src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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_VdwLJ_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  84     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  91     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m256d          dummy_mask,cutoff_mask;
  97     __m128           tmpmask0,tmpmask1;
  98     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  99     __m256d          one     = _mm256_set1_pd(1.0);
 100     __m256d          two     = _mm256_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm256_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     vftab            = kernel_data->table_elec->data;
 119     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 124     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 125     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 126     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 160
 161         fix0             = _mm256_setzero_pd();
 162         fiy0             = _mm256_setzero_pd();
 163         fiz0             = _mm256_setzero_pd();
 164         fix1             = _mm256_setzero_pd();
 165         fiy1             = _mm256_setzero_pd();
 166         fiz1             = _mm256_setzero_pd();
 167         fix2             = _mm256_setzero_pd();
 168         fiy2             = _mm256_setzero_pd();
 169         fiz2             = _mm256_setzero_pd();
 170         fix3             = _mm256_setzero_pd();
 171         fiy3             = _mm256_setzero_pd();
 172         fiz3             = _mm256_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_pd();
 176         vvdwsum          = _mm256_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm256_sub_pd(ix0,jx0);
 199             dy00             = _mm256_sub_pd(iy0,jy0);
 200             dz00             = _mm256_sub_pd(iz0,jz0);
 201             dx10             = _mm256_sub_pd(ix1,jx0);
 202             dy10             = _mm256_sub_pd(iy1,jy0);
 203             dz10             = _mm256_sub_pd(iz1,jz0);
 204             dx20             = _mm256_sub_pd(ix2,jx0);
 205             dy20             = _mm256_sub_pd(iy2,jy0);
 206             dz20             = _mm256_sub_pd(iz2,jz0);
 207             dx30             = _mm256_sub_pd(ix3,jx0);
 208             dy30             = _mm256_sub_pd(iy3,jy0);
 209             dz30             = _mm256_sub_pd(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 215             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 220
 221             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 225                                                                  charge+jnrC+0,charge+jnrD+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228             vdwjidx0C        = 2*vdwtype[jnrC+0];
 229             vdwjidx0D        = 2*vdwtype[jnrD+0];
 230
 231             fjx0             = _mm256_setzero_pd();
 232             fjy0             = _mm256_setzero_pd();
 233             fjz0             = _mm256_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 241                                             vdwioffsetptr0+vdwjidx0B,
 242                                             vdwioffsetptr0+vdwjidx0C,
 243                                             vdwioffsetptr0+vdwjidx0D,
 244                                             &c6_00,&c12_00);
 245
 246             /* LENNARD-JONES DISPERSION/REPULSION */
 247
 248             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 249             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 250             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 251             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 252             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 256
 257             fscal            = fvdw;
 258
 259             /* Calculate temporary vectorial force */
 260             tx               = _mm256_mul_pd(fscal,dx00);
 261             ty               = _mm256_mul_pd(fscal,dy00);
 262             tz               = _mm256_mul_pd(fscal,dz00);
 263
 264             /* Update vectorial force */
 265             fix0             = _mm256_add_pd(fix0,tx);
 266             fiy0             = _mm256_add_pd(fiy0,ty);
 267             fiz0             = _mm256_add_pd(fiz0,tz);
 268
 269             fjx0             = _mm256_add_pd(fjx0,tx);
 270             fjy0             = _mm256_add_pd(fjy0,ty);
 271             fjz0             = _mm256_add_pd(fjz0,tz);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r10              = _mm256_mul_pd(rsq10,rinv10);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq10             = _mm256_mul_pd(iq1,jq0);
 281
 282             /* Calculate table index by multiplying r with table scale and truncate to integer */
 283             rt               = _mm256_mul_pd(r10,vftabscale);
 284             vfitab           = _mm256_cvttpd_epi32(rt);
 285             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 286             vfitab           = _mm_slli_epi32(vfitab,2);
 287
 288             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 289             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 290             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 291             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 292             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 293             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 294             Heps             = _mm256_mul_pd(vfeps,H);
 295             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 296             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 297             velec            = _mm256_mul_pd(qq10,VV);
 298             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 299             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm256_add_pd(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm256_mul_pd(fscal,dx10);
 308             ty               = _mm256_mul_pd(fscal,dy10);
 309             tz               = _mm256_mul_pd(fscal,dz10);
 310
 311             /* Update vectorial force */
 312             fix1             = _mm256_add_pd(fix1,tx);
 313             fiy1             = _mm256_add_pd(fiy1,ty);
 314             fiz1             = _mm256_add_pd(fiz1,tz);
 315
 316             fjx0             = _mm256_add_pd(fjx0,tx);
 317             fjy0             = _mm256_add_pd(fjy0,ty);
 318             fjz0             = _mm256_add_pd(fjz0,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r20              = _mm256_mul_pd(rsq20,rinv20);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm256_mul_pd(iq2,jq0);
 328
 329             /* Calculate table index by multiplying r with table scale and truncate to integer */
