src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_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_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_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     __m256d          dummy_mask,cutoff_mask;
  90     __m128           tmpmask0,tmpmask1;
  91     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  92     __m256d          one     = _mm256_set1_pd(1.0);
  93     __m256d          two     = _mm256_set1_pd(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm256_set1_pd(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     /* Setup water-specific parameters */
 112     inr              = nlist->iinr[0];
 113     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 114     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 115     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 116     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = jnrC = jnrD = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122     j_coord_offsetC = 0;
 123     j_coord_offsetD = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     for(iidx=0;iidx<4*DIM;iidx++)
 129     {
 130         scratch[iidx] = 0.0;
 131     }
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 149                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 150
 151         fix0             = _mm256_setzero_pd();
 152         fiy0             = _mm256_setzero_pd();
 153         fiz0             = _mm256_setzero_pd();
 154         fix1             = _mm256_setzero_pd();
 155         fiy1             = _mm256_setzero_pd();
 156         fiz1             = _mm256_setzero_pd();
 157         fix2             = _mm256_setzero_pd();
 158         fiy2             = _mm256_setzero_pd();
 159         fiz2             = _mm256_setzero_pd();
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm256_setzero_pd();
 163         vvdwsum          = _mm256_setzero_pd();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176             j_coord_offsetC  = DIM*jnrC;
 177             j_coord_offsetD  = DIM*jnrD;
 178
 179             /* load j atom coordinates */
 180             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 182                                                  &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm256_sub_pd(ix0,jx0);
 186             dy00             = _mm256_sub_pd(iy0,jy0);
 187             dz00             = _mm256_sub_pd(iz0,jz0);
 188             dx10             = _mm256_sub_pd(ix1,jx0);
 189             dy10             = _mm256_sub_pd(iy1,jy0);
 190             dz10             = _mm256_sub_pd(iz1,jz0);
 191             dx20             = _mm256_sub_pd(ix2,jx0);
 192             dy20             = _mm256_sub_pd(iy2,jy0);
 193             dz20             = _mm256_sub_pd(iz2,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 197             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 198             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 199
 200             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 201             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 202             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 203
 204             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 205             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 206             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 210                                                                  charge+jnrC+0,charge+jnrD+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213             vdwjidx0C        = 2*vdwtype[jnrC+0];
 214             vdwjidx0D        = 2*vdwtype[jnrD+0];
 215
 216             fjx0             = _mm256_setzero_pd();
 217             fjy0             = _mm256_setzero_pd();
 218             fjz0             = _mm256_setzero_pd();
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm256_mul_pd(iq0,jq0);
 226             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 227                                             vdwioffsetptr0+vdwjidx0B,
 228                                             vdwioffsetptr0+vdwjidx0C,
 229                                             vdwioffsetptr0+vdwjidx0D,
 230                                             &c6_00,&c12_00);
 231
 232             /* COULOMB ELECTROSTATICS */
 233             velec            = _mm256_mul_pd(qq00,rinv00);
 234             felec            = _mm256_mul_pd(velec,rinvsq00);
 235
 236             /* LENNARD-JONES DISPERSION/REPULSION */
 237
 238             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 239             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 240             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 241             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 242             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm256_add_pd(velecsum,velec);
 246             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 247
 248             fscal            = _mm256_add_pd(felec,fvdw);
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm256_mul_pd(fscal,dx00);
 252             ty               = _mm256_mul_pd(fscal,dy00);
 253             tz               = _mm256_mul_pd(fscal,dz00);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm256_add_pd(fix0,tx);
 257             fiy0             = _mm256_add_pd(fiy0,ty);
 258             fiz0             = _mm256_add_pd(fiz0,tz);
 259
 260             fjx0             = _mm256_add_pd(fjx0,tx);
 261             fjy0             = _mm256_add_pd(fjy0,ty);
 262             fjz0             = _mm256_add_pd(fjz0,tz);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             qq10             = _mm256_mul_pd(iq1,jq0);
 270
 271             /* COULOMB ELECTROSTATICS */
 272             velec            = _mm256_mul_pd(qq10,rinv10);
 273             felec            = _mm256_mul_pd(velec,rinvsq10);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velecsum         = _mm256_add_pd(velecsum,velec);
