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

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  79     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  80     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  81     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  82     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  84     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  85     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  86     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  98     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_vdw->data;
 126     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 131     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 132     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     jq0              = _mm256_set1_pd(charge[inr+0]);
 136     jq1              = _mm256_set1_pd(charge[inr+1]);
 137     jq2              = _mm256_set1_pd(charge[inr+2]);
 138     vdwjidx0A        = 2*vdwtype[inr+0];
 139     qq00             = _mm256_mul_pd(iq0,jq0);
 140     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 141     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 142     qq01             = _mm256_mul_pd(iq0,jq1);
 143     qq02             = _mm256_mul_pd(iq0,jq2);
 144     qq10             = _mm256_mul_pd(iq1,jq0);
 145     qq11             = _mm256_mul_pd(iq1,jq1);
 146     qq12             = _mm256_mul_pd(iq1,jq2);
 147     qq20             = _mm256_mul_pd(iq2,jq0);
 148     qq21             = _mm256_mul_pd(iq2,jq1);
 149     qq22             = _mm256_mul_pd(iq2,jq2);
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm256_setzero_pd();
 185         fiy0             = _mm256_setzero_pd();
 186         fiz0             = _mm256_setzero_pd();
 187         fix1             = _mm256_setzero_pd();
 188         fiy1             = _mm256_setzero_pd();
 189         fiz1             = _mm256_setzero_pd();
 190         fix2             = _mm256_setzero_pd();
 191         fiy2             = _mm256_setzero_pd();
 192         fiz2             = _mm256_setzero_pd();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_pd();
 196         vvdwsum          = _mm256_setzero_pd();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm256_sub_pd(ix0,jx0);
 219             dy00             = _mm256_sub_pd(iy0,jy0);
 220             dz00             = _mm256_sub_pd(iz0,jz0);
 221             dx01             = _mm256_sub_pd(ix0,jx1);
 222             dy01             = _mm256_sub_pd(iy0,jy1);
 223             dz01             = _mm256_sub_pd(iz0,jz1);
 224             dx02             = _mm256_sub_pd(ix0,jx2);
 225             dy02             = _mm256_sub_pd(iy0,jy2);
 226             dz02             = _mm256_sub_pd(iz0,jz2);
 227             dx10             = _mm256_sub_pd(ix1,jx0);
 228             dy10             = _mm256_sub_pd(iy1,jy0);
 229             dz10             = _mm256_sub_pd(iz1,jz0);
 230             dx11             = _mm256_sub_pd(ix1,jx1);
 231             dy11             = _mm256_sub_pd(iy1,jy1);
 232             dz11             = _mm256_sub_pd(iz1,jz1);
 233             dx12             = _mm256_sub_pd(ix1,jx2);
 234             dy12             = _mm256_sub_pd(iy1,jy2);
 235             dz12             = _mm256_sub_pd(iz1,jz2);
 236             dx20             = _mm256_sub_pd(ix2,jx0);
 237             dy20             = _mm256_sub_pd(iy2,jy0);
 238             dz20             = _mm256_sub_pd(iz2,jz0);
 239             dx21             = _mm256_sub_pd(ix2,jx1);
 240             dy21             = _mm256_sub_pd(iy2,jy1);
 241             dz21             = _mm256_sub_pd(iz2,jz1);
 242             dx22             = _mm256_sub_pd(ix2,jx2);
 243             dy22             = _mm256_sub_pd(iy2,jy2);
 244             dz22             = _mm256_sub_pd(iz2,jz2);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 248             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 249             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 250             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 251             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 252             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 253             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 254             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 255             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 256
 257             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 258             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 259             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 260             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 261             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 262             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 263             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 264             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 265             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 266
 267             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 268             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 269             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 270             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 271             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 272             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 273             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 274             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 275             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 276
 277             fjx0             = _mm256_setzero_pd();
 278             fjy0             = _mm256_setzero_pd();
 279             fjz0             = _mm256_setzero_pd();
 280             fjx1             = _mm256_setzero_pd();
 281             fjy1             = _mm256_setzero_pd();
 282             fjz1             = _mm256_setzero_pd();
 283             fjx2             = _mm256_setzero_pd();
 284             fjy2             = _mm256_setzero_pd();
 285             fjz2             = _mm256_setzero_pd();
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             r00              = _mm256_mul_pd(rsq00,rinv00);
 292
 293             /* Calculate table index by multiplying r with table scale and truncate to integer */
 294             rt               = _mm256_mul_pd(r00,vftabscale);
 295             vfitab           = _mm256_cvttpd_epi32(rt);
 296             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 297             vfitab           = _mm_slli_epi32(vfitab,3);
 298
 299             /* COULOMB ELECTROSTATICS */
 300             velec            = _mm256_mul_pd(qq00,rinv00);
 301             felec            = _mm256_mul_pd(velec,rinvsq00);
 302
 303             /* CUBIC SPLINE TABLE DISPERSION */
 304             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 305             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 306             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 307             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 308             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 309             Heps             = _mm256_mul_pd(vfeps,H);
 310             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 311             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 312             vvdw6            = _mm256_mul_pd(c6_00,VV);
 313             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 314             fvdw6            = _mm256_mul_pd(c6_00,FF);
 315
 316             /* CUBIC SPLINE TABLE REPULSION */
 317             vfitab           = _mm_add_epi32(vfitab,ifour);
 318             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 319             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 320             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 321             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 322             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 323             Heps             = _mm256_mul_pd(vfeps,H);
 324             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 325             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 326             vvdw12           = _mm256_mul_pd(c12_00,VV);
 327             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 328             fvdw12           = _mm256_mul_pd(c12_00,FF);
 329             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 330             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm256_add_pd(velecsum,velec);
 334             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 335
 336             fscal            = _mm256_add_pd(felec,fvdw);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_pd(fscal,dx00);
