src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     int              nvdwtype;
  84     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  85     int              *vdwtype;
  86     real             *vdwparam;
  87     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  88     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  89     __m128i          vfitab;
  90     __m128i          ifour       = _mm_set1_epi32(4);
  91     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  92     real             *vftab;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     vftab            = kernel_data->table_elec_vdw->data;
 116     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 121     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 122     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 123     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 156                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 157
 158         fix0             = _mm256_setzero_pd();
 159         fiy0             = _mm256_setzero_pd();
 160         fiz0             = _mm256_setzero_pd();
 161         fix1             = _mm256_setzero_pd();
 162         fiy1             = _mm256_setzero_pd();
 163         fiz1             = _mm256_setzero_pd();
 164         fix2             = _mm256_setzero_pd();
 165         fiy2             = _mm256_setzero_pd();
 166         fiz2             = _mm256_setzero_pd();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm256_setzero_pd();
 170         vvdwsum          = _mm256_setzero_pd();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm256_sub_pd(ix0,jx0);
 193             dy00             = _mm256_sub_pd(iy0,jy0);
 194             dz00             = _mm256_sub_pd(iz0,jz0);
 195             dx10             = _mm256_sub_pd(ix1,jx0);
 196             dy10             = _mm256_sub_pd(iy1,jy0);
 197             dz10             = _mm256_sub_pd(iz1,jz0);
 198             dx20             = _mm256_sub_pd(ix2,jx0);
 199             dy20             = _mm256_sub_pd(iy2,jy0);
 200             dz20             = _mm256_sub_pd(iz2,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 206
 207             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 208             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 209             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 213                                                                  charge+jnrC+0,charge+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             fjx0             = _mm256_setzero_pd();
 220             fjy0             = _mm256_setzero_pd();
 221             fjz0             = _mm256_setzero_pd();
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             r00              = _mm256_mul_pd(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm256_mul_pd(iq0,jq0);
 231             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 232                                             vdwioffsetptr0+vdwjidx0B,
 233                                             vdwioffsetptr0+vdwjidx0C,
 234                                             vdwioffsetptr0+vdwjidx0D,
 235                                             &c6_00,&c12_00);
 236
 237             /* Calculate table index by multiplying r with table scale and truncate to integer */
 238             rt               = _mm256_mul_pd(r00,vftabscale);
 239             vfitab           = _mm256_cvttpd_epi32(rt);
 240             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 241             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 242
 243             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 244             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 245             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 246             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 247             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 248             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 249             Heps             = _mm256_mul_pd(vfeps,H);
 250             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 251             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 252             velec            = _mm256_mul_pd(qq00,VV);
 253             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 254             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 255
 256             /* CUBIC SPLINE TABLE DISPERSION */
 257             vfitab           = _mm_add_epi32(vfitab,ifour);
 258             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 259             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 260             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 261             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 262             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 263             Heps             = _mm256_mul_pd(vfeps,H);
 264             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 265             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 266             vvdw6            = _mm256_mul_pd(c6_00,VV);
 267             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 268             fvdw6            = _mm256_mul_pd(c6_00,FF);
 269
 270             /* CUBIC SPLINE TABLE REPULSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 272             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 273             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 274             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 275             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 276             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 277             Heps             = _mm256_mul_pd(vfeps,H);
 278             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 279             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 280             vvdw12           = _mm256_mul_pd(c12_00,VV);