 330             rt               = _mm256_mul_pd(r20,vftabscale);
 331             vfitab           = _mm256_cvttpd_epi32(rt);
 332             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 333             vfitab           = _mm_slli_epi32(vfitab,2);
 334
 335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 336             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 337             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 338             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 339             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 340             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 341             Heps             = _mm256_mul_pd(vfeps,H);
 342             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 343             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 344             velec            = _mm256_mul_pd(qq20,VV);
 345             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 346             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm256_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_pd(fscal,dx20);
 355             ty               = _mm256_mul_pd(fscal,dy20);
 356             tz               = _mm256_mul_pd(fscal,dz20);
 357
 358             /* Update vectorial force */
 359             fix2             = _mm256_add_pd(fix2,tx);
 360             fiy2             = _mm256_add_pd(fiy2,ty);
 361             fiz2             = _mm256_add_pd(fiz2,tz);
 362
 363             fjx0             = _mm256_add_pd(fjx0,tx);
 364             fjy0             = _mm256_add_pd(fjy0,ty);
 365             fjz0             = _mm256_add_pd(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r30              = _mm256_mul_pd(rsq30,rinv30);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq30             = _mm256_mul_pd(iq3,jq0);
 375
 376             /* Calculate table index by multiplying r with table scale and truncate to integer */
 377             rt               = _mm256_mul_pd(r30,vftabscale);
 378             vfitab           = _mm256_cvttpd_epi32(rt);
 379             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 380             vfitab           = _mm_slli_epi32(vfitab,2);
 381
 382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 383             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 384             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 385             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 386             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 387             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 388             Heps             = _mm256_mul_pd(vfeps,H);
 389             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 390             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 391             velec            = _mm256_mul_pd(qq30,VV);
 392             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 393             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm256_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm256_mul_pd(fscal,dx30);
 402             ty               = _mm256_mul_pd(fscal,dy30);
 403             tz               = _mm256_mul_pd(fscal,dz30);
 404
 405             /* Update vectorial force */
 406             fix3             = _mm256_add_pd(fix3,tx);
 407             fiy3             = _mm256_add_pd(fiy3,ty);
 408             fiz3             = _mm256_add_pd(fiz3,tz);
 409
 410             fjx0             = _mm256_add_pd(fjx0,tx);
 411             fjy0             = _mm256_add_pd(fjy0,ty);
 412             fjz0             = _mm256_add_pd(fjz0,tz);
 413
 414             fjptrA             = f+j_coord_offsetA;
 415             fjptrB             = f+j_coord_offsetB;
 416             fjptrC             = f+j_coord_offsetC;
 417             fjptrD             = f+j_coord_offsetD;
 418
 419             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 164 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             /* Sign of each element will be negative for non-real atoms.
 433              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 434              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 435              */
 436             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 437
 438             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 439             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 440             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 441
 442             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 443             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 444             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 445             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450
 451             /* load j atom coordinates */
 452             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 453                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 454                                                  &jx0,&jy0,&jz0);
 455
 456             /* Calculate displacement vector */
 457             dx00             = _mm256_sub_pd(ix0,jx0);
 458             dy00             = _mm256_sub_pd(iy0,jy0);
 459             dz00             = _mm256_sub_pd(iz0,jz0);
 460             dx10             = _mm256_sub_pd(ix1,jx0);
 461             dy10             = _mm256_sub_pd(iy1,jy0);
 462             dz10             = _mm256_sub_pd(iz1,jz0);
 463             dx20             = _mm256_sub_pd(ix2,jx0);
 464             dy20             = _mm256_sub_pd(iy2,jy0);
 465             dz20             = _mm256_sub_pd(iz2,jz0);
 466             dx30             = _mm256_sub_pd(ix3,jx0);
 467             dy30             = _mm256_sub_pd(iy3,jy0);
 468             dz30             = _mm256_sub_pd(iz3,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 472             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 473             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 474             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 475
 476             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 477             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 478             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 479
 480             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 481
 482             /* Load parameters for j particles */
 483             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 484                                                                  charge+jnrC+0,charge+jnrD+0);
