src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_GeomW3W3_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecRF_VdwCSTab_GeomW3W3_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_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     krf              = _mm256_set1_pd(fr->ic->k_rf);
 135     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 136     crf              = _mm256_set1_pd(fr->ic->c_rf);
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     vftab            = kernel_data->table_vdw->data;
 142     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 147     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 148     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     jq0              = _mm256_set1_pd(charge[inr+0]);
 152     jq1              = _mm256_set1_pd(charge[inr+1]);
 153     jq2              = _mm256_set1_pd(charge[inr+2]);
 154     vdwjidx0A        = 2*vdwtype[inr+0];
 155     qq00             = _mm256_mul_pd(iq0,jq0);
 156     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 157     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 158     qq01             = _mm256_mul_pd(iq0,jq1);
 159     qq02             = _mm256_mul_pd(iq0,jq2);
 160     qq10             = _mm256_mul_pd(iq1,jq0);
 161     qq11             = _mm256_mul_pd(iq1,jq1);
 162     qq12             = _mm256_mul_pd(iq1,jq2);
 163     qq20             = _mm256_mul_pd(iq2,jq0);
 164     qq21             = _mm256_mul_pd(iq2,jq1);
 165     qq22             = _mm256_mul_pd(iq2,jq2);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 199
 200         fix0             = _mm256_setzero_pd();
 201         fiy0             = _mm256_setzero_pd();
 202         fiz0             = _mm256_setzero_pd();
 203         fix1             = _mm256_setzero_pd();
 204         fiy1             = _mm256_setzero_pd();
 205         fiz1             = _mm256_setzero_pd();
 206         fix2             = _mm256_setzero_pd();
 207         fiy2             = _mm256_setzero_pd();
 208         fiz2             = _mm256_setzero_pd();
 209
 210         /* Reset potential sums */
 211         velecsum         = _mm256_setzero_pd();
 212         vvdwsum          = _mm256_setzero_pd();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             j_coord_offsetA  = DIM*jnrA;
 224             j_coord_offsetB  = DIM*jnrB;
 225             j_coord_offsetC  = DIM*jnrC;
 226             j_coord_offsetD  = DIM*jnrD;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm256_sub_pd(ix0,jx0);
 235             dy00             = _mm256_sub_pd(iy0,jy0);
 236             dz00             = _mm256_sub_pd(iz0,jz0);
 237             dx01             = _mm256_sub_pd(ix0,jx1);
 238             dy01             = _mm256_sub_pd(iy0,jy1);
 239             dz01             = _mm256_sub_pd(iz0,jz1);
 240             dx02             = _mm256_sub_pd(ix0,jx2);
 241             dy02             = _mm256_sub_pd(iy0,jy2);
 242             dz02             = _mm256_sub_pd(iz0,jz2);
 243             dx10             = _mm256_sub_pd(ix1,jx0);
 244             dy10             = _mm256_sub_pd(iy1,jy0);
 245             dz10             = _mm256_sub_pd(iz1,jz0);
 246             dx11             = _mm256_sub_pd(ix1,jx1);
 247             dy11             = _mm256_sub_pd(iy1,jy1);
 248             dz11             = _mm256_sub_pd(iz1,jz1);
 249             dx12             = _mm256_sub_pd(ix1,jx2);
 250             dy12             = _mm256_sub_pd(iy1,jy2);
 251             dz12             = _mm256_sub_pd(iz1,jz2);
 252             dx20             = _mm256_sub_pd(ix2,jx0);
 253             dy20             = _mm256_sub_pd(iy2,jy0);
 254             dz20             = _mm256_sub_pd(iz2,jz0);
 255             dx21             = _mm256_sub_pd(ix2,jx1);
 256             dy21             = _mm256_sub_pd(iy2,jy1);
 257             dz21             = _mm256_sub_pd(iz2,jz1);
 258             dx22             = _mm256_sub_pd(ix2,jx2);
 259             dy22             = _mm256_sub_pd(iy2,jy2);
 260             dz22             = _mm256_sub_pd(iz2,jz2);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 264             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 265             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 266             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 267             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 268             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 269             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 270             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 271             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 272
 273             rinv00           = avx256_invsqrt_d(rsq00);
 274             rinv01           = avx256_invsqrt_d(rsq01);
 275             rinv02           = avx256_invsqrt_d(rsq02);
 276             rinv10           = avx256_invsqrt_d(rsq10);
 277             rinv11           = avx256_invsqrt_d(rsq11);
 278             rinv12           = avx256_invsqrt_d(rsq12);
 279             rinv20           = avx256_invsqrt_d(rsq20);
 280             rinv21           = avx256_invsqrt_d(rsq21);
 281             rinv22           = avx256_invsqrt_d(rsq22);
 282
 283             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 284             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 285             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 286             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 287             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 288             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 289             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 290             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 291             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 292
 293             fjx0             = _mm256_setzero_pd();
 294             fjy0             = _mm256_setzero_pd();
 295             fjz0             = _mm256_setzero_pd();
 296             fjx1             = _mm256_setzero_pd();
 297             fjy1             = _mm256_setzero_pd();
 298             fjz1             = _mm256_setzero_pd();
 299             fjx2             = _mm256_setzero_pd();
 300             fjy2             = _mm256_setzero_pd();
 301             fjz2             = _mm256_setzero_pd();
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             r00              = _mm256_mul_pd(rsq00,rinv00);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm256_mul_pd(r00,vftabscale);
 311             vfitab           = _mm256_cvttpd_epi32(rt);
 312             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 313             vfitab           = _mm_slli_epi32(vfitab,3);
 314
 315             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 317             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 318
 319             /* CUBIC SPLINE TABLE DISPERSION */
 320             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 321             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 322             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 323             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 324             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 325             Heps             = _mm256_mul_pd(vfeps,H);
 326             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 327             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 328             vvdw6            = _mm256_mul_pd(c6_00,VV);
 329             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 330             fvdw6            = _mm256_mul_pd(c6_00,FF);
 331
 332             /* CUBIC SPLINE TABLE REPULSION */
 333             vfitab           = _mm_add_epi32(vfitab,ifour);
 334             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 335             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 336             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 337             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 339             Heps             = _mm256_mul_pd(vfeps,H);
 340             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 341             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 342             vvdw12           = _mm256_mul_pd(c12_00,VV);
 343             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 344             fvdw12           = _mm256_mul_pd(c12_00,FF);
 345             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 346             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm256_add_pd(velecsum,velec);