 281             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 282             fvdw12           = _mm256_mul_pd(c12_00,FF);
 283             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 284             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm256_add_pd(velecsum,velec);
 288             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = _mm256_add_pd(felec,fvdw);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_pd(fscal,dx00);
 294             ty               = _mm256_mul_pd(fscal,dy00);
 295             tz               = _mm256_mul_pd(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm256_add_pd(fix0,tx);
 299             fiy0             = _mm256_add_pd(fiy0,ty);
 300             fiz0             = _mm256_add_pd(fiz0,tz);
 301
 302             fjx0             = _mm256_add_pd(fjx0,tx);
 303             fjy0             = _mm256_add_pd(fjy0,ty);
 304             fjz0             = _mm256_add_pd(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             r10              = _mm256_mul_pd(rsq10,rinv10);
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm256_mul_pd(iq1,jq0);
 314
 315             /* Calculate table index by multiplying r with table scale and truncate to integer */
 316             rt               = _mm256_mul_pd(r10,vftabscale);
 317             vfitab           = _mm256_cvttpd_epi32(rt);
 318             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 319             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 320
 321             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 322             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 323             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 324             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 325             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 326             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 327             Heps             = _mm256_mul_pd(vfeps,H);
 328             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 329             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 330             velec            = _mm256_mul_pd(qq10,VV);
 331             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 332             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm256_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_pd(fscal,dx10);
 341             ty               = _mm256_mul_pd(fscal,dy10);
 342             tz               = _mm256_mul_pd(fscal,dz10);
 343
 344             /* Update vectorial force */
 345             fix1             = _mm256_add_pd(fix1,tx);
 346             fiy1             = _mm256_add_pd(fiy1,ty);
 347             fiz1             = _mm256_add_pd(fiz1,tz);
 348
 349             fjx0             = _mm256_add_pd(fjx0,tx);
 350             fjy0             = _mm256_add_pd(fjy0,ty);
 351             fjz0             = _mm256_add_pd(fjz0,tz);
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             r20              = _mm256_mul_pd(rsq20,rinv20);
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq20             = _mm256_mul_pd(iq2,jq0);
 361
 362             /* Calculate table index by multiplying r with table scale and truncate to integer */
 363             rt               = _mm256_mul_pd(r20,vftabscale);
 364             vfitab           = _mm256_cvttpd_epi32(rt);
 365             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 366             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 367
 368             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 369             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 370             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 371             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 372             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 373             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 374             Heps             = _mm256_mul_pd(vfeps,H);
 375             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 376             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 377             velec            = _mm256_mul_pd(qq20,VV);
 378             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 379             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 380
 381             /* Update potential sum for this i atom from the interaction with this j atom. */
 382             velecsum         = _mm256_add_pd(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm256_mul_pd(fscal,dx20);
 388             ty               = _mm256_mul_pd(fscal,dy20);
 389             tz               = _mm256_mul_pd(fscal,dz20);
 390
 391             /* Update vectorial force */
 392             fix2             = _mm256_add_pd(fix2,tx);
 393             fiy2             = _mm256_add_pd(fiy2,ty);
 394             fiz2             = _mm256_add_pd(fiz2,tz);
 395
 396             fjx0             = _mm256_add_pd(fjx0,tx);
 397             fjy0             = _mm256_add_pd(fjy0,ty);
 398             fjz0             = _mm256_add_pd(fjz0,tz);
 399
 400             fjptrA             = f+j_coord_offsetA;
 401             fjptrB             = f+j_coord_offsetB;
 402             fjptrC             = f+j_coord_offsetC;
 403             fjptrD             = f+j_coord_offsetD;
 404
 405             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 406
 407             /* Inner loop uses 162 flops */
 408         }
 409
 410         if(jidx<j_index_end)
 411         {
 412
 413             /* Get j neighbor index, and coordinate index */
 414             jnrlistA         = jjnr[jidx];
 415             jnrlistB         = jjnr[jidx+1];
 416             jnrlistC         = jjnr[jidx+2];
 417             jnrlistD         = jjnr[jidx+3];
 418             /* Sign of each element will be negative for non-real atoms.
