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_256_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 111     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 112     __m128i          vfitab;
 113     __m128i          ifour       = _mm_set1_epi32(4);
 114     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 115     real             *vftab;
 116     __m256d          dummy_mask,cutoff_mask;
 117     __m128           tmpmask0,tmpmask1;
 118     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 119     __m256d          one     = _mm256_set1_pd(1.0);
 120     __m256d          two     = _mm256_set1_pd(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_pd(fr->ic->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137
 138     vftab            = kernel_data->table_vdw->data;
 139     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 144     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 145     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 146     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 147
 148     jq0              = _mm256_set1_pd(charge[inr+0]);
 149     jq1              = _mm256_set1_pd(charge[inr+1]);
 150     jq2              = _mm256_set1_pd(charge[inr+2]);
 151     vdwjidx0A        = 2*vdwtype[inr+0];
 152     qq00             = _mm256_mul_pd(iq0,jq0);
 153     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 154     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 155     qq01             = _mm256_mul_pd(iq0,jq1);
 156     qq02             = _mm256_mul_pd(iq0,jq2);
 157     qq10             = _mm256_mul_pd(iq1,jq0);
 158     qq11             = _mm256_mul_pd(iq1,jq1);
 159     qq12             = _mm256_mul_pd(iq1,jq2);
 160     qq20             = _mm256_mul_pd(iq2,jq0);
 161     qq21             = _mm256_mul_pd(iq2,jq1);
 162     qq22             = _mm256_mul_pd(iq2,jq2);
 163
 164     /* Avoid stupid compiler warnings */
 165     jnrA = jnrB = jnrC = jnrD = 0;
 166     j_coord_offsetA = 0;
 167     j_coord_offsetB = 0;
 168     j_coord_offsetC = 0;
 169     j_coord_offsetD = 0;
 170
 171     outeriter        = 0;
 172     inneriter        = 0;
 173
 174     for(iidx=0;iidx<4*DIM;iidx++)
 175     {
 176         scratch[iidx] = 0.0;
 177     }
 178
 179     /* Start outer loop over neighborlists */
 180     for(iidx=0; iidx<nri; iidx++)
 181     {
 182         /* Load shift vector for this list */
 183         i_shift_offset   = DIM*shiftidx[iidx];
 184
 185         /* Load limits for loop over neighbors */
 186         j_index_start    = jindex[iidx];
 187         j_index_end      = jindex[iidx+1];
 188
 189         /* Get outer coordinate index */
 190         inr              = iinr[iidx];
 191         i_coord_offset   = DIM*inr;
 192
 193         /* Load i particle coords and add shift vector */
 194         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 195                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 196
 197         fix0             = _mm256_setzero_pd();
 198         fiy0             = _mm256_setzero_pd();
 199         fiz0             = _mm256_setzero_pd();
 200         fix1             = _mm256_setzero_pd();
 201         fiy1             = _mm256_setzero_pd();
 202         fiz1             = _mm256_setzero_pd();
 203         fix2             = _mm256_setzero_pd();
 204         fiy2             = _mm256_setzero_pd();
 205         fiz2             = _mm256_setzero_pd();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm256_setzero_pd();
 209         vvdwsum          = _mm256_setzero_pd();
 210
 211         /* Start inner kernel loop */
 212         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 213         {
 214
 215             /* Get j neighbor index, and coordinate index */
 216             jnrA             = jjnr[jidx];
 217             jnrB             = jjnr[jidx+1];
 218             jnrC             = jjnr[jidx+2];
 219             jnrD             = jjnr[jidx+3];
 220             j_coord_offsetA  = DIM*jnrA;
 221             j_coord_offsetB  = DIM*jnrB;
 222             j_coord_offsetC  = DIM*jnrC;
 223             j_coord_offsetD  = DIM*jnrD;
 224
 225             /* load j atom coordinates */
 226             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 227                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 228                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm256_sub_pd(ix0,jx0);
 232             dy00             = _mm256_sub_pd(iy0,jy0);
 233             dz00             = _mm256_sub_pd(iz0,jz0);
 234             dx01             = _mm256_sub_pd(ix0,jx1);
 235             dy01             = _mm256_sub_pd(iy0,jy1);
 236             dz01             = _mm256_sub_pd(iz0,jz1);
 237             dx02             = _mm256_sub_pd(ix0,jx2);
 238             dy02             = _mm256_sub_pd(iy0,jy2);
 239             dz02             = _mm256_sub_pd(iz0,jz2);
 240             dx10             = _mm256_sub_pd(ix1,jx0);
 241             dy10             = _mm256_sub_pd(iy1,jy0);
 242             dz10             = _mm256_sub_pd(iz1,jz0);
 243             dx11             = _mm256_sub_pd(ix1,jx1);
 244             dy11             = _mm256_sub_pd(iy1,jy1);
 245             dz11             = _mm256_sub_pd(iz1,jz1);
 246             dx12             = _mm256_sub_pd(ix1,jx2);
 247             dy12             = _mm256_sub_pd(iy1,jy2);
 248             dz12             = _mm256_sub_pd(iz1,jz2);
 249             dx20             = _mm256_sub_pd(ix2,jx0);
 250             dy20             = _mm256_sub_pd(iy2,jy0);
 251             dz20             = _mm256_sub_pd(iz2,jz0);
 252             dx21             = _mm256_sub_pd(ix2,jx1);
 253             dy21             = _mm256_sub_pd(iy2,jy1);
 254             dz21             = _mm256_sub_pd(iz2,jz1);
 255             dx22             = _mm256_sub_pd(ix2,jx2);
 256             dy22             = _mm256_sub_pd(iy2,jy2);
 257             dz22             = _mm256_sub_pd(iz2,jz2);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 261             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 262             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 263             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 264             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 265             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 266             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 267             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 268             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 269
 270             rinv00           = avx256_invsqrt_d(rsq00);
 271             rinv01           = avx256_invsqrt_d(rsq01);
 272             rinv02           = avx256_invsqrt_d(rsq02);
 273             rinv10           = avx256_invsqrt_d(rsq10);
 274             rinv11           = avx256_invsqrt_d(rsq11);
 275             rinv12           = avx256_invsqrt_d(rsq12);
 276             rinv20           = avx256_invsqrt_d(rsq20);
 277             rinv21           = avx256_invsqrt_d(rsq21);
 278             rinv22           = avx256_invsqrt_d(rsq22);
 279
 280             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 281             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 282             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 283             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 284             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 285             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 286             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 287             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 288             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 289
 290             fjx0             = _mm256_setzero_pd();
 291             fjy0             = _mm256_setzero_pd();
 292             fjz0             = _mm256_setzero_pd();
 293             fjx1             = _mm256_setzero_pd();
 294             fjy1             = _mm256_setzero_pd();
 295             fjz1             = _mm256_setzero_pd();
 296             fjx2             = _mm256_setzero_pd();
 297             fjy2             = _mm256_setzero_pd();
 298             fjz2             = _mm256_setzero_pd();
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r00              = _mm256_mul_pd(rsq00,rinv00);
 305
 306             /* Calculate table index by multiplying r with table scale and truncate to integer */
 307             rt               = _mm256_mul_pd(r00,vftabscale);
 308             vfitab           = _mm256_cvttpd_epi32(rt);
 309             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 310             vfitab           = _mm_slli_epi32(vfitab,3);
 311
 312             /* COULOMB ELECTROSTATICS */
 313             velec            = _mm256_mul_pd(qq00,rinv00);
 314             felec            = _mm256_mul_pd(velec,rinvsq00);
 315
 316             /* CUBIC SPLINE TABLE DISPERSION */
 317             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 318             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 319             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 320             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 321             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 322             Heps             = _mm256_mul_pd(vfeps,H);
 323             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 324             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 325             vvdw6            = _mm256_mul_pd(c6_00,VV);
 326             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 327             fvdw6            = _mm256_mul_pd(c6_00,FF);
 328
 329             /* CUBIC SPLINE TABLE REPULSION */
 330             vfitab           = _mm_add_epi32(vfitab,ifour);