 350             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 351
 352             fscal            = _mm256_add_pd(felec,fvdw);
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm256_mul_pd(fscal,dx00);
 356             ty               = _mm256_mul_pd(fscal,dy00);
 357             tz               = _mm256_mul_pd(fscal,dz00);
 358
 359             /* Update vectorial force */
 360             fix0             = _mm256_add_pd(fix0,tx);
 361             fiy0             = _mm256_add_pd(fiy0,ty);
 362             fiz0             = _mm256_add_pd(fiz0,tz);
 363
 364             fjx0             = _mm256_add_pd(fjx0,tx);
 365             fjy0             = _mm256_add_pd(fjy0,ty);
 366             fjz0             = _mm256_add_pd(fjz0,tz);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* REACTION-FIELD ELECTROSTATICS */
 373             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 374             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm256_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm256_mul_pd(fscal,dx01);
 383             ty               = _mm256_mul_pd(fscal,dy01);
 384             tz               = _mm256_mul_pd(fscal,dz01);
 385
 386             /* Update vectorial force */
 387             fix0             = _mm256_add_pd(fix0,tx);
 388             fiy0             = _mm256_add_pd(fiy0,ty);
 389             fiz0             = _mm256_add_pd(fiz0,tz);
 390
 391             fjx1             = _mm256_add_pd(fjx1,tx);
 392             fjy1             = _mm256_add_pd(fjy1,ty);
 393             fjz1             = _mm256_add_pd(fjz1,tz);
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             /* REACTION-FIELD ELECTROSTATICS */
 400             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 401             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velecsum         = _mm256_add_pd(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_pd(fscal,dx02);
 410             ty               = _mm256_mul_pd(fscal,dy02);
 411             tz               = _mm256_mul_pd(fscal,dz02);
 412
 413             /* Update vectorial force */
 414             fix0             = _mm256_add_pd(fix0,tx);
 415             fiy0             = _mm256_add_pd(fiy0,ty);
 416             fiz0             = _mm256_add_pd(fiz0,tz);
 417
 418             fjx2             = _mm256_add_pd(fjx2,tx);
 419             fjy2             = _mm256_add_pd(fjy2,ty);
 420             fjz2             = _mm256_add_pd(fjz2,tz);
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             /* REACTION-FIELD ELECTROSTATICS */
 427             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 428             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velecsum         = _mm256_add_pd(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm256_mul_pd(fscal,dx10);
 437             ty               = _mm256_mul_pd(fscal,dy10);
 438             tz               = _mm256_mul_pd(fscal,dz10);
 439
 440             /* Update vectorial force */
 441             fix1             = _mm256_add_pd(fix1,tx);
 442             fiy1             = _mm256_add_pd(fiy1,ty);
 443             fiz1             = _mm256_add_pd(fiz1,tz);
 444
 445             fjx0             = _mm256_add_pd(fjx0,tx);
 446             fjy0             = _mm256_add_pd(fjy0,ty);
 447             fjz0             = _mm256_add_pd(fjz0,tz);
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             /* REACTION-FIELD ELECTROSTATICS */
 454             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 455             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velecsum         = _mm256_add_pd(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm256_mul_pd(fscal,dx11);
 464             ty               = _mm256_mul_pd(fscal,dy11);
 465             tz               = _mm256_mul_pd(fscal,dz11);
 466
 467             /* Update vectorial force */
 468             fix1             = _mm256_add_pd(fix1,tx);
 469             fiy1             = _mm256_add_pd(fiy1,ty);
 470             fiz1             = _mm256_add_pd(fiz1,tz);
 471
 472             fjx1             = _mm256_add_pd(fjx1,tx);
 473             fjy1             = _mm256_add_pd(fjy1,ty);
 474             fjz1             = _mm256_add_pd(fjz1,tz);
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             /* REACTION-FIELD ELECTROSTATICS */
 481             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 482             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velecsum         = _mm256_add_pd(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_pd(fscal,dx12);
 491             ty               = _mm256_mul_pd(fscal,dy12);
 492             tz               = _mm256_mul_pd(fscal,dz12);
 493
 494             /* Update vectorial force */
 495             fix1             = _mm256_add_pd(fix1,tx);
 496             fiy1             = _mm256_add_pd(fiy1,ty);
 497             fiz1             = _mm256_add_pd(fiz1,tz);
 498
 499             fjx2             = _mm256_add_pd(fjx2,tx);
 500             fjy2             = _mm256_add_pd(fjy2,ty);
 501             fjz2             = _mm256_add_pd(fjz2,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             /* REACTION-FIELD ELECTROSTATICS */
 508             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 509             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velecsum         = _mm256_add_pd(velecsum,velec);
 513
 514             fscal            = felec;
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm256_mul_pd(fscal,dx20);
 518             ty               = _mm256_mul_pd(fscal,dy20);
 519             tz               = _mm256_mul_pd(fscal,dz20);
 520
 521             /* Update vectorial force */
 522             fix2             = _mm256_add_pd(fix2,tx);
 523             fiy2             = _mm256_add_pd(fiy2,ty);
 524             fiz2             = _mm256_add_pd(fiz2,tz);
 525
 526             fjx0             = _mm256_add_pd(fjx0,tx);
 527             fjy0             = _mm256_add_pd(fjy0,ty);
 528             fjz0             = _mm256_add_pd(fjz0,tz);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             /* REACTION-FIELD ELECTROSTATICS */
 535             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 536             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velecsum         = _mm256_add_pd(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm256_mul_pd(fscal,dx21);
 545             ty               = _mm256_mul_pd(fscal,dy21);
 546             tz               = _mm256_mul_pd(fscal,dz21);
 547
 548             /* Update vectorial force */
 549             fix2             = _mm256_add_pd(fix2,tx);
 550             fiy2             = _mm256_add_pd(fiy2,ty);
 551             fiz2             = _mm256_add_pd(fiz2,tz);
 552
 553             fjx1             = _mm256_add_pd(fjx1,tx);
 554             fjy1             = _mm256_add_pd(fjy1,ty);
 555             fjz1             = _mm256_add_pd(fjz1,tz);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             /* REACTION-FIELD ELECTROSTATICS */
 562             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 563             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velecsum         = _mm256_add_pd(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm256_mul_pd(fscal,dx22);
 572             ty               = _mm256_mul_pd(fscal,dy22);
 573             tz               = _mm256_mul_pd(fscal,dz22);
 574
 575             /* Update vectorial force */
 576             fix2             = _mm256_add_pd(fix2,tx);
 577             fiy2             = _mm256_add_pd(fiy2,ty);
 578             fiz2             = _mm256_add_pd(fiz2,tz);
 579
 580             fjx2             = _mm256_add_pd(fjx2,tx);
 581             fjy2             = _mm256_add_pd(fjy2,ty);
 582             fjz2             = _mm256_add_pd(fjz2,tz);
 583
 584             fjptrA             = f+j_coord_offsetA;
 585             fjptrB             = f+j_coord_offsetB;
 586             fjptrC             = f+j_coord_offsetC;
 587             fjptrD             = f+j_coord_offsetD;
 588
 589             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 590                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 591
 592             /* Inner loop uses 323 flops */
 593         }
 594
 595         if(jidx<j_index_end)
 596         {
 597
 598             /* Get j neighbor index, and coordinate index */
 599             jnrlistA         = jjnr[jidx];
 600             jnrlistB         = jjnr[jidx+1];
 601             jnrlistC         = jjnr[jidx+2];
 602             jnrlistD         = jjnr[jidx+3];
 603             /* Sign of each element will be negative for non-real atoms.