 277
 278             fscal            = felec;
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm256_mul_pd(fscal,dx10);
 282             ty               = _mm256_mul_pd(fscal,dy10);
 283             tz               = _mm256_mul_pd(fscal,dz10);
 284
 285             /* Update vectorial force */
 286             fix1             = _mm256_add_pd(fix1,tx);
 287             fiy1             = _mm256_add_pd(fiy1,ty);
 288             fiz1             = _mm256_add_pd(fiz1,tz);
 289
 290             fjx0             = _mm256_add_pd(fjx0,tx);
 291             fjy0             = _mm256_add_pd(fjy0,ty);
 292             fjz0             = _mm256_add_pd(fjz0,tz);
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq20             = _mm256_mul_pd(iq2,jq0);
 300
 301             /* COULOMB ELECTROSTATICS */
 302             velec            = _mm256_mul_pd(qq20,rinv20);
 303             felec            = _mm256_mul_pd(velec,rinvsq20);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velecsum         = _mm256_add_pd(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_pd(fscal,dx20);
 312             ty               = _mm256_mul_pd(fscal,dy20);
 313             tz               = _mm256_mul_pd(fscal,dz20);
 314
 315             /* Update vectorial force */
 316             fix2             = _mm256_add_pd(fix2,tx);
 317             fiy2             = _mm256_add_pd(fiy2,ty);
 318             fiz2             = _mm256_add_pd(fiz2,tz);
 319
 320             fjx0             = _mm256_add_pd(fjx0,tx);
 321             fjy0             = _mm256_add_pd(fjy0,ty);
 322             fjz0             = _mm256_add_pd(fjz0,tz);
 323
 324             fjptrA             = f+j_coord_offsetA;
 325             fjptrB             = f+j_coord_offsetB;
 326             fjptrC             = f+j_coord_offsetC;
 327             fjptrD             = f+j_coord_offsetD;
 328
 329             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 330
 331             /* Inner loop uses 96 flops */
 332         }
 333
 334         if(jidx<j_index_end)
 335         {
 336
 337             /* Get j neighbor index, and coordinate index */
 338             jnrlistA         = jjnr[jidx];
 339             jnrlistB         = jjnr[jidx+1];
 340             jnrlistC         = jjnr[jidx+2];
 341             jnrlistD         = jjnr[jidx+3];
 342             /* Sign of each element will be negative for non-real atoms.
 343              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 344              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 345              */
 346             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 347
 348             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 349             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 350             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 351
 352             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 353             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 354             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 355             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 356             j_coord_offsetA  = DIM*jnrA;
 357             j_coord_offsetB  = DIM*jnrB;
 358             j_coord_offsetC  = DIM*jnrC;
 359             j_coord_offsetD  = DIM*jnrD;
 360
 361             /* load j atom coordinates */
 362             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 363                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 364                                                  &jx0,&jy0,&jz0);
 365
 366             /* Calculate displacement vector */
 367             dx00             = _mm256_sub_pd(ix0,jx0);
 368             dy00             = _mm256_sub_pd(iy0,jy0);
 369             dz00             = _mm256_sub_pd(iz0,jz0);
 370             dx10             = _mm256_sub_pd(ix1,jx0);
 371             dy10             = _mm256_sub_pd(iy1,jy0);
 372             dz10             = _mm256_sub_pd(iz1,jz0);
 373             dx20             = _mm256_sub_pd(ix2,jx0);
 374             dy20             = _mm256_sub_pd(iy2,jy0);
 375             dz20             = _mm256_sub_pd(iz2,jz0);
 376
 377             /* Calculate squared distance and things based on it */
 378             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 379             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 380             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 381
 382             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 383             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 384             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 385
 386             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 387             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 388             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 389
 390             /* Load parameters for j particles */
 391             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 392                                                                  charge+jnrC+0,charge+jnrD+0);
 393             vdwjidx0A        = 2*vdwtype[jnrA+0];
 394             vdwjidx0B        = 2*vdwtype[jnrB+0];
 395             vdwjidx0C        = 2*vdwtype[jnrC+0];
 396             vdwjidx0D        = 2*vdwtype[jnrD+0];
 397
 398             fjx0             = _mm256_setzero_pd();
 399             fjy0             = _mm256_setzero_pd();
 400             fjz0             = _mm256_setzero_pd();
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq00             = _mm256_mul_pd(iq0,jq0);
 408             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 409                                             vdwioffsetptr0+vdwjidx0B,
 410                                             vdwioffsetptr0+vdwjidx0C,
 411                                             vdwioffsetptr0+vdwjidx0D,
 412                                             &c6_00,&c12_00);
