src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_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_ElecRF_VdwLJ_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     __m256d          dummy_mask,cutoff_mask;
  93     __m128           tmpmask0,tmpmask1;
  94     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  95     __m256d          one     = _mm256_set1_pd(1.0);
  96     __m256d          two     = _mm256_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm256_set1_pd(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm256_set1_pd(fr->ic->k_rf);
 111     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 112     crf              = _mm256_set1_pd(fr->ic->c_rf);
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 120     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 121     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 122     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 156
 157         fix0             = _mm256_setzero_pd();
 158         fiy0             = _mm256_setzero_pd();
 159         fiz0             = _mm256_setzero_pd();
 160         fix1             = _mm256_setzero_pd();
 161         fiy1             = _mm256_setzero_pd();
 162         fiz1             = _mm256_setzero_pd();
 163         fix2             = _mm256_setzero_pd();
 164         fiy2             = _mm256_setzero_pd();
 165         fiz2             = _mm256_setzero_pd();
 166         fix3             = _mm256_setzero_pd();
 167         fiy3             = _mm256_setzero_pd();
 168         fiz3             = _mm256_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm256_setzero_pd();
 172         vvdwsum          = _mm256_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 191                                                  &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm256_sub_pd(ix0,jx0);
 195             dy00             = _mm256_sub_pd(iy0,jy0);
 196             dz00             = _mm256_sub_pd(iz0,jz0);
 197             dx10             = _mm256_sub_pd(ix1,jx0);
 198             dy10             = _mm256_sub_pd(iy1,jy0);
 199             dz10             = _mm256_sub_pd(iz1,jz0);
 200             dx20             = _mm256_sub_pd(ix2,jx0);
 201             dy20             = _mm256_sub_pd(iy2,jy0);
 202             dz20             = _mm256_sub_pd(iz2,jz0);
 203             dx30             = _mm256_sub_pd(ix3,jx0);
 204             dy30             = _mm256_sub_pd(iy3,jy0);
 205             dz30             = _mm256_sub_pd(iz3,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 211             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 212
 213             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 214             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 215             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 216
 217             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 218             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 219             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 220             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 224                                                                  charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             fjx0             = _mm256_setzero_pd();
 231             fjy0             = _mm256_setzero_pd();
 232             fjz0             = _mm256_setzero_pd();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 240                                             vdwioffsetptr0+vdwjidx0B,
 241                                             vdwioffsetptr0+vdwjidx0C,
 242                                             vdwioffsetptr0+vdwjidx0D,
 243                                             &c6_00,&c12_00);
 244
 245             /* LENNARD-JONES DISPERSION/REPULSION */
 246
 247             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 248             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 249             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 250             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 251             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 252
 253             /* Update potential sum for this i atom from the interaction with this j atom. */
 254             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 255
 256             fscal            = fvdw;
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm256_mul_pd(fscal,dx00);
 260             ty               = _mm256_mul_pd(fscal,dy00);
 261             tz               = _mm256_mul_pd(fscal,dz00);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm256_add_pd(fix0,tx);
 265             fiy0             = _mm256_add_pd(fiy0,ty);
 266             fiz0             = _mm256_add_pd(fiz0,tz);
 267
 268             fjx0             = _mm256_add_pd(fjx0,tx);
 269             fjy0             = _mm256_add_pd(fjy0,ty);
 270             fjz0             = _mm256_add_pd(fjz0,tz);
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq10             = _mm256_mul_pd(iq1,jq0);
 278
 279             /* REACTION-FIELD ELECTROSTATICS */
 280             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 281             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm256_add_pd(velecsum,velec);
 285
 286             fscal            = felec;
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm256_mul_pd(fscal,dx10);
 290             ty               = _mm256_mul_pd(fscal,dy10);
 291             tz               = _mm256_mul_pd(fscal,dz10);
 292
 293             /* Update vectorial force */
 294             fix1             = _mm256_add_pd(fix1,tx);
 295             fiy1             = _mm256_add_pd(fiy1,ty);
 296             fiz1             = _mm256_add_pd(fiz1,tz);
 297
 298             fjx0             = _mm256_add_pd(fjx0,tx);
 299             fjy0             = _mm256_add_pd(fjy0,ty);
 300             fjz0             = _mm256_add_pd(fjz0,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq20             = _mm256_mul_pd(iq2,jq0);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 311             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm256_add_pd(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_pd(fscal,dx20);