 604              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 605              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 606              */
 607             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 608
 609             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 610             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 611             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 612
 613             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 614             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 615             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 616             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 617             j_coord_offsetA  = DIM*jnrA;
 618             j_coord_offsetB  = DIM*jnrB;
 619             j_coord_offsetC  = DIM*jnrC;
 620             j_coord_offsetD  = DIM*jnrD;
 621
 622             /* load j atom coordinates */
 623             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 624                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 625                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 626
 627             /* Calculate displacement vector */
 628             dx00             = _mm256_sub_pd(ix0,jx0);
 629             dy00             = _mm256_sub_pd(iy0,jy0);
 630             dz00             = _mm256_sub_pd(iz0,jz0);
 631             dx01             = _mm256_sub_pd(ix0,jx1);
 632             dy01             = _mm256_sub_pd(iy0,jy1);
 633             dz01             = _mm256_sub_pd(iz0,jz1);
 634             dx02             = _mm256_sub_pd(ix0,jx2);
 635             dy02             = _mm256_sub_pd(iy0,jy2);
 636             dz02             = _mm256_sub_pd(iz0,jz2);
 637             dx10             = _mm256_sub_pd(ix1,jx0);
 638             dy10             = _mm256_sub_pd(iy1,jy0);
 639             dz10             = _mm256_sub_pd(iz1,jz0);
 640             dx11             = _mm256_sub_pd(ix1,jx1);
 641             dy11             = _mm256_sub_pd(iy1,jy1);
 642             dz11             = _mm256_sub_pd(iz1,jz1);
 643             dx12             = _mm256_sub_pd(ix1,jx2);
 644             dy12             = _mm256_sub_pd(iy1,jy2);
 645             dz12             = _mm256_sub_pd(iz1,jz2);
 646             dx20             = _mm256_sub_pd(ix2,jx0);
 647             dy20             = _mm256_sub_pd(iy2,jy0);
 648             dz20             = _mm256_sub_pd(iz2,jz0);
 649             dx21             = _mm256_sub_pd(ix2,jx1);
 650             dy21             = _mm256_sub_pd(iy2,jy1);
 651             dz21             = _mm256_sub_pd(iz2,jz1);
 652             dx22             = _mm256_sub_pd(ix2,jx2);
 653             dy22             = _mm256_sub_pd(iy2,jy2);
 654             dz22             = _mm256_sub_pd(iz2,jz2);
 655
 656             /* Calculate squared distance and things based on it */
 657             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 658             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 659             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 660             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 661             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 662             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 663             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 664             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 665             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 666
 667             rinv00           = avx256_invsqrt_d(rsq00);
 668             rinv01           = avx256_invsqrt_d(rsq01);
 669             rinv02           = avx256_invsqrt_d(rsq02);
 670             rinv10           = avx256_invsqrt_d(rsq10);
 671             rinv11           = avx256_invsqrt_d(rsq11);
 672             rinv12           = avx256_invsqrt_d(rsq12);
 673             rinv20           = avx256_invsqrt_d(rsq20);
 674             rinv21           = avx256_invsqrt_d(rsq21);
 675             rinv22           = avx256_invsqrt_d(rsq22);
 676
 677             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 678             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 679             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 680             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 681             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 682             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 683             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 684             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 685             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 686
 687             fjx0             = _mm256_setzero_pd();
 688             fjy0             = _mm256_setzero_pd();
 689             fjz0             = _mm256_setzero_pd();
 690             fjx1             = _mm256_setzero_pd();
 691             fjy1             = _mm256_setzero_pd();
 692             fjz1             = _mm256_setzero_pd();
 693             fjx2             = _mm256_setzero_pd();
 694             fjy2             = _mm256_setzero_pd();
 695             fjz2             = _mm256_setzero_pd();
 696
 697             /**************************
 698              * CALCULATE INTERACTIONS *
 699              **************************/
 700
 701             r00              = _mm256_mul_pd(rsq00,rinv00);
 702             r00              = _mm256_andnot_pd(dummy_mask,r00);
 703
 704             /* Calculate table index by multiplying r with table scale and truncate to integer */
 705             rt               = _mm256_mul_pd(r00,vftabscale);
 706             vfitab           = _mm256_cvttpd_epi32(rt);
 707             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 708             vfitab           = _mm_slli_epi32(vfitab,3);
 709
 710             /* REACTION-FIELD ELECTROSTATICS */
 711             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 712             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 713
 714             /* CUBIC SPLINE TABLE DISPERSION */
 715             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 716             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 717             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 718             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 719             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 720             Heps             = _mm256_mul_pd(vfeps,H);
 721             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 722             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 723             vvdw6            = _mm256_mul_pd(c6_00,VV);
 724             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 725             fvdw6            = _mm256_mul_pd(c6_00,FF);
 726
 727             /* CUBIC SPLINE TABLE REPULSION */
 728             vfitab           = _mm_add_epi32(vfitab,ifour);
 729             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 730             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 731             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 732             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 733             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 734             Heps             = _mm256_mul_pd(vfeps,H);
 735             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 736             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 737             vvdw12           = _mm256_mul_pd(c12_00,VV);
 738             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 739             fvdw12           = _mm256_mul_pd(c12_00,FF);
 740             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 741             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 742
 743             /* Update potential sum for this i atom from the interaction with this j atom. */
 744             velec            = _mm256_andnot_pd(dummy_mask,velec);
 745             velecsum         = _mm256_add_pd(velecsum,velec);
 746             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 747             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 748
 749             fscal            = _mm256_add_pd(felec,fvdw);
 750
 751             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 752
 753             /* Calculate temporary vectorial force */
 754             tx               = _mm256_mul_pd(fscal,dx00);
 755             ty               = _mm256_mul_pd(fscal,dy00);
 756             tz               = _mm256_mul_pd(fscal,dz00);
 757
 758             /* Update vectorial force */
 759             fix0             = _mm256_add_pd(fix0,tx);
 760             fiy0             = _mm256_add_pd(fiy0,ty);
 761             fiz0             = _mm256_add_pd(fiz0,tz);
 762
 763             fjx0             = _mm256_add_pd(fjx0,tx);
 764             fjy0             = _mm256_add_pd(fjy0,ty);
 765             fjz0             = _mm256_add_pd(fjz0,tz);
 766
 767             /**************************
 768              * CALCULATE INTERACTIONS *
 769              **************************/
 770
 771             /* REACTION-FIELD ELECTROSTATICS */
 772             velec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_add_pd(rinv01,_mm256_mul_pd(krf,rsq01)),crf));
 773             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
 774
 775             /* Update potential sum for this i atom from the interaction with this j atom. */
 776             velec            = _mm256_andnot_pd(dummy_mask,velec);
 777             velecsum         = _mm256_add_pd(velecsum,velec);
 778
 779             fscal            = felec;
 780
 781             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 782
 783             /* Calculate temporary vectorial force */
 784             tx               = _mm256_mul_pd(fscal,dx01);
 785             ty               = _mm256_mul_pd(fscal,dy01);
 786             tz               = _mm256_mul_pd(fscal,dz01);
 787
 788             /* Update vectorial force */
 789             fix0             = _mm256_add_pd(fix0,tx);
 790             fiy0             = _mm256_add_pd(fiy0,ty);
 791             fiz0             = _mm256_add_pd(fiz0,tz);
 792
 793             fjx1             = _mm256_add_pd(fjx1,tx);
 794             fjy1             = _mm256_add_pd(fjy1,ty);
 795             fjz1             = _mm256_add_pd(fjz1,tz);
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             /* REACTION-FIELD ELECTROSTATICS */