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487             vdwjidx0C        = 2*vdwtype[jnrC+0];
 488             vdwjidx0D        = 2*vdwtype[jnrD+0];
 489
 490             fjx0             = _mm256_setzero_pd();
 491             fjy0             = _mm256_setzero_pd();
 492             fjz0             = _mm256_setzero_pd();
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 500                                             vdwioffsetptr0+vdwjidx0B,
 501                                             vdwioffsetptr0+vdwjidx0C,
 502                                             vdwioffsetptr0+vdwjidx0D,
 503                                             &c6_00,&c12_00);
 504
 505             /* LENNARD-JONES DISPERSION/REPULSION */
 506
 507             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 508             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 509             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 510             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 511             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 515             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 516
 517             fscal            = fvdw;
 518
 519             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx00);
 523             ty               = _mm256_mul_pd(fscal,dy00);
 524             tz               = _mm256_mul_pd(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm256_add_pd(fix0,tx);
 528             fiy0             = _mm256_add_pd(fiy0,ty);
 529             fiz0             = _mm256_add_pd(fiz0,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r10              = _mm256_mul_pd(rsq10,rinv10);
 540             r10              = _mm256_andnot_pd(dummy_mask,r10);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq10             = _mm256_mul_pd(iq1,jq0);
 544
 545             /* Calculate table index by multiplying r with table scale and truncate to integer */
 546             rt               = _mm256_mul_pd(r10,vftabscale);
 547             vfitab           = _mm256_cvttpd_epi32(rt);
 548             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 549             vfitab           = _mm_slli_epi32(vfitab,2);
 550
 551             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 552             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 553             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 554             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 555             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 556             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 557             Heps             = _mm256_mul_pd(vfeps,H);
 558             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 559             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 560             velec            = _mm256_mul_pd(qq10,VV);
 561             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 562             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm256_andnot_pd(dummy_mask,velec);
 566             velecsum         = _mm256_add_pd(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm256_mul_pd(fscal,dx10);
 574             ty               = _mm256_mul_pd(fscal,dy10);
 575             tz               = _mm256_mul_pd(fscal,dz10);
 576
 577             /* Update vectorial force */
 578             fix1             = _mm256_add_pd(fix1,tx);
 579             fiy1             = _mm256_add_pd(fiy1,ty);
 580             fiz1             = _mm256_add_pd(fiz1,tz);
 581
 582             fjx0             = _mm256_add_pd(fjx0,tx);
 583             fjy0             = _mm256_add_pd(fjy0,ty);
 584             fjz0             = _mm256_add_pd(fjz0,tz);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             r20              = _mm256_mul_pd(rsq20,rinv20);
 591             r20              = _mm256_andnot_pd(dummy_mask,r20);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq20             = _mm256_mul_pd(iq2,jq0);
 595
 596             /* Calculate table index by multiplying r with table scale and truncate to integer */
 597             rt               = _mm256_mul_pd(r20,vftabscale);
 598             vfitab           = _mm256_cvttpd_epi32(rt);
 599             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 600             vfitab           = _mm_slli_epi32(vfitab,2);
 601
 602             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 603             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 604             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 605             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 606             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 607             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 608             Heps             = _mm256_mul_pd(vfeps,H);
 609             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 610             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 611             velec            = _mm256_mul_pd(qq20,VV);
 612             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 613             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 614
 615             /* Update potential sum for this i atom from the interaction with this j atom. */
 616             velec            = _mm256_andnot_pd(dummy_mask,velec);
 617             velecsum         = _mm256_add_pd(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm256_mul_pd(fscal,dx20);
 625             ty               = _mm256_mul_pd(fscal,dy20);
 626             tz               = _mm256_mul_pd(fscal,dz20);
 627
 628             /* Update vectorial force */
 629             fix2             = _mm256_add_pd(fix2,tx);
 630             fiy2             = _mm256_add_pd(fiy2,ty);
 631             fiz2             = _mm256_add_pd(fiz2,tz);
 632
 633             fjx0             = _mm256_add_pd(fjx0,tx);
 634             fjy0             = _mm256_add_pd(fjy0,ty);
 635             fjz0             = _mm256_add_pd(fjz0,tz);
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             r30              = _mm256_mul_pd(rsq30,rinv30);
 642             r30              = _mm256_andnot_pd(dummy_mask,r30);
 643
 644             /* Compute parameters for interactions between i and j atoms */
 645             qq30             = _mm256_mul_pd(iq3,jq0);
 646
 647             /* Calculate table index by multiplying r with table scale and truncate to integer */
 648             rt               = _mm256_mul_pd(r30,vftabscale);
 649             vfitab           = _mm256_cvttpd_epi32(rt);
 650             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 651             vfitab           = _mm_slli_epi32(vfitab,2);
 652