 419              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 420              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 421              */
 422             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 423
 424             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 425             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 426             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 427
 428             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 429             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 430             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 431             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 432             j_coord_offsetA  = DIM*jnrA;
 433             j_coord_offsetB  = DIM*jnrB;
 434             j_coord_offsetC  = DIM*jnrC;
 435             j_coord_offsetD  = DIM*jnrD;
 436
 437             /* load j atom coordinates */
 438             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 439                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 440                                                  &jx0,&jy0,&jz0);
 441
 442             /* Calculate displacement vector */
 443             dx00             = _mm256_sub_pd(ix0,jx0);
 444             dy00             = _mm256_sub_pd(iy0,jy0);
 445             dz00             = _mm256_sub_pd(iz0,jz0);
 446             dx10             = _mm256_sub_pd(ix1,jx0);
 447             dy10             = _mm256_sub_pd(iy1,jy0);
 448             dz10             = _mm256_sub_pd(iz1,jz0);
 449             dx20             = _mm256_sub_pd(ix2,jx0);
 450             dy20             = _mm256_sub_pd(iy2,jy0);
 451             dz20             = _mm256_sub_pd(iz2,jz0);
 452
 453             /* Calculate squared distance and things based on it */
 454             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 455             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 456             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 457
 458             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 459             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 460             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 461
 462             /* Load parameters for j particles */
 463             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 464                                                                  charge+jnrC+0,charge+jnrD+0);
 465             vdwjidx0A        = 2*vdwtype[jnrA+0];
 466             vdwjidx0B        = 2*vdwtype[jnrB+0];
 467             vdwjidx0C        = 2*vdwtype[jnrC+0];
 468             vdwjidx0D        = 2*vdwtype[jnrD+0];
 469
 470             fjx0             = _mm256_setzero_pd();
 471             fjy0             = _mm256_setzero_pd();
 472             fjz0             = _mm256_setzero_pd();
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             r00              = _mm256_mul_pd(rsq00,rinv00);
 479             r00              = _mm256_andnot_pd(dummy_mask,r00);
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             qq00             = _mm256_mul_pd(iq0,jq0);
 483             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 484                                             vdwioffsetptr0+vdwjidx0B,
 485                                             vdwioffsetptr0+vdwjidx0C,
 486                                             vdwioffsetptr0+vdwjidx0D,
 487                                             &c6_00,&c12_00);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm256_mul_pd(r00,vftabscale);
 491             vfitab           = _mm256_cvttpd_epi32(rt);
 492             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 493             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 494
 495             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 496             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 497             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 498             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 499             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 500             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 501             Heps             = _mm256_mul_pd(vfeps,H);
 502             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 503             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 504             velec            = _mm256_mul_pd(qq00,VV);
 505             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 506             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 507
 508             /* CUBIC SPLINE TABLE DISPERSION */
 509             vfitab           = _mm_add_epi32(vfitab,ifour);
 510             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 511             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 512             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 513             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 514             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 515             Heps             = _mm256_mul_pd(vfeps,H);
 516             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 517             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 518             vvdw6            = _mm256_mul_pd(c6_00,VV);
 519             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 520             fvdw6            = _mm256_mul_pd(c6_00,FF);
 521
 522             /* CUBIC SPLINE TABLE REPULSION */
 523             vfitab           = _mm_add_epi32(vfitab,ifour);
 524             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 525             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 526             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 527             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 528             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 529             Heps             = _mm256_mul_pd(vfeps,H);
 530             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 531             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 532             vvdw12           = _mm256_mul_pd(c12_00,VV);
 533             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 534             fvdw12           = _mm256_mul_pd(c12_00,FF);
 535             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 536             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velec            = _mm256_andnot_pd(dummy_mask,velec);
 540             velecsum         = _mm256_add_pd(velecsum,velec);
 541             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 542             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 543
 544             fscal            = _mm256_add_pd(felec,fvdw);
 545
 546             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_pd(fscal,dx00);
 550             ty               = _mm256_mul_pd(fscal,dy00);
 551             tz               = _mm256_mul_pd(fscal,dz00);
 552