 340             ty               = _mm256_mul_pd(fscal,dy00);
 341             tz               = _mm256_mul_pd(fscal,dz00);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm256_add_pd(fix0,tx);
 345             fiy0             = _mm256_add_pd(fiy0,ty);
 346             fiz0             = _mm256_add_pd(fiz0,tz);
 347
 348             fjx0             = _mm256_add_pd(fjx0,tx);
 349             fjy0             = _mm256_add_pd(fjy0,ty);
 350             fjz0             = _mm256_add_pd(fjz0,tz);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             /* COULOMB ELECTROSTATICS */
 357             velec            = _mm256_mul_pd(qq01,rinv01);
 358             felec            = _mm256_mul_pd(velec,rinvsq01);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm256_mul_pd(fscal,dx01);
 367             ty               = _mm256_mul_pd(fscal,dy01);
 368             tz               = _mm256_mul_pd(fscal,dz01);
 369
 370             /* Update vectorial force */
 371             fix0             = _mm256_add_pd(fix0,tx);
 372             fiy0             = _mm256_add_pd(fiy0,ty);
 373             fiz0             = _mm256_add_pd(fiz0,tz);
 374
 375             fjx1             = _mm256_add_pd(fjx1,tx);
 376             fjy1             = _mm256_add_pd(fjy1,ty);
 377             fjz1             = _mm256_add_pd(fjz1,tz);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             /* COULOMB ELECTROSTATICS */
 384             velec            = _mm256_mul_pd(qq02,rinv02);
 385             felec            = _mm256_mul_pd(velec,rinvsq02);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm256_add_pd(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm256_mul_pd(fscal,dx02);
 394             ty               = _mm256_mul_pd(fscal,dy02);
 395             tz               = _mm256_mul_pd(fscal,dz02);
 396
 397             /* Update vectorial force */
 398             fix0             = _mm256_add_pd(fix0,tx);
 399             fiy0             = _mm256_add_pd(fiy0,ty);
 400             fiz0             = _mm256_add_pd(fiz0,tz);
 401
 402             fjx2             = _mm256_add_pd(fjx2,tx);
 403             fjy2             = _mm256_add_pd(fjy2,ty);
 404             fjz2             = _mm256_add_pd(fjz2,tz);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             /* COULOMB ELECTROSTATICS */
 411             velec            = _mm256_mul_pd(qq10,rinv10);
 412             felec            = _mm256_mul_pd(velec,rinvsq10);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velecsum         = _mm256_add_pd(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_pd(fscal,dx10);
 421             ty               = _mm256_mul_pd(fscal,dy10);
 422             tz               = _mm256_mul_pd(fscal,dz10);
 423
 424             /* Update vectorial force */
 425             fix1             = _mm256_add_pd(fix1,tx);
 426             fiy1             = _mm256_add_pd(fiy1,ty);
 427             fiz1             = _mm256_add_pd(fiz1,tz);
 428
 429             fjx0             = _mm256_add_pd(fjx0,tx);
 430             fjy0             = _mm256_add_pd(fjy0,ty);
 431             fjz0             = _mm256_add_pd(fjz0,tz);
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             /* COULOMB ELECTROSTATICS */
 438             velec            = _mm256_mul_pd(qq11,rinv11);
 439             felec            = _mm256_mul_pd(velec,rinvsq11);
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velecsum         = _mm256_add_pd(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_pd(fscal,dx11);
 448             ty               = _mm256_mul_pd(fscal,dy11);
 449             tz               = _mm256_mul_pd(fscal,dz11);
 450
 451             /* Update vectorial force */
 452             fix1             = _mm256_add_pd(fix1,tx);
 453             fiy1             = _mm256_add_pd(fiy1,ty);
 454             fiz1             = _mm256_add_pd(fiz1,tz);
 455
 456             fjx1             = _mm256_add_pd(fjx1,tx);
 457             fjy1             = _mm256_add_pd(fjy1,ty);
 458             fjz1             = _mm256_add_pd(fjz1,tz);
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             /* COULOMB ELECTROSTATICS */
 465             velec            = _mm256_mul_pd(qq12,rinv12);
 466             felec            = _mm256_mul_pd(velec,rinvsq12);
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             velecsum         = _mm256_add_pd(velecsum,velec);
 470
 471             fscal            = felec;
 472
 473             /* Calculate temporary vectorial force */
 474             tx               = _mm256_mul_pd(fscal,dx12);
 475             ty               = _mm256_mul_pd(fscal,dy12);
 476             tz               = _mm256_mul_pd(fscal,dz12);
 477
 478             /* Update vectorial force */
 479             fix1             = _mm256_add_pd(fix1,tx);
 480             fiy1             = _mm256_add_pd(fiy1,ty);
 481             fiz1             = _mm256_add_pd(fiz1,tz);
 482
 483             fjx2             = _mm256_add_pd(fjx2,tx);
 484             fjy2             = _mm256_add_pd(fjy2,ty);
 485             fjz2             = _mm256_add_pd(fjz2,tz);
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             /* COULOMB ELECTROSTATICS */
 492             velec            = _mm256_mul_pd(qq20,rinv20);
 493             felec            = _mm256_mul_pd(velec,rinvsq20);
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velecsum         = _mm256_add_pd(velecsum,velec);
 497
 498             fscal            = felec;
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm256_mul_pd(fscal,dx20);
 502             ty               = _mm256_mul_pd(fscal,dy20);
 503             tz               = _mm256_mul_pd(fscal,dz20);
 504
 505             /* Update vectorial force */
 506             fix2             = _mm256_add_pd(fix2,tx);
 507             fiy2             = _mm256_add_pd(fiy2,ty);
 508             fiz2             = _mm256_add_pd(fiz2,tz);
 509
 510             fjx0             = _mm256_add_pd(fjx0,tx);
 511             fjy0             = _mm256_add_pd(fjy0,ty);
 512             fjz0             = _mm256_add_pd(fjz0,tz);
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             /* COULOMB ELECTROSTATICS */
 519             velec            = _mm256_mul_pd(qq21,rinv21);
 520             felec            = _mm256_mul_pd(velec,rinvsq21);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velecsum         = _mm256_add_pd(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm256_mul_pd(fscal,dx21);
 529             ty               = _mm256_mul_pd(fscal,dy21);
 530             tz               = _mm256_mul_pd(fscal,dz21);
 531
 532             /* Update vectorial force */
 533             fix2             = _mm256_add_pd(fix2,tx);
 534             fiy2             = _mm256_add_pd(fiy2,ty);
 535             fiz2             = _mm256_add_pd(fiz2,tz);
 536
 537             fjx1             = _mm256_add_pd(fjx1,tx);
 538             fjy1             = _mm256_add_pd(fjy1,ty);
 539             fjz1             = _mm256_add_pd(fjz1,tz);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             /* COULOMB ELECTROSTATICS */
 546             velec            = _mm256_mul_pd(qq22,rinv22);
 547             felec            = _mm256_mul_pd(velec,rinvsq22);
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velecsum         = _mm256_add_pd(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm256_mul_pd(fscal,dx22);
 556             ty               = _mm256_mul_pd(fscal,dy22);
 557             tz               = _mm256_mul_pd(fscal,dz22);
 558
 559             /* Update vectorial force */
 560             fix2             = _mm256_add_pd(fix2,tx);
 561             fiy2             = _mm256_add_pd(fiy2,ty);
 562             fiz2             = _mm256_add_pd(fiz2,tz);
 563
 564             fjx2             = _mm256_add_pd(fjx2,tx);
 565             fjy2             = _mm256_add_pd(fjy2,ty);
 566             fjz2             = _mm256_add_pd(fjz2,tz);
 567
 568             fjptrA             = f+j_coord_offsetA;
 569             fjptrB             = f+j_coord_offsetB;
 570             fjptrC             = f+j_coord_offsetC;
 571             fjptrD             = f+j_coord_offsetD;
 572
 573             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 574                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 575
 576             /* Inner loop uses 278 flops */
 577         }
 578
 579         if(jidx<j_index_end)
 580         {
 581
 582             /* Get j neighbor index, and coordinate index */
 583             jnrlistA         = jjnr[jidx];
 584             jnrlistB         = jjnr[jidx+1];
 585             jnrlistC         = jjnr[jidx+2];
 586             jnrlistD         = jjnr[jidx+3];
 587             /* Sign of each element will be negative for non-real atoms.