 331             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 332             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 333             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 334             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 335             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 336             Heps             = _mm256_mul_pd(vfeps,H);
 337             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 338             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 339             vvdw12           = _mm256_mul_pd(c12_00,VV);
 340             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 341             fvdw12           = _mm256_mul_pd(c12_00,FF);
 342             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 343             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm256_add_pd(velecsum,velec);
 347             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 348
 349             fscal            = _mm256_add_pd(felec,fvdw);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx00);
 353             ty               = _mm256_mul_pd(fscal,dy00);
 354             tz               = _mm256_mul_pd(fscal,dz00);
 355
 356             /* Update vectorial force */
 357             fix0             = _mm256_add_pd(fix0,tx);
 358             fiy0             = _mm256_add_pd(fiy0,ty);
 359             fiz0             = _mm256_add_pd(fiz0,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             /* COULOMB ELECTROSTATICS */
 370             velec            = _mm256_mul_pd(qq01,rinv01);
 371             felec            = _mm256_mul_pd(velec,rinvsq01);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velecsum         = _mm256_add_pd(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm256_mul_pd(fscal,dx01);
 380             ty               = _mm256_mul_pd(fscal,dy01);
 381             tz               = _mm256_mul_pd(fscal,dz01);
 382
 383             /* Update vectorial force */
 384             fix0             = _mm256_add_pd(fix0,tx);
 385             fiy0             = _mm256_add_pd(fiy0,ty);
 386             fiz0             = _mm256_add_pd(fiz0,tz);
 387
 388             fjx1             = _mm256_add_pd(fjx1,tx);
 389             fjy1             = _mm256_add_pd(fjy1,ty);
 390             fjz1             = _mm256_add_pd(fjz1,tz);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             /* COULOMB ELECTROSTATICS */
 397             velec            = _mm256_mul_pd(qq02,rinv02);
 398             felec            = _mm256_mul_pd(velec,rinvsq02);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm256_add_pd(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx02);
 407             ty               = _mm256_mul_pd(fscal,dy02);
 408             tz               = _mm256_mul_pd(fscal,dz02);
 409
 410             /* Update vectorial force */
 411             fix0             = _mm256_add_pd(fix0,tx);
 412             fiy0             = _mm256_add_pd(fiy0,ty);
 413             fiz0             = _mm256_add_pd(fiz0,tz);
 414
 415             fjx2             = _mm256_add_pd(fjx2,tx);
 416             fjy2             = _mm256_add_pd(fjy2,ty);
 417             fjz2             = _mm256_add_pd(fjz2,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             /* COULOMB ELECTROSTATICS */
 424             velec            = _mm256_mul_pd(qq10,rinv10);
 425             felec            = _mm256_mul_pd(velec,rinvsq10);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velecsum         = _mm256_add_pd(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm256_mul_pd(fscal,dx10);
 434             ty               = _mm256_mul_pd(fscal,dy10);
 435             tz               = _mm256_mul_pd(fscal,dz10);
 436
 437             /* Update vectorial force */
 438             fix1             = _mm256_add_pd(fix1,tx);
 439             fiy1             = _mm256_add_pd(fiy1,ty);
 440             fiz1             = _mm256_add_pd(fiz1,tz);
 441
 442             fjx0             = _mm256_add_pd(fjx0,tx);
 443             fjy0             = _mm256_add_pd(fjy0,ty);
 444             fjz0             = _mm256_add_pd(fjz0,tz);
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             /* COULOMB ELECTROSTATICS */
 451             velec            = _mm256_mul_pd(qq11,rinv11);
 452             felec            = _mm256_mul_pd(velec,rinvsq11);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velecsum         = _mm256_add_pd(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             /* Calculate temporary vectorial force */
 460             tx               = _mm256_mul_pd(fscal,dx11);
 461             ty               = _mm256_mul_pd(fscal,dy11);
 462             tz               = _mm256_mul_pd(fscal,dz11);
 463
 464             /* Update vectorial force */
 465             fix1             = _mm256_add_pd(fix1,tx);
 466             fiy1             = _mm256_add_pd(fiy1,ty);
 467             fiz1             = _mm256_add_pd(fiz1,tz);
 468
 469             fjx1             = _mm256_add_pd(fjx1,tx);
 470             fjy1             = _mm256_add_pd(fjy1,ty);
 471             fjz1             = _mm256_add_pd(fjz1,tz);
 472
 473             /**************************
 474              * CALCULATE INTERACTIONS *
 475              **************************/
 476
 477             /* COULOMB ELECTROSTATICS */
 478             velec            = _mm256_mul_pd(qq12,rinv12);
 479             felec            = _mm256_mul_pd(velec,rinvsq12);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velecsum         = _mm256_add_pd(velecsum,velec);
 483
 484             fscal            = felec;
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm256_mul_pd(fscal,dx12);
 488             ty               = _mm256_mul_pd(fscal,dy12);
 489             tz               = _mm256_mul_pd(fscal,dz12);
 490
 491             /* Update vectorial force */
 492             fix1             = _mm256_add_pd(fix1,tx);
 493             fiy1             = _mm256_add_pd(fiy1,ty);
 494             fiz1             = _mm256_add_pd(fiz1,tz);
 495
 496             fjx2             = _mm256_add_pd(fjx2,tx);
 497             fjy2             = _mm256_add_pd(fjy2,ty);
 498             fjz2             = _mm256_add_pd(fjz2,tz);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* COULOMB ELECTROSTATICS */
 505             velec            = _mm256_mul_pd(qq20,rinv20);
 506             felec            = _mm256_mul_pd(velec,rinvsq20);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velecsum         = _mm256_add_pd(velecsum,velec);
 510
 511             fscal            = felec;
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm256_mul_pd(fscal,dx20);
 515             ty               = _mm256_mul_pd(fscal,dy20);
 516             tz               = _mm256_mul_pd(fscal,dz20);
 517
 518             /* Update vectorial force */
 519             fix2             = _mm256_add_pd(fix2,tx);
 520             fiy2             = _mm256_add_pd(fiy2,ty);
 521             fiz2             = _mm256_add_pd(fiz2,tz);
 522
 523             fjx0             = _mm256_add_pd(fjx0,tx);
 524             fjy0             = _mm256_add_pd(fjy0,ty);
 525             fjz0             = _mm256_add_pd(fjz0,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             /* COULOMB ELECTROSTATICS */
 532             velec            = _mm256_mul_pd(qq21,rinv21);
 533             felec            = _mm256_mul_pd(velec,rinvsq21);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velecsum         = _mm256_add_pd(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm256_mul_pd(fscal,dx21);
 542             ty               = _mm256_mul_pd(fscal,dy21);
 543             tz               = _mm256_mul_pd(fscal,dz21);
 544
 545             /* Update vectorial force */
 546             fix2             = _mm256_add_pd(fix2,tx);
 547             fiy2             = _mm256_add_pd(fiy2,ty);
 548             fiz2             = _mm256_add_pd(fiz2,tz);
 549
 550             fjx1             = _mm256_add_pd(fjx1,tx);
 551             fjy1             = _mm256_add_pd(fjy1,ty);
 552             fjz1             = _mm256_add_pd(fjz1,tz);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             /* COULOMB ELECTROSTATICS */
 559             velec            = _mm256_mul_pd(qq22,rinv22);
 560             felec            = _mm256_mul_pd(velec,rinvsq22);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velecsum         = _mm256_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm256_mul_pd(fscal,dx22);
 569             ty               = _mm256_mul_pd(fscal,dy22);
 570             tz               = _mm256_mul_pd(fscal,dz22);
 571
 572             /* Update vectorial force */
 573             fix2             = _mm256_add_pd(fix2,tx);
 574             fiy2             = _mm256_add_pd(fiy2,ty);
 575             fiz2             = _mm256_add_pd(fiz2,tz);
 576
 577             fjx2             = _mm256_add_pd(fjx2,tx);
 578             fjy2             = _mm256_add_pd(fjy2,ty);
 579             fjz2             = _mm256_add_pd(fjz2,tz);
 580
 581             fjptrA             = f+j_coord_offsetA;
 582             fjptrB             = f+j_coord_offsetB;
 583             fjptrC             = f+j_coord_offsetC;
 584             fjptrD             = f+j_coord_offsetD;
 585
 586             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 587                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 588
 589             /* Inner loop uses 278 flops */
 590         }
 591
 592         if(jidx<j_index_end)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrlistA         = jjnr[jidx];
 597             jnrlistB         = jjnr[jidx+1];
 598             jnrlistC         = jjnr[jidx+2];
 599             jnrlistD         = jjnr[jidx+3];
 600             /* Sign of each element will be negative for non-real atoms.