 413
 414             /* COULOMB ELECTROSTATICS */
 415             velec            = _mm256_mul_pd(qq00,rinv00);
 416             felec            = _mm256_mul_pd(velec,rinvsq00);
 417
 418             /* LENNARD-JONES DISPERSION/REPULSION */
 419
 420             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 421             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 422             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 423             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 424             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm256_andnot_pd(dummy_mask,velec);
 428             velecsum         = _mm256_add_pd(velecsum,velec);
 429             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 430             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 431
 432             fscal            = _mm256_add_pd(felec,fvdw);
 433
 434             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm256_mul_pd(fscal,dx00);
 438             ty               = _mm256_mul_pd(fscal,dy00);
 439             tz               = _mm256_mul_pd(fscal,dz00);
 440
 441             /* Update vectorial force */
 442             fix0             = _mm256_add_pd(fix0,tx);
 443             fiy0             = _mm256_add_pd(fiy0,ty);
 444             fiz0             = _mm256_add_pd(fiz0,tz);
 445
 446             fjx0             = _mm256_add_pd(fjx0,tx);
 447             fjy0             = _mm256_add_pd(fjy0,ty);
 448             fjz0             = _mm256_add_pd(fjz0,tz);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq10             = _mm256_mul_pd(iq1,jq0);
 456
 457             /* COULOMB ELECTROSTATICS */
 458             velec            = _mm256_mul_pd(qq10,rinv10);
 459             felec            = _mm256_mul_pd(velec,rinvsq10);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velec            = _mm256_andnot_pd(dummy_mask,velec);
 463             velecsum         = _mm256_add_pd(velecsum,velec);
 464
 465             fscal            = felec;
 466
 467             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm256_mul_pd(fscal,dx10);
 471             ty               = _mm256_mul_pd(fscal,dy10);
 472             tz               = _mm256_mul_pd(fscal,dz10);
 473
 474             /* Update vectorial force */
 475             fix1             = _mm256_add_pd(fix1,tx);
 476             fiy1             = _mm256_add_pd(fiy1,ty);
 477             fiz1             = _mm256_add_pd(fiz1,tz);
 478
 479             fjx0             = _mm256_add_pd(fjx0,tx);
 480             fjy0             = _mm256_add_pd(fjy0,ty);
 481             fjz0             = _mm256_add_pd(fjz0,tz);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq20             = _mm256_mul_pd(iq2,jq0);
 489
 490             /* COULOMB ELECTROSTATICS */
 491             velec            = _mm256_mul_pd(qq20,rinv20);
 492             felec            = _mm256_mul_pd(velec,rinvsq20);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm256_andnot_pd(dummy_mask,velec);
 496             velecsum         = _mm256_add_pd(velecsum,velec);
 497
 498             fscal            = felec;
 499
 500             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_pd(fscal,dx20);
 504             ty               = _mm256_mul_pd(fscal,dy20);
 505             tz               = _mm256_mul_pd(fscal,dz20);
 506
 507             /* Update vectorial force */
 508             fix2             = _mm256_add_pd(fix2,tx);
 509             fiy2             = _mm256_add_pd(fiy2,ty);
 510             fiz2             = _mm256_add_pd(fiz2,tz);
 511
 512             fjx0             = _mm256_add_pd(fjx0,tx);
 513             fjy0             = _mm256_add_pd(fjy0,ty);
 514             fjz0             = _mm256_add_pd(fjz0,tz);
 515
 516             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 517             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 518             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 519             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 520
 521             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 522
 523             /* Inner loop uses 96 flops */
 524         }
 525
 526         /* End of innermost loop */
 527
 528         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 529                                                  f+i_coord_offset,fshift+i_shift_offset);
 530
 531         ggid                        = gid[iidx];
 532         /* Update potential energies */
 533         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 534         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 535
 536         /* Increment number of inner iterations */
 537         inneriter                  += j_index_end - j_index_start;
 538
 539         /* Outer loop uses 20 flops */
 540     }
 541
 542     /* Increment number of outer iterations */
 543     outeriter        += nri;
 544
 545     /* Update outer/inner flops */
 546
 547     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 548 }
 549 /*
 550  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
 551  * Electrostatics interaction: Coulomb
 552  * VdW interaction:            LennardJones
 553  * Geometry:                   Water3-Particle
 554  * Calculate force/pot:        Force
 555  */
 556 void
 557 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
 558                     (t_nblist * gmx_restrict                nlist,
 559                      rvec * gmx_restrict                    xx,
 560                      rvec * gmx_restrict                    ff,
 561                      t_forcerec * gmx_restrict              fr,
 562                      t_mdatoms * gmx_restrict               mdatoms,
 563                      nb_kernel_data_t * gmx_restrict        kernel_data,
 564                      t_nrnb * gmx_restrict                  nrnb)
 565 {
 566     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 567      * just 0 for non-waters.