 588              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 589              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 590              */
 591             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 592
 593             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 594             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 595             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 596
 597             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 598             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 599             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 600             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 601             j_coord_offsetA  = DIM*jnrA;
 602             j_coord_offsetB  = DIM*jnrB;
 603             j_coord_offsetC  = DIM*jnrC;
 604             j_coord_offsetD  = DIM*jnrD;
 605
 606             /* load j atom coordinates */
 607             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 608                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 609                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 610
 611             /* Calculate displacement vector */
 612             dx00             = _mm256_sub_pd(ix0,jx0);
 613             dy00             = _mm256_sub_pd(iy0,jy0);
 614             dz00             = _mm256_sub_pd(iz0,jz0);
 615             dx01             = _mm256_sub_pd(ix0,jx1);
 616             dy01             = _mm256_sub_pd(iy0,jy1);
 617             dz01             = _mm256_sub_pd(iz0,jz1);
 618             dx02             = _mm256_sub_pd(ix0,jx2);
 619             dy02             = _mm256_sub_pd(iy0,jy2);
 620             dz02             = _mm256_sub_pd(iz0,jz2);
 621             dx10             = _mm256_sub_pd(ix1,jx0);
 622             dy10             = _mm256_sub_pd(iy1,jy0);
 623             dz10             = _mm256_sub_pd(iz1,jz0);
 624             dx11             = _mm256_sub_pd(ix1,jx1);
 625             dy11             = _mm256_sub_pd(iy1,jy1);
 626             dz11             = _mm256_sub_pd(iz1,jz1);
 627             dx12             = _mm256_sub_pd(ix1,jx2);
 628             dy12             = _mm256_sub_pd(iy1,jy2);
 629             dz12             = _mm256_sub_pd(iz1,jz2);
 630             dx20             = _mm256_sub_pd(ix2,jx0);
 631             dy20             = _mm256_sub_pd(iy2,jy0);
 632             dz20             = _mm256_sub_pd(iz2,jz0);
 633             dx21             = _mm256_sub_pd(ix2,jx1);
 634             dy21             = _mm256_sub_pd(iy2,jy1);
 635             dz21             = _mm256_sub_pd(iz2,jz1);
 636             dx22             = _mm256_sub_pd(ix2,jx2);
 637             dy22             = _mm256_sub_pd(iy2,jy2);
 638             dz22             = _mm256_sub_pd(iz2,jz2);
 639
 640             /* Calculate squared distance and things based on it */
 641             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 642             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 643             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 644             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 645             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 646             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 647             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 648             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 649             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 650
 651             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 652             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 653             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 654             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 655             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 656             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 657             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 658             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 659             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 660
 661             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 662             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 663             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 664             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 665             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 666             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 667             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 668             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 669             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 670
 671             fjx0             = _mm256_setzero_pd();
 672             fjy0             = _mm256_setzero_pd();
 673             fjz0             = _mm256_setzero_pd();
 674             fjx1             = _mm256_setzero_pd();
 675             fjy1             = _mm256_setzero_pd();
 676             fjz1             = _mm256_setzero_pd();
 677             fjx2             = _mm256_setzero_pd();
 678             fjy2             = _mm256_setzero_pd();
 679             fjz2             = _mm256_setzero_pd();
 680
 681             /**************************
 682              * CALCULATE INTERACTIONS *
 683              **************************/
 684
 685             r00              = _mm256_mul_pd(rsq00,rinv00);
 686             r00              = _mm256_andnot_pd(dummy_mask,r00);
 687
 688             /* Calculate table index by multiplying r with table scale and truncate to integer */
 689             rt               = _mm256_mul_pd(r00,vftabscale);
 690             vfitab           = _mm256_cvttpd_epi32(rt);
 691             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 692             vfitab           = _mm_slli_epi32(vfitab,3);
 693
 694             /* COULOMB ELECTROSTATICS */
 695             velec            = _mm256_mul_pd(qq00,rinv00);
 696             felec            = _mm256_mul_pd(velec,rinvsq00);
 697
 698             /* CUBIC SPLINE TABLE DISPERSION */
 699             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 700             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 701             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 702             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 703             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 704             Heps             = _mm256_mul_pd(vfeps,H);
 705             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 706             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 707             vvdw6            = _mm256_mul_pd(c6_00,VV);
 708             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 709             fvdw6            = _mm256_mul_pd(c6_00,FF);
 710
 711             /* CUBIC SPLINE TABLE REPULSION */
 712             vfitab           = _mm_add_epi32(vfitab,ifour);
 713             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 714             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 715             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 716             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 717             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 718             Heps             = _mm256_mul_pd(vfeps,H);
 719             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 720             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 721             vvdw12           = _mm256_mul_pd(c12_00,VV);
 722             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 723             fvdw12           = _mm256_mul_pd(c12_00,FF);
 724             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 725             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 726
 727             /* Update potential sum for this i atom from the interaction with this j atom. */
 728             velec            = _mm256_andnot_pd(dummy_mask,velec);
 729             velecsum         = _mm256_add_pd(velecsum,velec);
 730             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 731             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 732
 733             fscal            = _mm256_add_pd(felec,fvdw);
 734
 735             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 736
 737             /* Calculate temporary vectorial force */
 738             tx               = _mm256_mul_pd(fscal,dx00);
 739             ty               = _mm256_mul_pd(fscal,dy00);
 740             tz               = _mm256_mul_pd(fscal,dz00);
 741
 742             /* Update vectorial force */
 743             fix0             = _mm256_add_pd(fix0,tx);
 744             fiy0             = _mm256_add_pd(fiy0,ty);
 745             fiz0             = _mm256_add_pd(fiz0,tz);
 746
 747             fjx0             = _mm256_add_pd(fjx0,tx);
 748             fjy0             = _mm256_add_pd(fjy0,ty);
 749             fjz0             = _mm256_add_pd(fjz0,tz);
 750
 751             /**************************
 752              * CALCULATE INTERACTIONS *
 753              **************************/
 754
 755             /* COULOMB ELECTROSTATICS */
 756             velec            = _mm256_mul_pd(qq01,rinv01);
 757             felec            = _mm256_mul_pd(velec,rinvsq01);
 758
 759             /* Update potential sum for this i atom from the interaction with this j atom. */
 760             velec            = _mm256_andnot_pd(dummy_mask,velec);
 761             velecsum         = _mm256_add_pd(velecsum,velec);
 762
 763             fscal            = felec;
 764
 765             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 766
 767             /* Calculate temporary vectorial force */
 768             tx               = _mm256_mul_pd(fscal,dx01);
 769             ty               = _mm256_mul_pd(fscal,dy01);
 770             tz               = _mm256_mul_pd(fscal,dz01);
 771
 772             /* Update vectorial force */
 773             fix0             = _mm256_add_pd(fix0,tx);
 774             fiy0             = _mm256_add_pd(fiy0,ty);
 775             fiz0             = _mm256_add_pd(fiz0,tz);
 776
 777             fjx1             = _mm256_add_pd(fjx1,tx);
 778             fjy1             = _mm256_add_pd(fjy1,ty);
 779             fjz1             = _mm256_add_pd(fjz1,tz);