 320             ty               = _mm256_mul_pd(fscal,dy20);
 321             tz               = _mm256_mul_pd(fscal,dz20);
 322
 323             /* Update vectorial force */
 324             fix2             = _mm256_add_pd(fix2,tx);
 325             fiy2             = _mm256_add_pd(fiy2,ty);
 326             fiz2             = _mm256_add_pd(fiz2,tz);
 327
 328             fjx0             = _mm256_add_pd(fjx0,tx);
 329             fjy0             = _mm256_add_pd(fjy0,ty);
 330             fjz0             = _mm256_add_pd(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq30             = _mm256_mul_pd(iq3,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 341             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm256_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_pd(fscal,dx30);
 350             ty               = _mm256_mul_pd(fscal,dy30);
 351             tz               = _mm256_mul_pd(fscal,dz30);
 352
 353             /* Update vectorial force */
 354             fix3             = _mm256_add_pd(fix3,tx);
 355             fiy3             = _mm256_add_pd(fiy3,ty);
 356             fiz3             = _mm256_add_pd(fiz3,tz);
 357
 358             fjx0             = _mm256_add_pd(fjx0,tx);
 359             fjy0             = _mm256_add_pd(fjy0,ty);
 360             fjz0             = _mm256_add_pd(fjz0,tz);
 361
 362             fjptrA             = f+j_coord_offsetA;
 363             fjptrB             = f+j_coord_offsetB;
 364             fjptrC             = f+j_coord_offsetC;
 365             fjptrD             = f+j_coord_offsetD;
 366
 367             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 368
 369             /* Inner loop uses 131 flops */
 370         }
 371
 372         if(jidx<j_index_end)
 373         {
 374
 375             /* Get j neighbor index, and coordinate index */
 376             jnrlistA         = jjnr[jidx];
 377             jnrlistB         = jjnr[jidx+1];
 378             jnrlistC         = jjnr[jidx+2];
 379             jnrlistD         = jjnr[jidx+3];
 380             /* Sign of each element will be negative for non-real atoms.
 381              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 382              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 383              */
 384             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 385
 386             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 387             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 388             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 389
 390             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 391             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 392             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 393             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 394             j_coord_offsetA  = DIM*jnrA;
 395             j_coord_offsetB  = DIM*jnrB;
 396             j_coord_offsetC  = DIM*jnrC;
 397             j_coord_offsetD  = DIM*jnrD;
 398
 399             /* load j atom coordinates */
 400             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 401                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 402                                                  &jx0,&jy0,&jz0);
 403
 404             /* Calculate displacement vector */
 405             dx00             = _mm256_sub_pd(ix0,jx0);
 406             dy00             = _mm256_sub_pd(iy0,jy0);
 407             dz00             = _mm256_sub_pd(iz0,jz0);
 408             dx10             = _mm256_sub_pd(ix1,jx0);
 409             dy10             = _mm256_sub_pd(iy1,jy0);
 410             dz10             = _mm256_sub_pd(iz1,jz0);
 411             dx20             = _mm256_sub_pd(ix2,jx0);
 412             dy20             = _mm256_sub_pd(iy2,jy0);
 413             dz20             = _mm256_sub_pd(iz2,jz0);
 414             dx30             = _mm256_sub_pd(ix3,jx0);
 415             dy30             = _mm256_sub_pd(iy3,jy0);
 416             dz30             = _mm256_sub_pd(iz3,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 420             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 421             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 422             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 423
 424             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 425             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 426             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 427
 428             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 429             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 430             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 431             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 432
 433             /* Load parameters for j particles */
 434             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 435                                                                  charge+jnrC+0,charge+jnrD+0);
 436             vdwjidx0A        = 2*vdwtype[jnrA+0];
 437             vdwjidx0B        = 2*vdwtype[jnrB+0];
 438             vdwjidx0C        = 2*vdwtype[jnrC+0];
 439             vdwjidx0D        = 2*vdwtype[jnrD+0];
 440
 441             fjx0             = _mm256_setzero_pd();
 442             fjy0             = _mm256_setzero_pd();
 443             fjz0             = _mm256_setzero_pd();
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 451                                             vdwioffsetptr0+vdwjidx0B,
 452                                             vdwioffsetptr0+vdwjidx0C,
 453                                             vdwioffsetptr0+vdwjidx0D,
 454                                             &c6_00,&c12_00);
 455
 456             /* LENNARD-JONES DISPERSION/REPULSION */
 457
 458             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 459             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 460             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 461             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 462             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 466             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 467