 568      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 569      * jnr indices corresponding to data put in the four positions in the SIMD register.
 570      */
 571     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 572     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 573     int              jnrA,jnrB,jnrC,jnrD;
 574     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 575     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 576     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 577     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 578     real             rcutoff_scalar;
 579     real             *shiftvec,*fshift,*x,*f;
 580     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 581     real             scratch[4*DIM];
 582     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 583     real *           vdwioffsetptr0;
 584     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 585     real *           vdwioffsetptr1;
 586     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 587     real *           vdwioffsetptr2;
 588     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 589     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 590     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 591     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 592     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 593     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 594     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 595     real             *charge;
 596     int              nvdwtype;
 597     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 598     int              *vdwtype;
 599     real             *vdwparam;
 600     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 601     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 602     __m256d          dummy_mask,cutoff_mask;
 603     __m128           tmpmask0,tmpmask1;
 604     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 605     __m256d          one     = _mm256_set1_pd(1.0);
 606     __m256d          two     = _mm256_set1_pd(2.0);
 607     x                = xx[0];
 608     f                = ff[0];
 609
 610     nri              = nlist->nri;
 611     iinr             = nlist->iinr;
 612     jindex           = nlist->jindex;
 613     jjnr             = nlist->jjnr;
 614     shiftidx         = nlist->shift;
 615     gid              = nlist->gid;
 616     shiftvec         = fr->shift_vec[0];
 617     fshift           = fr->fshift[0];
 618     facel            = _mm256_set1_pd(fr->epsfac);
 619     charge           = mdatoms->chargeA;
 620     nvdwtype         = fr->ntype;
 621     vdwparam         = fr->nbfp;
 622     vdwtype          = mdatoms->typeA;
 623
 624     /* Setup water-specific parameters */
 625     inr              = nlist->iinr[0];
 626     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 627     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 628     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 629     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 630
 631     /* Avoid stupid compiler warnings */
 632     jnrA = jnrB = jnrC = jnrD = 0;
 633     j_coord_offsetA = 0;
 634     j_coord_offsetB = 0;
 635     j_coord_offsetC = 0;
 636     j_coord_offsetD = 0;
 637
 638     outeriter        = 0;
 639     inneriter        = 0;
 640
 641     for(iidx=0;iidx<4*DIM;iidx++)
 642     {
 643         scratch[iidx] = 0.0;
 644     }
 645
 646     /* Start outer loop over neighborlists */
 647     for(iidx=0; iidx<nri; iidx++)
 648     {
 649         /* Load shift vector for this list */
 650         i_shift_offset   = DIM*shiftidx[iidx];
 651
 652         /* Load limits for loop over neighbors */
 653         j_index_start    = jindex[iidx];
 654         j_index_end      = jindex[iidx+1];
 655
 656         /* Get outer coordinate index */
 657         inr              = iinr[iidx];
 658         i_coord_offset   = DIM*inr;
 659
 660         /* Load i particle coords and add shift vector */
 661         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 662                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 663
 664         fix0             = _mm256_setzero_pd();
 665         fiy0             = _mm256_setzero_pd();
 666         fiz0             = _mm256_setzero_pd();
 667         fix1             = _mm256_setzero_pd();
 668         fiy1             = _mm256_setzero_pd();
 669         fiz1             = _mm256_setzero_pd();
 670         fix2             = _mm256_setzero_pd();
 671         fiy2             = _mm256_setzero_pd();
 672         fiz2             = _mm256_setzero_pd();
 673
 674         /* Start inner kernel loop */
 675         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 676         {
 677
 678             /* Get j neighbor index, and coordinate index */
 679             jnrA             = jjnr[jidx];