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             /* COULOMB ELECTROSTATICS */
 786             velec            = _mm256_mul_pd(qq02,rinv02);
 787             felec            = _mm256_mul_pd(velec,rinvsq02);
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm256_andnot_pd(dummy_mask,velec);
 791             velecsum         = _mm256_add_pd(velecsum,velec);
 792
 793             fscal            = felec;
 794
 795             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 796
 797             /* Calculate temporary vectorial force */
 798             tx               = _mm256_mul_pd(fscal,dx02);
 799             ty               = _mm256_mul_pd(fscal,dy02);
 800             tz               = _mm256_mul_pd(fscal,dz02);
 801
 802             /* Update vectorial force */
 803             fix0             = _mm256_add_pd(fix0,tx);
 804             fiy0             = _mm256_add_pd(fiy0,ty);
 805             fiz0             = _mm256_add_pd(fiz0,tz);
 806
 807             fjx2             = _mm256_add_pd(fjx2,tx);
 808             fjy2             = _mm256_add_pd(fjy2,ty);
 809             fjz2             = _mm256_add_pd(fjz2,tz);
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* COULOMB ELECTROSTATICS */
 816             velec            = _mm256_mul_pd(qq10,rinv10);
 817             felec            = _mm256_mul_pd(velec,rinvsq10);
 818
 819             /* Update potential sum for this i atom from the interaction with this j atom. */
 820             velec            = _mm256_andnot_pd(dummy_mask,velec);
 821             velecsum         = _mm256_add_pd(velecsum,velec);
 822
 823             fscal            = felec;
 824
 825             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 826
 827             /* Calculate temporary vectorial force */
 828             tx               = _mm256_mul_pd(fscal,dx10);
 829             ty               = _mm256_mul_pd(fscal,dy10);
 830             tz               = _mm256_mul_pd(fscal,dz10);
 831
 832             /* Update vectorial force */
 833             fix1             = _mm256_add_pd(fix1,tx);
 834             fiy1             = _mm256_add_pd(fiy1,ty);
 835             fiz1             = _mm256_add_pd(fiz1,tz);
 836
 837             fjx0             = _mm256_add_pd(fjx0,tx);
 838             fjy0             = _mm256_add_pd(fjy0,ty);
 839             fjz0             = _mm256_add_pd(fjz0,tz);
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             /* COULOMB ELECTROSTATICS */
 846             velec            = _mm256_mul_pd(qq11,rinv11);
 847             felec            = _mm256_mul_pd(velec,rinvsq11);
 848
 849             /* Update potential sum for this i atom from the interaction with this j atom. */
 850             velec            = _mm256_andnot_pd(dummy_mask,velec);
 851             velecsum         = _mm256_add_pd(velecsum,velec);
 852
 853             fscal            = felec;
 854
 855             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm256_mul_pd(fscal,dx11);
 859             ty               = _mm256_mul_pd(fscal,dy11);
 860             tz               = _mm256_mul_pd(fscal,dz11);
 861
 862             /* Update vectorial force */
 863             fix1             = _mm256_add_pd(fix1,tx);
 864             fiy1             = _mm256_add_pd(fiy1,ty);
 865             fiz1             = _mm256_add_pd(fiz1,tz);
 866
 867             fjx1             = _mm256_add_pd(fjx1,tx);
 868             fjy1             = _mm256_add_pd(fjy1,ty);
 869             fjz1             = _mm256_add_pd(fjz1,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             /* COULOMB ELECTROSTATICS */
 876             velec            = _mm256_mul_pd(qq12,rinv12);
 877             felec            = _mm256_mul_pd(velec,rinvsq12);
 878
 879             /* Update potential sum for this i atom from the interaction with this j atom. */
 880             velec            = _mm256_andnot_pd(dummy_mask,velec);
 881             velecsum         = _mm256_add_pd(velecsum,velec);
 882
 883             fscal            = felec;
 884
 885             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm256_mul_pd(fscal,dx12);
 889             ty               = _mm256_mul_pd(fscal,dy12);
 890             tz               = _mm256_mul_pd(fscal,dz12);
 891
 892             /* Update vectorial force */
 893             fix1             = _mm256_add_pd(fix1,tx);
 894             fiy1             = _mm256_add_pd(fiy1,ty);
 895             fiz1             = _mm256_add_pd(fiz1,tz);
 896
 897             fjx2             = _mm256_add_pd(fjx2,tx);
 898             fjy2             = _mm256_add_pd(fjy2,ty);
 899             fjz2             = _mm256_add_pd(fjz2,tz);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             /* COULOMB ELECTROSTATICS */
 906             velec            = _mm256_mul_pd(qq20,rinv20);
 907             felec            = _mm256_mul_pd(velec,rinvsq20);
 908
 909             /* Update potential sum for this i atom from the interaction with this j atom. */
 910             velec            = _mm256_andnot_pd(dummy_mask,velec);
 911             velecsum         = _mm256_add_pd(velecsum,velec);
 912
 913             fscal            = felec;
 914
 915             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 916
 917             /* Calculate temporary vectorial force */
 918             tx               = _mm256_mul_pd(fscal,dx20);
 919             ty               = _mm256_mul_pd(fscal,dy20);
 920             tz               = _mm256_mul_pd(fscal,dz20);
 921
 922             /* Update vectorial force */
 923             fix2             = _mm256_add_pd(fix2,tx);
 924             fiy2             = _mm256_add_pd(fiy2,ty);
 925             fiz2             = _mm256_add_pd(fiz2,tz);
 926
 927             fjx0             = _mm256_add_pd(fjx0,tx);
 928             fjy0             = _mm256_add_pd(fjy0,ty);
 929             fjz0             = _mm256_add_pd(fjz0,tz);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             /* COULOMB ELECTROSTATICS */
 936             velec            = _mm256_mul_pd(qq21,rinv21);
 937             felec            = _mm256_mul_pd(velec,rinvsq21);
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm256_andnot_pd(dummy_mask,velec);
 941             velecsum         = _mm256_add_pd(velecsum,velec);
 942
 943             fscal            = felec;
 944
 945             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 946
 947             /* Calculate temporary vectorial force */
 948             tx               = _mm256_mul_pd(fscal,dx21);
 949             ty               = _mm256_mul_pd(fscal,dy21);
 950             tz               = _mm256_mul_pd(fscal,dz21);
 951
 952             /* Update vectorial force */
 953             fix2             = _mm256_add_pd(fix2,tx);
 954             fiy2             = _mm256_add_pd(fiy2,ty);
 955             fiz2             = _mm256_add_pd(fiz2,tz);
 956
 957             fjx1             = _mm256_add_pd(fjx1,tx);
 958             fjy1             = _mm256_add_pd(fjy1,ty);
 959             fjz1             = _mm256_add_pd(fjz1,tz);
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             /* COULOMB ELECTROSTATICS */
 966             velec            = _mm256_mul_pd(qq22,rinv22);
 967             felec            = _mm256_mul_pd(velec,rinvsq22);
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_pd(dummy_mask,velec);
 971             velecsum         = _mm256_add_pd(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_pd(fscal,dx22);
 979             ty               = _mm256_mul_pd(fscal,dy22);
 980             tz               = _mm256_mul_pd(fscal,dz22);
 981
 982             /* Update vectorial force */
 983             fix2             = _mm256_add_pd(fix2,tx);
 984             fiy2             = _mm256_add_pd(fiy2,ty);
 985             fiz2             = _mm256_add_pd(fiz2,tz);
 986
 987             fjx2             = _mm256_add_pd(fjx2,tx);
 988             fjy2             = _mm256_add_pd(fjy2,ty);
 989             fjz2             = _mm256_add_pd(fjz2,tz);
 990
 991             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 992             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 993             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 994             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 995
 996             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 997                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 998
 999             /* Inner loop uses 279 flops */
1000         }
1001
1002         /* End of innermost loop */
1003
1004         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1005                                                  f+i_coord_offset,fshift+i_shift_offset);
1006
1007         ggid                        = gid[iidx];
1008         /* Update potential energies */
1009         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1010         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1011
1012         /* Increment number of inner iterations */
1013         inneriter                  += j_index_end - j_index_start;
1014
1015         /* Outer loop uses 20 flops */
1016     }
1017
1018     /* Increment number of outer iterations */
1019     outeriter        += nri;
1020
1021     /* Update outer/inner flops */
1022
1023     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*279);
1024 }
1025 /*
1026  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_256_double
1027  * Electrostatics interaction: Coulomb
1028  * VdW interaction:            CubicSplineTable
1029  * Geometry:                   Water3-Water3
1030  * Calculate force/pot:        Force
1031  */
1032 void
1033 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_256_double
1034                     (t_nblist * gmx_restrict                nlist,
1035                      rvec * gmx_restrict                    xx,
1036                      rvec * gmx_restrict                    ff,
1037                      t_forcerec * gmx_restrict              fr,
1038                      t_mdatoms * gmx_restrict               mdatoms,
1039                      nb_kernel_data_t * gmx_restrict        kernel_data,
1040                      t_nrnb * gmx_restrict                  nrnb)
1041 {
1042     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1043      * just 0 for non-waters.
1044      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1045      * jnr indices corresponding to data put in the four positions in the SIMD register.