 553             /* Update vectorial force */
 554             fix0             = _mm256_add_pd(fix0,tx);
 555             fiy0             = _mm256_add_pd(fiy0,ty);
 556             fiz0             = _mm256_add_pd(fiz0,tz);
 557
 558             fjx0             = _mm256_add_pd(fjx0,tx);
 559             fjy0             = _mm256_add_pd(fjy0,ty);
 560             fjz0             = _mm256_add_pd(fjz0,tz);
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             r10              = _mm256_mul_pd(rsq10,rinv10);
 567             r10              = _mm256_andnot_pd(dummy_mask,r10);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq10             = _mm256_mul_pd(iq1,jq0);
 571
 572             /* Calculate table index by multiplying r with table scale and truncate to integer */
 573             rt               = _mm256_mul_pd(r10,vftabscale);
 574             vfitab           = _mm256_cvttpd_epi32(rt);
 575             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 576             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 577
 578             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 579             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 580             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 581             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 582             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 583             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 584             Heps             = _mm256_mul_pd(vfeps,H);
 585             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 586             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 587             velec            = _mm256_mul_pd(qq10,VV);
 588             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 589             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm256_andnot_pd(dummy_mask,velec);
 593             velecsum         = _mm256_add_pd(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm256_mul_pd(fscal,dx10);
 601             ty               = _mm256_mul_pd(fscal,dy10);
 602             tz               = _mm256_mul_pd(fscal,dz10);
 603
 604             /* Update vectorial force */
 605             fix1             = _mm256_add_pd(fix1,tx);
 606             fiy1             = _mm256_add_pd(fiy1,ty);
 607             fiz1             = _mm256_add_pd(fiz1,tz);
 608
 609             fjx0             = _mm256_add_pd(fjx0,tx);
 610             fjy0             = _mm256_add_pd(fjy0,ty);
 611             fjz0             = _mm256_add_pd(fjz0,tz);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r20              = _mm256_mul_pd(rsq20,rinv20);
 618             r20              = _mm256_andnot_pd(dummy_mask,r20);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq20             = _mm256_mul_pd(iq2,jq0);
 622
 623             /* Calculate table index by multiplying r with table scale and truncate to integer */
 624             rt               = _mm256_mul_pd(r20,vftabscale);
 625             vfitab           = _mm256_cvttpd_epi32(rt);
 626             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 627             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 628
 629             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 630             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 631             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 632             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 633             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 634             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 635             Heps             = _mm256_mul_pd(vfeps,H);
 636             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 637             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 638             velec            = _mm256_mul_pd(qq20,VV);
 639             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 640             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm256_andnot_pd(dummy_mask,velec);
 644             velecsum         = _mm256_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_pd(fscal,dx20);
 652             ty               = _mm256_mul_pd(fscal,dy20);
 653             tz               = _mm256_mul_pd(fscal,dz20);
 654
 655             /* Update vectorial force */
 656             fix2             = _mm256_add_pd(fix2,tx);
 657             fiy2             = _mm256_add_pd(fiy2,ty);
 658             fiz2             = _mm256_add_pd(fiz2,tz);
 659
 660             fjx0             = _mm256_add_pd(fjx0,tx);
 661             fjy0             = _mm256_add_pd(fjy0,ty);
 662             fjz0             = _mm256_add_pd(fjz0,tz);
 663
 664             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 665             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 666             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 667             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 668
 669             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 670
 671             /* Inner loop uses 165 flops */
 672         }
 673
 674         /* End of innermost loop */
 675
 676         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 677                                                  f+i_coord_offset,fshift+i_shift_offset);
 678
 679         ggid                        = gid[iidx];
 680         /* Update potential energies */
 681         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 682         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 683
 684         /* Increment number of inner iterations */
 685         inneriter                  += j_index_end - j_index_start;
 686
 687         /* Outer loop uses 20 flops */
 688     }
 689
 690     /* Increment number of outer iterations */
 691     outeriter        += nri;
 692
 693     /* Update outer/inner flops */
 694
 695     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 696 }
 697 /*
 698  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 699  * Electrostatics interaction: CubicSplineTable
 700  * VdW interaction:            CubicSplineTable
 701  * Geometry:                   Water3-Particle
 702  * Calculate force/pot:        Force
 703  */
 704 void
 705 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 706                     (t_nblist * gmx_restrict                nlist,
 707                      rvec * gmx_restrict                    xx,
 708                      rvec * gmx_restrict                    ff,
 709                      t_forcerec * gmx_restrict              fr,
 710                      t_mdatoms * gmx_restrict               mdatoms,
 711                      nb_kernel_data_t * gmx_restrict        kernel_data,
 712                      t_nrnb * gmx_restrict                  nrnb)
 713 {
 714     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 715      * just 0 for non-waters.
 716      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 717      * jnr indices corresponding to data put in the four positions in the SIMD register.