 601              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 602              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 603              */
 604             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 605
 606             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 607             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 608             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 609
 610             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 611             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 612             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 613             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 614             j_coord_offsetA  = DIM*jnrA;
 615             j_coord_offsetB  = DIM*jnrB;
 616             j_coord_offsetC  = DIM*jnrC;
 617             j_coord_offsetD  = DIM*jnrD;
 618
 619             /* load j atom coordinates */
 620             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 621                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 622                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 623
 624             /* Calculate displacement vector */
 625             dx00             = _mm256_sub_pd(ix0,jx0);
 626             dy00             = _mm256_sub_pd(iy0,jy0);
 627             dz00             = _mm256_sub_pd(iz0,jz0);
 628             dx01             = _mm256_sub_pd(ix0,jx1);
 629             dy01             = _mm256_sub_pd(iy0,jy1);
 630             dz01             = _mm256_sub_pd(iz0,jz1);
 631             dx02             = _mm256_sub_pd(ix0,jx2);
 632             dy02             = _mm256_sub_pd(iy0,jy2);
 633             dz02             = _mm256_sub_pd(iz0,jz2);
 634             dx10             = _mm256_sub_pd(ix1,jx0);
 635             dy10             = _mm256_sub_pd(iy1,jy0);
 636             dz10             = _mm256_sub_pd(iz1,jz0);
 637             dx11             = _mm256_sub_pd(ix1,jx1);
 638             dy11             = _mm256_sub_pd(iy1,jy1);
 639             dz11             = _mm256_sub_pd(iz1,jz1);
 640             dx12             = _mm256_sub_pd(ix1,jx2);
 641             dy12             = _mm256_sub_pd(iy1,jy2);
 642             dz12             = _mm256_sub_pd(iz1,jz2);
 643             dx20             = _mm256_sub_pd(ix2,jx0);
 644             dy20             = _mm256_sub_pd(iy2,jy0);
 645             dz20             = _mm256_sub_pd(iz2,jz0);
 646             dx21             = _mm256_sub_pd(ix2,jx1);
 647             dy21             = _mm256_sub_pd(iy2,jy1);
 648             dz21             = _mm256_sub_pd(iz2,jz1);
 649             dx22             = _mm256_sub_pd(ix2,jx2);
 650             dy22             = _mm256_sub_pd(iy2,jy2);
 651             dz22             = _mm256_sub_pd(iz2,jz2);
 652
 653             /* Calculate squared distance and things based on it */
 654             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 655             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 656             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 657             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 658             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 659             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 660             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 661             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 662             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 663
 664             rinv00           = avx256_invsqrt_d(rsq00);
 665             rinv01           = avx256_invsqrt_d(rsq01);
 666             rinv02           = avx256_invsqrt_d(rsq02);
 667             rinv10           = avx256_invsqrt_d(rsq10);
 668             rinv11           = avx256_invsqrt_d(rsq11);
 669             rinv12           = avx256_invsqrt_d(rsq12);
 670             rinv20           = avx256_invsqrt_d(rsq20);
 671             rinv21           = avx256_invsqrt_d(rsq21);
 672             rinv22           = avx256_invsqrt_d(rsq22);
 673
 674             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 675             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 676             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 677             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 678             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 679             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 680             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 681             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 682             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 683
 684             fjx0             = _mm256_setzero_pd();
 685             fjy0             = _mm256_setzero_pd();
 686             fjz0             = _mm256_setzero_pd();
 687             fjx1             = _mm256_setzero_pd();
 688             fjy1             = _mm256_setzero_pd();
 689             fjz1             = _mm256_setzero_pd();
 690             fjx2             = _mm256_setzero_pd();
 691             fjy2             = _mm256_setzero_pd();
 692             fjz2             = _mm256_setzero_pd();
 693
 694             /**************************
 695              * CALCULATE INTERACTIONS *
 696              **************************/
 697
 698             r00              = _mm256_mul_pd(rsq00,rinv00);
 699             r00              = _mm256_andnot_pd(dummy_mask,r00);
 700
 701             /* Calculate table index by multiplying r with table scale and truncate to integer */
 702             rt               = _mm256_mul_pd(r00,vftabscale);
 703             vfitab           = _mm256_cvttpd_epi32(rt);
 704             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 705             vfitab           = _mm_slli_epi32(vfitab,3);
 706
 707             /* COULOMB ELECTROSTATICS */
 708             velec            = _mm256_mul_pd(qq00,rinv00);
 709             felec            = _mm256_mul_pd(velec,rinvsq00);
 710
 711             /* CUBIC SPLINE TABLE DISPERSION */
 712             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 713             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 714             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 715             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 716             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 717             Heps             = _mm256_mul_pd(vfeps,H);
 718             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 719             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 720             vvdw6            = _mm256_mul_pd(c6_00,VV);
 721             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 722             fvdw6            = _mm256_mul_pd(c6_00,FF);
 723
 724             /* CUBIC SPLINE TABLE REPULSION */
 725             vfitab           = _mm_add_epi32(vfitab,ifour);
 726             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 727             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 728             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 729             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 730             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 731             Heps             = _mm256_mul_pd(vfeps,H);
 732             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 733             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 734             vvdw12           = _mm256_mul_pd(c12_00,VV);
 735             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 736             fvdw12           = _mm256_mul_pd(c12_00,FF);
 737             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 738             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 739
 740             /* Update potential sum for this i atom from the interaction with this j atom. */
 741             velec            = _mm256_andnot_pd(dummy_mask,velec);
 742             velecsum         = _mm256_add_pd(velecsum,velec);
 743             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 744             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 745
 746             fscal            = _mm256_add_pd(felec,fvdw);
 747
 748             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm256_mul_pd(fscal,dx00);
 752             ty               = _mm256_mul_pd(fscal,dy00);
 753             tz               = _mm256_mul_pd(fscal,dz00);
 754
 755             /* Update vectorial force */
 756             fix0             = _mm256_add_pd(fix0,tx);
 757             fiy0             = _mm256_add_pd(fiy0,ty);
 758             fiz0             = _mm256_add_pd(fiz0,tz);
 759
 760             fjx0             = _mm256_add_pd(fjx0,tx);
 761             fjy0             = _mm256_add_pd(fjy0,ty);
 762             fjz0             = _mm256_add_pd(fjz0,tz);
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             /* COULOMB ELECTROSTATICS */
 769             velec            = _mm256_mul_pd(qq01,rinv01);
 770             felec            = _mm256_mul_pd(velec,rinvsq01);
 771
 772             /* Update potential sum for this i atom from the interaction with this j atom. */
 773             velec            = _mm256_andnot_pd(dummy_mask,velec);
 774             velecsum         = _mm256_add_pd(velecsum,velec);
 775
 776             fscal            = felec;
 777
 778             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 779
 780             /* Calculate temporary vectorial force */
 781             tx               = _mm256_mul_pd(fscal,dx01);
 782             ty               = _mm256_mul_pd(fscal,dy01);
 783             tz               = _mm256_mul_pd(fscal,dz01);
 784
 785             /* Update vectorial force */
 786             fix0             = _mm256_add_pd(fix0,tx);
 787             fiy0             = _mm256_add_pd(fiy0,ty);
 788             fiz0             = _mm256_add_pd(fiz0,tz);
 789
 790             fjx1             = _mm256_add_pd(fjx1,tx);
 791             fjy1             = _mm256_add_pd(fjy1,ty);
 792             fjz1             = _mm256_add_pd(fjz1,tz);
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             /* COULOMB ELECTROSTATICS */
 799             velec            = _mm256_mul_pd(qq02,rinv02);
 800             felec            = _mm256_mul_pd(velec,rinvsq02);