1046      */
1047     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1048     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1049     int              jnrA,jnrB,jnrC,jnrD;
1050     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1051     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1052     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1053     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1054     real             rcutoff_scalar;
1055     real             *shiftvec,*fshift,*x,*f;
1056     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1057     real             scratch[4*DIM];
1058     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1059     real *           vdwioffsetptr0;
1060     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1061     real *           vdwioffsetptr1;
1062     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1063     real *           vdwioffsetptr2;
1064     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1065     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1066     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1067     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1068     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1069     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1070     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1071     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1072     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1073     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1074     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1075     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1076     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1077     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1078     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1079     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1080     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1081     real             *charge;
1082     int              nvdwtype;
1083     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1084     int              *vdwtype;
1085     real             *vdwparam;
1086     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1087     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1088     __m128i          vfitab;
1089     __m128i          ifour       = _mm_set1_epi32(4);
1090     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1091     real             *vftab;
1092     __m256d          dummy_mask,cutoff_mask;
1093     __m128           tmpmask0,tmpmask1;
1094     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1095     __m256d          one     = _mm256_set1_pd(1.0);
1096     __m256d          two     = _mm256_set1_pd(2.0);
1097     x                = xx[0];
1098     f                = ff[0];
1099
1100     nri              = nlist->nri;
1101     iinr             = nlist->iinr;
1102     jindex           = nlist->jindex;
1103     jjnr             = nlist->jjnr;
1104     shiftidx         = nlist->shift;
1105     gid              = nlist->gid;
1106     shiftvec         = fr->shift_vec[0];
1107     fshift           = fr->fshift[0];
1108     facel            = _mm256_set1_pd(fr->epsfac);
1109     charge           = mdatoms->chargeA;
1110     nvdwtype         = fr->ntype;
1111     vdwparam         = fr->nbfp;
1112     vdwtype          = mdatoms->typeA;
1113
1114     vftab            = kernel_data->table_vdw->data;
1115     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1116
1117     /* Setup water-specific parameters */
1118     inr              = nlist->iinr[0];
1119     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1120     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1121     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1122     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1123
1124     jq0              = _mm256_set1_pd(charge[inr+0]);
1125     jq1              = _mm256_set1_pd(charge[inr+1]);
1126     jq2              = _mm256_set1_pd(charge[inr+2]);
1127     vdwjidx0A        = 2*vdwtype[inr+0];
1128     qq00             = _mm256_mul_pd(iq0,jq0);
1129     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1130     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1131     qq01             = _mm256_mul_pd(iq0,jq1);
1132     qq02             = _mm256_mul_pd(iq0,jq2);
1133     qq10             = _mm256_mul_pd(iq1,jq0);
1134     qq11             = _mm256_mul_pd(iq1,jq1);
1135     qq12             = _mm256_mul_pd(iq1,jq2);
1136     qq20             = _mm256_mul_pd(iq2,jq0);
1137     qq21             = _mm256_mul_pd(iq2,jq1);
1138     qq22             = _mm256_mul_pd(iq2,jq2);
1139
1140     /* Avoid stupid compiler warnings */
1141     jnrA = jnrB = jnrC = jnrD = 0;
1142     j_coord_offsetA = 0;
1143     j_coord_offsetB = 0;
1144     j_coord_offsetC = 0;
1145     j_coord_offsetD = 0;
1146
1147     outeriter        = 0;
1148     inneriter        = 0;
1149
1150     for(iidx=0;iidx<4*DIM;iidx++)
1151     {
1152         scratch[iidx] = 0.0;
1153     }
1154
1155     /* Start outer loop over neighborlists */
1156     for(iidx=0; iidx<nri; iidx++)
1157     {
1158         /* Load shift vector for this list */
1159         i_shift_offset   = DIM*shiftidx[iidx];
1160
1161         /* Load limits for loop over neighbors */
1162         j_index_start    = jindex[iidx];
1163         j_index_end      = jindex[iidx+1];
1164
1165         /* Get outer coordinate index */
1166         inr              = iinr[iidx];
1167         i_coord_offset   = DIM*inr;
1168
1169         /* Load i particle coords and add shift vector */
1170         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1171                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1172
1173         fix0             = _mm256_setzero_pd();
1174         fiy0             = _mm256_setzero_pd();
1175         fiz0             = _mm256_setzero_pd();
1176         fix1             = _mm256_setzero_pd();
1177         fiy1             = _mm256_setzero_pd();
1178         fiz1             = _mm256_setzero_pd();
1179         fix2             = _mm256_setzero_pd();
1180         fiy2             = _mm256_setzero_pd();
1181         fiz2             = _mm256_setzero_pd();
1182
1183         /* Start inner kernel loop */
1184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1185         {
1186
1187             /* Get j neighbor index, and coordinate index */
1188             jnrA             = jjnr[jidx];
1189             jnrB             = jjnr[jidx+1];
1190             jnrC             = jjnr[jidx+2];
1191             jnrD             = jjnr[jidx+3];
1192             j_coord_offsetA  = DIM*jnrA;
1193             j_coord_offsetB  = DIM*jnrB;
1194             j_coord_offsetC  = DIM*jnrC;
1195             j_coord_offsetD  = DIM*jnrD;
1196
1197             /* load j atom coordinates */
1198             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1199                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1200                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1201
1202             /* Calculate displacement vector */
1203             dx00             = _mm256_sub_pd(ix0,jx0);
1204             dy00             = _mm256_sub_pd(iy0,jy0);
1205             dz00             = _mm256_sub_pd(iz0,jz0);
1206             dx01             = _mm256_sub_pd(ix0,jx1);
1207             dy01             = _mm256_sub_pd(iy0,jy1);
1208             dz01             = _mm256_sub_pd(iz0,jz1);
1209             dx02             = _mm256_sub_pd(ix0,jx2);
1210             dy02             = _mm256_sub_pd(iy0,jy2);
1211             dz02             = _mm256_sub_pd(iz0,jz2);
1212             dx10             = _mm256_sub_pd(ix1,jx0);
1213             dy10             = _mm256_sub_pd(iy1,jy0);
1214             dz10             = _mm256_sub_pd(iz1,jz0);
1215             dx11             = _mm256_sub_pd(ix1,jx1);
1216             dy11             = _mm256_sub_pd(iy1,jy1);
1217             dz11             = _mm256_sub_pd(iz1,jz1);
1218             dx12             = _mm256_sub_pd(ix1,jx2);
1219             dy12             = _mm256_sub_pd(iy1,jy2);
1220             dz12             = _mm256_sub_pd(iz1,jz2);
1221             dx20             = _mm256_sub_pd(ix2,jx0);
1222             dy20             = _mm256_sub_pd(iy2,jy0);
1223             dz20             = _mm256_sub_pd(iz2,jz0);
1224             dx21             = _mm256_sub_pd(ix2,jx1);
1225             dy21             = _mm256_sub_pd(iy2,jy1);
1226             dz21             = _mm256_sub_pd(iz2,jz1);
1227             dx22             = _mm256_sub_pd(ix2,jx2);
1228             dy22             = _mm256_sub_pd(iy2,jy2);
1229             dz22             = _mm256_sub_pd(iz2,jz2);
1230
1231             /* Calculate squared distance and things based on it */
1232             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1233             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1234             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1235             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1236             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1237             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1238             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1239             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1240             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1241
1242             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1243             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1244             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1245             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1246             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1247             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1248             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1249             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1250             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1251
1252             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1253             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1254             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1255             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1256             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1257             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1258             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1259             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1260             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1261
1262             fjx0             = _mm256_setzero_pd();
1263             fjy0             = _mm256_setzero_pd();
1264             fjz0             = _mm256_setzero_pd();
1265             fjx1             = _mm256_setzero_pd();
1266             fjy1             = _mm256_setzero_pd();
1267             fjz1             = _mm256_setzero_pd();
1268             fjx2             = _mm256_setzero_pd();
1269             fjy2             = _mm256_setzero_pd();
1270             fjz2             = _mm256_setzero_pd();
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             r00              = _mm256_mul_pd(rsq00,rinv00);
1277
1278             /* Calculate table index by multiplying r with table scale and truncate to integer */
1279             rt               = _mm256_mul_pd(r00,vftabscale);
1280             vfitab           = _mm256_cvttpd_epi32(rt);