 680             jnrB             = jjnr[jidx+1];
 681             jnrC             = jjnr[jidx+2];
 682             jnrD             = jjnr[jidx+3];
 683             j_coord_offsetA  = DIM*jnrA;
 684             j_coord_offsetB  = DIM*jnrB;
 685             j_coord_offsetC  = DIM*jnrC;
 686             j_coord_offsetD  = DIM*jnrD;
 687
 688             /* load j atom coordinates */
 689             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 690                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 691                                                  &jx0,&jy0,&jz0);
 692
 693             /* Calculate displacement vector */
 694             dx00             = _mm256_sub_pd(ix0,jx0);
 695             dy00             = _mm256_sub_pd(iy0,jy0);
 696             dz00             = _mm256_sub_pd(iz0,jz0);
 697             dx10             = _mm256_sub_pd(ix1,jx0);
 698             dy10             = _mm256_sub_pd(iy1,jy0);
 699             dz10             = _mm256_sub_pd(iz1,jz0);
 700             dx20             = _mm256_sub_pd(ix2,jx0);
 701             dy20             = _mm256_sub_pd(iy2,jy0);
 702             dz20             = _mm256_sub_pd(iz2,jz0);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 706             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 707             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 708
 709             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 710             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 711             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 712
 713             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 714             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 715             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 716
 717             /* Load parameters for j particles */
 718             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 719                                                                  charge+jnrC+0,charge+jnrD+0);
 720             vdwjidx0A        = 2*vdwtype[jnrA+0];
 721             vdwjidx0B        = 2*vdwtype[jnrB+0];
 722             vdwjidx0C        = 2*vdwtype[jnrC+0];
 723             vdwjidx0D        = 2*vdwtype[jnrD+0];
 724
 725             fjx0             = _mm256_setzero_pd();
 726             fjy0             = _mm256_setzero_pd();
 727             fjz0             = _mm256_setzero_pd();
 728
 729             /**************************
 730              * CALCULATE INTERACTIONS *
 731              **************************/
 732
 733             /* Compute parameters for interactions between i and j atoms */
 734             qq00             = _mm256_mul_pd(iq0,jq0);
 735             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 736                                             vdwioffsetptr0+vdwjidx0B,
 737                                             vdwioffsetptr0+vdwjidx0C,
 738                                             vdwioffsetptr0+vdwjidx0D,
 739                                             &c6_00,&c12_00);
 740
 741             /* COULOMB ELECTROSTATICS */
 742             velec            = _mm256_mul_pd(qq00,rinv00);
 743             felec            = _mm256_mul_pd(velec,rinvsq00);
 744
 745             /* LENNARD-JONES DISPERSION/REPULSION */
 746
 747             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 748             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 749
 750             fscal            = _mm256_add_pd(felec,fvdw);
 751
 752             /* Calculate temporary vectorial force */
 753             tx               = _mm256_mul_pd(fscal,dx00);
 754             ty               = _mm256_mul_pd(fscal,dy00);
 755             tz               = _mm256_mul_pd(fscal,dz00);
 756
 757             /* Update vectorial force */
 758             fix0             = _mm256_add_pd(fix0,tx);
 759             fiy0             = _mm256_add_pd(fiy0,ty);
 760             fiz0             = _mm256_add_pd(fiz0,tz);
 761
 762             fjx0             = _mm256_add_pd(fjx0,tx);
 763             fjy0             = _mm256_add_pd(fjy0,ty);
 764             fjz0             = _mm256_add_pd(fjz0,tz);
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq10             = _mm256_mul_pd(iq1,jq0);
 772
 773             /* COULOMB ELECTROSTATICS */
 774             velec            = _mm256_mul_pd(qq10,rinv10);
 775             felec            = _mm256_mul_pd(velec,rinvsq10);
 776
 777             fscal            = felec;
 778
 779             /* Calculate temporary vectorial force */
 780             tx               = _mm256_mul_pd(fscal,dx10);
 781             ty               = _mm256_mul_pd(fscal,dy10);
 782             tz               = _mm256_mul_pd(fscal,dz10);
 783
 784             /* Update vectorial force */
 785             fix1             = _mm256_add_pd(fix1,tx);