 468             fscal            = fvdw;
 469
 470             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 471
 472             /* Calculate temporary vectorial force */
 473             tx               = _mm256_mul_pd(fscal,dx00);
 474             ty               = _mm256_mul_pd(fscal,dy00);
 475             tz               = _mm256_mul_pd(fscal,dz00);
 476
 477             /* Update vectorial force */
 478             fix0             = _mm256_add_pd(fix0,tx);
 479             fiy0             = _mm256_add_pd(fiy0,ty);
 480             fiz0             = _mm256_add_pd(fiz0,tz);
 481
 482             fjx0             = _mm256_add_pd(fjx0,tx);
 483             fjy0             = _mm256_add_pd(fjy0,ty);
 484             fjz0             = _mm256_add_pd(fjz0,tz);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq10             = _mm256_mul_pd(iq1,jq0);
 492
 493             /* REACTION-FIELD ELECTROSTATICS */
 494             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 495             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velec            = _mm256_andnot_pd(dummy_mask,velec);
 499             velecsum         = _mm256_add_pd(velecsum,velec);
 500
 501             fscal            = felec;
 502
 503             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_pd(fscal,dx10);
 507             ty               = _mm256_mul_pd(fscal,dy10);
 508             tz               = _mm256_mul_pd(fscal,dz10);
 509
 510             /* Update vectorial force */
 511             fix1             = _mm256_add_pd(fix1,tx);
 512             fiy1             = _mm256_add_pd(fiy1,ty);
 513             fiz1             = _mm256_add_pd(fiz1,tz);
 514
 515             fjx0             = _mm256_add_pd(fjx0,tx);
 516             fjy0             = _mm256_add_pd(fjy0,ty);
 517             fjz0             = _mm256_add_pd(fjz0,tz);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq20             = _mm256_mul_pd(iq2,jq0);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 528             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm256_andnot_pd(dummy_mask,velec);
 532             velecsum         = _mm256_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_pd(fscal,dx20);
 540             ty               = _mm256_mul_pd(fscal,dy20);
 541             tz               = _mm256_mul_pd(fscal,dz20);
 542
 543             /* Update vectorial force */
 544             fix2             = _mm256_add_pd(fix2,tx);
 545             fiy2             = _mm256_add_pd(fiy2,ty);
 546             fiz2             = _mm256_add_pd(fiz2,tz);
 547
 548             fjx0             = _mm256_add_pd(fjx0,tx);
 549             fjy0             = _mm256_add_pd(fjy0,ty);
 550             fjz0             = _mm256_add_pd(fjz0,tz);
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq30             = _mm256_mul_pd(iq3,jq0);
 558
 559             /* REACTION-FIELD ELECTROSTATICS */
 560             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 561             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm256_andnot_pd(dummy_mask,velec);
 565             velecsum         = _mm256_add_pd(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm256_mul_pd(fscal,dx30);
 573             ty               = _mm256_mul_pd(fscal,dy30);
 574             tz               = _mm256_mul_pd(fscal,dz30);
 575
 576             /* Update vectorial force */
 577             fix3             = _mm256_add_pd(fix3,tx);
 578             fiy3             = _mm256_add_pd(fiy3,ty);
 579             fiz3             = _mm256_add_pd(fiz3,tz);
 580
 581             fjx0             = _mm256_add_pd(fjx0,tx);
 582             fjy0             = _mm256_add_pd(fjy0,ty);
 583             fjz0             = _mm256_add_pd(fjz0,tz);
 584
 585             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 586             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 587             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 588             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 589
 590             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 591
 592             /* Inner loop uses 131 flops */
 593         }
 594
 595         /* End of innermost loop */
 596
 597         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 598                                                  f+i_coord_offset,fshift+i_shift_offset);
 599
 600         ggid                        = gid[iidx];
 601         /* Update potential energies */
 602         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 603         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 604
 605         /* Increment number of inner iterations */
 606         inneriter                  += j_index_end - j_index_start;
 607
 608         /* Outer loop uses 26 flops */
 609     }
 610
 611     /* Increment number of outer iterations */
 612     outeriter        += nri;
 613
 614     /* Update outer/inner flops */
 615
 616     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 617 }
 618 /*
 619  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_double
 620  * Electrostatics interaction: ReactionField
 621  * VdW interaction:            LennardJones
 622  * Geometry:                   Water4-Particle
 623  * Calculate force/pot:        Force
 624  */
 625 void
 626 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_double
 627                     (t_nblist * gmx_restrict                nlist,
 628                      rvec * gmx_restrict                    xx,
 629                      rvec * gmx_restrict                    ff,
 630                      t_forcerec * gmx_restrict              fr,
 631                      t_mdatoms * gmx_restrict               mdatoms,
 632                      nb_kernel_data_t * gmx_restrict        kernel_data,
 633                      t_nrnb * gmx_restrict                  nrnb)
 634 {
 635     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 636      * just 0 for non-waters.