 653             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 654             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 655             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 656             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 657             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 658             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 659             Heps             = _mm256_mul_pd(vfeps,H);
 660             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 661             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 662             velec            = _mm256_mul_pd(qq30,VV);
 663             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 664             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 665
 666             /* Update potential sum for this i atom from the interaction with this j atom. */
 667             velec            = _mm256_andnot_pd(dummy_mask,velec);
 668             velecsum         = _mm256_add_pd(velecsum,velec);
 669
 670             fscal            = felec;
 671
 672             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 673
 674             /* Calculate temporary vectorial force */
 675             tx               = _mm256_mul_pd(fscal,dx30);
 676             ty               = _mm256_mul_pd(fscal,dy30);
 677             tz               = _mm256_mul_pd(fscal,dz30);
 678
 679             /* Update vectorial force */
 680             fix3             = _mm256_add_pd(fix3,tx);
 681             fiy3             = _mm256_add_pd(fiy3,ty);
 682             fiz3             = _mm256_add_pd(fiz3,tz);
 683
 684             fjx0             = _mm256_add_pd(fjx0,tx);
 685             fjy0             = _mm256_add_pd(fjy0,ty);
 686             fjz0             = _mm256_add_pd(fjz0,tz);
 687
 688             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 689             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 690             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 691             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 692
 693             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 694
 695             /* Inner loop uses 167 flops */
 696         }
 697
 698         /* End of innermost loop */
 699
 700         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 701                                                  f+i_coord_offset,fshift+i_shift_offset);
 702
 703         ggid                        = gid[iidx];
 704         /* Update potential energies */
 705         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 706         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 707
 708         /* Increment number of inner iterations */
 709         inneriter                  += j_index_end - j_index_start;
 710
 711         /* Outer loop uses 26 flops */
 712     }
 713
 714     /* Increment number of outer iterations */
 715     outeriter        += nri;
 716
 717     /* Update outer/inner flops */
 718
 719     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 720 }
 721 /*
 722  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_double
 723  * Electrostatics interaction: CubicSplineTable
 724  * VdW interaction:            LennardJones
 725  * Geometry:                   Water4-Particle
 726  * Calculate force/pot:        Force
 727  */
 728 void
 729 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_double
 730                     (t_nblist * gmx_restrict                nlist,
 731                      rvec * gmx_restrict                    xx,
 732                      rvec * gmx_restrict                    ff,
 733                      t_forcerec * gmx_restrict              fr,
 734                      t_mdatoms * gmx_restrict               mdatoms,
 735                      nb_kernel_data_t * gmx_restrict        kernel_data,
 736                      t_nrnb * gmx_restrict                  nrnb)
 737 {
 738     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 739      * just 0 for non-waters.
 740      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 741      * jnr indices corresponding to data put in the four positions in the SIMD register.
 742      */
 743     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 744     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 745     int              jnrA,jnrB,jnrC,jnrD;
 746     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 747     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 748     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 749     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 750     real             rcutoff_scalar;
 751     real             *shiftvec,*fshift,*x,*f;
 752     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 753     real             scratch[4*DIM];
 754     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 755     real *           vdwioffsetptr0;
 756     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 757     real *           vdwioffsetptr1;
 758     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 759     real *           vdwioffsetptr2;
 760     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 761     real *           vdwioffsetptr3;
 762     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 763     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 764     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 765     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 766     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 767     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 768     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 769     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 770     real             *charge;
 771     int              nvdwtype;
 772     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 773     int              *vdwtype;
 774     real             *vdwparam;
 775     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 776     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 777     __m128i          vfitab;
 778     __m128i          ifour       = _mm_set1_epi32(4);
 779     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 780     real             *vftab;
 781     __m256d          dummy_mask,cutoff_mask;
 782     __m128           tmpmask0,tmpmask1;
 783     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 784     __m256d          one     = _mm256_set1_pd(1.0);
 785     __m256d          two     = _mm256_set1_pd(2.0);
 786     x                = xx[0];
 787     f                = ff[0];
 788
 789     nri              = nlist->nri;
 790     iinr             = nlist->iinr;
 791     jindex           = nlist->jindex;