1281             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1282             vfitab           = _mm_slli_epi32(vfitab,3);
1283
1284             /* COULOMB ELECTROSTATICS */
1285             velec            = _mm256_mul_pd(qq00,rinv00);
1286             felec            = _mm256_mul_pd(velec,rinvsq00);
1287
1288             /* CUBIC SPLINE TABLE DISPERSION */
1289             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1290             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1291             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1292             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1293             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1294             Heps             = _mm256_mul_pd(vfeps,H);
1295             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1296             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1297             fvdw6            = _mm256_mul_pd(c6_00,FF);
1298
1299             /* CUBIC SPLINE TABLE REPULSION */
1300             vfitab           = _mm_add_epi32(vfitab,ifour);
1301             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1302             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1303             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1304             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1305             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1306             Heps             = _mm256_mul_pd(vfeps,H);
1307             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1308             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1309             fvdw12           = _mm256_mul_pd(c12_00,FF);
1310             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1311
1312             fscal            = _mm256_add_pd(felec,fvdw);
1313
1314             /* Calculate temporary vectorial force */
1315             tx               = _mm256_mul_pd(fscal,dx00);
1316             ty               = _mm256_mul_pd(fscal,dy00);
1317             tz               = _mm256_mul_pd(fscal,dz00);
1318
1319             /* Update vectorial force */
1320             fix0             = _mm256_add_pd(fix0,tx);
1321             fiy0             = _mm256_add_pd(fiy0,ty);
1322             fiz0             = _mm256_add_pd(fiz0,tz);
1323
1324             fjx0             = _mm256_add_pd(fjx0,tx);
1325             fjy0             = _mm256_add_pd(fjy0,ty);
1326             fjz0             = _mm256_add_pd(fjz0,tz);
1327
1328             /**************************
1329              * CALCULATE INTERACTIONS *
1330              **************************/
1331
1332             /* COULOMB ELECTROSTATICS */
1333             velec            = _mm256_mul_pd(qq01,rinv01);
1334             felec            = _mm256_mul_pd(velec,rinvsq01);
1335
1336             fscal            = felec;
1337
1338             /* Calculate temporary vectorial force */
1339             tx               = _mm256_mul_pd(fscal,dx01);
1340             ty               = _mm256_mul_pd(fscal,dy01);
1341             tz               = _mm256_mul_pd(fscal,dz01);
1342
1343             /* Update vectorial force */
1344             fix0             = _mm256_add_pd(fix0,tx);
1345             fiy0             = _mm256_add_pd(fiy0,ty);
1346             fiz0             = _mm256_add_pd(fiz0,tz);
1347
1348             fjx1             = _mm256_add_pd(fjx1,tx);
1349             fjy1             = _mm256_add_pd(fjy1,ty);
1350             fjz1             = _mm256_add_pd(fjz1,tz);
1351
1352             /**************************
1353              * CALCULATE INTERACTIONS *
1354              **************************/
1355
1356             /* COULOMB ELECTROSTATICS */
1357             velec            = _mm256_mul_pd(qq02,rinv02);
1358             felec            = _mm256_mul_pd(velec,rinvsq02);
1359
1360             fscal            = felec;
1361
1362             /* Calculate temporary vectorial force */
1363             tx               = _mm256_mul_pd(fscal,dx02);
1364             ty               = _mm256_mul_pd(fscal,dy02);
1365             tz               = _mm256_mul_pd(fscal,dz02);
1366
1367             /* Update vectorial force */
1368             fix0             = _mm256_add_pd(fix0,tx);
1369             fiy0             = _mm256_add_pd(fiy0,ty);
1370             fiz0             = _mm256_add_pd(fiz0,tz);
1371
1372             fjx2             = _mm256_add_pd(fjx2,tx);
1373             fjy2             = _mm256_add_pd(fjy2,ty);
1374             fjz2             = _mm256_add_pd(fjz2,tz);
1375
1376             /**************************
1377              * CALCULATE INTERACTIONS *
1378              **************************/
1379
1380             /* COULOMB ELECTROSTATICS */
1381             velec            = _mm256_mul_pd(qq10,rinv10);
1382             felec            = _mm256_mul_pd(velec,rinvsq10);
1383
1384             fscal            = felec;
1385
1386             /* Calculate temporary vectorial force */
1387             tx               = _mm256_mul_pd(fscal,dx10);
1388             ty               = _mm256_mul_pd(fscal,dy10);
1389             tz               = _mm256_mul_pd(fscal,dz10);
1390
1391             /* Update vectorial force */
1392             fix1             = _mm256_add_pd(fix1,tx);
1393             fiy1             = _mm256_add_pd(fiy1,ty);
1394             fiz1             = _mm256_add_pd(fiz1,tz);
1395
1396             fjx0             = _mm256_add_pd(fjx0,tx);
1397             fjy0             = _mm256_add_pd(fjy0,ty);
1398             fjz0             = _mm256_add_pd(fjz0,tz);
1399
1400             /**************************
1401              * CALCULATE INTERACTIONS *
1402              **************************/
1403
1404             /* COULOMB ELECTROSTATICS */
1405             velec            = _mm256_mul_pd(qq11,rinv11);
1406             felec            = _mm256_mul_pd(velec,rinvsq11);
1407
1408             fscal            = felec;
1409
1410             /* Calculate temporary vectorial force */
1411             tx               = _mm256_mul_pd(fscal,dx11);
1412             ty               = _mm256_mul_pd(fscal,dy11);
1413             tz               = _mm256_mul_pd(fscal,dz11);
1414
1415             /* Update vectorial force */
1416             fix1             = _mm256_add_pd(fix1,tx);
1417             fiy1             = _mm256_add_pd(fiy1,ty);
1418             fiz1             = _mm256_add_pd(fiz1,tz);
1419
1420             fjx1             = _mm256_add_pd(fjx1,tx);
1421             fjy1             = _mm256_add_pd(fjy1,ty);
1422             fjz1             = _mm256_add_pd(fjz1,tz);
1423
1424             /**************************
1425              * CALCULATE INTERACTIONS *
1426              **************************/
1427
1428             /* COULOMB ELECTROSTATICS */
1429             velec            = _mm256_mul_pd(qq12,rinv12);
1430             felec            = _mm256_mul_pd(velec,rinvsq12);
1431
1432             fscal            = felec;
1433
1434             /* Calculate temporary vectorial force */
1435             tx               = _mm256_mul_pd(fscal,dx12);
1436             ty               = _mm256_mul_pd(fscal,dy12);
1437             tz               = _mm256_mul_pd(fscal,dz12);
1438
1439             /* Update vectorial force */
1440             fix1             = _mm256_add_pd(fix1,tx);
1441             fiy1             = _mm256_add_pd(fiy1,ty);
1442             fiz1             = _mm256_add_pd(fiz1,tz);
1443
1444             fjx2             = _mm256_add_pd(fjx2,tx);
1445             fjy2             = _mm256_add_pd(fjy2,ty);
1446             fjz2             = _mm256_add_pd(fjz2,tz);
1447
1448             /**************************
1449              * CALCULATE INTERACTIONS *
1450              **************************/
1451
1452             /* COULOMB ELECTROSTATICS */
1453             velec            = _mm256_mul_pd(qq20,rinv20);
1454             felec            = _mm256_mul_pd(velec,rinvsq20);
1455
1456             fscal            = felec;
1457
1458             /* Calculate temporary vectorial force */
1459             tx               = _mm256_mul_pd(fscal,dx20);
1460             ty               = _mm256_mul_pd(fscal,dy20);
1461             tz               = _mm256_mul_pd(fscal,dz20);
1462
1463             /* Update vectorial force */
1464             fix2             = _mm256_add_pd(fix2,tx);
1465             fiy2             = _mm256_add_pd(fiy2,ty);
1466             fiz2             = _mm256_add_pd(fiz2,tz);
1467
1468             fjx0             = _mm256_add_pd(fjx0,tx);
1469             fjy0             = _mm256_add_pd(fjy0,ty);
1470             fjz0             = _mm256_add_pd(fjz0,tz);
1471
1472             /**************************
1473              * CALCULATE INTERACTIONS *
1474              **************************/
1475
1476             /* COULOMB ELECTROSTATICS */
1477             velec            = _mm256_mul_pd(qq21,rinv21);
1478             felec            = _mm256_mul_pd(velec,rinvsq21);
1479
1480             fscal            = felec;
1481
1482             /* Calculate temporary vectorial force */
1483             tx               = _mm256_mul_pd(fscal,dx21);
1484             ty               = _mm256_mul_pd(fscal,dy21);
1485             tz               = _mm256_mul_pd(fscal,dz21);
1486
1487             /* Update vectorial force */
1488             fix2             = _mm256_add_pd(fix2,tx);
1489             fiy2             = _mm256_add_pd(fiy2,ty);
1490             fiz2             = _mm256_add_pd(fiz2,tz);
1491
1492             fjx1             = _mm256_add_pd(fjx1,tx);
1493             fjy1             = _mm256_add_pd(fjy1,ty);
1494             fjz1             = _mm256_add_pd(fjz1,tz);
1495
1496             /**************************
1497              * CALCULATE INTERACTIONS *
1498              **************************/
1499
1500             /* COULOMB ELECTROSTATICS */
1501             velec            = _mm256_mul_pd(qq22,rinv22);
1502             felec            = _mm256_mul_pd(velec,rinvsq22);
1503
1504             fscal            = felec;
1505
1506             /* Calculate temporary vectorial force */
1507             tx               = _mm256_mul_pd(fscal,dx22);
1508             ty               = _mm256_mul_pd(fscal,dy22);
1509             tz               = _mm256_mul_pd(fscal,dz22);
1510
1511             /* Update vectorial force */
1512             fix2             = _mm256_add_pd(fix2,tx);
1513             fiy2             = _mm256_add_pd(fiy2,ty);
1514             fiz2             = _mm256_add_pd(fiz2,tz);
1515
1516             fjx2             = _mm256_add_pd(fjx2,tx);
1517             fjy2             = _mm256_add_pd(fjy2,ty);
1518             fjz2             = _mm256_add_pd(fjz2,tz);
1519
1520             fjptrA             = f+j_coord_offsetA;
1521             fjptrB             = f+j_coord_offsetB;
1522             fjptrC             = f+j_coord_offsetC;
1523             fjptrD             = f+j_coord_offsetD;
1524
1525             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1526                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1527
1528             /* Inner loop uses 261 flops */
1529         }
1530
1531         if(jidx<j_index_end)
1532         {
1533
1534             /* Get j neighbor index, and coordinate index */
1535             jnrlistA         = jjnr[jidx];
1536             jnrlistB         = jjnr[jidx+1];
1537             jnrlistC         = jjnr[jidx+2];
1538             jnrlistD         = jjnr[jidx+3];
1539             /* Sign of each element will be negative for non-real atoms.