 801
 802             /* Update potential sum for this i atom from the interaction with this j atom. */
 803             velec            = _mm256_andnot_pd(dummy_mask,velec);
 804             velecsum         = _mm256_add_pd(velecsum,velec);
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_pd(fscal,dx02);
 812             ty               = _mm256_mul_pd(fscal,dy02);
 813             tz               = _mm256_mul_pd(fscal,dz02);
 814
 815             /* Update vectorial force */
 816             fix0             = _mm256_add_pd(fix0,tx);
 817             fiy0             = _mm256_add_pd(fiy0,ty);
 818             fiz0             = _mm256_add_pd(fiz0,tz);
 819
 820             fjx2             = _mm256_add_pd(fjx2,tx);
 821             fjy2             = _mm256_add_pd(fjy2,ty);
 822             fjz2             = _mm256_add_pd(fjz2,tz);
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             /* COULOMB ELECTROSTATICS */
 829             velec            = _mm256_mul_pd(qq10,rinv10);
 830             felec            = _mm256_mul_pd(velec,rinvsq10);
 831
 832             /* Update potential sum for this i atom from the interaction with this j atom. */
 833             velec            = _mm256_andnot_pd(dummy_mask,velec);
 834             velecsum         = _mm256_add_pd(velecsum,velec);
 835
 836             fscal            = felec;
 837
 838             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm256_mul_pd(fscal,dx10);
 842             ty               = _mm256_mul_pd(fscal,dy10);
 843             tz               = _mm256_mul_pd(fscal,dz10);
 844
 845             /* Update vectorial force */
 846             fix1             = _mm256_add_pd(fix1,tx);
 847             fiy1             = _mm256_add_pd(fiy1,ty);
 848             fiz1             = _mm256_add_pd(fiz1,tz);
 849
 850             fjx0             = _mm256_add_pd(fjx0,tx);
 851             fjy0             = _mm256_add_pd(fjy0,ty);
 852             fjz0             = _mm256_add_pd(fjz0,tz);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             /* COULOMB ELECTROSTATICS */
 859             velec            = _mm256_mul_pd(qq11,rinv11);
 860             felec            = _mm256_mul_pd(velec,rinvsq11);
 861
 862             /* Update potential sum for this i atom from the interaction with this j atom. */
 863             velec            = _mm256_andnot_pd(dummy_mask,velec);
 864             velecsum         = _mm256_add_pd(velecsum,velec);
 865
 866             fscal            = felec;
 867
 868             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 869
 870             /* Calculate temporary vectorial force */
 871             tx               = _mm256_mul_pd(fscal,dx11);
 872             ty               = _mm256_mul_pd(fscal,dy11);
 873             tz               = _mm256_mul_pd(fscal,dz11);
 874
 875             /* Update vectorial force */
 876             fix1             = _mm256_add_pd(fix1,tx);
 877             fiy1             = _mm256_add_pd(fiy1,ty);
 878             fiz1             = _mm256_add_pd(fiz1,tz);
 879
 880             fjx1             = _mm256_add_pd(fjx1,tx);
 881             fjy1             = _mm256_add_pd(fjy1,ty);
 882             fjz1             = _mm256_add_pd(fjz1,tz);
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             /* COULOMB ELECTROSTATICS */
 889             velec            = _mm256_mul_pd(qq12,rinv12);
 890             felec            = _mm256_mul_pd(velec,rinvsq12);
 891
 892             /* Update potential sum for this i atom from the interaction with this j atom. */
 893             velec            = _mm256_andnot_pd(dummy_mask,velec);
 894             velecsum         = _mm256_add_pd(velecsum,velec);
 895
 896             fscal            = felec;
 897
 898             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm256_mul_pd(fscal,dx12);
 902             ty               = _mm256_mul_pd(fscal,dy12);
 903             tz               = _mm256_mul_pd(fscal,dz12);
 904
 905             /* Update vectorial force */
 906             fix1             = _mm256_add_pd(fix1,tx);
 907             fiy1             = _mm256_add_pd(fiy1,ty);
 908             fiz1             = _mm256_add_pd(fiz1,tz);
 909
 910             fjx2             = _mm256_add_pd(fjx2,tx);
 911             fjy2             = _mm256_add_pd(fjy2,ty);
 912             fjz2             = _mm256_add_pd(fjz2,tz);
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             /* COULOMB ELECTROSTATICS */
 919             velec            = _mm256_mul_pd(qq20,rinv20);
 920             felec            = _mm256_mul_pd(velec,rinvsq20);
 921
 922             /* Update potential sum for this i atom from the interaction with this j atom. */
 923             velec            = _mm256_andnot_pd(dummy_mask,velec);
 924             velecsum         = _mm256_add_pd(velecsum,velec);
 925
 926             fscal            = felec;
 927
 928             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 929
 930             /* Calculate temporary vectorial force */
 931             tx               = _mm256_mul_pd(fscal,dx20);
 932             ty               = _mm256_mul_pd(fscal,dy20);
 933             tz               = _mm256_mul_pd(fscal,dz20);
 934
 935             /* Update vectorial force */
 936             fix2             = _mm256_add_pd(fix2,tx);
 937             fiy2             = _mm256_add_pd(fiy2,ty);
 938             fiz2             = _mm256_add_pd(fiz2,tz);
 939
 940             fjx0             = _mm256_add_pd(fjx0,tx);
 941             fjy0             = _mm256_add_pd(fjy0,ty);
 942             fjz0             = _mm256_add_pd(fjz0,tz);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             /* COULOMB ELECTROSTATICS */
 949             velec            = _mm256_mul_pd(qq21,rinv21);
 950             felec            = _mm256_mul_pd(velec,rinvsq21);
 951
 952             /* Update potential sum for this i atom from the interaction with this j atom. */
 953             velec            = _mm256_andnot_pd(dummy_mask,velec);
 954             velecsum         = _mm256_add_pd(velecsum,velec);
 955
 956             fscal            = felec;
 957
 958             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm256_mul_pd(fscal,dx21);
 962             ty               = _mm256_mul_pd(fscal,dy21);
 963             tz               = _mm256_mul_pd(fscal,dz21);
 964
 965             /* Update vectorial force */
 966             fix2             = _mm256_add_pd(fix2,tx);
 967             fiy2             = _mm256_add_pd(fiy2,ty);
 968             fiz2             = _mm256_add_pd(fiz2,tz);
 969
 970             fjx1             = _mm256_add_pd(fjx1,tx);
 971             fjy1             = _mm256_add_pd(fjy1,ty);
 972             fjz1             = _mm256_add_pd(fjz1,tz);
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             /* COULOMB ELECTROSTATICS */
 979             velec            = _mm256_mul_pd(qq22,rinv22);
 980             felec            = _mm256_mul_pd(velec,rinvsq22);
 981
 982             /* Update potential sum for this i atom from the interaction with this j atom. */
 983             velec            = _mm256_andnot_pd(dummy_mask,velec);
 984             velecsum         = _mm256_add_pd(velecsum,velec);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_pd(fscal,dx22);
 992             ty               = _mm256_mul_pd(fscal,dy22);
 993             tz               = _mm256_mul_pd(fscal,dz22);
 994
 995             /* Update vectorial force */
 996             fix2             = _mm256_add_pd(fix2,tx);
 997             fiy2             = _mm256_add_pd(fiy2,ty);
 998             fiz2             = _mm256_add_pd(fiz2,tz);
 999
1000             fjx2             = _mm256_add_pd(fjx2,tx);
1001             fjy2             = _mm256_add_pd(fjy2,ty);
1002             fjz2             = _mm256_add_pd(fjz2,tz);
1003
1004             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1005             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1006             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1007             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1008
1009             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1010                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1011
1012             /* Inner loop uses 279 flops */
1013         }
1014
1015         /* End of innermost loop */
1016
1017         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1018                                                  f+i_coord_offset,fshift+i_shift_offset);
1019
1020         ggid                        = gid[iidx];
1021         /* Update potential energies */
1022         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1023         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1024
1025         /* Increment number of inner iterations */
1026         inneriter                  += j_index_end - j_index_start;
1027
1028         /* Outer loop uses 20 flops */
1029     }
1030
1031     /* Increment number of outer iterations */
1032     outeriter        += nri;
1033
1034     /* Update outer/inner flops */
1035
1036     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*279);
1037 }
1038 /*
1039  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_256_double
1040  * Electrostatics interaction: Coulomb
1041  * VdW interaction:            CubicSplineTable
1042  * Geometry:                   Water3-Water3
1043  * Calculate force/pot:        Force
1044  */
1045 void
1046 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_256_double
1047                     (t_nblist                    * gmx_restrict       nlist,
1048                      rvec                        * gmx_restrict          xx,
1049                      rvec                        * gmx_restrict          ff,
1050                      struct t_forcerec           * gmx_restrict          fr,
1051                      t_mdatoms                   * gmx_restrict     mdatoms,
1052                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1053                      t_nrnb                      * gmx_restrict        nrnb)
1054 {
1055     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1056      * just 0 for non-waters.
1057      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1058      * jnr indices corresponding to data put in the four positions in the SIMD register.