 637      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 638      * jnr indices corresponding to data put in the four positions in the SIMD register.
 639      */
 640     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 641     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 642     int              jnrA,jnrB,jnrC,jnrD;
 643     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 644     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 645     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 646     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 647     real             rcutoff_scalar;
 648     real             *shiftvec,*fshift,*x,*f;
 649     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 650     real             scratch[4*DIM];
 651     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 652     real *           vdwioffsetptr0;
 653     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 654     real *           vdwioffsetptr1;
 655     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 656     real *           vdwioffsetptr2;
 657     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 658     real *           vdwioffsetptr3;
 659     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 660     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 661     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 662     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 663     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 664     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 665     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 666     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 667     real             *charge;
 668     int              nvdwtype;
 669     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 670     int              *vdwtype;
 671     real             *vdwparam;
 672     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 673     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 674     __m256d          dummy_mask,cutoff_mask;
 675     __m128           tmpmask0,tmpmask1;
 676     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 677     __m256d          one     = _mm256_set1_pd(1.0);
 678     __m256d          two     = _mm256_set1_pd(2.0);
 679     x                = xx[0];
 680     f                = ff[0];
 681
 682     nri              = nlist->nri;
 683     iinr             = nlist->iinr;
 684     jindex           = nlist->jindex;
 685     jjnr             = nlist->jjnr;
 686     shiftidx         = nlist->shift;
 687     gid              = nlist->gid;
 688     shiftvec         = fr->shift_vec[0];
 689     fshift           = fr->fshift[0];
 690     facel            = _mm256_set1_pd(fr->epsfac);
 691     charge           = mdatoms->chargeA;
 692     krf              = _mm256_set1_pd(fr->ic->k_rf);
 693     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 694     crf              = _mm256_set1_pd(fr->ic->c_rf);
 695     nvdwtype         = fr->ntype;
 696     vdwparam         = fr->nbfp;
 697     vdwtype          = mdatoms->typeA;
 698
 699     /* Setup water-specific parameters */
 700     inr              = nlist->iinr[0];
 701     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 702     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 703     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 704     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 705
 706     /* Avoid stupid compiler warnings */
 707     jnrA = jnrB = jnrC = jnrD = 0;
 708     j_coord_offsetA = 0;
 709     j_coord_offsetB = 0;
 710     j_coord_offsetC = 0;
 711     j_coord_offsetD = 0;
 712
 713     outeriter        = 0;
 714     inneriter        = 0;
 715
 716     for(iidx=0;iidx<4*DIM;iidx++)
 717     {
 718         scratch[iidx] = 0.0;
 719     }
 720
 721     /* Start outer loop over neighborlists */
 722     for(iidx=0; iidx<nri; iidx++)
 723     {
 724         /* Load shift vector for this list */
 725         i_shift_offset   = DIM*shiftidx[iidx];
 726
 727         /* Load limits for loop over neighbors */
 728         j_index_start    = jindex[iidx];
 729         j_index_end      = jindex[iidx+1];
 730
 731         /* Get outer coordinate index */
 732         inr              = iinr[iidx];
 733         i_coord_offset   = DIM*inr;
 734
 735         /* Load i particle coords and add shift vector */
 736         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 737                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 738
 739         fix0             = _mm256_setzero_pd();
 740         fiy0             = _mm256_setzero_pd();
 741         fiz0             = _mm256_setzero_pd();
 742         fix1             = _mm256_setzero_pd();
 743         fiy1             = _mm256_setzero_pd();
 744         fiz1             = _mm256_setzero_pd();
 745         fix2             = _mm256_setzero_pd();
 746         fiy2             = _mm256_setzero_pd();
 747         fiz2             = _mm256_setzero_pd();
 748         fix3             = _mm256_setzero_pd();
 749         fiy3             = _mm256_setzero_pd();
 750         fiz3             = _mm256_setzero_pd();
 751
 752         /* Start inner kernel loop */
 753         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 754         {
 755
 756             /* Get j neighbor index, and coordinate index */