1540              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1541              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1542              */
1543             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1544
1545             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1546             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1547             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1548
1549             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1550             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1551             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1552             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1553             j_coord_offsetA  = DIM*jnrA;
1554             j_coord_offsetB  = DIM*jnrB;
1555             j_coord_offsetC  = DIM*jnrC;
1556             j_coord_offsetD  = DIM*jnrD;
1557
1558             /* load j atom coordinates */
1559             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1560                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1561                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1562
1563             /* Calculate displacement vector */
1564             dx00             = _mm256_sub_pd(ix0,jx0);
1565             dy00             = _mm256_sub_pd(iy0,jy0);
1566             dz00             = _mm256_sub_pd(iz0,jz0);
1567             dx01             = _mm256_sub_pd(ix0,jx1);
1568             dy01             = _mm256_sub_pd(iy0,jy1);
1569             dz01             = _mm256_sub_pd(iz0,jz1);
1570             dx02             = _mm256_sub_pd(ix0,jx2);
1571             dy02             = _mm256_sub_pd(iy0,jy2);
1572             dz02             = _mm256_sub_pd(iz0,jz2);
1573             dx10             = _mm256_sub_pd(ix1,jx0);
1574             dy10             = _mm256_sub_pd(iy1,jy0);
1575             dz10             = _mm256_sub_pd(iz1,jz0);
1576             dx11             = _mm256_sub_pd(ix1,jx1);
1577             dy11             = _mm256_sub_pd(iy1,jy1);
1578             dz11             = _mm256_sub_pd(iz1,jz1);
1579             dx12             = _mm256_sub_pd(ix1,jx2);
1580             dy12             = _mm256_sub_pd(iy1,jy2);
1581             dz12             = _mm256_sub_pd(iz1,jz2);
1582             dx20             = _mm256_sub_pd(ix2,jx0);
1583             dy20             = _mm256_sub_pd(iy2,jy0);
1584             dz20             = _mm256_sub_pd(iz2,jz0);
1585             dx21             = _mm256_sub_pd(ix2,jx1);
1586             dy21             = _mm256_sub_pd(iy2,jy1);
1587             dz21             = _mm256_sub_pd(iz2,jz1);
1588             dx22             = _mm256_sub_pd(ix2,jx2);
1589             dy22             = _mm256_sub_pd(iy2,jy2);
1590             dz22             = _mm256_sub_pd(iz2,jz2);
1591
1592             /* Calculate squared distance and things based on it */
1593             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1594             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1595             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1596             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1597             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1598             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1599             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1600             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1601             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1602
1603             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1604             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1605             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1606             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1607             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1608             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1609             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1610             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1611             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1612
1613             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1614             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1615             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1616             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1617             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1618             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1619             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1620             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1621             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1622
1623             fjx0             = _mm256_setzero_pd();
1624             fjy0             = _mm256_setzero_pd();
1625             fjz0             = _mm256_setzero_pd();
1626             fjx1             = _mm256_setzero_pd();
1627             fjy1             = _mm256_setzero_pd();
1628             fjz1             = _mm256_setzero_pd();
1629             fjx2             = _mm256_setzero_pd();
1630             fjy2             = _mm256_setzero_pd();
1631             fjz2             = _mm256_setzero_pd();
1632
1633             /**************************
1634              * CALCULATE INTERACTIONS *
1635              **************************/
1636
1637             r00              = _mm256_mul_pd(rsq00,rinv00);
1638             r00              = _mm256_andnot_pd(dummy_mask,r00);
1639
1640             /* Calculate table index by multiplying r with table scale and truncate to integer */
1641             rt               = _mm256_mul_pd(r00,vftabscale);
1642             vfitab           = _mm256_cvttpd_epi32(rt);
1643             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1644             vfitab           = _mm_slli_epi32(vfitab,3);
1645
1646             /* COULOMB ELECTROSTATICS */
1647             velec            = _mm256_mul_pd(qq00,rinv00);
1648             felec            = _mm256_mul_pd(velec,rinvsq00);
1649
1650             /* CUBIC SPLINE TABLE DISPERSION */
1651             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1652             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1653             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1654             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1655             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1656             Heps             = _mm256_mul_pd(vfeps,H);
1657             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1658             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1659             fvdw6            = _mm256_mul_pd(c6_00,FF);
1660
1661             /* CUBIC SPLINE TABLE REPULSION */
1662             vfitab           = _mm_add_epi32(vfitab,ifour);
1663             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1664             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1665             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1666             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1667             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1668             Heps             = _mm256_mul_pd(vfeps,H);
1669             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1670             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1671             fvdw12           = _mm256_mul_pd(c12_00,FF);
1672             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1673
1674             fscal            = _mm256_add_pd(felec,fvdw);
1675
1676             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1677
1678             /* Calculate temporary vectorial force */
1679             tx               = _mm256_mul_pd(fscal,dx00);
1680             ty               = _mm256_mul_pd(fscal,dy00);
1681             tz               = _mm256_mul_pd(fscal,dz00);
1682
1683             /* Update vectorial force */
1684             fix0             = _mm256_add_pd(fix0,tx);
1685             fiy0             = _mm256_add_pd(fiy0,ty);
1686             fiz0             = _mm256_add_pd(fiz0,tz);
1687
1688             fjx0             = _mm256_add_pd(fjx0,tx);
1689             fjy0             = _mm256_add_pd(fjy0,ty);
1690             fjz0             = _mm256_add_pd(fjz0,tz);
1691
1692             /**************************
1693              * CALCULATE INTERACTIONS *
1694              **************************/
1695
1696             /* COULOMB ELECTROSTATICS */
1697             velec            = _mm256_mul_pd(qq01,rinv01);
1698             felec            = _mm256_mul_pd(velec,rinvsq01);
1699
1700             fscal            = felec;
1701
1702             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1703
1704             /* Calculate temporary vectorial force */
1705             tx               = _mm256_mul_pd(fscal,dx01);
1706             ty               = _mm256_mul_pd(fscal,dy01);
1707             tz               = _mm256_mul_pd(fscal,dz01);
1708
1709             /* Update vectorial force */
1710             fix0             = _mm256_add_pd(fix0,tx);