 792     jjnr             = nlist->jjnr;
 793     shiftidx         = nlist->shift;
 794     gid              = nlist->gid;
 795     shiftvec         = fr->shift_vec[0];
 796     fshift           = fr->fshift[0];
 797     facel            = _mm256_set1_pd(fr->epsfac);
 798     charge           = mdatoms->chargeA;
 799     nvdwtype         = fr->ntype;
 800     vdwparam         = fr->nbfp;
 801     vdwtype          = mdatoms->typeA;
 802
 803     vftab            = kernel_data->table_elec->data;
 804     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 805
 806     /* Setup water-specific parameters */
 807     inr              = nlist->iinr[0];
 808     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 809     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 810     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 811     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 812
 813     /* Avoid stupid compiler warnings */
 814     jnrA = jnrB = jnrC = jnrD = 0;
 815     j_coord_offsetA = 0;
 816     j_coord_offsetB = 0;
 817     j_coord_offsetC = 0;
 818     j_coord_offsetD = 0;
 819
 820     outeriter        = 0;
 821     inneriter        = 0;
 822
 823     for(iidx=0;iidx<4*DIM;iidx++)
 824     {
 825         scratch[iidx] = 0.0;
 826     }
 827
 828     /* Start outer loop over neighborlists */
 829     for(iidx=0; iidx<nri; iidx++)
 830     {
 831         /* Load shift vector for this list */
 832         i_shift_offset   = DIM*shiftidx[iidx];
 833
 834         /* Load limits for loop over neighbors */
 835         j_index_start    = jindex[iidx];
 836         j_index_end      = jindex[iidx+1];
 837
 838         /* Get outer coordinate index */
 839         inr              = iinr[iidx];
 840         i_coord_offset   = DIM*inr;
 841
 842         /* Load i particle coords and add shift vector */
 843         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 844                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 845
 846         fix0             = _mm256_setzero_pd();
 847         fiy0             = _mm256_setzero_pd();
 848         fiz0             = _mm256_setzero_pd();
 849         fix1             = _mm256_setzero_pd();
 850         fiy1             = _mm256_setzero_pd();
 851         fiz1             = _mm256_setzero_pd();
 852         fix2             = _mm256_setzero_pd();
 853         fiy2             = _mm256_setzero_pd();
 854         fiz2             = _mm256_setzero_pd();
 855         fix3             = _mm256_setzero_pd();
 856         fiy3             = _mm256_setzero_pd();
 857         fiz3             = _mm256_setzero_pd();
 858
 859         /* Start inner kernel loop */
 860         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 861         {
 862
 863             /* Get j neighbor index, and coordinate index */
 864             jnrA             = jjnr[jidx];
 865             jnrB             = jjnr[jidx+1];
 866             jnrC             = jjnr[jidx+2];
 867             jnrD             = jjnr[jidx+3];
 868             j_coord_offsetA  = DIM*jnrA;
 869             j_coord_offsetB  = DIM*jnrB;
 870             j_coord_offsetC  = DIM*jnrC;
 871             j_coord_offsetD  = DIM*jnrD;
 872
 873             /* load j atom coordinates */
 874             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 875                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 876                                                  &jx0,&jy0,&jz0);
 877
 878             /* Calculate displacement vector */
 879             dx00             = _mm256_sub_pd(ix0,jx0);
 880             dy00             = _mm256_sub_pd(iy0,jy0);
 881             dz00             = _mm256_sub_pd(iz0,jz0);
 882             dx10             = _mm256_sub_pd(ix1,jx0);
 883             dy10             = _mm256_sub_pd(iy1,jy0);
 884             dz10             = _mm256_sub_pd(iz1,jz0);
 885             dx20             = _mm256_sub_pd(ix2,jx0);
 886             dy20             = _mm256_sub_pd(iy2,jy0);
 887             dz20             = _mm256_sub_pd(iz2,jz0);
 888             dx30             = _mm256_sub_pd(ix3,jx0);
 889             dy30             = _mm256_sub_pd(iy3,jy0);
 890             dz30             = _mm256_sub_pd(iz3,jz0);
 891
 892             /* Calculate squared distance and things based on it */
 893             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 894             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 895             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 896             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 897
 898             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 899             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 900             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 901
 902             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 903
 904             /* Load parameters for j particles */
 905             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 906                                                                  charge+jnrC+0,charge+jnrD+0);
 907             vdwjidx0A        = 2*vdwtype[jnrA+0];
 908             vdwjidx0B        = 2*vdwtype[jnrB+0];
 909             vdwjidx0C        = 2*vdwtype[jnrC+0];
 910             vdwjidx0D        = 2*vdwtype[jnrD+0];
 911
 912             fjx0             = _mm256_setzero_pd();
 913             fjy0             = _mm256_setzero_pd();
 914             fjz0             = _mm256_setzero_pd();
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 922                                             vdwioffsetptr0+vdwjidx0B,
 923                                             vdwioffsetptr0+vdwjidx0C,
 924                                             vdwioffsetptr0+vdwjidx0D,
 925                                             &c6_00,&c12_00);
 926
 927             /* LENNARD-JONES DISPERSION/REPULSION */
 928
 929             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 930             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 931
 932             fscal            = fvdw;
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm256_mul_pd(fscal,dx00);
 936             ty               = _mm256_mul_pd(fscal,dy00);
 937             tz               = _mm256_mul_pd(fscal,dz00);
 938
 939             /* Update vectorial force */
 940             fix0             = _mm256_add_pd(fix0,tx);
 941             fiy0             = _mm256_add_pd(fiy0,ty);
 942             fiz0             = _mm256_add_pd(fiz0,tz);
 943
 944             fjx0             = _mm256_add_pd(fjx0,tx);
 945             fjy0             = _mm256_add_pd(fjy0,ty);
 946             fjz0             = _mm256_add_pd(fjz0,tz);