 786             fiy1             = _mm256_add_pd(fiy1,ty);
 787             fiz1             = _mm256_add_pd(fiz1,tz);
 788
 789             fjx0             = _mm256_add_pd(fjx0,tx);
 790             fjy0             = _mm256_add_pd(fjy0,ty);
 791             fjz0             = _mm256_add_pd(fjz0,tz);
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             /* Compute parameters for interactions between i and j atoms */
 798             qq20             = _mm256_mul_pd(iq2,jq0);
 799
 800             /* COULOMB ELECTROSTATICS */
 801             velec            = _mm256_mul_pd(qq20,rinv20);
 802             felec            = _mm256_mul_pd(velec,rinvsq20);
 803
 804             fscal            = felec;
 805
 806             /* Calculate temporary vectorial force */
 807             tx               = _mm256_mul_pd(fscal,dx20);
 808             ty               = _mm256_mul_pd(fscal,dy20);
 809             tz               = _mm256_mul_pd(fscal,dz20);
 810
 811             /* Update vectorial force */
 812             fix2             = _mm256_add_pd(fix2,tx);
 813             fiy2             = _mm256_add_pd(fiy2,ty);
 814             fiz2             = _mm256_add_pd(fiz2,tz);
 815
 816             fjx0             = _mm256_add_pd(fjx0,tx);
 817             fjy0             = _mm256_add_pd(fjy0,ty);
 818             fjz0             = _mm256_add_pd(fjz0,tz);
 819
 820             fjptrA             = f+j_coord_offsetA;
 821             fjptrB             = f+j_coord_offsetB;
 822             fjptrC             = f+j_coord_offsetC;
 823             fjptrD             = f+j_coord_offsetD;
 824
 825             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 826
 827             /* Inner loop uses 88 flops */
 828         }
 829
 830         if(jidx<j_index_end)
 831         {
 832
 833             /* Get j neighbor index, and coordinate index */
 834             jnrlistA         = jjnr[jidx];
 835             jnrlistB         = jjnr[jidx+1];
 836             jnrlistC         = jjnr[jidx+2];
 837             jnrlistD         = jjnr[jidx+3];
 838             /* Sign of each element will be negative for non-real atoms.
 839              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 840              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 841              */
 842             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 843
 844             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 845             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 846             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 847
 848             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 849             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 850             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 851             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 852             j_coord_offsetA  = DIM*jnrA;
 853             j_coord_offsetB  = DIM*jnrB;
 854             j_coord_offsetC  = DIM*jnrC;
 855             j_coord_offsetD  = DIM*jnrD;
 856
 857             /* load j atom coordinates */
 858             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 859                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 860                                                  &jx0,&jy0,&jz0);
 861
 862             /* Calculate displacement vector */
 863             dx00             = _mm256_sub_pd(ix0,jx0);
 864             dy00             = _mm256_sub_pd(iy0,jy0);
 865             dz00             = _mm256_sub_pd(iz0,jz0);
 866             dx10             = _mm256_sub_pd(ix1,jx0);
 867             dy10             = _mm256_sub_pd(iy1,jy0);
 868             dz10             = _mm256_sub_pd(iz1,jz0);
 869             dx20             = _mm256_sub_pd(ix2,jx0);
 870             dy20             = _mm256_sub_pd(iy2,jy0);
 871             dz20             = _mm256_sub_pd(iz2,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 875             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 876             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 877
 878             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 879             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 880             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 881
 882             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 883             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 884             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 885
 886             /* Load parameters for j particles */
 887             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 888                                                                  charge+jnrC+0,charge+jnrD+0);
 889             vdwjidx0A        = 2*vdwtype[jnrA+0];
 890             vdwjidx0B        = 2*vdwtype[jnrB+0];