1711             fiy0             = _mm256_add_pd(fiy0,ty);
1712             fiz0             = _mm256_add_pd(fiz0,tz);
1713
1714             fjx1             = _mm256_add_pd(fjx1,tx);
1715             fjy1             = _mm256_add_pd(fjy1,ty);
1716             fjz1             = _mm256_add_pd(fjz1,tz);
1717
1718             /**************************
1719              * CALCULATE INTERACTIONS *
1720              **************************/
1721
1722             /* COULOMB ELECTROSTATICS */
1723             velec            = _mm256_mul_pd(qq02,rinv02);
1724             felec            = _mm256_mul_pd(velec,rinvsq02);
1725
1726             fscal            = felec;
1727
1728             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1729
1730             /* Calculate temporary vectorial force */
1731             tx               = _mm256_mul_pd(fscal,dx02);
1732             ty               = _mm256_mul_pd(fscal,dy02);
1733             tz               = _mm256_mul_pd(fscal,dz02);
1734
1735             /* Update vectorial force */
1736             fix0             = _mm256_add_pd(fix0,tx);
1737             fiy0             = _mm256_add_pd(fiy0,ty);
1738             fiz0             = _mm256_add_pd(fiz0,tz);
1739
1740             fjx2             = _mm256_add_pd(fjx2,tx);
1741             fjy2             = _mm256_add_pd(fjy2,ty);
1742             fjz2             = _mm256_add_pd(fjz2,tz);
1743
1744             /**************************
1745              * CALCULATE INTERACTIONS *
1746              **************************/
1747
1748             /* COULOMB ELECTROSTATICS */
1749             velec            = _mm256_mul_pd(qq10,rinv10);
1750             felec            = _mm256_mul_pd(velec,rinvsq10);
1751
1752             fscal            = felec;
1753
1754             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1755
1756             /* Calculate temporary vectorial force */
1757             tx               = _mm256_mul_pd(fscal,dx10);
1758             ty               = _mm256_mul_pd(fscal,dy10);
1759             tz               = _mm256_mul_pd(fscal,dz10);
1760
1761             /* Update vectorial force */
1762             fix1             = _mm256_add_pd(fix1,tx);
1763             fiy1             = _mm256_add_pd(fiy1,ty);
1764             fiz1             = _mm256_add_pd(fiz1,tz);
1765
1766             fjx0             = _mm256_add_pd(fjx0,tx);
1767             fjy0             = _mm256_add_pd(fjy0,ty);
1768             fjz0             = _mm256_add_pd(fjz0,tz);
1769
1770             /**************************
1771              * CALCULATE INTERACTIONS *
1772              **************************/
1773
1774             /* COULOMB ELECTROSTATICS */
1775             velec            = _mm256_mul_pd(qq11,rinv11);
1776             felec            = _mm256_mul_pd(velec,rinvsq11);
1777
1778             fscal            = felec;
1779
1780             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1781
1782             /* Calculate temporary vectorial force */
1783             tx               = _mm256_mul_pd(fscal,dx11);
1784             ty               = _mm256_mul_pd(fscal,dy11);
1785             tz               = _mm256_mul_pd(fscal,dz11);
1786
1787             /* Update vectorial force */
1788             fix1             = _mm256_add_pd(fix1,tx);
1789             fiy1             = _mm256_add_pd(fiy1,ty);
1790             fiz1             = _mm256_add_pd(fiz1,tz);
1791
1792             fjx1             = _mm256_add_pd(fjx1,tx);
1793             fjy1             = _mm256_add_pd(fjy1,ty);
1794             fjz1             = _mm256_add_pd(fjz1,tz);
1795
1796             /**************************
1797              * CALCULATE INTERACTIONS *
1798              **************************/
1799
1800             /* COULOMB ELECTROSTATICS */
1801             velec            = _mm256_mul_pd(qq12,rinv12);
1802             felec            = _mm256_mul_pd(velec,rinvsq12);
1803
1804             fscal            = felec;
1805
1806             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1807
1808             /* Calculate temporary vectorial force */
1809             tx               = _mm256_mul_pd(fscal,dx12);
1810             ty               = _mm256_mul_pd(fscal,dy12);
1811             tz               = _mm256_mul_pd(fscal,dz12);
1812
1813             /* Update vectorial force */
1814             fix1             = _mm256_add_pd(fix1,tx);
1815             fiy1             = _mm256_add_pd(fiy1,ty);
1816             fiz1             = _mm256_add_pd(fiz1,tz);
1817
1818             fjx2             = _mm256_add_pd(fjx2,tx);
1819             fjy2             = _mm256_add_pd(fjy2,ty);
1820             fjz2             = _mm256_add_pd(fjz2,tz);
1821
1822             /**************************
1823              * CALCULATE INTERACTIONS *
1824              **************************/
1825
1826             /* COULOMB ELECTROSTATICS */
1827             velec            = _mm256_mul_pd(qq20,rinv20);
1828             felec            = _mm256_mul_pd(velec,rinvsq20);
1829
1830             fscal            = felec;
1831
1832             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1833
1834             /* Calculate temporary vectorial force */
1835             tx               = _mm256_mul_pd(fscal,dx20);
1836             ty               = _mm256_mul_pd(fscal,dy20);
1837             tz               = _mm256_mul_pd(fscal,dz20);
1838
1839             /* Update vectorial force */
1840             fix2             = _mm256_add_pd(fix2,tx);
1841             fiy2             = _mm256_add_pd(fiy2,ty);
1842             fiz2             = _mm256_add_pd(fiz2,tz);
1843
1844             fjx0             = _mm256_add_pd(fjx0,tx);
1845             fjy0             = _mm256_add_pd(fjy0,ty);
1846             fjz0             = _mm256_add_pd(fjz0,tz);
1847
1848             /**************************
1849              * CALCULATE INTERACTIONS *
1850              **************************/
1851
1852             /* COULOMB ELECTROSTATICS */
1853             velec            = _mm256_mul_pd(qq21,rinv21);
1854             felec            = _mm256_mul_pd(velec,rinvsq21);
1855
1856             fscal            = felec;
1857
1858             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1859
1860             /* Calculate temporary vectorial force */
1861             tx               = _mm256_mul_pd(fscal,dx21);
1862             ty               = _mm256_mul_pd(fscal,dy21);
1863             tz               = _mm256_mul_pd(fscal,dz21);
1864
1865             /* Update vectorial force */
1866             fix2             = _mm256_add_pd(fix2,tx);
1867             fiy2             = _mm256_add_pd(fiy2,ty);
1868             fiz2             = _mm256_add_pd(fiz2,tz);
1869
1870             fjx1             = _mm256_add_pd(fjx1,tx);
1871             fjy1             = _mm256_add_pd(fjy1,ty);
1872             fjz1             = _mm256_add_pd(fjz1,tz);
1873
1874             /**************************
1875              * CALCULATE INTERACTIONS *
1876              **************************/
1877
1878             /* COULOMB ELECTROSTATICS */
1879             velec            = _mm256_mul_pd(qq22,rinv22);
1880             felec            = _mm256_mul_pd(velec,rinvsq22);
1881
1882             fscal            = felec;
1883
1884             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1885
1886             /* Calculate temporary vectorial force */
1887             tx               = _mm256_mul_pd(fscal,dx22);
1888             ty               = _mm256_mul_pd(fscal,dy22);
1889             tz               = _mm256_mul_pd(fscal,dz22);
1890
1891             /* Update vectorial force */
1892             fix2             = _mm256_add_pd(fix2,tx);
1893             fiy2             = _mm256_add_pd(fiy2,ty);
1894             fiz2             = _mm256_add_pd(fiz2,tz);
1895
1896             fjx2             = _mm256_add_pd(fjx2,tx);
1897             fjy2             = _mm256_add_pd(fjy2,ty);
1898             fjz2             = _mm256_add_pd(fjz2,tz);
1899
1900             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1901             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1902             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1903             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1904
1905             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1906                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1907
1908             /* Inner loop uses 262 flops */
1909         }
1910
1911         /* End of innermost loop */
1912
1913         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1914                                                  f+i_coord_offset,fshift+i_shift_offset);
1915
1916         /* Increment number of inner iterations */
1917         inneriter                  += j_index_end - j_index_start;
1918
1919         /* Outer loop uses 18 flops */
1920     }
1921
1922     /* Increment number of outer iterations */
1923     outeriter        += nri;
1924
1925     /* Update outer/inner flops */
1926
1927     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*262);
1928 }