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             r10              = _mm256_mul_pd(rsq10,rinv10);
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq10             = _mm256_mul_pd(iq1,jq0);
 956
 957             /* Calculate table index by multiplying r with table scale and truncate to integer */
 958             rt               = _mm256_mul_pd(r10,vftabscale);
 959             vfitab           = _mm256_cvttpd_epi32(rt);
 960             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 961             vfitab           = _mm_slli_epi32(vfitab,2);
 962
 963             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 964             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 965             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 966             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 967             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 968             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 969             Heps             = _mm256_mul_pd(vfeps,H);
 970             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 971             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 972             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 973
 974             fscal            = felec;
 975
 976             /* Calculate temporary vectorial force */
 977             tx               = _mm256_mul_pd(fscal,dx10);
 978             ty               = _mm256_mul_pd(fscal,dy10);
 979             tz               = _mm256_mul_pd(fscal,dz10);
 980
 981             /* Update vectorial force */
 982             fix1             = _mm256_add_pd(fix1,tx);
 983             fiy1             = _mm256_add_pd(fiy1,ty);
 984             fiz1             = _mm256_add_pd(fiz1,tz);
 985
 986             fjx0             = _mm256_add_pd(fjx0,tx);
 987             fjy0             = _mm256_add_pd(fjy0,ty);
 988             fjz0             = _mm256_add_pd(fjz0,tz);
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             r20              = _mm256_mul_pd(rsq20,rinv20);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq20             = _mm256_mul_pd(iq2,jq0);
 998
 999             /* Calculate table index by multiplying r with table scale and truncate to integer */
1000             rt               = _mm256_mul_pd(r20,vftabscale);
1001             vfitab           = _mm256_cvttpd_epi32(rt);
1002             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1003             vfitab           = _mm_slli_epi32(vfitab,2);
1004
1005             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1006             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1007             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1008             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1009             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1010             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1011             Heps             = _mm256_mul_pd(vfeps,H);
1012             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1013             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1014             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1015
1016             fscal            = felec;
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm256_mul_pd(fscal,dx20);
1020             ty               = _mm256_mul_pd(fscal,dy20);
1021             tz               = _mm256_mul_pd(fscal,dz20);
1022
1023             /* Update vectorial force */
1024             fix2             = _mm256_add_pd(fix2,tx);
1025             fiy2             = _mm256_add_pd(fiy2,ty);
1026             fiz2             = _mm256_add_pd(fiz2,tz);
1027
1028             fjx0             = _mm256_add_pd(fjx0,tx);
1029             fjy0             = _mm256_add_pd(fjy0,ty);
1030             fjz0             = _mm256_add_pd(fjz0,tz);
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             r30              = _mm256_mul_pd(rsq30,rinv30);
1037
1038             /* Compute parameters for interactions between i and j atoms */
1039             qq30             = _mm256_mul_pd(iq3,jq0);
1040
1041             /* Calculate table index by multiplying r with table scale and truncate to integer */
1042             rt               = _mm256_mul_pd(r30,vftabscale);
1043             vfitab           = _mm256_cvttpd_epi32(rt);
1044             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1045             vfitab           = _mm_slli_epi32(vfitab,2);
1046
1047             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1048             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1049             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1050             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1051             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1052             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1053             Heps             = _mm256_mul_pd(vfeps,H);
1054             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1055             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1056             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1057
1058             fscal            = felec;
1059
1060             /* Calculate temporary vectorial force */
1061             tx               = _mm256_mul_pd(fscal,dx30);
1062             ty               = _mm256_mul_pd(fscal,dy30);
1063             tz               = _mm256_mul_pd(fscal,dz30);
1064
1065             /* Update vectorial force */
1066             fix3             = _mm256_add_pd(fix3,tx);
1067             fiy3             = _mm256_add_pd(fiy3,ty);
1068             fiz3             = _mm256_add_pd(fiz3,tz);
1069
1070             fjx0             = _mm256_add_pd(fjx0,tx);
1071             fjy0             = _mm256_add_pd(fjy0,ty);
1072             fjz0             = _mm256_add_pd(fjz0,tz);
1073
1074             fjptrA             = f+j_coord_offsetA;
1075             fjptrB             = f+j_coord_offsetB;
1076             fjptrC             = f+j_coord_offsetC;
1077             fjptrD             = f+j_coord_offsetD;
1078
1079             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1080
1081             /* Inner loop uses 147 flops */
1082         }
1083
1084         if(jidx<j_index_end)
1085         {
1086
1087             /* Get j neighbor index, and coordinate index */
1088             jnrlistA         = jjnr[jidx];
1089             jnrlistB         = jjnr[jidx+1];
1090             jnrlistC         = jjnr[jidx+2];
1091             jnrlistD         = jjnr[jidx+3];
1092             /* Sign of each element will be negative for non-real atoms.