 757             jnrA             = jjnr[jidx];
 758             jnrB             = jjnr[jidx+1];
 759             jnrC             = jjnr[jidx+2];
 760             jnrD             = jjnr[jidx+3];
 761             j_coord_offsetA  = DIM*jnrA;
 762             j_coord_offsetB  = DIM*jnrB;
 763             j_coord_offsetC  = DIM*jnrC;
 764             j_coord_offsetD  = DIM*jnrD;
 765
 766             /* load j atom coordinates */
 767             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 768                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 769                                                  &jx0,&jy0,&jz0);
 770
 771             /* Calculate displacement vector */
 772             dx00             = _mm256_sub_pd(ix0,jx0);
 773             dy00             = _mm256_sub_pd(iy0,jy0);
 774             dz00             = _mm256_sub_pd(iz0,jz0);
 775             dx10             = _mm256_sub_pd(ix1,jx0);
 776             dy10             = _mm256_sub_pd(iy1,jy0);
 777             dz10             = _mm256_sub_pd(iz1,jz0);
 778             dx20             = _mm256_sub_pd(ix2,jx0);
 779             dy20             = _mm256_sub_pd(iy2,jy0);
 780             dz20             = _mm256_sub_pd(iz2,jz0);
 781             dx30             = _mm256_sub_pd(ix3,jx0);
 782             dy30             = _mm256_sub_pd(iy3,jy0);
 783             dz30             = _mm256_sub_pd(iz3,jz0);
 784
 785             /* Calculate squared distance and things based on it */
 786             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 787             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 788             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 789             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 790
 791             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 792             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 793             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 794
 795             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 796             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 797             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 798             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 799
 800             /* Load parameters for j particles */
 801             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 802                                                                  charge+jnrC+0,charge+jnrD+0);
 803             vdwjidx0A        = 2*vdwtype[jnrA+0];
 804             vdwjidx0B        = 2*vdwtype[jnrB+0];
 805             vdwjidx0C        = 2*vdwtype[jnrC+0];
 806             vdwjidx0D        = 2*vdwtype[jnrD+0];
 807
 808             fjx0             = _mm256_setzero_pd();
 809             fjy0             = _mm256_setzero_pd();
 810             fjz0             = _mm256_setzero_pd();
 811
 812             /**************************
 813              * CALCULATE INTERACTIONS *
 814              **************************/
 815
 816             /* Compute parameters for interactions between i and j atoms */
 817             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 818                                             vdwioffsetptr0+vdwjidx0B,
 819                                             vdwioffsetptr0+vdwjidx0C,
 820                                             vdwioffsetptr0+vdwjidx0D,
 821                                             &c6_00,&c12_00);
 822
 823             /* LENNARD-JONES DISPERSION/REPULSION */
 824
 825             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 826             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 827
 828             fscal            = fvdw;
 829
 830             /* Calculate temporary vectorial force */
 831             tx               = _mm256_mul_pd(fscal,dx00);
 832             ty               = _mm256_mul_pd(fscal,dy00);
 833             tz               = _mm256_mul_pd(fscal,dz00);
 834
 835             /* Update vectorial force */
 836             fix0             = _mm256_add_pd(fix0,tx);
 837             fiy0             = _mm256_add_pd(fiy0,ty);
 838             fiz0             = _mm256_add_pd(fiz0,tz);
 839
 840             fjx0             = _mm256_add_pd(fjx0,tx);
 841             fjy0             = _mm256_add_pd(fjy0,ty);
 842             fjz0             = _mm256_add_pd(fjz0,tz);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq10             = _mm256_mul_pd(iq1,jq0);
 850
 851             /* REACTION-FIELD ELECTROSTATICS */
 852             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 853
 854             fscal            = felec;
 855
 856             /* Calculate temporary vectorial force */
 857             tx               = _mm256_mul_pd(fscal,dx10);
 858             ty               = _mm256_mul_pd(fscal,dy10);
 859             tz               = _mm256_mul_pd(fscal,dz10);
 860
 861             /* Update vectorial force */
 862             fix1             = _mm256_add_pd(fix1,tx);
 863             fiy1             = _mm256_add_pd(fiy1,ty);
 864             fiz1             = _mm256_add_pd(fiz1,tz);
 865
 866             fjx0             = _mm256_add_pd(fjx0,tx);
 867             fjy0             = _mm256_add_pd(fjy0,ty);