1059      */
1060     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1061     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1062     int              jnrA,jnrB,jnrC,jnrD;
1063     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1064     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1065     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1066     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1067     real             rcutoff_scalar;
1068     real             *shiftvec,*fshift,*x,*f;
1069     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1070     real             scratch[4*DIM];
1071     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1072     real *           vdwioffsetptr0;
1073     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1074     real *           vdwioffsetptr1;
1075     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1076     real *           vdwioffsetptr2;
1077     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1078     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1079     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1080     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1081     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1082     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1083     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1084     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1085     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1086     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1087     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1088     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1089     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1090     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1091     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1092     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1093     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1094     real             *charge;
1095     int              nvdwtype;
1096     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1097     int              *vdwtype;
1098     real             *vdwparam;
1099     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1100     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1101     __m128i          vfitab;
1102     __m128i          ifour       = _mm_set1_epi32(4);
1103     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1104     real             *vftab;
1105     __m256d          dummy_mask,cutoff_mask;
1106     __m128           tmpmask0,tmpmask1;
1107     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1108     __m256d          one     = _mm256_set1_pd(1.0);
1109     __m256d          two     = _mm256_set1_pd(2.0);
1110     x                = xx[0];
1111     f                = ff[0];
1112
1113     nri              = nlist->nri;
1114     iinr             = nlist->iinr;
1115     jindex           = nlist->jindex;
1116     jjnr             = nlist->jjnr;
1117     shiftidx         = nlist->shift;
1118     gid              = nlist->gid;
1119     shiftvec         = fr->shift_vec[0];
1120     fshift           = fr->fshift[0];
1121     facel            = _mm256_set1_pd(fr->ic->epsfac);
1122     charge           = mdatoms->chargeA;
1123     nvdwtype         = fr->ntype;
1124     vdwparam         = fr->nbfp;
1125     vdwtype          = mdatoms->typeA;
1126
1127     vftab            = kernel_data->table_vdw->data;
1128     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1129
1130     /* Setup water-specific parameters */
1131     inr              = nlist->iinr[0];
1132     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1136
1137     jq0              = _mm256_set1_pd(charge[inr+0]);
1138     jq1              = _mm256_set1_pd(charge[inr+1]);
1139     jq2              = _mm256_set1_pd(charge[inr+2]);
1140     vdwjidx0A        = 2*vdwtype[inr+0];
1141     qq00             = _mm256_mul_pd(iq0,jq0);
1142     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1143     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1144     qq01             = _mm256_mul_pd(iq0,jq1);
1145     qq02             = _mm256_mul_pd(iq0,jq2);
1146     qq10             = _mm256_mul_pd(iq1,jq0);
1147     qq11             = _mm256_mul_pd(iq1,jq1);
1148     qq12             = _mm256_mul_pd(iq1,jq2);
1149     qq20             = _mm256_mul_pd(iq2,jq0);
1150     qq21             = _mm256_mul_pd(iq2,jq1);
1151     qq22             = _mm256_mul_pd(iq2,jq2);
1152
1153     /* Avoid stupid compiler warnings */
1154     jnrA = jnrB = jnrC = jnrD = 0;
1155     j_coord_offsetA = 0;
1156     j_coord_offsetB = 0;
1157     j_coord_offsetC = 0;
1158     j_coord_offsetD = 0;
1159
1160     outeriter        = 0;
1161     inneriter        = 0;
1162
1163     for(iidx=0;iidx<4*DIM;iidx++)
1164     {
1165         scratch[iidx] = 0.0;
1166     }
1167
1168     /* Start outer loop over neighborlists */
1169     for(iidx=0; iidx<nri; iidx++)
1170     {
1171         /* Load shift vector for this list */
1172         i_shift_offset   = DIM*shiftidx[iidx];
1173
1174         /* Load limits for loop over neighbors */
1175         j_index_start    = jindex[iidx];
1176         j_index_end      = jindex[iidx+1];
1177
1178         /* Get outer coordinate index */
1179         inr              = iinr[iidx];
1180         i_coord_offset   = DIM*inr;
1181
1182         /* Load i particle coords and add shift vector */
1183         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1184                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1185
1186         fix0             = _mm256_setzero_pd();
1187         fiy0             = _mm256_setzero_pd();
1188         fiz0             = _mm256_setzero_pd();
1189         fix1             = _mm256_setzero_pd();
1190         fiy1             = _mm256_setzero_pd();
1191         fiz1             = _mm256_setzero_pd();
1192         fix2             = _mm256_setzero_pd();
1193         fiy2             = _mm256_setzero_pd();
1194         fiz2             = _mm256_setzero_pd();
1195
1196         /* Start inner kernel loop */
1197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1198         {
1199
1200             /* Get j neighbor index, and coordinate index */
1201             jnrA             = jjnr[jidx];
1202             jnrB             = jjnr[jidx+1];
1203             jnrC             = jjnr[jidx+2];
1204             jnrD             = jjnr[jidx+3];
1205             j_coord_offsetA  = DIM*jnrA;
1206             j_coord_offsetB  = DIM*jnrB;
1207             j_coord_offsetC  = DIM*jnrC;
1208             j_coord_offsetD  = DIM*jnrD;
1209
1210             /* load j atom coordinates */
1211             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1212                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1213                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1214
1215             /* Calculate displacement vector */
1216             dx00             = _mm256_sub_pd(ix0,jx0);
1217             dy00             = _mm256_sub_pd(iy0,jy0);
1218             dz00             = _mm256_sub_pd(iz0,jz0);
1219             dx01             = _mm256_sub_pd(ix0,jx1);
1220             dy01             = _mm256_sub_pd(iy0,jy1);
1221             dz01             = _mm256_sub_pd(iz0,jz1);
1222             dx02             = _mm256_sub_pd(ix0,jx2);
1223             dy02             = _mm256_sub_pd(iy0,jy2);
1224             dz02             = _mm256_sub_pd(iz0,jz2);
1225             dx10             = _mm256_sub_pd(ix1,jx0);
1226             dy10             = _mm256_sub_pd(iy1,jy0);
1227             dz10             = _mm256_sub_pd(iz1,jz0);
1228             dx11             = _mm256_sub_pd(ix1,jx1);
1229             dy11             = _mm256_sub_pd(iy1,jy1);
1230             dz11             = _mm256_sub_pd(iz1,jz1);
1231             dx12             = _mm256_sub_pd(ix1,jx2);
1232             dy12             = _mm256_sub_pd(iy1,jy2);
1233             dz12             = _mm256_sub_pd(iz1,jz2);
1234             dx20             = _mm256_sub_pd(ix2,jx0);
1235             dy20             = _mm256_sub_pd(iy2,jy0);
1236             dz20             = _mm256_sub_pd(iz2,jz0);
1237             dx21             = _mm256_sub_pd(ix2,jx1);
1238             dy21             = _mm256_sub_pd(iy2,jy1);
1239             dz21             = _mm256_sub_pd(iz2,jz1);
1240             dx22             = _mm256_sub_pd(ix2,jx2);
1241             dy22             = _mm256_sub_pd(iy2,jy2);
1242             dz22             = _mm256_sub_pd(iz2,jz2);
1243
1244             /* Calculate squared distance and things based on it */
1245             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1246             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1247             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1248             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1249             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1250             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1251             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1252             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1253             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1254
1255             rinv00           = avx256_invsqrt_d(rsq00);
1256             rinv01           = avx256_invsqrt_d(rsq01);
1257             rinv02           = avx256_invsqrt_d(rsq02);
1258             rinv10           = avx256_invsqrt_d(rsq10);
1259             rinv11           = avx256_invsqrt_d(rsq11);
1260             rinv12           = avx256_invsqrt_d(rsq12);
1261             rinv20           = avx256_invsqrt_d(rsq20);
1262             rinv21           = avx256_invsqrt_d(rsq21);
1263             rinv22           = avx256_invsqrt_d(rsq22);
1264
1265             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1266             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1267             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1268             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1269             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1270             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1271             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1272             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1273             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1274
1275             fjx0             = _mm256_setzero_pd();
1276             fjy0             = _mm256_setzero_pd();
1277             fjz0             = _mm256_setzero_pd();
1278             fjx1             = _mm256_setzero_pd();
1279             fjy1             = _mm256_setzero_pd();
1280             fjz1             = _mm256_setzero_pd();
1281             fjx2             = _mm256_setzero_pd();
1282             fjy2             = _mm256_setzero_pd();
1283             fjz2             = _mm256_setzero_pd();
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             r00              = _mm256_mul_pd(rsq00,rinv00);
1290
1291             /* Calculate table index by multiplying r with table scale and truncate to integer */
1292             rt               = _mm256_mul_pd(r00,vftabscale);
1293             vfitab           = _mm256_cvttpd_epi32(rt);
1294             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1295             vfitab           = _mm_slli_epi32(vfitab,3);
1296