1093              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1094              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1095              */
1096             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1097
1098             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1099             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1100             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1101
1102             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1103             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1104             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1105             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1106             j_coord_offsetA  = DIM*jnrA;
1107             j_coord_offsetB  = DIM*jnrB;
1108             j_coord_offsetC  = DIM*jnrC;
1109             j_coord_offsetD  = DIM*jnrD;
1110
1111             /* load j atom coordinates */
1112             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1113                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1114                                                  &jx0,&jy0,&jz0);
1115
1116             /* Calculate displacement vector */
1117             dx00             = _mm256_sub_pd(ix0,jx0);
1118             dy00             = _mm256_sub_pd(iy0,jy0);
1119             dz00             = _mm256_sub_pd(iz0,jz0);
1120             dx10             = _mm256_sub_pd(ix1,jx0);
1121             dy10             = _mm256_sub_pd(iy1,jy0);
1122             dz10             = _mm256_sub_pd(iz1,jz0);
1123             dx20             = _mm256_sub_pd(ix2,jx0);
1124             dy20             = _mm256_sub_pd(iy2,jy0);
1125             dz20             = _mm256_sub_pd(iz2,jz0);
1126             dx30             = _mm256_sub_pd(ix3,jx0);
1127             dy30             = _mm256_sub_pd(iy3,jy0);
1128             dz30             = _mm256_sub_pd(iz3,jz0);
1129
1130             /* Calculate squared distance and things based on it */
1131             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1132             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1133             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1134             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1135
1136             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1137             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1138             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1139
1140             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1141
1142             /* Load parameters for j particles */
1143             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1144                                                                  charge+jnrC+0,charge+jnrD+0);
1145             vdwjidx0A        = 2*vdwtype[jnrA+0];
1146             vdwjidx0B        = 2*vdwtype[jnrB+0];
1147             vdwjidx0C        = 2*vdwtype[jnrC+0];
1148             vdwjidx0D        = 2*vdwtype[jnrD+0];
1149
1150             fjx0             = _mm256_setzero_pd();
1151             fjy0             = _mm256_setzero_pd();
1152             fjz0             = _mm256_setzero_pd();
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1160                                             vdwioffsetptr0+vdwjidx0B,
1161                                             vdwioffsetptr0+vdwjidx0C,
1162                                             vdwioffsetptr0+vdwjidx0D,
1163                                             &c6_00,&c12_00);
1164
1165             /* LENNARD-JONES DISPERSION/REPULSION */
1166
1167             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1168             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1169
1170             fscal            = fvdw;
1171
1172             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm256_mul_pd(fscal,dx00);
1176             ty               = _mm256_mul_pd(fscal,dy00);
1177             tz               = _mm256_mul_pd(fscal,dz00);
1178
1179             /* Update vectorial force */
1180             fix0             = _mm256_add_pd(fix0,tx);
1181             fiy0             = _mm256_add_pd(fiy0,ty);
1182             fiz0             = _mm256_add_pd(fiz0,tz);
1183
1184             fjx0             = _mm256_add_pd(fjx0,tx);
1185             fjy0             = _mm256_add_pd(fjy0,ty);
1186             fjz0             = _mm256_add_pd(fjz0,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r10              = _mm256_mul_pd(rsq10,rinv10);
1193             r10              = _mm256_andnot_pd(dummy_mask,r10);
1194
1195             /* Compute parameters for interactions between i and j atoms */
1196             qq10             = _mm256_mul_pd(iq1,jq0);
1197
1198             /* Calculate table index by multiplying r with table scale and truncate to integer */
1199             rt               = _mm256_mul_pd(r10,vftabscale);
1200             vfitab           = _mm256_cvttpd_epi32(rt);
1201             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1202             vfitab           = _mm_slli_epi32(vfitab,2);
1203
1204             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1205             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1206             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1207             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1208             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1209             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1210             Heps             = _mm256_mul_pd(vfeps,H);
1211             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1212             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1213             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1218
1219             /* Calculate temporary vectorial force */
1220             tx               = _mm256_mul_pd(fscal,dx10);