 868             fjz0             = _mm256_add_pd(fjz0,tz);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq20             = _mm256_mul_pd(iq2,jq0);
 876
 877             /* REACTION-FIELD ELECTROSTATICS */
 878             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 879
 880             fscal            = felec;
 881
 882             /* Calculate temporary vectorial force */
 883             tx               = _mm256_mul_pd(fscal,dx20);
 884             ty               = _mm256_mul_pd(fscal,dy20);
 885             tz               = _mm256_mul_pd(fscal,dz20);
 886
 887             /* Update vectorial force */
 888             fix2             = _mm256_add_pd(fix2,tx);
 889             fiy2             = _mm256_add_pd(fiy2,ty);
 890             fiz2             = _mm256_add_pd(fiz2,tz);
 891
 892             fjx0             = _mm256_add_pd(fjx0,tx);
 893             fjy0             = _mm256_add_pd(fjy0,ty);
 894             fjz0             = _mm256_add_pd(fjz0,tz);
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq30             = _mm256_mul_pd(iq3,jq0);
 902
 903             /* REACTION-FIELD ELECTROSTATICS */
 904             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 905
 906             fscal            = felec;
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm256_mul_pd(fscal,dx30);
 910             ty               = _mm256_mul_pd(fscal,dy30);
 911             tz               = _mm256_mul_pd(fscal,dz30);
 912
 913             /* Update vectorial force */
 914             fix3             = _mm256_add_pd(fix3,tx);
 915             fiy3             = _mm256_add_pd(fiy3,ty);
 916             fiz3             = _mm256_add_pd(fiz3,tz);
 917
 918             fjx0             = _mm256_add_pd(fjx0,tx);
 919             fjy0             = _mm256_add_pd(fjy0,ty);
 920             fjz0             = _mm256_add_pd(fjz0,tz);
 921
 922             fjptrA             = f+j_coord_offsetA;
 923             fjptrB             = f+j_coord_offsetB;
 924             fjptrC             = f+j_coord_offsetC;
 925             fjptrD             = f+j_coord_offsetD;
 926
 927             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 928
 929             /* Inner loop uses 111 flops */
 930         }
 931
 932         if(jidx<j_index_end)
 933         {
 934
 935             /* Get j neighbor index, and coordinate index */
 936             jnrlistA         = jjnr[jidx];
 937             jnrlistB         = jjnr[jidx+1];
 938             jnrlistC         = jjnr[jidx+2];
 939             jnrlistD         = jjnr[jidx+3];
 940             /* Sign of each element will be negative for non-real atoms.
 941              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 942              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 943              */
 944             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 945
 946             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 947             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 948             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 949
 950             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 951             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 952             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 953             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 954             j_coord_offsetA  = DIM*jnrA;
 955             j_coord_offsetB  = DIM*jnrB;
 956             j_coord_offsetC  = DIM*jnrC;
 957             j_coord_offsetD  = DIM*jnrD;
 958
 959             /* load j atom coordinates */
 960             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 961                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 962                                                  &jx0,&jy0,&jz0);
 963
 964             /* Calculate displacement vector */
 965             dx00             = _mm256_sub_pd(ix0,jx0);
 966             dy00             = _mm256_sub_pd(iy0,jy0);
 967             dz00             = _mm256_sub_pd(iz0,jz0);
 968             dx10             = _mm256_sub_pd(ix1,jx0);
 969             dy10             = _mm256_sub_pd(iy1,jy0);
 970             dz10             = _mm256_sub_pd(iz1,jz0);
 971             dx20             = _mm256_sub_pd(ix2,jx0);
 972             dy20             = _mm256_sub_pd(iy2,jy0);
 973             dz20             = _mm256_sub_pd(iz2,jz0);
 974             dx30             = _mm256_sub_pd(ix3,jx0);
 975             dy30             = _mm256_sub_pd(iy3,jy0);
 976             dz30             = _mm256_sub_pd(iz3,jz0);
 977
 978             /* Calculate squared distance and things based on it */
 979             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 980             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 981             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 982             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 983
 984             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 985             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 986             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 987
 988             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 989             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 990             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 991             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 992