 802             velec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_add_pd(rinv02,_mm256_mul_pd(krf,rsq02)),crf));
 803             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
 804
 805             /* Update potential sum for this i atom from the interaction with this j atom. */
 806             velec            = _mm256_andnot_pd(dummy_mask,velec);
 807             velecsum         = _mm256_add_pd(velecsum,velec);
 808
 809             fscal            = felec;
 810
 811             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 812
 813             /* Calculate temporary vectorial force */
 814             tx               = _mm256_mul_pd(fscal,dx02);
 815             ty               = _mm256_mul_pd(fscal,dy02);
 816             tz               = _mm256_mul_pd(fscal,dz02);
 817
 818             /* Update vectorial force */
 819             fix0             = _mm256_add_pd(fix0,tx);
 820             fiy0             = _mm256_add_pd(fiy0,ty);
 821             fiz0             = _mm256_add_pd(fiz0,tz);
 822
 823             fjx2             = _mm256_add_pd(fjx2,tx);
 824             fjy2             = _mm256_add_pd(fjy2,ty);
 825             fjz2             = _mm256_add_pd(fjz2,tz);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* REACTION-FIELD ELECTROSTATICS */
 832             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 833             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 834
 835             /* Update potential sum for this i atom from the interaction with this j atom. */
 836             velec            = _mm256_andnot_pd(dummy_mask,velec);
 837             velecsum         = _mm256_add_pd(velecsum,velec);
 838
 839             fscal            = felec;
 840
 841             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 842
 843             /* Calculate temporary vectorial force */
 844             tx               = _mm256_mul_pd(fscal,dx10);
 845             ty               = _mm256_mul_pd(fscal,dy10);
 846             tz               = _mm256_mul_pd(fscal,dz10);
 847
 848             /* Update vectorial force */
 849             fix1             = _mm256_add_pd(fix1,tx);
 850             fiy1             = _mm256_add_pd(fiy1,ty);
 851             fiz1             = _mm256_add_pd(fiz1,tz);
 852
 853             fjx0             = _mm256_add_pd(fjx0,tx);
 854             fjy0             = _mm256_add_pd(fjy0,ty);
 855             fjz0             = _mm256_add_pd(fjz0,tz);
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             /* REACTION-FIELD ELECTROSTATICS */
 862             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_add_pd(rinv11,_mm256_mul_pd(krf,rsq11)),crf));
 863             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
 864
 865             /* Update potential sum for this i atom from the interaction with this j atom. */
 866             velec            = _mm256_andnot_pd(dummy_mask,velec);
 867             velecsum         = _mm256_add_pd(velecsum,velec);
 868
 869             fscal            = felec;
 870
 871             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm256_mul_pd(fscal,dx11);
 875             ty               = _mm256_mul_pd(fscal,dy11);
 876             tz               = _mm256_mul_pd(fscal,dz11);
 877
 878             /* Update vectorial force */
 879             fix1             = _mm256_add_pd(fix1,tx);
 880             fiy1             = _mm256_add_pd(fiy1,ty);
 881             fiz1             = _mm256_add_pd(fiz1,tz);
 882
 883             fjx1             = _mm256_add_pd(fjx1,tx);
 884             fjy1             = _mm256_add_pd(fjy1,ty);
 885             fjz1             = _mm256_add_pd(fjz1,tz);
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_add_pd(rinv12,_mm256_mul_pd(krf,rsq12)),crf));
 893             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
 894
 895             /* Update potential sum for this i atom from the interaction with this j atom. */
 896             velec            = _mm256_andnot_pd(dummy_mask,velec);
 897             velecsum         = _mm256_add_pd(velecsum,velec);
 898
 899             fscal            = felec;
 900
 901             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 902
 903             /* Calculate temporary vectorial force */
 904             tx               = _mm256_mul_pd(fscal,dx12);
 905             ty               = _mm256_mul_pd(fscal,dy12);
 906             tz               = _mm256_mul_pd(fscal,dz12);
 907
 908             /* Update vectorial force */
 909             fix1             = _mm256_add_pd(fix1,tx);
 910             fiy1             = _mm256_add_pd(fiy1,ty);
 911             fiz1             = _mm256_add_pd(fiz1,tz);
 912
 913             fjx2             = _mm256_add_pd(fjx2,tx);
 914             fjy2             = _mm256_add_pd(fjy2,ty);
 915             fjz2             = _mm256_add_pd(fjz2,tz);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 923             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 924
 925             /* Update potential sum for this i atom from the interaction with this j atom. */
 926             velec            = _mm256_andnot_pd(dummy_mask,velec);
 927             velecsum         = _mm256_add_pd(velecsum,velec);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm256_mul_pd(fscal,dx20);
 935             ty               = _mm256_mul_pd(fscal,dy20);
 936             tz               = _mm256_mul_pd(fscal,dz20);
 937
 938             /* Update vectorial force */
 939             fix2             = _mm256_add_pd(fix2,tx);
 940             fiy2             = _mm256_add_pd(fiy2,ty);
 941             fiz2             = _mm256_add_pd(fiz2,tz);
 942
 943             fjx0             = _mm256_add_pd(fjx0,tx);
 944             fjy0             = _mm256_add_pd(fjy0,ty);
 945             fjz0             = _mm256_add_pd(fjz0,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             /* REACTION-FIELD ELECTROSTATICS */
 952             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_add_pd(rinv21,_mm256_mul_pd(krf,rsq21)),crf));
 953             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
 954
 955             /* Update potential sum for this i atom from the interaction with this j atom. */
 956             velec            = _mm256_andnot_pd(dummy_mask,velec);
 957             velecsum         = _mm256_add_pd(velecsum,velec);
 958
 959             fscal            = felec;
 960
 961             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm256_mul_pd(fscal,dx21);
 965             ty               = _mm256_mul_pd(fscal,dy21);
 966             tz               = _mm256_mul_pd(fscal,dz21);
 967
 968             /* Update vectorial force */
 969             fix2             = _mm256_add_pd(fix2,tx);
 970             fiy2             = _mm256_add_pd(fiy2,ty);
 971             fiz2             = _mm256_add_pd(fiz2,tz);
 972
 973             fjx1             = _mm256_add_pd(fjx1,tx);
 974             fjy1             = _mm256_add_pd(fjy1,ty);
 975             fjz1             = _mm256_add_pd(fjz1,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_add_pd(rinv22,_mm256_mul_pd(krf,rsq22)),crf));
 983             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
 984
 985             /* Update potential sum for this i atom from the interaction with this j atom. */
 986             velec            = _mm256_andnot_pd(dummy_mask,velec);
 987             velecsum         = _mm256_add_pd(velecsum,velec);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_pd(fscal,dx22);
 995             ty               = _mm256_mul_pd(fscal,dy22);
 996             tz               = _mm256_mul_pd(fscal,dz22);
 997
 998             /* Update vectorial force */
 999             fix2             = _mm256_add_pd(fix2,tx);
1000             fiy2             = _mm256_add_pd(fiy2,ty);
1001             fiz2             = _mm256_add_pd(fiz2,tz);
1002
1003             fjx2             = _mm256_add_pd(fjx2,tx);
1004             fjy2             = _mm256_add_pd(fjy2,ty);
1005             fjz2             = _mm256_add_pd(fjz2,tz);
1006
1007             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1008             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1009             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1010             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1011
1012             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1013                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1014
1015             /* Inner loop uses 324 flops */
1016         }
1017
1018         /* End of innermost loop */
1019
1020         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1021                                                  f+i_coord_offset,fshift+i_shift_offset);
1022
1023         ggid                        = gid[iidx];
1024         /* Update potential energies */
1025         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1026         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1027
1028         /* Increment number of inner iterations */
1029         inneriter                  += j_index_end - j_index_start;
1030
1031         /* Outer loop uses 20 flops */
1032     }
1033
1034     /* Increment number of outer iterations */
1035     outeriter        += nri;
1036
1037     /* Update outer/inner flops */
1038
1039     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*324);
1040 }
1041 /*
1042  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_double
1043  * Electrostatics interaction: ReactionField
1044  * VdW interaction:            CubicSplineTable
1045  * Geometry:                   Water3-Water3
1046  * Calculate force/pot:        Force
1047  */
1048 void
1049 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_256_double
1050                     (t_nblist                    * gmx_restrict       nlist,
1051                      rvec                        * gmx_restrict          xx,
1052                      rvec                        * gmx_restrict          ff,
1053                      struct t_forcerec           * gmx_restrict          fr,
1054                      t_mdatoms                   * gmx_restrict     mdatoms,
1055                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1056                      t_nrnb                      * gmx_restrict        nrnb)
1057 {
1058     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1059      * just 0 for non-waters.
1060      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1061      * jnr indices corresponding to data put in the four positions in the SIMD register.