1297             /* COULOMB ELECTROSTATICS */
1298             velec            = _mm256_mul_pd(qq00,rinv00);
1299             felec            = _mm256_mul_pd(velec,rinvsq00);
1300
1301             /* CUBIC SPLINE TABLE DISPERSION */
1302             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1303             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1304             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1305             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1306             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1307             Heps             = _mm256_mul_pd(vfeps,H);
1308             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1309             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1310             fvdw6            = _mm256_mul_pd(c6_00,FF);
1311
1312             /* CUBIC SPLINE TABLE REPULSION */
1313             vfitab           = _mm_add_epi32(vfitab,ifour);
1314             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1315             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1316             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1317             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1318             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1319             Heps             = _mm256_mul_pd(vfeps,H);
1320             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1321             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1322             fvdw12           = _mm256_mul_pd(c12_00,FF);
1323             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1324
1325             fscal            = _mm256_add_pd(felec,fvdw);
1326
1327             /* Calculate temporary vectorial force */
1328             tx               = _mm256_mul_pd(fscal,dx00);
1329             ty               = _mm256_mul_pd(fscal,dy00);
1330             tz               = _mm256_mul_pd(fscal,dz00);
1331
1332             /* Update vectorial force */
1333             fix0             = _mm256_add_pd(fix0,tx);
1334             fiy0             = _mm256_add_pd(fiy0,ty);
1335             fiz0             = _mm256_add_pd(fiz0,tz);
1336
1337             fjx0             = _mm256_add_pd(fjx0,tx);
1338             fjy0             = _mm256_add_pd(fjy0,ty);
1339             fjz0             = _mm256_add_pd(fjz0,tz);
1340
1341             /**************************
1342              * CALCULATE INTERACTIONS *
1343              **************************/
1344
1345             /* COULOMB ELECTROSTATICS */
1346             velec            = _mm256_mul_pd(qq01,rinv01);
1347             felec            = _mm256_mul_pd(velec,rinvsq01);
1348
1349             fscal            = felec;
1350
1351             /* Calculate temporary vectorial force */
1352             tx               = _mm256_mul_pd(fscal,dx01);
1353             ty               = _mm256_mul_pd(fscal,dy01);
1354             tz               = _mm256_mul_pd(fscal,dz01);
1355
1356             /* Update vectorial force */
1357             fix0             = _mm256_add_pd(fix0,tx);
1358             fiy0             = _mm256_add_pd(fiy0,ty);
1359             fiz0             = _mm256_add_pd(fiz0,tz);
1360
1361             fjx1             = _mm256_add_pd(fjx1,tx);
1362             fjy1             = _mm256_add_pd(fjy1,ty);
1363             fjz1             = _mm256_add_pd(fjz1,tz);
1364
1365             /**************************
1366              * CALCULATE INTERACTIONS *
1367              **************************/
1368
1369             /* COULOMB ELECTROSTATICS */
1370             velec            = _mm256_mul_pd(qq02,rinv02);
1371             felec            = _mm256_mul_pd(velec,rinvsq02);
1372
1373             fscal            = felec;
1374
1375             /* Calculate temporary vectorial force */
1376             tx               = _mm256_mul_pd(fscal,dx02);
1377             ty               = _mm256_mul_pd(fscal,dy02);
1378             tz               = _mm256_mul_pd(fscal,dz02);
1379
1380             /* Update vectorial force */
1381             fix0             = _mm256_add_pd(fix0,tx);
1382             fiy0             = _mm256_add_pd(fiy0,ty);
1383             fiz0             = _mm256_add_pd(fiz0,tz);
1384
1385             fjx2             = _mm256_add_pd(fjx2,tx);
1386             fjy2             = _mm256_add_pd(fjy2,ty);
1387             fjz2             = _mm256_add_pd(fjz2,tz);
1388
1389             /**************************
1390              * CALCULATE INTERACTIONS *
1391              **************************/
1392
1393             /* COULOMB ELECTROSTATICS */
1394             velec            = _mm256_mul_pd(qq10,rinv10);
1395             felec            = _mm256_mul_pd(velec,rinvsq10);
1396
1397             fscal            = felec;
1398
1399             /* Calculate temporary vectorial force */
1400             tx               = _mm256_mul_pd(fscal,dx10);
1401             ty               = _mm256_mul_pd(fscal,dy10);
1402             tz               = _mm256_mul_pd(fscal,dz10);
1403
1404             /* Update vectorial force */
1405             fix1             = _mm256_add_pd(fix1,tx);
1406             fiy1             = _mm256_add_pd(fiy1,ty);
1407             fiz1             = _mm256_add_pd(fiz1,tz);
1408
1409             fjx0             = _mm256_add_pd(fjx0,tx);
1410             fjy0             = _mm256_add_pd(fjy0,ty);
1411             fjz0             = _mm256_add_pd(fjz0,tz);
1412
1413             /**************************
1414              * CALCULATE INTERACTIONS *
1415              **************************/
1416
1417             /* COULOMB ELECTROSTATICS */
1418             velec            = _mm256_mul_pd(qq11,rinv11);
1419             felec            = _mm256_mul_pd(velec,rinvsq11);
1420
1421             fscal            = felec;
1422
1423             /* Calculate temporary vectorial force */
1424             tx               = _mm256_mul_pd(fscal,dx11);
1425             ty               = _mm256_mul_pd(fscal,dy11);
1426             tz               = _mm256_mul_pd(fscal,dz11);
1427
1428             /* Update vectorial force */
1429             fix1             = _mm256_add_pd(fix1,tx);
1430             fiy1             = _mm256_add_pd(fiy1,ty);
1431             fiz1             = _mm256_add_pd(fiz1,tz);
1432
1433             fjx1             = _mm256_add_pd(fjx1,tx);
1434             fjy1             = _mm256_add_pd(fjy1,ty);
1435             fjz1             = _mm256_add_pd(fjz1,tz);
1436
1437             /**************************
1438              * CALCULATE INTERACTIONS *
1439              **************************/
1440
1441             /* COULOMB ELECTROSTATICS */
1442             velec            = _mm256_mul_pd(qq12,rinv12);
1443             felec            = _mm256_mul_pd(velec,rinvsq12);
1444
1445             fscal            = felec;
1446
1447             /* Calculate temporary vectorial force */
1448             tx               = _mm256_mul_pd(fscal,dx12);
1449             ty               = _mm256_mul_pd(fscal,dy12);
1450             tz               = _mm256_mul_pd(fscal,dz12);
1451
1452             /* Update vectorial force */
1453             fix1             = _mm256_add_pd(fix1,tx);
1454             fiy1             = _mm256_add_pd(fiy1,ty);
1455             fiz1             = _mm256_add_pd(fiz1,tz);
1456
1457             fjx2             = _mm256_add_pd(fjx2,tx);
1458             fjy2             = _mm256_add_pd(fjy2,ty);
1459             fjz2             = _mm256_add_pd(fjz2,tz);
1460
1461             /**************************
1462              * CALCULATE INTERACTIONS *
1463              **************************/
1464
1465             /* COULOMB ELECTROSTATICS */
1466             velec            = _mm256_mul_pd(qq20,rinv20);
1467             felec            = _mm256_mul_pd(velec,rinvsq20);
1468
1469             fscal            = felec;
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm256_mul_pd(fscal,dx20);
1473             ty               = _mm256_mul_pd(fscal,dy20);
1474             tz               = _mm256_mul_pd(fscal,dz20);
1475
1476             /* Update vectorial force */
1477             fix2             = _mm256_add_pd(fix2,tx);
1478             fiy2             = _mm256_add_pd(fiy2,ty);
1479             fiz2             = _mm256_add_pd(fiz2,tz);
1480
1481             fjx0             = _mm256_add_pd(fjx0,tx);
1482             fjy0             = _mm256_add_pd(fjy0,ty);
1483             fjz0             = _mm256_add_pd(fjz0,tz);
1484
1485             /**************************
1486              * CALCULATE INTERACTIONS *
1487              **************************/
1488
1489             /* COULOMB ELECTROSTATICS */
1490             velec            = _mm256_mul_pd(qq21,rinv21);
1491             felec            = _mm256_mul_pd(velec,rinvsq21);
1492
1493             fscal            = felec;
1494
1495             /* Calculate temporary vectorial force */
1496             tx               = _mm256_mul_pd(fscal,dx21);
1497             ty               = _mm256_mul_pd(fscal,dy21);
1498             tz               = _mm256_mul_pd(fscal,dz21);
1499
1500             /* Update vectorial force */
1501             fix2             = _mm256_add_pd(fix2,tx);
1502             fiy2             = _mm256_add_pd(fiy2,ty);
1503             fiz2             = _mm256_add_pd(fiz2,tz);
1504
1505             fjx1             = _mm256_add_pd(fjx1,tx);
1506             fjy1             = _mm256_add_pd(fjy1,ty);
1507             fjz1             = _mm256_add_pd(fjz1,tz);
1508
1509             /**************************
1510              * CALCULATE INTERACTIONS *
1511              **************************/
1512
1513             /* COULOMB ELECTROSTATICS */
1514             velec            = _mm256_mul_pd(qq22,rinv22);
1515             felec            = _mm256_mul_pd(velec,rinvsq22);
1516
1517             fscal            = felec;
1518
1519             /* Calculate temporary vectorial force */
1520             tx               = _mm256_mul_pd(fscal,dx22);
1521             ty               = _mm256_mul_pd(fscal,dy22);
1522             tz               = _mm256_mul_pd(fscal,dz22);
1523
1524             /* Update vectorial force */
1525             fix2             = _mm256_add_pd(fix2,tx);
1526             fiy2             = _mm256_add_pd(fiy2,ty);
1527             fiz2             = _mm256_add_pd(fiz2,tz);
1528
1529             fjx2             = _mm256_add_pd(fjx2,tx);
1530             fjy2             = _mm256_add_pd(fjy2,ty);
1531             fjz2             = _mm256_add_pd(fjz2,tz);
1532
1533             fjptrA             = f+j_coord_offsetA;
1534             fjptrB             = f+j_coord_offsetB;
1535             fjptrC             = f+j_coord_offsetC;
1536             fjptrD             = f+j_coord_offsetD;
1537
1538             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1539                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1540
1541             /* Inner loop uses 261 flops */
1542         }
1543
1544         if(jidx<j_index_end)
1545         {
1546
1547             /* Get j neighbor index, and coordinate index */
1548             jnrlistA         = jjnr[jidx];
1549             jnrlistB         = jjnr[jidx+1];
1550             jnrlistC         = jjnr[jidx+2];
1551             jnrlistD         = jjnr[jidx+3];
1552             /* Sign of each element will be negative for non-real atoms.