 891             vdwjidx0C        = 2*vdwtype[jnrC+0];
 892             vdwjidx0D        = 2*vdwtype[jnrD+0];
 893
 894             fjx0             = _mm256_setzero_pd();
 895             fjy0             = _mm256_setzero_pd();
 896             fjz0             = _mm256_setzero_pd();
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq00             = _mm256_mul_pd(iq0,jq0);
 904             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 905                                             vdwioffsetptr0+vdwjidx0B,
 906                                             vdwioffsetptr0+vdwjidx0C,
 907                                             vdwioffsetptr0+vdwjidx0D,
 908                                             &c6_00,&c12_00);
 909
 910             /* COULOMB ELECTROSTATICS */
 911             velec            = _mm256_mul_pd(qq00,rinv00);
 912             felec            = _mm256_mul_pd(velec,rinvsq00);
 913
 914             /* LENNARD-JONES DISPERSION/REPULSION */
 915
 916             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 917             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 918
 919             fscal            = _mm256_add_pd(felec,fvdw);
 920
 921             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm256_mul_pd(fscal,dx00);
 925             ty               = _mm256_mul_pd(fscal,dy00);
 926             tz               = _mm256_mul_pd(fscal,dz00);
 927
 928             /* Update vectorial force */
 929             fix0             = _mm256_add_pd(fix0,tx);
 930             fiy0             = _mm256_add_pd(fiy0,ty);
 931             fiz0             = _mm256_add_pd(fiz0,tz);
 932
 933             fjx0             = _mm256_add_pd(fjx0,tx);
 934             fjy0             = _mm256_add_pd(fjy0,ty);
 935             fjz0             = _mm256_add_pd(fjz0,tz);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq10             = _mm256_mul_pd(iq1,jq0);
 943
 944             /* COULOMB ELECTROSTATICS */
 945             velec            = _mm256_mul_pd(qq10,rinv10);
 946             felec            = _mm256_mul_pd(velec,rinvsq10);
 947
 948             fscal            = felec;
 949
 950             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm256_mul_pd(fscal,dx10);
 954             ty               = _mm256_mul_pd(fscal,dy10);
 955             tz               = _mm256_mul_pd(fscal,dz10);
 956
 957             /* Update vectorial force */
 958             fix1             = _mm256_add_pd(fix1,tx);
 959             fiy1             = _mm256_add_pd(fiy1,ty);
 960             fiz1             = _mm256_add_pd(fiz1,tz);
 961
 962             fjx0             = _mm256_add_pd(fjx0,tx);
 963             fjy0             = _mm256_add_pd(fjy0,ty);
 964             fjz0             = _mm256_add_pd(fjz0,tz);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq20             = _mm256_mul_pd(iq2,jq0);
 972
 973             /* COULOMB ELECTROSTATICS */
 974             velec            = _mm256_mul_pd(qq20,rinv20);
 975             felec            = _mm256_mul_pd(velec,rinvsq20);
 976
 977             fscal            = felec;
 978
 979             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 980
 981             /* Calculate temporary vectorial force */
 982             tx               = _mm256_mul_pd(fscal,dx20);
 983             ty               = _mm256_mul_pd(fscal,dy20);
 984             tz               = _mm256_mul_pd(fscal,dz20);
 985
 986             /* Update vectorial force */
 987             fix2             = _mm256_add_pd(fix2,tx);
 988             fiy2             = _mm256_add_pd(fiy2,ty);
 989             fiz2             = _mm256_add_pd(fiz2,tz);
 990
 991             fjx0             = _mm256_add_pd(fjx0,tx);
 992             fjy0             = _mm256_add_pd(fjy0,ty);
 993             fjz0             = _mm256_add_pd(fjz0,tz);
 994
 995             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 996             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 997             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 998             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 999
1000             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1001
1002             /* Inner loop uses 88 flops */
1003         }
1004
1005         /* End of innermost loop */
1006
1007         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1008                                                  f+i_coord_offset,fshift+i_shift_offset);
1009
1010         /* Increment number of inner iterations */
1011         inneriter                  += j_index_end - j_index_start;
1012
1013         /* Outer loop uses 18 flops */
1014     }
1015
1016     /* Increment number of outer iterations */
1017     outeriter        += nri;
1018
1019     /* Update outer/inner flops */
1020
1021     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1022 }