1062      */
1063     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1064     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1065     int              jnrA,jnrB,jnrC,jnrD;
1066     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1067     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1068     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1069     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1070     real             rcutoff_scalar;
1071     real             *shiftvec,*fshift,*x,*f;
1072     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1073     real             scratch[4*DIM];
1074     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1075     real *           vdwioffsetptr0;
1076     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1077     real *           vdwioffsetptr1;
1078     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1079     real *           vdwioffsetptr2;
1080     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1081     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1082     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1083     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1084     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1085     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1086     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1087     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1088     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1089     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1090     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1091     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1092     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1093     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1094     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1095     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1096     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1097     real             *charge;
1098     int              nvdwtype;
1099     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1100     int              *vdwtype;
1101     real             *vdwparam;
1102     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1103     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1104     __m128i          vfitab;
1105     __m128i          ifour       = _mm_set1_epi32(4);
1106     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1107     real             *vftab;
1108     __m256d          dummy_mask,cutoff_mask;
1109     __m128           tmpmask0,tmpmask1;
1110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1111     __m256d          one     = _mm256_set1_pd(1.0);
1112     __m256d          two     = _mm256_set1_pd(2.0);
1113     x                = xx[0];
1114     f                = ff[0];
1115
1116     nri              = nlist->nri;
1117     iinr             = nlist->iinr;
1118     jindex           = nlist->jindex;
1119     jjnr             = nlist->jjnr;
1120     shiftidx         = nlist->shift;
1121     gid              = nlist->gid;
1122     shiftvec         = fr->shift_vec[0];
1123     fshift           = fr->fshift[0];
1124     facel            = _mm256_set1_pd(fr->ic->epsfac);
1125     charge           = mdatoms->chargeA;
1126     krf              = _mm256_set1_pd(fr->ic->k_rf);
1127     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
1128     crf              = _mm256_set1_pd(fr->ic->c_rf);
1129     nvdwtype         = fr->ntype;
1130     vdwparam         = fr->nbfp;
1131     vdwtype          = mdatoms->typeA;
1132
1133     vftab            = kernel_data->table_vdw->data;
1134     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1135
1136     /* Setup water-specific parameters */
1137     inr              = nlist->iinr[0];
1138     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1139     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1140     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1142
1143     jq0              = _mm256_set1_pd(charge[inr+0]);
1144     jq1              = _mm256_set1_pd(charge[inr+1]);
1145     jq2              = _mm256_set1_pd(charge[inr+2]);
1146     vdwjidx0A        = 2*vdwtype[inr+0];
1147     qq00             = _mm256_mul_pd(iq0,jq0);
1148     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1149     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1150     qq01             = _mm256_mul_pd(iq0,jq1);
1151     qq02             = _mm256_mul_pd(iq0,jq2);
1152     qq10             = _mm256_mul_pd(iq1,jq0);
1153     qq11             = _mm256_mul_pd(iq1,jq1);
1154     qq12             = _mm256_mul_pd(iq1,jq2);
1155     qq20             = _mm256_mul_pd(iq2,jq0);
1156     qq21             = _mm256_mul_pd(iq2,jq1);
1157     qq22             = _mm256_mul_pd(iq2,jq2);
1158
1159     /* Avoid stupid compiler warnings */
1160     jnrA = jnrB = jnrC = jnrD = 0;
1161     j_coord_offsetA = 0;
1162     j_coord_offsetB = 0;
1163     j_coord_offsetC = 0;
1164     j_coord_offsetD = 0;
1165
1166     outeriter        = 0;
1167     inneriter        = 0;
1168
1169     for(iidx=0;iidx<4*DIM;iidx++)
1170     {
1171         scratch[iidx] = 0.0;
1172     }
1173
1174     /* Start outer loop over neighborlists */
1175     for(iidx=0; iidx<nri; iidx++)
1176     {
1177         /* Load shift vector for this list */
1178         i_shift_offset   = DIM*shiftidx[iidx];
1179
1180         /* Load limits for loop over neighbors */
1181         j_index_start    = jindex[iidx];
1182         j_index_end      = jindex[iidx+1];
1183
1184         /* Get outer coordinate index */
1185         inr              = iinr[iidx];
1186         i_coord_offset   = DIM*inr;
1187
1188         /* Load i particle coords and add shift vector */
1189         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1190                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1191
1192         fix0             = _mm256_setzero_pd();
1193         fiy0             = _mm256_setzero_pd();
1194         fiz0             = _mm256_setzero_pd();
1195         fix1             = _mm256_setzero_pd();
1196         fiy1             = _mm256_setzero_pd();
1197         fiz1             = _mm256_setzero_pd();
1198         fix2             = _mm256_setzero_pd();
1199         fiy2             = _mm256_setzero_pd();
1200         fiz2             = _mm256_setzero_pd();
1201
1202         /* Start inner kernel loop */
1203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1204         {
1205
1206             /* Get j neighbor index, and coordinate index */
1207             jnrA             = jjnr[jidx];
1208             jnrB             = jjnr[jidx+1];
1209             jnrC             = jjnr[jidx+2];
1210             jnrD             = jjnr[jidx+3];
1211             j_coord_offsetA  = DIM*jnrA;
1212             j_coord_offsetB  = DIM*jnrB;
1213             j_coord_offsetC  = DIM*jnrC;
1214             j_coord_offsetD  = DIM*jnrD;
1215
1216             /* load j atom coordinates */
1217             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1219                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1220
1221             /* Calculate displacement vector */
1222             dx00             = _mm256_sub_pd(ix0,jx0);
1223             dy00             = _mm256_sub_pd(iy0,jy0);
1224             dz00             = _mm256_sub_pd(iz0,jz0);
1225             dx01             = _mm256_sub_pd(ix0,jx1);
1226             dy01             = _mm256_sub_pd(iy0,jy1);
1227             dz01             = _mm256_sub_pd(iz0,jz1);
1228             dx02             = _mm256_sub_pd(ix0,jx2);
1229             dy02             = _mm256_sub_pd(iy0,jy2);
1230             dz02             = _mm256_sub_pd(iz0,jz2);
1231             dx10             = _mm256_sub_pd(ix1,jx0);
1232             dy10             = _mm256_sub_pd(iy1,jy0);
1233             dz10             = _mm256_sub_pd(iz1,jz0);
1234             dx11             = _mm256_sub_pd(ix1,jx1);
1235             dy11             = _mm256_sub_pd(iy1,jy1);
1236             dz11             = _mm256_sub_pd(iz1,jz1);
1237             dx12             = _mm256_sub_pd(ix1,jx2);
1238             dy12             = _mm256_sub_pd(iy1,jy2);
1239             dz12             = _mm256_sub_pd(iz1,jz2);
1240             dx20             = _mm256_sub_pd(ix2,jx0);
1241             dy20             = _mm256_sub_pd(iy2,jy0);
1242             dz20             = _mm256_sub_pd(iz2,jz0);
1243             dx21             = _mm256_sub_pd(ix2,jx1);
1244             dy21             = _mm256_sub_pd(iy2,jy1);
1245             dz21             = _mm256_sub_pd(iz2,jz1);
1246             dx22             = _mm256_sub_pd(ix2,jx2);
1247             dy22             = _mm256_sub_pd(iy2,jy2);
1248             dz22             = _mm256_sub_pd(iz2,jz2);
1249
1250             /* Calculate squared distance and things based on it */
1251             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1252             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1253             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1254             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1255             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1256             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1257             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1258             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1259             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1260
1261             rinv00           = avx256_invsqrt_d(rsq00);
1262             rinv01           = avx256_invsqrt_d(rsq01);
1263             rinv02           = avx256_invsqrt_d(rsq02);
1264             rinv10           = avx256_invsqrt_d(rsq10);
1265             rinv11           = avx256_invsqrt_d(rsq11);
1266             rinv12           = avx256_invsqrt_d(rsq12);
1267             rinv20           = avx256_invsqrt_d(rsq20);
1268             rinv21           = avx256_invsqrt_d(rsq21);
1269             rinv22           = avx256_invsqrt_d(rsq22);
1270
1271             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1272             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1273             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1274             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1275             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1276             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1277             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1278             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1279             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1280
1281             fjx0             = _mm256_setzero_pd();
1282             fjy0             = _mm256_setzero_pd();
1283             fjz0             = _mm256_setzero_pd();
1284             fjx1             = _mm256_setzero_pd();
1285             fjy1             = _mm256_setzero_pd();