 718      */
 719     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 720     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 721     int              jnrA,jnrB,jnrC,jnrD;
 722     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 723     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 724     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 725     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 726     real             rcutoff_scalar;
 727     real             *shiftvec,*fshift,*x,*f;
 728     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 729     real             scratch[4*DIM];
 730     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 731     real *           vdwioffsetptr0;
 732     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 733     real *           vdwioffsetptr1;
 734     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 735     real *           vdwioffsetptr2;
 736     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 737     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 738     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 739     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 740     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 741     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 742     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 743     real             *charge;
 744     int              nvdwtype;
 745     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 746     int              *vdwtype;
 747     real             *vdwparam;
 748     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 749     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 750     __m128i          vfitab;
 751     __m128i          ifour       = _mm_set1_epi32(4);
 752     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 753     real             *vftab;
 754     __m256d          dummy_mask,cutoff_mask;
 755     __m128           tmpmask0,tmpmask1;
 756     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 757     __m256d          one     = _mm256_set1_pd(1.0);
 758     __m256d          two     = _mm256_set1_pd(2.0);
 759     x                = xx[0];
 760     f                = ff[0];
 761
 762     nri              = nlist->nri;
 763     iinr             = nlist->iinr;
 764     jindex           = nlist->jindex;
 765     jjnr             = nlist->jjnr;
 766     shiftidx         = nlist->shift;
 767     gid              = nlist->gid;
 768     shiftvec         = fr->shift_vec[0];
 769     fshift           = fr->fshift[0];
 770     facel            = _mm256_set1_pd(fr->epsfac);
 771     charge           = mdatoms->chargeA;
 772     nvdwtype         = fr->ntype;
 773     vdwparam         = fr->nbfp;
 774     vdwtype          = mdatoms->typeA;
 775
 776     vftab            = kernel_data->table_elec_vdw->data;
 777     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 778
 779     /* Setup water-specific parameters */
 780     inr              = nlist->iinr[0];
 781     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 782     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 783     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 784     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 785
 786     /* Avoid stupid compiler warnings */
 787     jnrA = jnrB = jnrC = jnrD = 0;
 788     j_coord_offsetA = 0;
 789     j_coord_offsetB = 0;
 790     j_coord_offsetC = 0;
 791     j_coord_offsetD = 0;
 792
 793     outeriter        = 0;
 794     inneriter        = 0;
 795
 796     for(iidx=0;iidx<4*DIM;iidx++)
 797     {
 798         scratch[iidx] = 0.0;
 799     }
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806
 807         /* Load limits for loop over neighbors */
 808         j_index_start    = jindex[iidx];
 809         j_index_end      = jindex[iidx+1];
 810
 811         /* Get outer coordinate index */
 812         inr              = iinr[iidx];
 813         i_coord_offset   = DIM*inr;
 814
 815         /* Load i particle coords and add shift vector */
 816         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 817                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 818
 819         fix0             = _mm256_setzero_pd();
 820         fiy0             = _mm256_setzero_pd();
 821         fiz0             = _mm256_setzero_pd();
 822         fix1             = _mm256_setzero_pd();
 823         fiy1             = _mm256_setzero_pd();
 824         fiz1             = _mm256_setzero_pd();
 825         fix2             = _mm256_setzero_pd();
 826         fiy2             = _mm256_setzero_pd();
 827         fiz2             = _mm256_setzero_pd();
 828
 829         /* Start inner kernel loop */
 830         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 831         {
 832
 833             /* Get j neighbor index, and coordinate index */
 834             jnrA             = jjnr[jidx];
 835             jnrB             = jjnr[jidx+1];
 836             jnrC             = jjnr[jidx+2];
 837             jnrD             = jjnr[jidx+3];
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840             j_coord_offsetC  = DIM*jnrC;
 841             j_coord_offsetD  = DIM*jnrD;
 842
 843             /* load j atom coordinates */
 844             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 845                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 846                                                  &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm256_sub_pd(ix0,jx0);
 850             dy00             = _mm256_sub_pd(iy0,jy0);
 851             dz00             = _mm256_sub_pd(iz0,jz0);
 852             dx10             = _mm256_sub_pd(ix1,jx0);
 853             dy10             = _mm256_sub_pd(iy1,jy0);
 854             dz10             = _mm256_sub_pd(iz1,jz0);
 855             dx20             = _mm256_sub_pd(ix2,jx0);
 856             dy20             = _mm256_sub_pd(iy2,jy0);
 857             dz20             = _mm256_sub_pd(iz2,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 861             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 862             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 863
 864             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 865             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 866             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 867
 868             /* Load parameters for j particles */
 869             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 870                                                                  charge+jnrC+0,charge+jnrD+0);
 871             vdwjidx0A        = 2*vdwtype[jnrA+0];
 872             vdwjidx0B        = 2*vdwtype[jnrB+0];
 873             vdwjidx0C        = 2*vdwtype[jnrC+0];
 874             vdwjidx0D        = 2*vdwtype[jnrD+0];
 875
 876             fjx0             = _mm256_setzero_pd();
 877             fjy0             = _mm256_setzero_pd();
 878             fjz0             = _mm256_setzero_pd();
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             r00              = _mm256_mul_pd(rsq00,rinv00);
 885
 886             /* Compute parameters for interactions between i and j atoms */
 887             qq00             = _mm256_mul_pd(iq0,jq0);
 888             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 889                                             vdwioffsetptr0+vdwjidx0B,
 890                                             vdwioffsetptr0+vdwjidx0C,
 891                                             vdwioffsetptr0+vdwjidx0D,
 892                                             &c6_00,&c12_00);
 893
 894             /* Calculate table index by multiplying r with table scale and truncate to integer */