 993             /* Load parameters for j particles */
 994             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 995                                                                  charge+jnrC+0,charge+jnrD+0);
 996             vdwjidx0A        = 2*vdwtype[jnrA+0];
 997             vdwjidx0B        = 2*vdwtype[jnrB+0];
 998             vdwjidx0C        = 2*vdwtype[jnrC+0];
 999             vdwjidx0D        = 2*vdwtype[jnrD+0];
1000
1001             fjx0             = _mm256_setzero_pd();
1002             fjy0             = _mm256_setzero_pd();
1003             fjz0             = _mm256_setzero_pd();
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1011                                             vdwioffsetptr0+vdwjidx0B,
1012                                             vdwioffsetptr0+vdwjidx0C,
1013                                             vdwioffsetptr0+vdwjidx0D,
1014                                             &c6_00,&c12_00);
1015
1016             /* LENNARD-JONES DISPERSION/REPULSION */
1017
1018             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1019             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1020
1021             fscal            = fvdw;
1022
1023             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm256_mul_pd(fscal,dx00);
1027             ty               = _mm256_mul_pd(fscal,dy00);
1028             tz               = _mm256_mul_pd(fscal,dz00);
1029
1030             /* Update vectorial force */
1031             fix0             = _mm256_add_pd(fix0,tx);
1032             fiy0             = _mm256_add_pd(fiy0,ty);
1033             fiz0             = _mm256_add_pd(fiz0,tz);
1034
1035             fjx0             = _mm256_add_pd(fjx0,tx);
1036             fjy0             = _mm256_add_pd(fjy0,ty);
1037             fjz0             = _mm256_add_pd(fjz0,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq10             = _mm256_mul_pd(iq1,jq0);
1045
1046             /* REACTION-FIELD ELECTROSTATICS */
1047             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1048
1049             fscal            = felec;
1050
1051             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_pd(fscal,dx10);
1055             ty               = _mm256_mul_pd(fscal,dy10);
1056             tz               = _mm256_mul_pd(fscal,dz10);
1057
1058             /* Update vectorial force */
1059             fix1             = _mm256_add_pd(fix1,tx);
1060             fiy1             = _mm256_add_pd(fiy1,ty);
1061             fiz1             = _mm256_add_pd(fiz1,tz);
1062
1063             fjx0             = _mm256_add_pd(fjx0,tx);
1064             fjy0             = _mm256_add_pd(fjy0,ty);
1065             fjz0             = _mm256_add_pd(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq20             = _mm256_mul_pd(iq2,jq0);
1073
1074             /* REACTION-FIELD ELECTROSTATICS */
1075             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm256_mul_pd(fscal,dx20);
1083             ty               = _mm256_mul_pd(fscal,dy20);
1084             tz               = _mm256_mul_pd(fscal,dz20);
1085
1086             /* Update vectorial force */
1087             fix2             = _mm256_add_pd(fix2,tx);
1088             fiy2             = _mm256_add_pd(fiy2,ty);
1089             fiz2             = _mm256_add_pd(fiz2,tz);
1090
1091             fjx0             = _mm256_add_pd(fjx0,tx);
1092             fjy0             = _mm256_add_pd(fjy0,ty);
1093             fjz0             = _mm256_add_pd(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq30             = _mm256_mul_pd(iq3,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm256_mul_pd(fscal,dx30);
1111             ty               = _mm256_mul_pd(fscal,dy30);
1112             tz               = _mm256_mul_pd(fscal,dz30);
1113
1114             /* Update vectorial force */
1115             fix3             = _mm256_add_pd(fix3,tx);
1116             fiy3             = _mm256_add_pd(fiy3,ty);
1117             fiz3             = _mm256_add_pd(fiz3,tz);
1118
1119             fjx0             = _mm256_add_pd(fjx0,tx);
1120             fjy0             = _mm256_add_pd(fjy0,ty);
1121             fjz0             = _mm256_add_pd(fjz0,tz);
1122
1123             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1124             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1125             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1126             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1127
1128             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 111 flops */
1131         }
1132
1133         /* End of innermost loop */
1134
1135         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1136                                                  f+i_coord_offset,fshift+i_shift_offset);
1137
1138         /* Increment number of inner iterations */
1139         inneriter                  += j_index_end - j_index_start;
1140
1141         /* Outer loop uses 24 flops */
1142     }
1143
1144     /* Increment number of outer iterations */
1145     outeriter        += nri;
1146
1147     /* Update outer/inner flops */
1148
1149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*111);
1150 }