1286             fjz1             = _mm256_setzero_pd();
1287             fjx2             = _mm256_setzero_pd();
1288             fjy2             = _mm256_setzero_pd();
1289             fjz2             = _mm256_setzero_pd();
1290
1291             /**************************
1292              * CALCULATE INTERACTIONS *
1293              **************************/
1294
1295             r00              = _mm256_mul_pd(rsq00,rinv00);
1296
1297             /* Calculate table index by multiplying r with table scale and truncate to integer */
1298             rt               = _mm256_mul_pd(r00,vftabscale);
1299             vfitab           = _mm256_cvttpd_epi32(rt);
1300             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1301             vfitab           = _mm_slli_epi32(vfitab,3);
1302
1303             /* REACTION-FIELD ELECTROSTATICS */
1304             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1305
1306             /* CUBIC SPLINE TABLE DISPERSION */
1307             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1308             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1309             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1310             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1311             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1312             Heps             = _mm256_mul_pd(vfeps,H);
1313             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1314             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1315             fvdw6            = _mm256_mul_pd(c6_00,FF);
1316
1317             /* CUBIC SPLINE TABLE REPULSION */
1318             vfitab           = _mm_add_epi32(vfitab,ifour);
1319             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1320             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1321             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1322             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1323             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1324             Heps             = _mm256_mul_pd(vfeps,H);
1325             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1326             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1327             fvdw12           = _mm256_mul_pd(c12_00,FF);
1328             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1329
1330             fscal            = _mm256_add_pd(felec,fvdw);
1331
1332             /* Calculate temporary vectorial force */
1333             tx               = _mm256_mul_pd(fscal,dx00);
1334             ty               = _mm256_mul_pd(fscal,dy00);
1335             tz               = _mm256_mul_pd(fscal,dz00);
1336
1337             /* Update vectorial force */
1338             fix0             = _mm256_add_pd(fix0,tx);
1339             fiy0             = _mm256_add_pd(fiy0,ty);
1340             fiz0             = _mm256_add_pd(fiz0,tz);
1341
1342             fjx0             = _mm256_add_pd(fjx0,tx);
1343             fjy0             = _mm256_add_pd(fjy0,ty);
1344             fjz0             = _mm256_add_pd(fjz0,tz);
1345
1346             /**************************
1347              * CALCULATE INTERACTIONS *
1348              **************************/
1349
1350             /* REACTION-FIELD ELECTROSTATICS */
1351             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1352
1353             fscal            = felec;
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm256_mul_pd(fscal,dx01);
1357             ty               = _mm256_mul_pd(fscal,dy01);
1358             tz               = _mm256_mul_pd(fscal,dz01);
1359
1360             /* Update vectorial force */
1361             fix0             = _mm256_add_pd(fix0,tx);
1362             fiy0             = _mm256_add_pd(fiy0,ty);
1363             fiz0             = _mm256_add_pd(fiz0,tz);
1364
1365             fjx1             = _mm256_add_pd(fjx1,tx);
1366             fjy1             = _mm256_add_pd(fjy1,ty);
1367             fjz1             = _mm256_add_pd(fjz1,tz);
1368
1369             /**************************
1370              * CALCULATE INTERACTIONS *
1371              **************************/
1372
1373             /* REACTION-FIELD ELECTROSTATICS */
1374             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1375
1376             fscal            = felec;
1377
1378             /* Calculate temporary vectorial force */
1379             tx               = _mm256_mul_pd(fscal,dx02);
1380             ty               = _mm256_mul_pd(fscal,dy02);
1381             tz               = _mm256_mul_pd(fscal,dz02);
1382
1383             /* Update vectorial force */
1384             fix0             = _mm256_add_pd(fix0,tx);
1385             fiy0             = _mm256_add_pd(fiy0,ty);
1386             fiz0             = _mm256_add_pd(fiz0,tz);
1387
1388             fjx2             = _mm256_add_pd(fjx2,tx);
1389             fjy2             = _mm256_add_pd(fjy2,ty);
1390             fjz2             = _mm256_add_pd(fjz2,tz);
1391
1392             /**************************
1393              * CALCULATE INTERACTIONS *
1394              **************************/
1395
1396             /* REACTION-FIELD ELECTROSTATICS */
1397             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1398
1399             fscal            = felec;
1400
1401             /* Calculate temporary vectorial force */
1402             tx               = _mm256_mul_pd(fscal,dx10);
1403             ty               = _mm256_mul_pd(fscal,dy10);
1404             tz               = _mm256_mul_pd(fscal,dz10);
1405
1406             /* Update vectorial force */
1407             fix1             = _mm256_add_pd(fix1,tx);
1408             fiy1             = _mm256_add_pd(fiy1,ty);
1409             fiz1             = _mm256_add_pd(fiz1,tz);
1410
1411             fjx0             = _mm256_add_pd(fjx0,tx);
1412             fjy0             = _mm256_add_pd(fjy0,ty);
1413             fjz0             = _mm256_add_pd(fjz0,tz);
1414
1415             /**************************
1416              * CALCULATE INTERACTIONS *
1417              **************************/
1418
1419             /* REACTION-FIELD ELECTROSTATICS */
1420             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1421
1422             fscal            = felec;
1423
1424             /* Calculate temporary vectorial force */
1425             tx               = _mm256_mul_pd(fscal,dx11);
1426             ty               = _mm256_mul_pd(fscal,dy11);
1427             tz               = _mm256_mul_pd(fscal,dz11);
1428
1429             /* Update vectorial force */
1430             fix1             = _mm256_add_pd(fix1,tx);
1431             fiy1             = _mm256_add_pd(fiy1,ty);
1432             fiz1             = _mm256_add_pd(fiz1,tz);
1433
1434             fjx1             = _mm256_add_pd(fjx1,tx);
1435             fjy1             = _mm256_add_pd(fjy1,ty);
1436             fjz1             = _mm256_add_pd(fjz1,tz);
1437
1438             /**************************
1439              * CALCULATE INTERACTIONS *
1440              **************************/
1441
1442             /* REACTION-FIELD ELECTROSTATICS */
1443             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
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             /* REACTION-FIELD ELECTROSTATICS */
1466             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1467
1468             fscal            = felec;
1469
1470             /* Calculate temporary vectorial force */
1471             tx               = _mm256_mul_pd(fscal,dx20);
1472             ty               = _mm256_mul_pd(fscal,dy20);
1473             tz               = _mm256_mul_pd(fscal,dz20);
1474
1475             /* Update vectorial force */
1476             fix2             = _mm256_add_pd(fix2,tx);
1477             fiy2             = _mm256_add_pd(fiy2,ty);
1478             fiz2             = _mm256_add_pd(fiz2,tz);
1479
1480             fjx0             = _mm256_add_pd(fjx0,tx);
1481             fjy0             = _mm256_add_pd(fjy0,ty);
1482             fjz0             = _mm256_add_pd(fjz0,tz);
1483
1484             /**************************
1485              * CALCULATE INTERACTIONS *
1486              **************************/
1487
1488             /* REACTION-FIELD ELECTROSTATICS */
1489             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1490
1491             fscal            = felec;
1492
1493             /* Calculate temporary vectorial force */
1494             tx               = _mm256_mul_pd(fscal,dx21);
1495             ty               = _mm256_mul_pd(fscal,dy21);
1496             tz               = _mm256_mul_pd(fscal,dz21);
1497
1498             /* Update vectorial force */
1499             fix2             = _mm256_add_pd(fix2,tx);
1500             fiy2             = _mm256_add_pd(fiy2,ty);
1501             fiz2             = _mm256_add_pd(fiz2,tz);
1502
1503             fjx1             = _mm256_add_pd(fjx1,tx);
1504             fjy1             = _mm256_add_pd(fjy1,ty);
1505             fjz1             = _mm256_add_pd(fjz1,tz);
1506
1507             /**************************
1508              * CALCULATE INTERACTIONS *
1509              **************************/
1510
1511             /* REACTION-FIELD ELECTROSTATICS */
1512             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1513
1514             fscal            = felec;
1515
1516             /* Calculate temporary vectorial force */
1517             tx               = _mm256_mul_pd(fscal,dx22);
1518             ty               = _mm256_mul_pd(fscal,dy22);
1519             tz               = _mm256_mul_pd(fscal,dz22);
1520
1521             /* Update vectorial force */
1522             fix2             = _mm256_add_pd(fix2,tx);
1523             fiy2             = _mm256_add_pd(fiy2,ty);
1524             fiz2             = _mm256_add_pd(fiz2,tz);
1525
1526             fjx2             = _mm256_add_pd(fjx2,tx);
1527             fjy2             = _mm256_add_pd(fjy2,ty);
1528             fjz2             = _mm256_add_pd(fjz2,tz);
1529
1530             fjptrA             = f+j_coord_offsetA;
1531             fjptrB             = f+j_coord_offsetB;
1532             fjptrC             = f+j_coord_offsetC;
1533             fjptrD             = f+j_coord_offsetD;
1534
1535             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1536                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1537
1538             /* Inner loop uses 270 flops */
1539         }
1540
1541         if(jidx<j_index_end)
1542         {
1543
1544             /* Get j neighbor index, and coordinate index */
1545             jnrlistA         = jjnr[jidx];
1546             jnrlistB         = jjnr[jidx+1];
1547             jnrlistC         = jjnr[jidx+2];
1548             jnrlistD         = jjnr[jidx+3];
1549             /* Sign of each element will be negative for non-real atoms.