1553              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1554              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1555              */
1556             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1557
1558             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1559             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1560             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1561
1562             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1563             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1564             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1565             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1566             j_coord_offsetA  = DIM*jnrA;
1567             j_coord_offsetB  = DIM*jnrB;
1568             j_coord_offsetC  = DIM*jnrC;
1569             j_coord_offsetD  = DIM*jnrD;
1570
1571             /* load j atom coordinates */
1572             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1573                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1574                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1575
1576             /* Calculate displacement vector */
1577             dx00             = _mm256_sub_pd(ix0,jx0);
1578             dy00             = _mm256_sub_pd(iy0,jy0);
1579             dz00             = _mm256_sub_pd(iz0,jz0);
1580             dx01             = _mm256_sub_pd(ix0,jx1);
1581             dy01             = _mm256_sub_pd(iy0,jy1);
1582             dz01             = _mm256_sub_pd(iz0,jz1);
1583             dx02             = _mm256_sub_pd(ix0,jx2);
1584             dy02             = _mm256_sub_pd(iy0,jy2);
1585             dz02             = _mm256_sub_pd(iz0,jz2);
1586             dx10             = _mm256_sub_pd(ix1,jx0);
1587             dy10             = _mm256_sub_pd(iy1,jy0);
1588             dz10             = _mm256_sub_pd(iz1,jz0);
1589             dx11             = _mm256_sub_pd(ix1,jx1);
1590             dy11             = _mm256_sub_pd(iy1,jy1);
1591             dz11             = _mm256_sub_pd(iz1,jz1);
1592             dx12             = _mm256_sub_pd(ix1,jx2);
1593             dy12             = _mm256_sub_pd(iy1,jy2);
1594             dz12             = _mm256_sub_pd(iz1,jz2);
1595             dx20             = _mm256_sub_pd(ix2,jx0);
1596             dy20             = _mm256_sub_pd(iy2,jy0);
1597             dz20             = _mm256_sub_pd(iz2,jz0);
1598             dx21             = _mm256_sub_pd(ix2,jx1);
1599             dy21             = _mm256_sub_pd(iy2,jy1);
1600             dz21             = _mm256_sub_pd(iz2,jz1);
1601             dx22             = _mm256_sub_pd(ix2,jx2);
1602             dy22             = _mm256_sub_pd(iy2,jy2);
1603             dz22             = _mm256_sub_pd(iz2,jz2);
1604
1605             /* Calculate squared distance and things based on it */
1606             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1607             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1608             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1609             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1610             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1611             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1612             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1613             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1614             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1615
1616             rinv00           = avx256_invsqrt_d(rsq00);
1617             rinv01           = avx256_invsqrt_d(rsq01);
1618             rinv02           = avx256_invsqrt_d(rsq02);
1619             rinv10           = avx256_invsqrt_d(rsq10);
1620             rinv11           = avx256_invsqrt_d(rsq11);
1621             rinv12           = avx256_invsqrt_d(rsq12);
1622             rinv20           = avx256_invsqrt_d(rsq20);
1623             rinv21           = avx256_invsqrt_d(rsq21);
1624             rinv22           = avx256_invsqrt_d(rsq22);
1625
1626             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1627             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1628             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1629             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1630             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1631             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1632             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1633             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1634             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1635
1636             fjx0             = _mm256_setzero_pd();
1637             fjy0             = _mm256_setzero_pd();
1638             fjz0             = _mm256_setzero_pd();
1639             fjx1             = _mm256_setzero_pd();
1640             fjy1             = _mm256_setzero_pd();
1641             fjz1             = _mm256_setzero_pd();
1642             fjx2             = _mm256_setzero_pd();
1643             fjy2             = _mm256_setzero_pd();
1644             fjz2             = _mm256_setzero_pd();
1645
1646             /**************************
1647              * CALCULATE INTERACTIONS *
1648              **************************/
1649
1650             r00              = _mm256_mul_pd(rsq00,rinv00);
1651             r00              = _mm256_andnot_pd(dummy_mask,r00);
1652
1653             /* Calculate table index by multiplying r with table scale and truncate to integer */
1654             rt               = _mm256_mul_pd(r00,vftabscale);
1655             vfitab           = _mm256_cvttpd_epi32(rt);
1656             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1657             vfitab           = _mm_slli_epi32(vfitab,3);
1658
1659             /* COULOMB ELECTROSTATICS */
1660             velec            = _mm256_mul_pd(qq00,rinv00);
1661             felec            = _mm256_mul_pd(velec,rinvsq00);
1662
1663             /* CUBIC SPLINE TABLE DISPERSION */
1664             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1665             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1666             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1667             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1668             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1669             Heps             = _mm256_mul_pd(vfeps,H);
1670             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1671             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1672             fvdw6            = _mm256_mul_pd(c6_00,FF);
1673
1674             /* CUBIC SPLINE TABLE REPULSION */
1675             vfitab           = _mm_add_epi32(vfitab,ifour);
1676             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1677             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1678             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1679             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1680             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1681             Heps             = _mm256_mul_pd(vfeps,H);
1682             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1683             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1684             fvdw12           = _mm256_mul_pd(c12_00,FF);
1685             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1686
1687             fscal            = _mm256_add_pd(felec,fvdw);
1688
1689             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1690
1691             /* Calculate temporary vectorial force */
1692             tx               = _mm256_mul_pd(fscal,dx00);
1693             ty               = _mm256_mul_pd(fscal,dy00);
1694             tz               = _mm256_mul_pd(fscal,dz00);
1695
1696             /* Update vectorial force */
1697             fix0             = _mm256_add_pd(fix0,tx);
1698             fiy0             = _mm256_add_pd(fiy0,ty);
1699             fiz0             = _mm256_add_pd(fiz0,tz);
1700
1701             fjx0             = _mm256_add_pd(fjx0,tx);
1702             fjy0             = _mm256_add_pd(fjy0,ty);
1703             fjz0             = _mm256_add_pd(fjz0,tz);
1704
1705             /**************************
1706              * CALCULATE INTERACTIONS *
1707              **************************/
1708
1709             /* COULOMB ELECTROSTATICS */
1710             velec            = _mm256_mul_pd(qq01,rinv01);
1711             felec            = _mm256_mul_pd(velec,rinvsq01);
1712
1713             fscal            = felec;
1714
1715             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1716
1717             /* Calculate temporary vectorial force */
1718             tx               = _mm256_mul_pd(fscal,dx01);
1719             ty               = _mm256_mul_pd(fscal,dy01);
1720             tz               = _mm256_mul_pd(fscal,dz01);
1721
1722             /* Update vectorial force */
1723             fix0             = _mm256_add_pd(fix0,tx);