 895             rt               = _mm256_mul_pd(r00,vftabscale);
 896             vfitab           = _mm256_cvttpd_epi32(rt);
 897             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 898             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 899
 900             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 901             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 902             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 903             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 904             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 905             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 906             Heps             = _mm256_mul_pd(vfeps,H);
 907             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 908             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 909             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 910
 911             /* CUBIC SPLINE TABLE DISPERSION */
 912             vfitab           = _mm_add_epi32(vfitab,ifour);
 913             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 914             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 915             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 916             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 917             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 918             Heps             = _mm256_mul_pd(vfeps,H);
 919             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 920             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 921             fvdw6            = _mm256_mul_pd(c6_00,FF);
 922
 923             /* CUBIC SPLINE TABLE REPULSION */
 924             vfitab           = _mm_add_epi32(vfitab,ifour);
 925             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 926             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 927             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 928             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 929             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 930             Heps             = _mm256_mul_pd(vfeps,H);
 931             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 932             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 933             fvdw12           = _mm256_mul_pd(c12_00,FF);
 934             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 935
 936             fscal            = _mm256_add_pd(felec,fvdw);
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm256_mul_pd(fscal,dx00);
 940             ty               = _mm256_mul_pd(fscal,dy00);
 941             tz               = _mm256_mul_pd(fscal,dz00);
 942
 943             /* Update vectorial force */
 944             fix0             = _mm256_add_pd(fix0,tx);
 945             fiy0             = _mm256_add_pd(fiy0,ty);
 946             fiz0             = _mm256_add_pd(fiz0,tz);
 947
 948             fjx0             = _mm256_add_pd(fjx0,tx);
 949             fjy0             = _mm256_add_pd(fjy0,ty);
 950             fjz0             = _mm256_add_pd(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             r10              = _mm256_mul_pd(rsq10,rinv10);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq10             = _mm256_mul_pd(iq1,jq0);
 960
 961             /* Calculate table index by multiplying r with table scale and truncate to integer */
 962             rt               = _mm256_mul_pd(r10,vftabscale);
 963             vfitab           = _mm256_cvttpd_epi32(rt);
 964             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 965             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 966
 967             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 968             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 969             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 970             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 971             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 972             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 973             Heps             = _mm256_mul_pd(vfeps,H);
 974             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 975             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 976             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 977
 978             fscal            = felec;
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm256_mul_pd(fscal,dx10);
 982             ty               = _mm256_mul_pd(fscal,dy10);
 983             tz               = _mm256_mul_pd(fscal,dz10);
 984
 985             /* Update vectorial force */
 986             fix1             = _mm256_add_pd(fix1,tx);
 987             fiy1             = _mm256_add_pd(fiy1,ty);
 988             fiz1             = _mm256_add_pd(fiz1,tz);
 989
 990             fjx0             = _mm256_add_pd(fjx0,tx);
 991             fjy0             = _mm256_add_pd(fjy0,ty);
 992             fjz0             = _mm256_add_pd(fjz0,tz);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             r20              = _mm256_mul_pd(rsq20,rinv20);
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq20             = _mm256_mul_pd(iq2,jq0);
1002
1003             /* Calculate table index by multiplying r with table scale and truncate to integer */
1004             rt               = _mm256_mul_pd(r20,vftabscale);
1005             vfitab           = _mm256_cvttpd_epi32(rt);
1006             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1007             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1008
1009             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1010             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1011             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1012             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1013             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1014             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1015             Heps             = _mm256_mul_pd(vfeps,H);
1016             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1017             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1018             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1019
1020             fscal            = felec;
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm256_mul_pd(fscal,dx20);
1024             ty               = _mm256_mul_pd(fscal,dy20);
1025             tz               = _mm256_mul_pd(fscal,dz20);
1026
1027             /* Update vectorial force */
1028             fix2             = _mm256_add_pd(fix2,tx);
1029             fiy2             = _mm256_add_pd(fiy2,ty);
1030             fiz2             = _mm256_add_pd(fiz2,tz);
1031
1032             fjx0             = _mm256_add_pd(fjx0,tx);
1033             fjy0             = _mm256_add_pd(fjy0,ty);
1034             fjz0             = _mm256_add_pd(fjz0,tz);
1035
1036             fjptrA             = f+j_coord_offsetA;
1037             fjptrB             = f+j_coord_offsetB;
1038             fjptrC             = f+j_coord_offsetC;
1039             fjptrD             = f+j_coord_offsetD;
1040
1041             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1042
1043             /* Inner loop uses 142 flops */
1044         }
1045
1046         if(jidx<j_index_end)
1047         {
1048
1049             /* Get j neighbor index, and coordinate index */
1050             jnrlistA         = jjnr[jidx];
1051             jnrlistB         = jjnr[jidx+1];
1052             jnrlistC         = jjnr[jidx+2];
1053             jnrlistD         = jjnr[jidx+3];
1054             /* Sign of each element will be negative for non-real atoms.