1550              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1551              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1552              */
1553             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1554
1555             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1556             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1557             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1558
1559             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1560             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1561             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1562             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1563             j_coord_offsetA  = DIM*jnrA;
1564             j_coord_offsetB  = DIM*jnrB;
1565             j_coord_offsetC  = DIM*jnrC;
1566             j_coord_offsetD  = DIM*jnrD;
1567
1568             /* load j atom coordinates */
1569             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1570                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1571                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1572
1573             /* Calculate displacement vector */
1574             dx00             = _mm256_sub_pd(ix0,jx0);
1575             dy00             = _mm256_sub_pd(iy0,jy0);
1576             dz00             = _mm256_sub_pd(iz0,jz0);
1577             dx01             = _mm256_sub_pd(ix0,jx1);
1578             dy01             = _mm256_sub_pd(iy0,jy1);
1579             dz01             = _mm256_sub_pd(iz0,jz1);
1580             dx02             = _mm256_sub_pd(ix0,jx2);
1581             dy02             = _mm256_sub_pd(iy0,jy2);
1582             dz02             = _mm256_sub_pd(iz0,jz2);
1583             dx10             = _mm256_sub_pd(ix1,jx0);
1584             dy10             = _mm256_sub_pd(iy1,jy0);
1585             dz10             = _mm256_sub_pd(iz1,jz0);
1586             dx11             = _mm256_sub_pd(ix1,jx1);
1587             dy11             = _mm256_sub_pd(iy1,jy1);
1588             dz11             = _mm256_sub_pd(iz1,jz1);
1589             dx12             = _mm256_sub_pd(ix1,jx2);
1590             dy12             = _mm256_sub_pd(iy1,jy2);
1591             dz12             = _mm256_sub_pd(iz1,jz2);
1592             dx20             = _mm256_sub_pd(ix2,jx0);
1593             dy20             = _mm256_sub_pd(iy2,jy0);
1594             dz20             = _mm256_sub_pd(iz2,jz0);
1595             dx21             = _mm256_sub_pd(ix2,jx1);
1596             dy21             = _mm256_sub_pd(iy2,jy1);
1597             dz21             = _mm256_sub_pd(iz2,jz1);
1598             dx22             = _mm256_sub_pd(ix2,jx2);
1599             dy22             = _mm256_sub_pd(iy2,jy2);
1600             dz22             = _mm256_sub_pd(iz2,jz2);
1601
1602             /* Calculate squared distance and things based on it */
1603             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1604             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1605             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1606             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1607             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1608             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1609             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1610             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1611             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1612
1613             rinv00           = avx256_invsqrt_d(rsq00);
1614             rinv01           = avx256_invsqrt_d(rsq01);
1615             rinv02           = avx256_invsqrt_d(rsq02);
1616             rinv10           = avx256_invsqrt_d(rsq10);
1617             rinv11           = avx256_invsqrt_d(rsq11);
1618             rinv12           = avx256_invsqrt_d(rsq12);
1619             rinv20           = avx256_invsqrt_d(rsq20);
1620             rinv21           = avx256_invsqrt_d(rsq21);
1621             rinv22           = avx256_invsqrt_d(rsq22);
1622
1623             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1624             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1625             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1626             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1627             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1628             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1629             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1630             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1631             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1632
1633             fjx0             = _mm256_setzero_pd();
1634             fjy0             = _mm256_setzero_pd();
1635             fjz0             = _mm256_setzero_pd();
1636             fjx1             = _mm256_setzero_pd();
1637             fjy1             = _mm256_setzero_pd();
1638             fjz1             = _mm256_setzero_pd();
1639             fjx2             = _mm256_setzero_pd();
1640             fjy2             = _mm256_setzero_pd();
1641             fjz2             = _mm256_setzero_pd();
1642
1643             /**************************
1644              * CALCULATE INTERACTIONS *
1645              **************************/
1646
1647             r00              = _mm256_mul_pd(rsq00,rinv00);
1648             r00              = _mm256_andnot_pd(dummy_mask,r00);
1649
1650             /* Calculate table index by multiplying r with table scale and truncate to integer */
1651             rt               = _mm256_mul_pd(r00,vftabscale);
1652             vfitab           = _mm256_cvttpd_epi32(rt);
1653             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1654             vfitab           = _mm_slli_epi32(vfitab,3);
1655
1656             /* REACTION-FIELD ELECTROSTATICS */
1657             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1658
1659             /* CUBIC SPLINE TABLE DISPERSION */
1660             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1661             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1662             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1663             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1664             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1665             Heps             = _mm256_mul_pd(vfeps,H);
1666             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1667             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1668             fvdw6            = _mm256_mul_pd(c6_00,FF);
1669
1670             /* CUBIC SPLINE TABLE REPULSION */
1671             vfitab           = _mm_add_epi32(vfitab,ifour);
1672             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1673             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1674             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1675             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1676             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1677             Heps             = _mm256_mul_pd(vfeps,H);
1678             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1679             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1680             fvdw12           = _mm256_mul_pd(c12_00,FF);
1681             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1682
1683             fscal            = _mm256_add_pd(felec,fvdw);
1684
1685             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1686
1687             /* Calculate temporary vectorial force */
1688             tx               = _mm256_mul_pd(fscal,dx00);
1689             ty               = _mm256_mul_pd(fscal,dy00);
1690             tz               = _mm256_mul_pd(fscal,dz00);
1691
1692             /* Update vectorial force */
1693             fix0             = _mm256_add_pd(fix0,tx);
1694             fiy0             = _mm256_add_pd(fiy0,ty);
1695             fiz0             = _mm256_add_pd(fiz0,tz);
1696
1697             fjx0             = _mm256_add_pd(fjx0,tx);
1698             fjy0             = _mm256_add_pd(fjy0,ty);
1699             fjz0             = _mm256_add_pd(fjz0,tz);
1700
1701             /**************************
1702              * CALCULATE INTERACTIONS *
1703              **************************/
1704
1705             /* REACTION-FIELD ELECTROSTATICS */
1706             felec            = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_mul_pd(rinv01,rinvsq01),krf2));
1707
1708             fscal            = felec;
1709
1710             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1711
1712             /* Calculate temporary vectorial force */
1713             tx               = _mm256_mul_pd(fscal,dx01);
1714             ty               = _mm256_mul_pd(fscal,dy01);
1715             tz               = _mm256_mul_pd(fscal,dz01);
1716
1717             /* Update vectorial force */