1221             ty               = _mm256_mul_pd(fscal,dy10);
1222             tz               = _mm256_mul_pd(fscal,dz10);
1223
1224             /* Update vectorial force */
1225             fix1             = _mm256_add_pd(fix1,tx);
1226             fiy1             = _mm256_add_pd(fiy1,ty);
1227             fiz1             = _mm256_add_pd(fiz1,tz);
1228
1229             fjx0             = _mm256_add_pd(fjx0,tx);
1230             fjy0             = _mm256_add_pd(fjy0,ty);
1231             fjz0             = _mm256_add_pd(fjz0,tz);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             r20              = _mm256_mul_pd(rsq20,rinv20);
1238             r20              = _mm256_andnot_pd(dummy_mask,r20);
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq20             = _mm256_mul_pd(iq2,jq0);
1242
1243             /* Calculate table index by multiplying r with table scale and truncate to integer */
1244             rt               = _mm256_mul_pd(r20,vftabscale);
1245             vfitab           = _mm256_cvttpd_epi32(rt);
1246             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1247             vfitab           = _mm_slli_epi32(vfitab,2);
1248
1249             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1250             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1251             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1252             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1253             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1254             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1255             Heps             = _mm256_mul_pd(vfeps,H);
1256             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1257             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1258             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1259
1260             fscal            = felec;
1261
1262             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1263
1264             /* Calculate temporary vectorial force */
1265             tx               = _mm256_mul_pd(fscal,dx20);
1266             ty               = _mm256_mul_pd(fscal,dy20);
1267             tz               = _mm256_mul_pd(fscal,dz20);
1268
1269             /* Update vectorial force */
1270             fix2             = _mm256_add_pd(fix2,tx);
1271             fiy2             = _mm256_add_pd(fiy2,ty);
1272             fiz2             = _mm256_add_pd(fiz2,tz);
1273
1274             fjx0             = _mm256_add_pd(fjx0,tx);
1275             fjy0             = _mm256_add_pd(fjy0,ty);
1276             fjz0             = _mm256_add_pd(fjz0,tz);
1277
1278             /**************************
1279              * CALCULATE INTERACTIONS *
1280              **************************/
1281
1282             r30              = _mm256_mul_pd(rsq30,rinv30);
1283             r30              = _mm256_andnot_pd(dummy_mask,r30);
1284
1285             /* Compute parameters for interactions between i and j atoms */
1286             qq30             = _mm256_mul_pd(iq3,jq0);
1287
1288             /* Calculate table index by multiplying r with table scale and truncate to integer */
1289             rt               = _mm256_mul_pd(r30,vftabscale);
1290             vfitab           = _mm256_cvttpd_epi32(rt);
1291             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1292             vfitab           = _mm_slli_epi32(vfitab,2);
1293
1294             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1295             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1296             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1297             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1298             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1299             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1300             Heps             = _mm256_mul_pd(vfeps,H);
1301             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1302             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1303             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1304
1305             fscal            = felec;
1306
1307             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1308
1309             /* Calculate temporary vectorial force */
1310             tx               = _mm256_mul_pd(fscal,dx30);
1311             ty               = _mm256_mul_pd(fscal,dy30);
1312             tz               = _mm256_mul_pd(fscal,dz30);
1313
1314             /* Update vectorial force */
1315             fix3             = _mm256_add_pd(fix3,tx);
1316             fiy3             = _mm256_add_pd(fiy3,ty);
1317             fiz3             = _mm256_add_pd(fiz3,tz);
1318
1319             fjx0             = _mm256_add_pd(fjx0,tx);
1320             fjy0             = _mm256_add_pd(fjy0,ty);
1321             fjz0             = _mm256_add_pd(fjz0,tz);
1322
1323             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1324             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1325             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1326             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1327
1328             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1329
1330             /* Inner loop uses 150 flops */
1331         }
1332
1333         /* End of innermost loop */
1334
1335         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1336                                                  f+i_coord_offset,fshift+i_shift_offset);
1337
1338         /* Increment number of inner iterations */
1339         inneriter                  += j_index_end - j_index_start;
1340
1341         /* Outer loop uses 24 flops */
1342     }
1343
1344     /* Increment number of outer iterations */
1345     outeriter        += nri;
1346
1347     /* Update outer/inner flops */
1348
1349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1350 }