1055              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1056              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1057              */
1058             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1059
1060             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1061             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1062             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1063
1064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1068             j_coord_offsetA  = DIM*jnrA;
1069             j_coord_offsetB  = DIM*jnrB;
1070             j_coord_offsetC  = DIM*jnrC;
1071             j_coord_offsetD  = DIM*jnrD;
1072
1073             /* load j atom coordinates */
1074             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1075                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1076                                                  &jx0,&jy0,&jz0);
1077
1078             /* Calculate displacement vector */
1079             dx00             = _mm256_sub_pd(ix0,jx0);
1080             dy00             = _mm256_sub_pd(iy0,jy0);
1081             dz00             = _mm256_sub_pd(iz0,jz0);
1082             dx10             = _mm256_sub_pd(ix1,jx0);
1083             dy10             = _mm256_sub_pd(iy1,jy0);
1084             dz10             = _mm256_sub_pd(iz1,jz0);
1085             dx20             = _mm256_sub_pd(ix2,jx0);
1086             dy20             = _mm256_sub_pd(iy2,jy0);
1087             dz20             = _mm256_sub_pd(iz2,jz0);
1088
1089             /* Calculate squared distance and things based on it */
1090             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1091             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1092             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1093
1094             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1095             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1096             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1097
1098             /* Load parameters for j particles */
1099             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1100                                                                  charge+jnrC+0,charge+jnrD+0);
1101             vdwjidx0A        = 2*vdwtype[jnrA+0];
1102             vdwjidx0B        = 2*vdwtype[jnrB+0];
1103             vdwjidx0C        = 2*vdwtype[jnrC+0];
1104             vdwjidx0D        = 2*vdwtype[jnrD+0];
1105
1106             fjx0             = _mm256_setzero_pd();
1107             fjy0             = _mm256_setzero_pd();
1108             fjz0             = _mm256_setzero_pd();
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             r00              = _mm256_mul_pd(rsq00,rinv00);
1115             r00              = _mm256_andnot_pd(dummy_mask,r00);
1116
1117             /* Compute parameters for interactions between i and j atoms */
1118             qq00             = _mm256_mul_pd(iq0,jq0);
1119             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1120                                             vdwioffsetptr0+vdwjidx0B,
1121                                             vdwioffsetptr0+vdwjidx0C,
1122                                             vdwioffsetptr0+vdwjidx0D,
1123                                             &c6_00,&c12_00);
1124
1125             /* Calculate table index by multiplying r with table scale and truncate to integer */
1126             rt               = _mm256_mul_pd(r00,vftabscale);
1127             vfitab           = _mm256_cvttpd_epi32(rt);
1128             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1129             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1130
1131             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1132             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1133             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1134             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1135             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1136             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1137             Heps             = _mm256_mul_pd(vfeps,H);
1138             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1139             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1140             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1141
1142             /* CUBIC SPLINE TABLE DISPERSION */
1143             vfitab           = _mm_add_epi32(vfitab,ifour);
1144             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1145             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1146             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1147             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1148             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1149             Heps             = _mm256_mul_pd(vfeps,H);
1150             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1151             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1152             fvdw6            = _mm256_mul_pd(c6_00,FF);
1153
1154             /* CUBIC SPLINE TABLE REPULSION */
1155             vfitab           = _mm_add_epi32(vfitab,ifour);
1156             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1157             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1158             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1159             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1160             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1161             Heps             = _mm256_mul_pd(vfeps,H);
1162             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1163             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1164             fvdw12           = _mm256_mul_pd(c12_00,FF);
1165             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1166
1167             fscal            = _mm256_add_pd(felec,fvdw);
1168
1169             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1170