1718             fix0             = _mm256_add_pd(fix0,tx);
1719             fiy0             = _mm256_add_pd(fiy0,ty);
1720             fiz0             = _mm256_add_pd(fiz0,tz);
1721
1722             fjx1             = _mm256_add_pd(fjx1,tx);
1723             fjy1             = _mm256_add_pd(fjy1,ty);
1724             fjz1             = _mm256_add_pd(fjz1,tz);
1725
1726             /**************************
1727              * CALCULATE INTERACTIONS *
1728              **************************/
1729
1730             /* REACTION-FIELD ELECTROSTATICS */
1731             felec            = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_mul_pd(rinv02,rinvsq02),krf2));
1732
1733             fscal            = felec;
1734
1735             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1736
1737             /* Calculate temporary vectorial force */
1738             tx               = _mm256_mul_pd(fscal,dx02);
1739             ty               = _mm256_mul_pd(fscal,dy02);
1740             tz               = _mm256_mul_pd(fscal,dz02);
1741
1742             /* Update vectorial force */
1743             fix0             = _mm256_add_pd(fix0,tx);
1744             fiy0             = _mm256_add_pd(fiy0,ty);
1745             fiz0             = _mm256_add_pd(fiz0,tz);
1746
1747             fjx2             = _mm256_add_pd(fjx2,tx);
1748             fjy2             = _mm256_add_pd(fjy2,ty);
1749             fjz2             = _mm256_add_pd(fjz2,tz);
1750
1751             /**************************
1752              * CALCULATE INTERACTIONS *
1753              **************************/
1754
1755             /* REACTION-FIELD ELECTROSTATICS */
1756             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1757
1758             fscal            = felec;
1759
1760             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1761
1762             /* Calculate temporary vectorial force */
1763             tx               = _mm256_mul_pd(fscal,dx10);
1764             ty               = _mm256_mul_pd(fscal,dy10);
1765             tz               = _mm256_mul_pd(fscal,dz10);
1766
1767             /* Update vectorial force */
1768             fix1             = _mm256_add_pd(fix1,tx);
1769             fiy1             = _mm256_add_pd(fiy1,ty);
1770             fiz1             = _mm256_add_pd(fiz1,tz);
1771
1772             fjx0             = _mm256_add_pd(fjx0,tx);
1773             fjy0             = _mm256_add_pd(fjy0,ty);
1774             fjz0             = _mm256_add_pd(fjz0,tz);
1775
1776             /**************************
1777              * CALCULATE INTERACTIONS *
1778              **************************/
1779
1780             /* REACTION-FIELD ELECTROSTATICS */
1781             felec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_mul_pd(rinv11,rinvsq11),krf2));
1782
1783             fscal            = felec;
1784
1785             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1786
1787             /* Calculate temporary vectorial force */
1788             tx               = _mm256_mul_pd(fscal,dx11);
1789             ty               = _mm256_mul_pd(fscal,dy11);
1790             tz               = _mm256_mul_pd(fscal,dz11);
1791
1792             /* Update vectorial force */
1793             fix1             = _mm256_add_pd(fix1,tx);
1794             fiy1             = _mm256_add_pd(fiy1,ty);
1795             fiz1             = _mm256_add_pd(fiz1,tz);
1796
1797             fjx1             = _mm256_add_pd(fjx1,tx);
1798             fjy1             = _mm256_add_pd(fjy1,ty);
1799             fjz1             = _mm256_add_pd(fjz1,tz);
1800
1801             /**************************
1802              * CALCULATE INTERACTIONS *
1803              **************************/
1804
1805             /* REACTION-FIELD ELECTROSTATICS */
1806             felec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_mul_pd(rinv12,rinvsq12),krf2));
1807
1808             fscal            = felec;
1809
1810             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1811
1812             /* Calculate temporary vectorial force */
1813             tx               = _mm256_mul_pd(fscal,dx12);
1814             ty               = _mm256_mul_pd(fscal,dy12);
1815             tz               = _mm256_mul_pd(fscal,dz12);
1816
1817             /* Update vectorial force */
1818             fix1             = _mm256_add_pd(fix1,tx);
1819             fiy1             = _mm256_add_pd(fiy1,ty);
1820             fiz1             = _mm256_add_pd(fiz1,tz);
1821
1822             fjx2             = _mm256_add_pd(fjx2,tx);
1823             fjy2             = _mm256_add_pd(fjy2,ty);
1824             fjz2             = _mm256_add_pd(fjz2,tz);
1825
1826             /**************************
1827              * CALCULATE INTERACTIONS *
1828              **************************/
1829
1830             /* REACTION-FIELD ELECTROSTATICS */
1831             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1832
1833             fscal            = felec;
1834
1835             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1836
1837             /* Calculate temporary vectorial force */
1838             tx               = _mm256_mul_pd(fscal,dx20);
1839             ty               = _mm256_mul_pd(fscal,dy20);
1840             tz               = _mm256_mul_pd(fscal,dz20);
1841
1842             /* Update vectorial force */
1843             fix2             = _mm256_add_pd(fix2,tx);
1844             fiy2             = _mm256_add_pd(fiy2,ty);
1845             fiz2             = _mm256_add_pd(fiz2,tz);
1846
1847             fjx0             = _mm256_add_pd(fjx0,tx);
1848             fjy0             = _mm256_add_pd(fjy0,ty);
1849             fjz0             = _mm256_add_pd(fjz0,tz);
1850
1851             /**************************
1852              * CALCULATE INTERACTIONS *
1853              **************************/
1854
1855             /* REACTION-FIELD ELECTROSTATICS */
1856             felec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_mul_pd(rinv21,rinvsq21),krf2));
1857
1858             fscal            = felec;
1859
1860             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1861
1862             /* Calculate temporary vectorial force */
1863             tx               = _mm256_mul_pd(fscal,dx21);
1864             ty               = _mm256_mul_pd(fscal,dy21);
1865             tz               = _mm256_mul_pd(fscal,dz21);
1866
1867             /* Update vectorial force */
1868             fix2             = _mm256_add_pd(fix2,tx);
1869             fiy2             = _mm256_add_pd(fiy2,ty);
1870             fiz2             = _mm256_add_pd(fiz2,tz);
1871
1872             fjx1             = _mm256_add_pd(fjx1,tx);
1873             fjy1             = _mm256_add_pd(fjy1,ty);
1874             fjz1             = _mm256_add_pd(fjz1,tz);
1875
1876             /**************************
1877              * CALCULATE INTERACTIONS *
1878              **************************/
1879
1880             /* REACTION-FIELD ELECTROSTATICS */
1881             felec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_mul_pd(rinv22,rinvsq22),krf2));
1882
1883             fscal            = felec;
1884
1885             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1886
1887             /* Calculate temporary vectorial force */
1888             tx               = _mm256_mul_pd(fscal,dx22);
1889             ty               = _mm256_mul_pd(fscal,dy22);
1890             tz               = _mm256_mul_pd(fscal,dz22);
1891
1892             /* Update vectorial force */
1893             fix2             = _mm256_add_pd(fix2,tx);
1894             fiy2             = _mm256_add_pd(fiy2,ty);
1895             fiz2             = _mm256_add_pd(fiz2,tz);
1896
1897             fjx2             = _mm256_add_pd(fjx2,tx);
1898             fjy2             = _mm256_add_pd(fjy2,ty);
1899             fjz2             = _mm256_add_pd(fjz2,tz);
1900
1901             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1902             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1903             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1904             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1905
1906             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1907                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1908
1909             /* Inner loop uses 271 flops */
1910         }
1911
1912         /* End of innermost loop */
1913
1914         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1915                                                  f+i_coord_offset,fshift+i_shift_offset);
1916
1917         /* Increment number of inner iterations */
1918         inneriter                  += j_index_end - j_index_start;
1919
1920         /* Outer loop uses 18 flops */
1921     }
1922
1923     /* Increment number of outer iterations */
1924     outeriter        += nri;
1925
1926     /* Update outer/inner flops */
1927
1928     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*271);
1929 }