1724             fiy0             = _mm256_add_pd(fiy0,ty);
1725             fiz0             = _mm256_add_pd(fiz0,tz);
1726
1727             fjx1             = _mm256_add_pd(fjx1,tx);
1728             fjy1             = _mm256_add_pd(fjy1,ty);
1729             fjz1             = _mm256_add_pd(fjz1,tz);
1730
1731             /**************************
1732              * CALCULATE INTERACTIONS *
1733              **************************/
1734
1735             /* COULOMB ELECTROSTATICS */
1736             velec            = _mm256_mul_pd(qq02,rinv02);
1737             felec            = _mm256_mul_pd(velec,rinvsq02);
1738
1739             fscal            = felec;
1740
1741             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1742
1743             /* Calculate temporary vectorial force */
1744             tx               = _mm256_mul_pd(fscal,dx02);
1745             ty               = _mm256_mul_pd(fscal,dy02);
1746             tz               = _mm256_mul_pd(fscal,dz02);
1747
1748             /* Update vectorial force */
1749             fix0             = _mm256_add_pd(fix0,tx);
1750             fiy0             = _mm256_add_pd(fiy0,ty);
1751             fiz0             = _mm256_add_pd(fiz0,tz);
1752
1753             fjx2             = _mm256_add_pd(fjx2,tx);
1754             fjy2             = _mm256_add_pd(fjy2,ty);
1755             fjz2             = _mm256_add_pd(fjz2,tz);
1756
1757             /**************************
1758              * CALCULATE INTERACTIONS *
1759              **************************/
1760
1761             /* COULOMB ELECTROSTATICS */
1762             velec            = _mm256_mul_pd(qq10,rinv10);
1763             felec            = _mm256_mul_pd(velec,rinvsq10);
1764
1765             fscal            = felec;
1766
1767             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1768
1769             /* Calculate temporary vectorial force */
1770             tx               = _mm256_mul_pd(fscal,dx10);
1771             ty               = _mm256_mul_pd(fscal,dy10);
1772             tz               = _mm256_mul_pd(fscal,dz10);
1773
1774             /* Update vectorial force */
1775             fix1             = _mm256_add_pd(fix1,tx);
1776             fiy1             = _mm256_add_pd(fiy1,ty);
1777             fiz1             = _mm256_add_pd(fiz1,tz);
1778
1779             fjx0             = _mm256_add_pd(fjx0,tx);
1780             fjy0             = _mm256_add_pd(fjy0,ty);
1781             fjz0             = _mm256_add_pd(fjz0,tz);
1782
1783             /**************************
1784              * CALCULATE INTERACTIONS *
1785              **************************/
1786
1787             /* COULOMB ELECTROSTATICS */
1788             velec            = _mm256_mul_pd(qq11,rinv11);
1789             felec            = _mm256_mul_pd(velec,rinvsq11);
1790
1791             fscal            = felec;
1792
1793             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1794
1795             /* Calculate temporary vectorial force */
1796             tx               = _mm256_mul_pd(fscal,dx11);
1797             ty               = _mm256_mul_pd(fscal,dy11);
1798             tz               = _mm256_mul_pd(fscal,dz11);
1799
1800             /* Update vectorial force */
1801             fix1             = _mm256_add_pd(fix1,tx);
1802             fiy1             = _mm256_add_pd(fiy1,ty);
1803             fiz1             = _mm256_add_pd(fiz1,tz);
1804
1805             fjx1             = _mm256_add_pd(fjx1,tx);
1806             fjy1             = _mm256_add_pd(fjy1,ty);
1807             fjz1             = _mm256_add_pd(fjz1,tz);
1808
1809             /**************************
1810              * CALCULATE INTERACTIONS *
1811              **************************/
1812
1813             /* COULOMB ELECTROSTATICS */
1814             velec            = _mm256_mul_pd(qq12,rinv12);
1815             felec            = _mm256_mul_pd(velec,rinvsq12);
1816
1817             fscal            = felec;
1818
1819             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1820
1821             /* Calculate temporary vectorial force */
1822             tx               = _mm256_mul_pd(fscal,dx12);
1823             ty               = _mm256_mul_pd(fscal,dy12);
1824             tz               = _mm256_mul_pd(fscal,dz12);
1825
1826             /* Update vectorial force */
1827             fix1             = _mm256_add_pd(fix1,tx);
1828             fiy1             = _mm256_add_pd(fiy1,ty);
1829             fiz1             = _mm256_add_pd(fiz1,tz);
1830
1831             fjx2             = _mm256_add_pd(fjx2,tx);
1832             fjy2             = _mm256_add_pd(fjy2,ty);
1833             fjz2             = _mm256_add_pd(fjz2,tz);
1834
1835             /**************************
1836              * CALCULATE INTERACTIONS *
1837              **************************/
1838
1839             /* COULOMB ELECTROSTATICS */
1840             velec            = _mm256_mul_pd(qq20,rinv20);
1841             felec            = _mm256_mul_pd(velec,rinvsq20);
1842
1843             fscal            = felec;
1844
1845             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1846
1847             /* Calculate temporary vectorial force */
1848             tx               = _mm256_mul_pd(fscal,dx20);
1849             ty               = _mm256_mul_pd(fscal,dy20);
1850             tz               = _mm256_mul_pd(fscal,dz20);
1851
1852             /* Update vectorial force */
1853             fix2             = _mm256_add_pd(fix2,tx);
1854             fiy2             = _mm256_add_pd(fiy2,ty);
1855             fiz2             = _mm256_add_pd(fiz2,tz);
1856
1857             fjx0             = _mm256_add_pd(fjx0,tx);
1858             fjy0             = _mm256_add_pd(fjy0,ty);
1859             fjz0             = _mm256_add_pd(fjz0,tz);
1860
1861             /**************************
1862              * CALCULATE INTERACTIONS *
1863              **************************/
1864
1865             /* COULOMB ELECTROSTATICS */
1866             velec            = _mm256_mul_pd(qq21,rinv21);
1867             felec            = _mm256_mul_pd(velec,rinvsq21);
1868
1869             fscal            = felec;
1870
1871             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1872
1873             /* Calculate temporary vectorial force */
1874             tx               = _mm256_mul_pd(fscal,dx21);
1875             ty               = _mm256_mul_pd(fscal,dy21);
1876             tz               = _mm256_mul_pd(fscal,dz21);
1877
1878             /* Update vectorial force */
1879             fix2             = _mm256_add_pd(fix2,tx);
1880             fiy2             = _mm256_add_pd(fiy2,ty);
1881             fiz2             = _mm256_add_pd(fiz2,tz);
1882
1883             fjx1             = _mm256_add_pd(fjx1,tx);
1884             fjy1             = _mm256_add_pd(fjy1,ty);
1885             fjz1             = _mm256_add_pd(fjz1,tz);
1886
1887             /**************************
1888              * CALCULATE INTERACTIONS *
1889              **************************/
1890
1891             /* COULOMB ELECTROSTATICS */
1892             velec            = _mm256_mul_pd(qq22,rinv22);
1893             felec            = _mm256_mul_pd(velec,rinvsq22);
1894
1895             fscal            = felec;
1896
1897             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1898
1899             /* Calculate temporary vectorial force */
1900             tx               = _mm256_mul_pd(fscal,dx22);
1901             ty               = _mm256_mul_pd(fscal,dy22);
1902             tz               = _mm256_mul_pd(fscal,dz22);
1903
1904             /* Update vectorial force */
1905             fix2             = _mm256_add_pd(fix2,tx);
1906             fiy2             = _mm256_add_pd(fiy2,ty);
1907             fiz2             = _mm256_add_pd(fiz2,tz);
1908
1909             fjx2             = _mm256_add_pd(fjx2,tx);
1910             fjy2             = _mm256_add_pd(fjy2,ty);
1911             fjz2             = _mm256_add_pd(fjz2,tz);
1912
1913             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1914             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1915             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1916             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1917
1918             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1919                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1920
1921             /* Inner loop uses 262 flops */
1922         }
1923
1924         /* End of innermost loop */
1925
1926         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1927                                                  f+i_coord_offset,fshift+i_shift_offset);
1928
1929         /* Increment number of inner iterations */
1930         inneriter                  += j_index_end - j_index_start;
1931
1932         /* Outer loop uses 18 flops */
1933     }
1934
1935     /* Increment number of outer iterations */
1936     outeriter        += nri;
1937
1938     /* Update outer/inner flops */
1939
1940     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*262);
1941 }