1171             /* Calculate temporary vectorial force */
1172             tx               = _mm256_mul_pd(fscal,dx00);
1173             ty               = _mm256_mul_pd(fscal,dy00);
1174             tz               = _mm256_mul_pd(fscal,dz00);
1175
1176             /* Update vectorial force */
1177             fix0             = _mm256_add_pd(fix0,tx);
1178             fiy0             = _mm256_add_pd(fiy0,ty);
1179             fiz0             = _mm256_add_pd(fiz0,tz);
1180
1181             fjx0             = _mm256_add_pd(fjx0,tx);
1182             fjy0             = _mm256_add_pd(fjy0,ty);
1183             fjz0             = _mm256_add_pd(fjz0,tz);
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             r10              = _mm256_mul_pd(rsq10,rinv10);
1190             r10              = _mm256_andnot_pd(dummy_mask,r10);
1191
1192             /* Compute parameters for interactions between i and j atoms */
1193             qq10             = _mm256_mul_pd(iq1,jq0);
1194
1195             /* Calculate table index by multiplying r with table scale and truncate to integer */
1196             rt               = _mm256_mul_pd(r10,vftabscale);
1197             vfitab           = _mm256_cvttpd_epi32(rt);
1198             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1199             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1200
1201             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1202             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1203             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1204             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1205             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1206             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1207             Heps             = _mm256_mul_pd(vfeps,H);
1208             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1209             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1210             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1215
1216             /* Calculate temporary vectorial force */
1217             tx               = _mm256_mul_pd(fscal,dx10);
1218             ty               = _mm256_mul_pd(fscal,dy10);
1219             tz               = _mm256_mul_pd(fscal,dz10);
1220
1221             /* Update vectorial force */
1222             fix1             = _mm256_add_pd(fix1,tx);
1223             fiy1             = _mm256_add_pd(fiy1,ty);
1224             fiz1             = _mm256_add_pd(fiz1,tz);
1225
1226             fjx0             = _mm256_add_pd(fjx0,tx);
1227             fjy0             = _mm256_add_pd(fjy0,ty);
1228             fjz0             = _mm256_add_pd(fjz0,tz);
1229
1230             /**************************
1231              * CALCULATE INTERACTIONS *
1232              **************************/
1233
1234             r20              = _mm256_mul_pd(rsq20,rinv20);
1235             r20              = _mm256_andnot_pd(dummy_mask,r20);
1236
1237             /* Compute parameters for interactions between i and j atoms */
1238             qq20             = _mm256_mul_pd(iq2,jq0);
1239
1240             /* Calculate table index by multiplying r with table scale and truncate to integer */
1241             rt               = _mm256_mul_pd(r20,vftabscale);
1242             vfitab           = _mm256_cvttpd_epi32(rt);
1243             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1244             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1245
1246             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1247             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1248             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1249             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1250             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1251             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1252             Heps             = _mm256_mul_pd(vfeps,H);
1253             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1254             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1255             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1256
1257             fscal            = felec;
1258
1259             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1260
1261             /* Calculate temporary vectorial force */
1262             tx               = _mm256_mul_pd(fscal,dx20);
1263             ty               = _mm256_mul_pd(fscal,dy20);
1264             tz               = _mm256_mul_pd(fscal,dz20);
1265
1266             /* Update vectorial force */
1267             fix2             = _mm256_add_pd(fix2,tx);
1268             fiy2             = _mm256_add_pd(fiy2,ty);
1269             fiz2             = _mm256_add_pd(fiz2,tz);
1270
1271             fjx0             = _mm256_add_pd(fjx0,tx);
1272             fjy0             = _mm256_add_pd(fjy0,ty);
1273             fjz0             = _mm256_add_pd(fjz0,tz);
1274
1275             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1276             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1277             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1278             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1279
1280             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1281
1282             /* Inner loop uses 145 flops */
1283         }
1284
1285         /* End of innermost loop */
1286
1287         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1288                                                  f+i_coord_offset,fshift+i_shift_offset);
1289
1290         /* Increment number of inner iterations */
1291         inneriter                  += j_index_end - j_index_start;
1292
1293         /* Outer loop uses 18 flops */
1294     }
1295
1296     /* Increment number of outer iterations */
1297     outeriter        += nri;
1298
1299     /* Update outer/inner flops */
1300
1301     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1302 }